ArticlePDF Available

Abstract and Figures

On-farm scoring of behavioural indicators of animal welfare is challenging but the increasing availability of low cost technology now makes automated monitoring of animal behaviour feasible. We discuss some of the issues with using automated methods to measure animal behaviour within the context of assessing animal welfare. Automated feeders (eg for dairy calves) can help measure the degree that animals are hungry and have potential to identify sick animals even in group housing. Such equipment is best used for longitudinal studies of individual animals rather than making comparisons between farms. Devices attached to animals (eg accelerometers or GPS devices) can help measure the activity levels of animals with a high degree of accuracy and can easily be transported between farms, making them best suited for welfare assessment at the group level. Automated image analysis has great potential to assess movement within groups of animals, but following individual animals can be difficult. The techniques have been validated against traditional methods (eg direct observation). The accuracy of measures taken automatically varies between methods but can be increased by combining measures. Technological developments have provided us with a variety of tools that can be used to monitor behaviour automatically, and these have great potential to improve our ability to monitor animal welfare indicators on-farm. However, it is important that methods be developed to measure a wider range of behaviour patterns. Animal welfare assessment schemes should not place undue emphasis on behavioural indicators solely on the basis that they can be monitored automatically.
Content may be subject to copyright.
339
© 2012 Universities Federation for Animal Welfare
The Old School, Brewhouse Hill, Wheathampstead,
Hertfordshire AL4 8AN, UK
www.ufaw.org.uk
Animal Welfare 2012, 21: 339-350
ISSN 0962-7286
doi: 10.7120/09627286.21.3.339
Automated monitoring of behavioural-based animal welfare indicators
J Rushen*, N Chapinaland AM de Passillé
Pacific Agri-Food Research Centre, Agriculture and Agri-Food Canada, PO 1000, 6947 Highway 7, Agassiz, BC, Canada V0M 1A0
Animal Welfare Program, University of British Columbia, Vancouver, BC, Canada V6T 1Z4
* Contact for correspondence and requests for reprints: jeff.rushen@agr.gc.ca
Abstract
On-farm scoring of behavioural indicators of animal welfare is challenging but the increasing availability of low cost technology now
makes automated monitoring of animal behaviour feasible. We discuss some of the issues with using automated methods to measure
animal behaviour within the context of assessing animal welfare. Automated feeders (eg for dairy calves) can help measure the degree
that animals are hungry and have potential to identify sick animals even in group housing. Such equipment is best used for longitu-
dinal studies of individual animals rather than making comparisons between farms. Devices attached to animals (eg accelerometers
or GPS devices) can help measure the activity levels of animals with a high degree of accuracy and can easily be transported between
farms, making them best suited for welfare assessment at the group level. Automated image analysis has great potential to assess
movement within groups of animals, but following individual animals can be difficult. The techniques have been validated against tradi-
tional methods (eg direct observation). The accuracy of measures taken automatically varies between methods but can be increased
by combining measures. Technological developments have provided us with a variety of tools that can be used to monitor behaviour
automatically, and these have great potential to improve our ability to monitor animal welfare indicators on-farm. However, it is
important that methods be developed to measure a wider range of behaviour patterns. Animal welfare assessment schemes should
not place undue emphasis on behavioural indicators solely on the basis that they can be monitored automatically.
Keywords:animal behaviour, animal welfare, automated monitoring, behavioural indicators, farm animals, on-farm assessment
Introduction
An article search, using the words ‘automation’ or
‘automatic’ and ‘animal welfare’ reveals that the number
of scientific articles using these words increased from 5 in
1997 to 80 in 2010. It is clear that the implications of farm
automation for animal welfare are being recognised,
including the potential of automation to monitor animal
welfare. At its simplest, automated assessment of animal
welfare involves using automation to take measures on
some aspect of an animal, which a human then interprets
in terms relevant to animal welfare; for example, using
machines to measure the activity of an animal, which a
person then uses to decide whether the animal is lame. In
addition, computer algorithms are being developed to
make higher order inferences from data collected auto-
matically, eg judging whether an animal is lame or not. In
this paper, we focus on using automated methods of
measuring animal behaviour. We do not deal intensively
with the technical aspects of the equipment available, nor
do we attempt a comprehensive review of all of the uses
to which automation has been put. Instead, we focus upon
some of the issues in using automation to increase the use
of on-farm behavioural recording in the context of
assessing animal welfare at both the level of the indi-
vidual animals and at the farm and group level. We
include self-assessments of animal welfare carried out by
farmers themselves as well as third-party audits.
Why automate? Most recent analyses of the concept of
animal welfare accept that behavioural issues are a key
aspect. This is apparent in the Five Freedoms
(www.fawc.org.uk/freedoms.htm), which include the
‘freedom’ to perform most normal patterns of behaviour.
Furthermore, behavioural measures, such as the occurrence
of aggression or stereotypic behaviour, are important indica-
tors of welfare problems. Including behavioural-based
welfare criteria is, therefore, essential for an overall welfare
assessment (Blokhuis et al 2010). Despite this, current on-
farm welfare assessment schemes often focus heavily on
health issues and include few behavioural measures. We
suggest that this is due mainly to the difficulty, time involved
and cost in taking behavioural measures during farm visits
(Edwards 2007; Sørensen et al 2007); the occurrence of
behaviour patterns is often erratic over time or else their
recording requires long periods of observation, while on-farm
assessments need to be done in a short period of time
(Edwards 2007; Webster 2009). These problems are likely to
Universities Federation for Animal Welfare Science in the Service of Animal Welfare
340 Rushen et al
increase as farm size increases. The availability of equipment
which can measure the behaviour of animals automatically
may help resolve this problem (Blokhuis et al 2010).
Another reason is that automation may prove superior to
people at measuring some behaviours. On-farm welfare
assessment requires that we minimise differences between
observers (Edwards 2007; Webster 2009) but behavioural
measures can be challenging to take reliably and this
requires considerable training of observers. This is evident,
for example, with gait scoring of animals to detect lameness
(Butterworth et al 2007; Flower & Weary 2009). While
observers can see some gait changes associated with
lameness, other changes, which can be detected with the
appropriate equipment, are much harder to observe (eg van
Nuffel et al 2009). The practical difficulties in detecting
lameness in cattle on-farm are well known, with farmers
repeatedly being shown to substantially underestimate the
number of lame cows on their farms (eg Espejo et al 2006).
Hopefully, increased use of automation to record behaviour
patterns will result in more reliable measures being taken.
Finally, assessments done by people tend to provide only a
snapshot of the state of welfare on the farm at a particular
time-point (Webster 2009). Use of automation may allow
longer term monitoring of the animals’ behaviour.
The main types of automation that we consider are the use
of equipment that is installed on farms, devices that can be
attached temporarily to animals, and the use of computer
vision and computer ‘hearing’.
On-farm automation
Computer-controlled feeders, which recognise individual
animals, usually by means of radio frequency identifica-
tion (RFID), are increasingly being used in the dairy,
beef and swine industries and can automatically record
aspects of feeding behaviour. Data from such feeders can
help identify problems associated with hunger and may
help detect animals that are sick.
The absence of hunger is one of the least controversial of the
‘Five Freedoms’ as an aspect of good animal welfare.
Nevertheless, some commonly used management practices
do result in farm animals being hungry for varying periods of
time (D’Eath et al 2009), and some method of assessing the
degree of hunger felt by animals would be valuable. For
example, a controversial issue in the raising of dairy calves
involves how much milk to feed unweaned calves and the
best age to wean them off milk. It is common for calves to be
fed milk or milk replacer in quantities which are substantially
lower than the amount they drink when allowed free access
(Khan et al 2011). In addition, calves may be weaned off milk
at an early age and have difficulty adapting to solid feed.
Calves fed these low amounts of milk visit automated milk
feeders far more often, and these are usually unrewarded
visits during which milk is not available (eg Jensen & Holm
2003; De Paula Vieira et al 2008; Borderas et al 2009a).
Unrewarded visits also increase when animals are being
weaned and there is a negative correlation between energy
intake and the frequency of visits to the milk feeders (de
Passillé et al 2011) (Figure 1). Thus, automatic monitoring of
the frequency of unrewarded visits to milk feeders that calves
make can detect periods when the calves are hungry due to
inadequate feeding and can identify individual animals that
are having difficulty adapting to post-weaning diets.
Farm animals are increasingly being housed in groups but a
potential disadvantage of group housing is that illness is
harder to detect. Changes in feeding behaviour of animals can
be used to identify animals that are sick (Millman 2007;
Weary et al 2009) and these can be detected automatically.
Early research with beef cattle feeding from specialised
feeders showed that drops in feed intake or time spent feeding
could accurately identify steers suffering from respiratory
disease substantially earlier than the normal inspections
(Quimby 2000). Automatically recorded changes in feeding
behaviour can also help identify dairy cows suffering from
peri-parturient diseases such as metritis, ketosis or lameness
(Huzzey et al 2007; Gonzalez et al 2008; Proudfoot et al
2010) (Figure 2) and dairy calves suffering from a variety of
illnesses (Svensson & Jensen 2007; Borderas et al 2009b).
Electronic sow feeders also have the potential to be used this
way (Cornou et al 2008). In addition, changes in drinking
behaviour and water intake, monitored automatically, may
also be useful to identify sick animals (Madsen & Kristensen
2005; Lukas et al 2008; Kruse et al 2011).
Automated feeders are not the only on-farm equipment
that can be used in this way. Automated milking systems
for dairy cattle automatically collect data on the milking
of dairy cows and have potential for monitoring poor
health. For example, lameness in dairy cows is apparent
in a reduced frequency of visits to the robot (Bach et al
2007; Borderas et al 2008), although this appears to have
low specificity as many other low attending cows are not
lame (Borderas et al 2008).
Use of force plates to measure the weight animals place on
their feet when walking can detect gait abnormalities in
poultry (Corr et al 2007; Sandilands et al 2011). A force-
plate system for measuring the force that cows exert when
walking is commercially available (Rajkondawar et al
2006) and been installed in some dairies but the sensitivity
of the measure for detecting lameness is low (Bicalho et al
2007) perhaps because the time when the force is exerted is
very short. Measuring weight distribution when the animals
are standing still is easier and can detect lame cows (Rushen
et al 2007; Chapinal et al 2010; Pastell et al 2010).
Research with weigh scales installed in automated milking
systems showed that automated measures of weight distri-
bution could identify lame cows significantly faster than
was achieved through routine veterinary inspection (Pastell
& Kujala 2007). Furthermore, these measures of weight
distribution are sensitive to the degree of pain associated
with the lameness (Rushen et al 2007; Chapinal et al 2010).
Advantages and disadvantages
Using data from equipment that is already installed on the
farm is relatively cheap since the only costs for using these
in welfare assessment are those associated with data extrac-
tion and manipulation. A disadvantage is that the data most
likely belong to the farmer and so there may be issues in
© 2012 Universities Federation for Animal Welfare
Automated monitoring of behavioural-based animal welfare indicators 341
using these data for third-party audits. However, the infor-
mation is very useful for farmers who wish to improve the
welfare of animals on their farms. An important disadvan-
tage is that, at present, most automated equipment involves
feeders, so there is less opportunity to record other forms of
behaviour in this way.
With RFID, this type of equipment can recognise indi-
vidual animals, and is most suited for longitudinal moni-
toring of individuals, in real time if necessary, where data
on each animal can be accumulated over relatively long
periods of time (eg Pastell & Kujala 2007). Although
force-plates could, in principle, be transported between
farms, most such equipment cannot be. Since not all
farms would have the same equipment, and those that do
may not manage it in the same way, there are limits on our
ability to make comparisons between farms. Thus, the
equipment is less useful for animal welfare assessment at
the group level than at the individual level.
Animal Welfare 2012, 21: 339-350
doi: 10.7120/09627286.21.3.339
Figure 1
The (a) mean daily frequency of visits to the milk feeder of dairy calves fed either 6 L per day of milk (black triangles) or 12 L per day
of milk (open squares) at each week of age, including during the 10-day weaning period and b) mean daily frequency of visits to the milk
feeder and the man daily intake of digestible energy of dairy calves during the weaning period. Figures redrawn from data presented in
de Passillé et al (2011).
342 Rushen et al
Devices attached to the animals
The second category of automation consists of devices that
can be attached temporarily to the animals specifically for
monitoring their behaviour. These are most commonly
accelerometers, but other devices such as pedometers, simple
tilt switch devices or GPS devices have also been used.
The amount of time that dairy cows spend standing up or
lying down each day is an important measure of their
welfare, since short lying times are a reflection of inade-
quate stalls and can lead to increased risk of lameness
(Rushen et al 2008). The time that cows spend lying down
can be measured by watching video, but obtaining a reliable
estimate in this way requires a considerable amount of
labour, and is quite impractical for on-farm visits. A number
of relatively cheap, small and accurate electronic devices
are now available that can be used to measure time spent
standing and lying (Ledgerwood et al 2010).
Accelerometers or tilt switch activated devices can be
attached to the legs of cattle and can measure the orientation
of the leg, with the assumption that when the leg is hori-
zontal, the animal is most likely lying down. Such devices
have recently been shown to be useful for on-farm measure-
ments. For example, Ito et al (2009, 2010) attached
accelerometers to over 2,000 cows on 43 Canadian dairy
farms and measured the time the cows lay down over
five days. The results showed variation between and within
farms in the average time that the cows lay down each day
(Figure 3). The average lying time was longer for lame cows,
suggesting that unusually long average lying times on a farm
may be indicative of a high prevalence of lameness.
Accelerometers attached to the legs can also measure the
pattern of acceleration associated with stepping (de Passillé
et al 2010; Ringgenberg et al 2010; Tanida et al 2011; Figure
4). At the simplest level, this provides a measure of the
number of steps taken by an animal, and commercial devices
(eg IceTag® from Ice Robotics Inc, Edinburgh, UK) are now
available to do this. Step counting has been used to detect
lameness in dairy cows (Chapinal et al 2010), to assess the
adequacy of different flooring surfaces (Ouweltjes et al
2011) and to assess the effects of changing flooring in barns
(Platz et al 2008). Accelerometers may also allow automated
gait scoring: one of the most obvious signs of lameness is
asymmetric stepping (Flower & Weary 2009) where there is
a difference within a pair of legs in the speed or duration of
a stride. Accelerometers attached to two legs can measure
differences between the legs in the variance of acceleration
and this is correlated with subjective assessment of asym-
metric stepping (Chapinal et al 2011). Accelerometers
attached to any part of a cow’s body can estimate walking
speed (Chapinal et al 2011). Measures of acceleration can
distinguish different gait types in dairy calves and can help
detect play running (de Passille et al 2010; Figure 4).
© 2012 Universities Federation for Animal Welfare
Figure 2
Mean daily time spent feeding at each day before and after calving (day 0) of dairy cows that were detected as suffering from severe
metritis or which remained free of metritis after calving. The vertical arrow shows the average time at which clinical signs of metritis
were present. Figure redrawn from data presented in Huzzey et al (2007).
Automated monitoring of behavioural-based animal welfare indicators 343
Animal Welfare 2012, 21: 339-350
doi: 10.7120/09627286.21.3.339
Figure 3
Mean daily duration of time spent lying down, as measured by accelerometers, of lactating dairy cows on individual Canadian dairy farms.
Figure redrawn from data presented in Ito et al (2009).
Measure of acceleration in vertical direction
of dairy calves’ legs when calves were walk-
ing (upper panel) or galloping (lower panel).
Individual steps can clearly be seen. Figure
redrawn from data presented in de Passillé
et al (2010).
Figure 4
344 Rushen et al
Accelerometers attached to various body parts have been
used to automatically detect the occurrence of other
behaviour patterns such as sleep patterns in dairy calves
(Hokkanen et al 2011) (Figure 5), activity around parturition
for sows (Cornou et al 2011) and feeding behaviour in goats
(Moreau et al 2009). Finally, measures of acceleration over
time can be used as a proxy measure for energy expenditure
(Gleiss et al 2011), and can be used to distinguish different
forms of locomotion in calves (de Passillé et al 2010).
Other devices have also been used to automatically record
animals’ locations, for example, local positioning has been
used to locate dairy cattle within a barn (Gygax et al 2007),
and GPS has helped assess the extent that zoo elephants
visit various parts of their enclosures (Leighty et al 2010).
Advantages and disadvantages
The main advantage of these devices for animal welfare
assessment at the farm or group level is that they can be
transported easily between farms or zoos, thus facilitating
comparisons. Many of the devices are self-contained with
their own power supply and memory storage, which means
that they can be used on free-roaming animals, such as beef
cattle on the range (Robert et al 2009) for which other
methods of data collection may be impractical.
These devices can be quite inexpensive, although this depends
upon the particular device being used. Since at least one
device is needed per animal, the total cost can be high when
there is a large number of animals, which will place a pressure
on reducing sample sizes during a welfare assessment.
One of the biggest disadvantages is the trade-off between cost
and the size of memory storage and power options. The
cheaper the device the less power can be carried and the
smaller the memory. The limit on memory means that
sampling frequency must be limited, which reduces the
ability of these devices to record the occurrence of short
duration behaviours over long periods of time. If the data are
recorded on the device itself, this reduces the chance of doing
real-time monitoring. The limits on memory can be overcome
by wireless collection of the data, which does allow for real-
time monitoring, but this increases the price and restricts the
area over which the animals can move, since they must be in
continuous or regular contact with the receiver.
Another disadvantage is that such devices, since they are
attached to the animals, are potentially invasive and may
influence the animals’ behaviour, or possibly cause
wounding, although this is rarely reported. The size of the
devices necessary to have a decent memory or power supply
will limit the extent that they can be used on smaller animals
such as poultry. Furthermore, some labour is needed to
attach and remove the devices from the animals, which can
increase the labour requirements of an animal welfare
inspection. Finally, the animals’ ability to remove the
devices must be considered: this can be particularly chal-
lenging in the case of inquisitive animals, such as group-
housed pigs, or elephants, for example (Leighty et al 2010).
Together, these problems place limits on the duration of
time that the devices can stay attached to the animals
meaning that they are less valuable for longitudinal studies
where animals are followed for a long period of time.
Perhaps one of the biggest disadvantages with these devices,
however, is the limited number of behaviour patterns for
which we have adequate tests of reliability and accuracy.
© 2012 Universities Federation for Animal Welfare
Figure 5
Amount of time that calves spent lying down, sleeping, in rapid eye movement sleep (REM) or non-rapid eye movement sleep (non-REM)
based on observations from video or predicted from the data from accelerometers attached to the calves’ necks. Redrawn from data
presented in Hokkanen et al (2011).
Automated monitoring of behavioural-based animal welfare indicators 345
Time spent lying and standing, general activity, and aspects
of gait can be reliably recorded in this way, and are clearly
important measures for animal welfare assessment.
Although there have been attempts to use accelerometers to
measure other behaviour patterns, we still lack sufficient
demonstrations of their ability to reliably and accurately
measure, for example, different forms of social behaviour
(eg Gygax et al 2007) or stereotyped behaviour etc.
Image and sound analysis
The ready availability of digital imagery along with the
development of computer programmes that can ‘read’ such
images, has resulted in the possibility of using automated
image analysis (‘computer vision’) to take measures of
animal behaviour. In addition, the ability of automation to
identify different sounds is also being explored.
A number of computer-assisted image analysis applica-
tions are being developed, such as for measuring space
use by cattle when getting up or lying down in order to
assess recommendations on stall size (Ceballos et al
2004), tracking the activity levels of individual chickens
and relating this to the degree of lameness (Aydin et al
2010), and tracking the movements of individual pigs in
group-housing systems (Ahrendt et al 2011).
Experimental studies of automated image analysis have
been done most often to aid in detecting lameness in dairy
cows. Two behavioural indicators of lameness in cows are
walking with an arched back, and poor tracking up, where
the back hoof is placed somewhat behind the front hoof
(Flower & Weary 2009). Computer programmes have
been used to detect both behaviour patterns (Flower et al
2005; Pluk et al 2010; Poursaberi et al 2010).
On-farm application of these approaches appears limited by
the difficulty in recognising a large number of individuals.
Image analysis has also been used to measure the ther-
moregulatory ‘clumping’ of pigs to assess the adequacy of
the pen temperature (Shao & Xin 2008), which does not
require the identification of individual animals but can be
used at the group level.
A promising and practical on-farm use of automated image
analysis comes from work using measures of ‘optical flow’
to examine movement patterns of broiler chickens.
Measures of optical flow are based on the changes in the
location of pixels in consecutive frames of a video, which
can be used to estimate the velocity of movement. Dawkins
et al (2009) placed webcams in ten commercial broiler
houses with flock size ranging from 3,000 to 40,000 birds,
and where traditional gait scoring on a sample of birds had
been used to estimate the prevalence of lameness within the
groups. The measures of optical flow were highly correlated
with the measures of gait scores. A major advantage of this
technique is that it does not require the identification of
individual animals but involves assessing movements of the
whole group of chickens. A subsequent study found that
measures of optical flow could identify periods of distur-
bance within flocks of laying hens, which would help
predict outbreaks of feather pecking (Lee et al 2011).
Finally, the development of computer programmes that can
identify and classify sounds is proving to be an interesting
development. This is particularly suitable for pigs which are
very vocal in expressing their emotional states: simple
measures of the amplitude of the sound produced by pigs
can give some information on the pigs’ responses to the
relative temperature and humidity within a barn (Borges
et al 2010) and computer programmes have been developed
to recognise pig vocalisations and separate these from back-
ground noise (Schön et al 2004). Exadaktylos et al (2008)
were able to develop a sound recognition programme that
could detect coughs by piglets. Preliminary results showed
that 82% of the sick cough sounds could be correctly iden-
tified. They concluded that the application could be used to
monitor the welfare in a pig house, and provide early iden-
tification of sick animals. Such an approach is also being
tested for dairy calves (Ferrari et al 2010).
Advantages and disadvantages
A major advantage of this type of automation is that it is
non-invasive. This is particularly important when consid-
ering smaller animals, such as poultry, where it would be
difficult to attach devices such as accelerometers. A
second important advantage is that the equipment is rela-
tively cheap, eg relatively simple webcams have been used
successfully (eg Dawkins et al 2009). Some forms of
image analysis do, however, require specialised cameras,
and the programmes for analysing the information can be
expensive to develop or buy.
The non-invasive nature of the equipment means that it can
be used for long-term monitoring of groups of animals, and is
also suitable for making comparisons between farms.
However, the difficulty in recognising a large number of indi-
vidual animals means that the approach is probably less
useful for long-term monitoring of individual animals than
for welfare assessment at the group level. Again, a major
limitation arises from the small number of behaviour patterns
that can be identified by computer vision. Measuring general
activity within groups appears relatively simple with
overhead cameras. For recording other forms of behaviour, eg
identification of lame cows through changes in gait, it may be
difficult to find a suitable location for the camera.
General issues with using automation to
measure behaviour
Validity and accuracy of automated measures
As with any scientific measurement, it is necessary to
establish the validity (ie are we measuring what we are
supposed to be measuring?) as well as the accuracy of the
measures (ie what is their sensitivity and specificity?).
The most common method of judging the validity of
measures collected automatically is by comparing them
with observations made by people, which can be either
direct or from video. DeVries et al (2003b) validated the
data generated by one automated system for recording
feeding behaviour (the GrowSafe monitoring system) and
compared this with measures taken from time-lapse video.
Animal Welfare 2012, 21: 339-350
doi: 10.7120/09627286.21.3.339
346 Rushen et al
The GrowSafe measure of the daily frequency of meals
showed perfect agreement with the result from the video
recordings. The duration of these meals was also highly
correlated with that estimated from the video. Comparison
with observations has also been used to validate automated
image analysis (Dawkins et al 2009) and devices such as
accelerometers (Ledgerwood et al 2010), although cross-
comparison with different types of loggers has also been
used (Ito et al 2009). However, in some cases it may be
difficult to validate automated measures since these are
effectively the only way of collecting data from, for
example, free-ranging animals.
It is also important to determine the accuracy (sensitivity
and specificity) with which the equipment can measure
behaviour. Automation is often assumed to be more
accurate than human observers, but, unfortunately,
machines do make mistakes, and while the electronics may
be very accurate in detecting the electronic signals, events
can occur which reduce the accuracy by which the elec-
tronic signals match the behaviour of interest. For example,
in their test of the accuracy with which automated feeders
detect feeding behaviour, DeVries et al (2003b) found
some instances in which the video showed that a cow was
present at the feed alley which was not recorded by
GrowSafe (12.6% of observations) and a few instances in
which the reverse was true (3.5% of observations).
Sometimes, however, automated feeding equipment can be
superior to human observation: Chapinal et al (2007)
reported that discrepancies between automated feeders and
human observers in measuring feeding behaviour of cattle
was due to the difficulties of seeing some aspects of
feeding from video recordings. Nevertheless, despite these
positive results it is important to stress that some estimate
of the likelihood of errors be determined.
Studies of accelerometers (Robert et al 2009; Ledgerwood
et al 2010; Ringgenberg et al 2010) generally report that
they measure standing and lying down with a high degree of
accuracy, especially when attached to the leg. The ability of
accelerometers to detect other behaviour patterns varies
according to the degree of fine discrimination between
similar behaviour patterns required. For example,
Hokkanen et al (2011) were able to identify 90% of the total
sleeping time of calves, with accelerometers attached to the
neck but were not as accurate in distinguishing the total
time the calves slept in either non-rapid eye movement
sleep or rapid eye movement sleep (Figure 5), although the
level of accuracy they report is still impressive.
However, the degree of accuracy of measures collected
automatically will depend upon the sampling schedule and
the method of ‘cleaning’ the data. For example, accelerom-
eters do not take measures continuously but instead take
samples. In many cases, the sampling rate is many times a
second, which can be considered essentially continuous.
However, since most devices in use are small and store the
data onboard, the limits on memory size mean that a longer
sampling interval (eg of several minutes) may be chosen
when recording over a long period of time. In these cases, a
sampling interval must be chosen which accurately
measures the behaviour in question but allows the device to
store the information for the period of time required. This
needs prior knowledge of the normal frequency and
duration with which the behaviour is performed.
Some editing of the data is also usually needed. For example,
automated measures of feeding behaviour often show that
the most common inter-visit intervals are very small, often
only a few minutes, and are unlikely to represent real inter-
meal intervals (DeVries et al 2003a). These brief intervals
can most easily be explained by temporary loss of contact
between the radio transmitter and the receiver, for example,
when the calf turns or lowers its head. These are usually
dealt with by removing very short inter-meal intervals. A
similar situation exists with devices such as accelerometers.
These effectively measure the orientation of the leg, and the
measures are based on the assumption that the animal is
lying down whenever the leg is horizontal. However, a hori-
zontal leg position can also occur briefly when the animal is
grooming, for example, and the most accepted method is to
remove very short occurrences on the assumption that large
animals are unlikely to lie down and then stand up again
within a short period of time. Ledgerwood et al (2010) found
that removal of very short bouts of lying down from the data
increased the accuracy of the measures, while Cornou et al
(2011) claimed that many misclassifications of sleeping
sows as ‘active’ could result from small movements that the
sow may make while sleeping. Of course, it may be that in
some circumstances cows do lie down for short periods of
time; for example, Ledgerwood et al (2010) suggested that
this occurs more often when cows are uncomfortable.
Hence, some care needs to be taken in ensuring the real and
meaningful behavioural events are not removed during the
data-cleaning process.
For other automated recording systems, the results of tests
of accuracy are less encouraging. For example, measures of
ground reaction force while cows are walking were shown
to have low sensitivity at detecting lameness in cows and
were inferior to subjective gait scoring (Bicalho et al 2007),
although recent developments have improved the ability of
pressure-sensitive walkways to detect lameness (Maertens
et al 2011). The accuracy of measures of weight distribution
or weight shifting while cows are standing are better, but
still relatively low. For example, Chapinal et al (2010)
found that the optimal result was a sensitivity of nearly 60%
and specificity of around 80% for detecting lame cows.
Tests of the ability of electronic sow feeders to detect illness
also report relatively low degrees of sensitivity (eg Cornou
et al 2008). Force plates have been shown to have low
accuracy in detecting leg problems in broilers, although
they are somewhat better than visual gait scores (Sandilands
et al 2011). Such low levels of accuracy mean that a large
number of animals would need to be tested in order to
obtain an accurate estimate of the farm prevalence (eg
Sandilands et al 2011), which does limit their value for on-
farm testing. It is likely that accuracy will be improved by
combining data from different sources: for example,
Chapinal et al (2010) found that by combining measures of
weight distribution with measures of walking speed and
© 2012 Universities Federation for Animal Welfare
Automated monitoring of behavioural-based animal welfare indicators 347
lying time, the sensitivity of detecting cows with sole ulcers
increased from nearly 60 to nearly 80%. Furthermore, the
accuracy can be improved by accounting for extraneous
factors that can systematically affect automated equipment.
For example, measures of how cattle distribute their weight
are affected by the time since milking (Chapinal et al 2009).
In general, the measure of accuracy or validity is valid only
for the conditions under which the test was done. For
example, IceTag accelerometers come with an algorithm for
calculating the number of steps taken by a dairy cow, which
appears quite accurate in doing this (Robert et al 2009;
Nielsen et al 2010). However, the measure is not accurate
for smaller dairy calves (Trénel et al 2009), probably
because the pattern of acceleration for the smaller animals
is very different from that of adults and a higher sampling
frequency is needed (de Passillé et al 2010).
Bias towards measures that can be automated
A real danger is that the enthusiasm for automated recording
will mean that more emphasis will be given to certain
behaviours in welfare assessment solely on the basis that
their measurement can be automated. For example, it is
relatively easy to automatically measure how much time
cows or pigs spend standing up, but an increased time spent
standing could occur because the animals are exploring
more or because they are fighting more, which have very
different implications for animal welfare. We need to
choose behavioural measures according to their relevance to
animal welfare and then develop methods of recording these
automatically, rather than choosing measures for their
ability to be recorded automatically.
Effects on relationship between people and animals
A valid concern in the use of any automation in animal
production is the effect on the relationship between people
and the animals, since automation generally reduces the
necessity for direct contact between them (Cornou 2009).
This is also true for the use of automation to assess the
welfare of animals, particularly welfare assessments done
by farmers, since the risk is that automation will reduce the
time that farmers spend watching their animals. On the
other hand, automation can give the farmers information
about the animals that they would not otherwise have, for
example, feed intakes of individual animals housed in
groups (Cornou 2009). Automated feeders can also help
detect sick animals within groups, which is difficult by
direct observation. Finally, farms are likely to continue
increasing in size for economic reasons, and farmers will
have less free time to observe their animals. Finding
automated methods of replacing human observers is a
necessary result of this rather than a contributing cause.
This is less of an issue in using automated recording in
third-party assessments, except that using automation may
reduce an assessor’s opportunities to make qualitative
judgements based on their direct observations. However,
most often automation is used not as a substitute for human
observers but to obtain data that would otherwise be prohib-
itively expensive to obtain.
Conclusion
To be useable for on-farm animal welfare assessment,
behavioural measures need to be valid, reliable and feasible
to take; the latter requirement usually means that recording
behaviour be cheap, not too time consuming, and not
interfere with the animals or the farm routines (Edwards
2007; Webster 2009). Our review leads us to conclude that
automatic measurement of animal behaviour has the
potential to meet all of these criteria. Most tests have been
able to establish validity and reliability, which is at least as
good as found between human observers. Feasibility has yet
to be fully established but while most examples of using
automation to record welfare-relevant behaviours come
from small-scale experimental studies, there have been
some real on-farm applications in welfare assessment
(Dawkins et al 2009; Ito et al 2009). However, the different
forms of automation have advantages and disadvantages;
with some being most useful for longitudinal monitoring of
individual animals, while others are best used for ‘cross-
sectional’ studies of group behaviours. Furthermore, the
claims for ‘objectivity’ need to be taken with a pinch of salt;
the need to clean data and the choice of sampling strategies
means that there is still an element of human judgement
involved in these measures. Perhaps the biggest problem so
far is the limited range of behaviours that have been
measured automatically, but technological developments,
especially in computer-vision will undoubtedly expand the
range; greater collaboration between ethologists and
engineers would certainly help. Behavioural measures need
to be chosen according to their relevance to animal welfare
rather than solely on their ability to be recorded automati-
cally. In general, however, we feel very positive about the
potential of automation to greatly extend the range of
behavioural measures that can be incorporated into on-farm
animal welfare assessment.
Acknowledgements
We gratefully acknowledge the input of our various collab-
orators in research especially Lene Munksgaard and Margit
Bak Jensen (University of Aarhus, Denmark), Laura
Hanninen and Matti Pastell (University of Helsinki,
Finland), Hajime Tanida and Yuki Koba (University of
Hiroshima, Japan). Our research in this area was supported
by Agriculture and Agri-Food Canada, the Dairy Farmers of
Canada, Global Animal Partnerships Inc, and the Natural
Sciences and Engineering Research Council of Canada.
References
Ahrendt P, Gregersen T and Karstoft H 2011 Development
of a real-time computer vision system for tracking loose-housed
pigs. Computers and Electronics in Agriculture 76: 169-174.
http://dx.doi.org/10.1016/j.compag.2011.01.011
Aydin A, Cangar O, Ozcan SE, Bahr C and Berckmans D
2010 Application of a fully automatic analysis tool to assess the
activity of broiler chickens with different gait scores. Computers
and Electronics in Agriculture 73: 194-199.
http://dx.doi.org/10.1016/j.compag.2010.05.004
Animal Welfare 2012, 21: 339-350
doi: 10.7120/09627286.21.3.339
348 Rushen et al
Bach A, Dinarés M, Devant M and Carré X 2007
Associations between lameness and production, feeding and milk-
ing attendance of Holstein cows milked with an automatic milking
system. Journal of Dairy Research 74: 40-46.
http://dx.doi.org/10.1017/S0022029906002184
Bicalho RC, Cheong SH, Cramer G and Guard CL 2007
Association between a visual and an automated locomotion score
in lactating Holstein cows. Journal of Dairy Science 90: 3294-3300.
http://dx.doi.org/10.3168/jds.2007-0076
Blokhuis HJ, Veissier I, Miele M and Jones B 2010 The Welfare
Quality® project and beyond: safeguarding farm animal well-being.
Acta Agriculturae Scandinavica A: Animal Sciences 60: 129-140
Borderas TF, Fournier A, Rushen J and de Passillé AM
2008 Effect of lameness on dairy cows’ visits to automatic milking
systems. Canadian Journal of Animal Science 88: 1-8.
http://dx.doi.org/10.4141/CJAS07014
Borderas TF, de Passillé AM and Rushen J 2009a Feeding
behavior of calves fed small or large amounts of milk. Journal of Dairy
Science 92: 2843-2852. http://dx.doi.org/10.3168/jds.2008-1886
Borderas TF, Rushen J, von Keyserlingk MAG and de
Passillé AM 2009b Automated measurement of changes in feeding
behavior of milk-fed calves associated with illness. Journal of Dairy
Science 92: 4549-4554. http://dx.doi.org/10.3168/jds.2009-2109
Borges G, Da Silva Miranda KO, Gates RS and Sales GT
2010 Environmental conditions effects in noise emission by piglets
in an intensive production farm. Proceedings of the American Society
of Agricultural and Biological Engineers Annual International Meeting
pp 5251-5257. 2010, Pittsburgh, PA, USA
Butterworth A, Knowles TG, Whittington P, Matthews L,
Rogers A and Bagshaw CS 2007 Validation of broiler chicken
gait scoring training in Thailand, Brazil and New Zealand. Animal
Welfare 16: 177-179
Ceballos A, Sanderson D, Rushen J and Weary DM 2004
Improving stall design: Use of 3-d kinematics to measure space
use by dairy cows when lying down. Journal of Dairy Science 87:
2042-2050. http://dx.doi.org/10.3168/jds.S0022-0302(04)70022-3
Chapinal N, de Passillé AM and Rushen J 2009 Weight dis-
tribution and gait in dairy cattle are affected by milking and late
pregnancy. Journal of Dairy Science 92: 581-588.
http://dx.doi.org/10.3168/jds.2008-1533
Chapinal N, de Passillé AM, Rushen J and Wagner S 2010
Automated methods for detecting lameness and measuring anal-
gesia in dairy cattle. Journal of Dairy Science 93: 2007-2013.
http://dx.doi.org/10.3168/jds.2009-2803
Chapinal N, de Passillé AM, Pastell M, Hänninen L,
Munksgaard L and Rushen J 2011 Measurement of accelera-
tion while walking as an automated method for gait assessment in
dairy cattle. Journal of Dairy Science 94: 2895-2901.
http://dx.doi.org/10.3168/jds.2010-3882
Chapinal N, Veira DM, Weary DM and von Keyserlingk
MAG 2007 Technical note: Validation of a system for monitoring
individual feeding and drinking behavior and intake in group-
housed cattle. Journal of Dairy Science 90: 5732-5736.
http://dx.doi.org/10.3168/jds.2007-0331
Cornou C 2009 Automation systems for farm animals: potential
impacts on the human-animal relationship and on animal welfare.
Anthrozoös 22: 213-220. http://dx.doi.org/10.2752/175303709
X457568
Cornou C, Vinther J and Kristensen AR 2008 Automatic
detection of oestrus and health disorders using data from elec-
tronic sow feeders. Livestock Science 118: 262-271.
http://dx.doi.org/10.1016/j.livsci.2008.02.004
Cornou C, Lundbye-Christensen S and Kristensen AR
2011 Modelling and monitoring sows’ activity types in farrowing
house using acceleration data. Computers and Electronics in
Agriculture 76: 316-324. http://dx.doi.org/10.1016/j.com-
pag.2011.02.010
Corr SA, McCorquodale C, McDonald J, Gentle M and
McGovern R 2007 A force plate study of avian gait. Journal of
Biomechanics 40: 2037-2043. http://dx.doi.org/10.1016/j.jbio-
mech.2006.09.014
Dawkins MS, Lee H, Waitt CD and Roberts SJ 2009 Optical
flow patterns in broiler chicken flocks as automated measures of
behaviour and gait. Applied Animal Behaviour Science 119: 203-209.
http://dx.doi.org/10.1016/j.applanim.2009.04.009
D’Eath RB, Tolkamp BJ, Kyriazakis I and Lawrence AB
2009 ‘Freedom from hunger’ and preventing obesity: the animal
welfare implications of reducing food quantity or quality. Animal
Behaviour 77: 275-288. http://dx.doi.org/10.1016/j.anbe-
hav.2008.10.028
de Passillé AM, Borderas TF and Rushen J 2011 Weaning
age of calves fed a high milk allowance by automated feeders:
Effects on feed, water, and energy intake, behavioral signs of
hunger, and weight gains. Journal of Dairy Science 94: 1401-1408.
http://dx.doi.org/10.3168/jds.2010-3441
de Passillé AM, Jensen MB, Chapinal N and Rushen J 2010
Technical note: use of accelerometers to describe gait patterns in
dairy calves. Journal of Dairy Science 93: 3287-3293.
http://dx.doi.org/10.3168/jds.2009-2758
De Paula Vieira A, Guesdon V, de Passillé AM, von
Keyserlingk MAG and Weary DM 2008 Behavioural indica-
tors of hunger in dairy calves. Applied Animal Behaviour Science 109:
180-189. http://dx.doi.org/10.1016/j.applanim.2007.03.006
DeVries TJ, von Keyserlingk MAG, Weary DM and
Beauchemin KA 2003a Measuring the feeding behavior of lac-
tating dairy cows in early to peak lactation. Journal of Dairy Science
86: 3354-3361. http://dx.doi.org/10.3168/jds.S0022-
0302(03)73938-1
DeVries TJ, von Keyserlingk MAG, Weary DM and
Beauchemin KA 2003b Technical note: validation of a system
for monitoring feeding behaviour of dairy cows. Journal of Dairy
Science 86: 3571-3574. http://dx.doi.org/10.3168/jds.S0022-
0302(03)73962-9
Edwards SA 2007 Experimental welfare assessment and on-farm
application. Animal Welfare 16: 111-115
Espejo LA, Endres MI and Salfer JA 2006 Prevalence of lame-
ness in high-producing Holstein cows housed in freestall barns in
Minnesota. Journal of Dairy Science 89: 3052-3058.
http://dx.doi.org/10.3168/jds.S0022-0302(06)72579-6
Exadaktylos V, Silva M, Aerts JM, Taylor CJ and
Berckmans D 2008 Real-time recognition of sick pig cough
sounds. Computers and Electronics in Agriculture 63: 207-214.
http://dx.doi.org/10.1016/j.compag.2008.02.010
Ferrari S, Piccinini R, Silva M, Exadaktylos V, Berckmans
D and Guarino M 2010 Cough sound description in relation to
respiratory diseases in dairy calves. Preventive Veterinary Medicine
96: 276-280. http://dx.doi.org/10.1016/j.prevetmed.2010.06.013
© 2012 Universities Federation for Animal Welfare
Automated monitoring of behavioural-based animal welfare indicators 349
Flower FC, Sanderson DJ and Weary DM 2005 Hoof
pathologies influence kinematic measures of dairy cow gait. Journal
of Dairy Science 88: 3166-3173. http://dx.doi.org/10.3168
/jds.S0022-0302(05)73000-9
Flower FC and Weary DM 2009 Gait assessment in dairy cat-
tle. Animal 3: 87-95. http://dx.doi.org/10.1017/S1
751731108003194
Gleiss AC, Wilson RP and Shepard ELC 2011 Making over-
all dynamic body acceleration work: on the theory of acceleration
as a proxy for energy expenditure. Methods in Ecology and Evolution
2: 23-33. http://dx.doi.org/10.1111/j.2041-210X.2010.00057.x
González LA, Tolkamp BJ, Coffey MP, Ferret A and
Kyriazakis I 2008 Changes in feeding behavior as possible indi-
cators for the automatic monitoring of health disorders in dairy
cows. Journal of Dairy Science 91: 1017-1028.
http://dx.doi.org/10.3168/jds.2007-0530
Gygax L, Neisen G and Bollhalder H 2007 Accuracy and val-
idation of a radar-based automatic local position measurement
system for tracking dairy cows in free-stall barns. Computers and
Electronics in Agriculture 56: 23-33. http://dx.doi.org/10.1016/j.com-
pag.2006.12.004
Hokkanen AH, Hänninen L, Tiusanen J and Pastell M 2011
Predicting sleep and lying time of calves with a support vector
machine classifier using accelerometer data. Applied Animal
Behaviour Science 134: 10-15. http://dx.doi.org/10.1016/j.applan-
im.2011.06.016
Huzzey JM, Veira DM, Weary DM and Von Keyserlingk
MAG 2007 Prepartum behavior and dry matter intake identify
dairy cows at risk for metritis. Journal of Dairy Science 90: 3220-
3233. http://dx.doi.org/10.3168/jds.2006-807
Ito K, Weary DM and von Keyserlingk MAG 2009 Lying
behavior: Assessing within- and between-herd variation in free-
stall-housed dairy cows. Journal of Dairy Science 92: 4412-4420.
http://dx.doi.org/10.3168/jds.2009-2235
Ito K, von Keyserlingk MAG, LeBlanc SJ and Weary DM
2010 Lying behavior as an indicator of lameness in dairy cows.
Journal of Dairy Science 93: 3553-3560. http://dx.doi.org/10.31
68/jds.2009-2951
Jensen MB and Holm L 2003 The effect of milk flow rate and
milk allowance on feeding related behaviour in dairy calves fed by
computer controlled milk feeders. Applied Animal Behaviour Science
82: 87-100. http://dx.doi.org/10.1016/S0168-1591(03)00054-6
Khan MA, Weary DM and von Keyserlingk MAG 2011
Invited review: effects of milk ration on solid feed intake, weaning
and performance in dairy heifers. Journal of Dairy Science 94: 1071-
1081. http://dx.doi.org/10.3168/jds.2010-3733
Kruse S, Traulsen I, Salau J and Krieter J 2011 A note on
using wavelet analysis for disease detection in lactating sows.
Computers and Electronics in Agriculture 77: 105-109.
http://dx.doi.org/10.1016/j.compag.2011.04.002
Ledgerwood DN, Winckler C and Tucker CB 2010
Evaluation of data loggers, sampling intervals, and editing techniques
for measuring the lying behavior of dairy cattle. Journal of Dairy
Science 93: 5129-5139. http://dx.doi.org/10.3168/jds.2009-2945
Lee HJ, Roberts SJ, Drake KA and Dawkins MS 2011
Prediction of feather damage in laying hens using optical flows and
Markov models. Journal of the Royal Society Interface 8: 489-499.
http://dx.doi.org/10.1098/rsif.2010.0268
Leighty KA, Soltis J and Savage A 2010 GPS assessment of
the use of exhibit space and resources by African elephants
(Loxodonta africana). Zoo Biology 29: 210-220
Lukas JM, Reneau JK and Linn JG 2008 Water intake and dry
matter intake changes as a feeding management tool and indicator
of health and estrus status in dairy cows. Journal of Dairy Science
91: 3385-3394. http://dx.doi.org/10.3168/jds.2007-0926
Madsen TN and Kristensen AR 2005 A model for monitoring
the condition of young pigs by their drinking behaviour. Computers
and Electronics in Agriculture 48: 138-154.
http://dx.doi.org/10.1016/j.compag.2005.02.014
Maertens W, Vangeyte J, Baert J, Jantuan A, Mertens KC,
De Campeneere S, Pluk A, Opsomer G, Van Weyenberg S
and Van Nuffel A 2011 Development of a real time cow gait
tracking and analysing tool to assess lameness using a pressure sen-
sitive walkway: The GAITWISE system. Biosystems Engineering 110:
29-39. http://dx.doi.org/10.1016/j.biosystemseng.2011.06.003
Millman ST 2007 Sickness behaviour and its relevance to animal
welfare assessment at the group level. Animal Welfare 16: 123-125
Moreau M, Siebert S, Buerkert A and Schlecht E 2009 Use
of a tri-axial accelerometer for automated recording and classifi-
cation of goats’ grazing behaviour. Applied Animal Behaviour Science
119: 158-170. http://dx.doi.org/10.1016/j.applanim.2009.04.008
Nielsen LR, Pedersen AR, Herskin MS and Munksgaard L
2010 Quantifying walking and standing behaviour of dairy cows
using a moving average based on output from an accelerometer.
Applied Animal Behaviour Science 127: 12-19.
http://dx.doi.org/10.1016/j.applanim.2010.08.004
Ouweltjes W, van der Werf JTN, Frankena K and van
Leeuwen JL 2011 Effects of flooring and restricted freestall
access on behavior and claw health of dairy heifers. Journal of Dairy
Science 94: 705-715. http://dx.doi.org/10.3168/jds.2010-3208
Pastell ME and Kujala M 2007 A probabilistic neural network
model for lameness detection. Journal of Dairy Science 90: 2283-
2292. http://dx.doi.org/10.3168/jds.2006-267
Pastell ME, Hänninen L, de Passillé AM and Rushen J 2010
Measures of weight distribution of dairy cows to detect lameness
and the presence of hoof lesions. Journal of Dairy Science 93: 954-
960. http://dx.doi.org/10.3168/jds.2009-2385
Platz S, Ahrens F, Bendel J, Meyer HHD and Erhard MH
2008 What happens with cow behavior when replacing concrete
slatted floor by rubber coating: a case study. Journal of Dairy
Science 91: 999-1004. http://dx.doi.org/10.3168/jds.2007-0584
Pluk A, Bahr C, Leroy T, Poursaberi A, Song X, Vranken
E, Maertens W, Van Nuffel A and Berckmans D 2010
Evaluation of step overlap as an automatic measure in dairy cow
locomotion. Transactions of the ASABE 53: 1305-1312
Poursaberi A, Bahr C, Pluk A, Van Nuffel A and
Berckmans D 2010 Real-time automatic lameness detection
based on back posture extraction in dairy cattle: Shape analysis of
cow with image processing techniques. Computers and Electronics
in Agriculture 74: 110-119. http://dx.doi.org/10.1016/j.com-
pag.2010.07.004
Proudfoot KL, Weary DM and von Keyserlingk MAG 2010
Behavior during transition differs for cows diagnosed with claw
horn lesions in mid lactation. Journal of Dairy Science 93: 3970-
3978. http://dx.doi.org/10.3168/jds.2009-2767
Animal Welfare 2012, 21: 339-350
doi: 10.7120/09627286.21.3.339
350 Rushen et al
Quimby W 2000 Application of feeding behaviour to predict
morbidity of newly received calves in a commercial feedlot.
Canadian Journal of Animal Science 81: 315-320.
http://dx.doi.org/10.4141/A00-098
Rajkondawar PG, Liu M, Dyer RM, Neerchal NK, Tasch
U, Lefcourt AM, Erez B and Varner MA 2006 Comparison
of models to identify lame cows based on gait and lesion scores,
and limb movement variables. Journal of Dairy Science 89: 4267-
4275. http://dx.doi.org/10.3168/jds.S0022-0302(06)72473-0
Ringgenberg N, Bergeron R and Devillers N 2010
Validation of accelerometers to automatically record sow pos-
tures and stepping behaviour. Applied Animal Behaviour Science
128: 37-44. http://dx.doi.org/10.1016/j.applanim.2010.09.018
Robert B, White BJ, Renter DG and Larson RL 2009
Evaluation of three-dimensional accelerometers to monitor and
classify behavior patterns in cattle. Computers and Electronics in
Agriculture 67: 80-84. http://dx.doi.org/10.1016/j.com-
pag.2009.03.002
Rushen J, Pombourcq E and de Passillé AM 2007 Validation of two
measures of lameness in dairy cows. Applied Animal Behaviour Science 106:
173-177. http://dx.doi.org/10.1016/j.applanim.2006.07.001
Rushen J, de Passillé AM, von Keyserlingk MAG and
Weary DM 2008 The Welfare of Cattle. Springer: Dordrecht, The
Netherlands. http://dx.doi.org/10.1007/978-1-4020-6558-3
Sandilands V, Brocklehurst S, Sparks N, Baker L,
McGovern R, Thorp B and Pearson D 2011 Assessing leg
health in chickens using a force plate and gait scoring: how many
birds is enough? Veterinary Record 168: 77.
http://dx.doi.org/10.1136/vr.c5978
Schön PC, Puppe B and Manteuffel G 2004 Automated
recording of stress vocalisations as a tool to document impaired
welfare in pigs. Animal Welfare 13: 105-110
Shao B and Xin H 2008 A real-time computer vision assess-
ment and control of thermal comfort for group-housed pigs.
Computers and Electronics in Agriculture 62: 15-21.
http://dx.doi.org/10.1016/j.compag.2007.09.006
Sørensen JT, Rousing T, Møller SH, Bonde M and
Hegelund L 2007 On-farm welfare assessment systems: What
are the recording costs? Animal Welfare 16: 237-239
Svensson C and Jensen MB 2007 Short communication:
Identification of diseased calves by use of data from automatic
milk feeders. Journal of Dairy Science 90: 994-997.
http://dx.doi.org/10.3168/jds.S0022-0302(07)71584-9
Tanida H, Koba Y, Rushen J and de Passillé AM 2011 Use
of three-dimensional acceleration sensing to assess dairy cow gait
and the effects of hoof trimming. Animal Science Journal 82: 792-
800. http://dx.doi.org/10.1111/j.1740-0929.2011.00903.x
Trénel P, Jensen MB, Decker EL and Skjøth F 2009
Technical note: quantifying and characterizing behavior in dairy
calves using the IceTag automatic recording device. Journal of Dairy
Science 92: 3397-3401. http://dx.doi.org/10.3168/jds.2009-2040
van Nuffel A, Sprenger M, Tuyttens FAM and Maertens
W2009 Cow gait scores and kinematic gait data: can people see
gait irregularities? Animal Welfare 18: 433-439
Weary DM, Huzzey JM and Von Keyserlingk MAG 2009
Board-invited Review: Using behavior to predict and identify ill
health in animals. Journal of Animal Science 87: 770-777.
http://dx.doi.org/10.2527/jas.2008-1297
Webster AJF 2009 The Virtuous Bicycle: a delivery vehicle for
improved farm animal welfare. Animal Welfare 18: 141-147
© 2012 Universities Federation for Animal Welfare
... Furthermore, the animals do not need to be equipped with a sensor or collar, as was the case in other studies (e.g., [20,44]). This non-invasive approach is a major advantage of computer vision [45]. ...
... There is a further limitation in assessing welfare based on behavior, e.g., 'highly active standing' could be due to the fox's interaction with enrichment (indicating good welfare) and also pacing (indicating poor welfare). Thus, the number of determinable behaviors is a limitation [45]. This limitation can be minimized, for example, by training a model with more postures. ...
Article
Full-text available
The behavior of animals is related to their health and welfare status. The latter plays a particular role in animal experiments, where continuous monitoring is essential for animal welfare. In this study, we focus on red foxes in an experimental setting and study their behavior. Although animal behavior is a complex concept, it can be described as a combination of body posture and activity. To measure body posture and activity, video monitoring can be used as a non-invasive and cost-efficient tool. While it is possible to analyze the video data resulting from the experiment manually, this method is time consuming and costly. We therefore use computer vision to detect and track the animals over several days. The detector is based on a neural network architecture. It is trained to detect red foxes and their body postures, i.e., ‘lying’, ‘sitting’, and ‘standing’. The trained algorithm has a mean average precision of 99.91%. The combination of activity and posture results in nearly continuous monitoring of animal behavior. Furthermore, the detector is suitable for real-time evaluation. In conclusion, evaluating the behavior of foxes in an experimental setting using computer vision is a powerful tool for cost-efficient real-time monitoring.
... Technically, an improvement of the sensitivity would be possible, but accompanied by a risk of a loss in specificitytherefore, finding the right balance between the performance criteria depends on how the respective interest measurements are implemented. It might be worth discussing this issue further, especially when using FPD as a welfare indicator rather than as benchmark criteria, as the focus then would benefit from a shift toward a higher sensitivity (49). ...
Article
Full-text available
Footpad dermatitis (FPD) is an indicator of animal welfare in turkeys, giving evidence of the animals' physical integrity and providing information on husbandry management. Automated systems for assessing FPD at slaughter can present a useful tool for objective data collection. However, using automated systems requires that they reliably assess the incidence. In this study, the feet of turkeys were scored for FPD by both an automated camera system and a human observer, using a five-scale score. The observer reliability between both was calculated (Krippendorff's alpha). The results were not acceptable, with an agreement coefficient of 0.44 in the initial situation. Therefore, pictures of 3,000 feet scored by the automated system were evaluated systematically to detect deficiencies. The reference area (metatarsal footpad) was not detected correctly in 55.0% of the feet, and false detections of the alteration on the footpad (FPD) were found in 32.9% of the feet. In 41.3% of the feet, the foot was not presented straight to the camera. According to these results, the algorithm of the automated system was modified, aiming to improve color detection and the distinction of the metatarsal footpad from the background. Pictures of the feet, now scored by the modified algorithm, were evaluated again. Observer reliability could be improved (Krippendorff's alpha = 0.61). However, detection of the metatarsal footpad (50.9% incorrect detections) and alterations (27.0% incorrect detections) remained a problem. We found that the performance of the camera system was affected by the angle at which the foot was presented to the camera (skew/straight; p < 0.05). Furthermore, the laterality of the foot (left/right) was found to have a significant effect (p < 0.001). We propose that the latter depends on the slaughter process. This study also highlights a high variability in observer reliability of human observers. Depending on the respective target parameter, the reliability coefficient (Krippendorff's alpha) ranged from 0.21 to 0.82. This stresses the importance of finding an objective alternative. Therefore, it was concluded that the automated detection system could be appropriate to reliably assess FPD at the slaughterhouse. However, there is still room to improve the existing method, especially when using FPD as a welfare indicator.
... For extended periods of time, it is difficult to supervise dogs in natural outdoor surroundings or multiroom structures. Sensors that monitor animal behavior automatically are an excellent solution for these purposes [2]. Additionally, machine learning enables novel methods for detecting potential health concerns early on by detecting indicators concealed in dogs' actions that are not obvious to the human eye. ...
... This is also relevant to the prioritization of animal welfare during trials. Additionally, systems for monitoring animal welfare designed and described for on-farm use (Rushen et al., 2012;Zehner et al., 2012;Matthews et al., 2016;Caria et al., 2017) would also be useful within research environments to enable the quick identification of sick or uncomfortable animals by continuously and closely monitoring specific welfare parameters (e.g., temperature, level of activity, social behavior, use of functional areas). The use of interactive and monitoring technology during trials to enhance animal welfare would contribute toward a higher score against this principle, provided that appropriate measures were taken to guarantee the welfare of partakers (e.g., temporarily or permanently withdrawing individual animals from the trial, or arresting the trial altogether when necessary). ...
Article
Full-text available
The principles of Replacement, Reduction and Refinement (3Rs) were developed to address the ethical dilemma that arises from the use of animals, without their consent, in procedures that may harm them but that are deemed necessary to achieve a greater good. While aiming to protect animals, the 3Rs are underpinned by a process-centered ethical perspective which regards them as instruments in a scientific apparatus. This paper explores the applicability of an animal-centered ethics to animal research, whereby animals would be regarded as autonomous subjects, legitimate stakeholders in and contributors to a research process, with their own interests and capable of consenting and dissenting to their involvement. This perspective derives from the ethical stance taken within the field of Animal-Computer Interaction (ACI), where researchers acknowledge that an animal-centered approach is essential to ensuring the best research outcomes. We propose the ethical principles of relevance, impartiality, welfare and consent, and a scoring system to help researchers and delegated authorities assess the extent to which a research procedure aligns with them. This could help researchers determine when being involved in research is indeed in an animal's best interests, when a procedure could be adjusted to increase its ethical standard or when the use of non-animal methods is more urgently advisable. We argue that the proposed principles should complement the 3Rs within an integrated ethical framework that recognizes animals' autonomy, interests and role, for a more nuanced ethical approach and for supporting the best possible research for the benefit animal partakers and wider society.
Chapter
Over the last 150 years, animal sheltering has evolved from its origins as a public health and safety industry to one that also highly prioritizes animal health and welfare. Animal welfare science has also expanded during this time and provides objective, evidence‐based approaches to the assessment of animals’ experiences both in and outside of the shelter setting, as well as means by which to guide quality of life judgments made by animal caretakers. Many questions around animal welfare and quality of life inherently involve ethical determinations. A variety of ethical decision‐making frameworks can be used for these purposes and may carry the benefits of inclusivity, transparency, objectivity, and minimization of moral stress.
Book
Full-text available
The aim of this publication is to summarize the current knowledge about the automatic milking systems and precision husbandry of dairy cows and calves. Recent knowledge are discussed. Automatic milking system (AMS) offer an innovative approach to improve productivity on dairy farms. AMS affects the future growth of farms, improves the working conditions of employees and the quality of life on family farms. The AMS impact on dairy cows is explored. What is the basis for adapting to the dairy cow milking is discussed in the first part of this publication. Precision methods for smart farming are written in the second part. The study used 232 references from current scientific and professional literature.
Article
Background: Accurate assessment of the welfare of non-human primates (NHPs) used and bred for scientific purposes is essential for effective implementation of obligations to optimise their well-being, for validation of refinement techniques and novel welfare indicators, and for ensuring the highest quality data is obtained from these animals. Despite the importance of welfare assessment in NHP research, there is little consensus on what should be measured. Greater harmonisation of welfare indicators between facilities would enable greater collaboration and data sharing to address welfare-related questions in the management and use of NHPs. Methods: A Delphi consultation was used to survey attendees of the 2019 NC3Rs Primate Welfare Meeting (73 respondents) to build consensus on which welfare indicators for macaques and marmosets are reliable, valid, and practicable, and how these can be measured. Results: Self-harm behaviour, social enrichment, cage dimensions, body weight, a health monitoring programme, appetite, staff training, and positive reinforcement training were considered valid, reliable, and practicable indicators for macaques (≥70% consensus) within a hypothetical scenario context involving 500 animals. Indicators ranked important for assessing marmoset welfare were body weight, NHP induced and environmentally induced injuries, cage furniture, huddled posture, mortality, blood in excreta, and physical enrichment. Participants working with macaques in infectious disease and breeding identified a greater range of indicators as valid and reliable than did those working in neuroscience and toxicology, where animal-based indicators were considered the most important. The findings for macaques were compared with a previous Delphi consultation, and the expert-defined consensus from the two surveys used to develop a prototype protocol for assessing macaque welfare in research settings. Conclusions: Together the Delphi results and proto-protocol enable those working with research NHPs to more effectively assess the welfare of the animals in their care and to collaborate to advance refinement of NHP management and use.
Article
Sensor-based behavioural observation methods improve our understanding of individual behaviour and welfare in large commercial groups, including poultry. Validating automatically generated data is essential to account for potential sources of error. Our study aimed to validate a sensor-based tracking system for broiler breeders (BB) and laying hens (LH) in commercially relevant housing systems. The BB study was conducted in 10 pens with 33 females and three males (Ross 308) per pen. Half of the pens contained a raised slatted area and two raised group nests (Raised), while in the remaining five pens, the nests and slats were on the floor (Floor). For the LH study, six pens with a commercial aviary were used, with 225 Dekalb White hens housed per pen (Aviary). Focal hens (BB, 10/pen; LH, 18/pen) were equipped with backpacks containing tracking devices that registered transitions between four (BB) or five (LH) resource-related zones covering all accessible areas within each housing system. The tracking data was compared against video observations for 20 focal BB on two days and 18 focal LH on three days (3 ×20 min/day). Three data cleaning methods tested with 30 values of a duration parameter were evaluated for reliability and stability with a cross-validation approach. Initial and post-cleaning performance were assessed with accuracy, precision, and sensitivity of recorded transitions and by calculating the reliability for two aspects of movement: total transitions (Lin’s Concordance Correlation Coefficient) and locations (mean proportion of matching duration). A mixed model was applied to evaluate the duration of stay after false and true tracking registrations. Initial location reliability was high (> 0.949) in all housing systems, while reliability of total transitions was low (< 0.264), particularly in both BB housing systems (< 0.064). The cross-validation revealed suitable cleaning procedures for Aviary and Raised but not for Floor, thus Floor was not used for further analysis. Cleaning improved total transitions (> 0.832) while reliability of locations remained high (> 0.949) in Aviary and Raised. The duration between registrations was affected by housing system (p < 0.001) and was longer for true compared to false registrations (p < 0.001). Initial tracking performance varied between movement aspects and housing systems. The difference in duration between true and false registrations allowed for the application of simple yet effective data cleaning in Aviary and Raised, ensuring that the generated data better represented the animal's actual movement with reduced error associated with the tracking system.
Article
Background: Accurate assessment of the welfare of non-human primates (NHPs) used and bred for scientific purposes is essential for effective implementation of obligations to optimise their well-being, for validation of refinement techniques and novel welfare indicators, and for ensuring the highest quality data is obtained from these animals. Despite the importance of welfare assessment in NHP research, there is little consensus on what should be measured. Greater harmonisation of welfare indicators between facilities would enable greater collaboration and data sharing to address welfare-related questions in the management and use of NHPs. Methods: A Delphi consultation was used to survey attendees of the 2019 NC3Rs Primate Welfare Meeting (73 respondents) to build consensus on which welfare indicators for macaques and marmosets are reliable, valid, and practicable, and how these can be measured. Results: Self-harm behaviour, social enrichment, cage dimensions, body weight, a health monitoring programme, appetite, staff training, and positive reinforcement training were considered valid, reliable, and practicable indicators for macaques (≥70% consensus) within a hypothetical scenario context involving 500 animals. Indicators ranked important for assessing marmoset welfare were body weight, NHP induced and environmentally induced injuries, cage furniture, huddled posture, mortality, blood in excreta, and physical enrichment. Participants working with macaques in infectious disease and breeding identified a greater range of indicators as valid and reliable than did those working in neuroscience and toxicology, where animal-based indicators were considered the most important. The findings for macaques were compared with a previous Delphi consultation, and the expert-defined consensus from the two surveys used to develop a prototype protocol for assessing macaque welfare in research settings. Conclusions: Together the Delphi results and proto-protocol enable those working with research NHPs to more effectively assess the welfare of the animals in their care and to collaborate to advance refinement of NHP management and use.
Book
Full-text available
Cieľom tejto publikácie je zhrnúť súčasné poznatky o systémoch automatického dojenia a precíznom chove dojníc a teliat. Diskutuje sa o najnovších poznatkoch. Automatický systém dojenia (AMS) ponúka inovatívny prístup na zlepšenie produktivity na mliečnych farmách. AMS ovplyvňuje budúci rast fariem, zlepšuje podmienky práce zamestnancov a kvalitu života na rodinných farmách. Skúma sa vplyv AMS na dojnice. Čo je základom prispôsobenia na dojenie kráv je popísané v prvej časti tejto publikácie. V druhej časti sú rozpísané precízne metódy pre inteligentné farmárčenie. V štúdii bolo použitých 232 odkazov z aktuálnej vedeckej a odbornej literatúry.
Article
Full-text available
Lameness is a major welfare problem for dairy cows and has important economic consequences. On-farm detection of lameness is difficult, and automated methods may be useful for early diagnoses. Lameness may reduce the efficiency of automated milking systems (AMS) if lame cows are less willing to visit the automatic milking unit voluntarily and poor attendance at milking units may help detect lameness. To determine whether a low frequency of visits in an AMS could serve as an indicator of lameness, data on the frequency of visits of 578 cows in 12 AMS on eight farms were collected. From each AMS, 22 cows (from a mean of 54 cows per AMS), were classified as either the 11 highest visitors or the 11 lowest visitors based on the total number of visits to the milking unit. These selected cows (n = 256) were videotaped while walking in a standard test area and their gait scored on a 5-point scale (1 = sound 5 = severely lame). Intra- and inter-observer reliability values between and within observers were high for gait scoring. Significant differences in gait scores between the two groups of cows (P < 0.05) were found in 9 out of 12 AMS: high-visiting cows had better gait scores than low-visiting cows. Four percent of high visitors were classified as slightly lame and 32% of low visitors were classified as either slightly or severely lame. The overall numerical rating score was the most effective in discriminating between high and low visitors, and scoring each individual component of gait did not greatly improve discrimination between the two groups of cows. The frequency that dairy cows visit an AMS is related to their locomotory ability, and data from the AMS may help in the early detection of lameness.
Conference Paper
Full-text available
In swine production, it is known that environmental factors have a great influence on animal behavior. An innovative method to quantify and classify uncomfortable situations for the animals is with automatic registration of noise levels emitted by these animals in intensive production systems. Thus, the objective of this work was to assess the influence of environment variables (temperature and air humidity) in the emission of noise levels by piglets in an intensive production system. The experiment was carried on a commercial swine farm during the period from April to June of 2008. The data collected include sound pressure levels (noise) emitted by piglets in nursering phase, and environmental variables (temperature and air relative humidity) inside the facility. Decibelimeters were used to record noise levels and HOBOS data loggers were used to measure and record environmental variables. Enthalpy and THI comfort indexes were used to characterize the thermal environment. Analysis of correlation and simple regression were used. Piglet's noise levels (NL piglets) showed a correlation to the environmental variables, air temperature and air humidity, so, it was possible, through a regression analysis, to estimate NLpiglets related to the environmental variables to the studied environment.
Article
Full-text available
The objective of this study was to use feeding behavior of newly received steers (average initial weight 191 kg) to detect morbidity in animals in a commercial feedlot. Two separate 32 d feeding trials were conducted in Wellton, Arizona, in July and November 1996. Radio frequency technology was used to record the total time spent within 50 cm of the feedbunk (animal presence every 5.25 s times 5.25 s) in 3 h intervals from 0600 to 2400 on a daily basis for 103 and 122 male calves in trial 1 and 2, respectively. Statistical procedures based on the cumulative sums (CUSUM) of the 3 h feeding intervals were used to detect morbid animals, compared with detection of animals deemed morbid by experienced pen riders. In trial 1, the CUSUM procedure detected animal morbidity 4.5 d earlier (P < 0.001) than the feedlot personnel. In trial 2, the CUSUM procedure detected animal morbidity 3.7 d earlier (P < 0.001) than feedlot pen riders. Overall accuracy, positive predictive value and sensitivity of the CUSUM prediction method were 87, 91, and 90%, respectively. Combined trial data suggest that feeding behavior during the first 30 d cattle are in a receiving pen, as collected with radio frequency technology and analyzed with CUSUM charts, may be used to detect animal morbidity approximately 4.1 d earlier (P < 0.001) than conventional methods typically employed in commercial feedlots.
Article
Full-text available
This article suggests a method for detecting oestrus, lameness and other health disorders for group housed sows fed by electronic sow feeders (ESF). The detection method is based on the measure of the individual eating rank, modeled using a univariate dynamic linear model. Differences between the predicted values of the model and the observations are monitored using a control chart: a V-mask is applied on the cumulative sum of the standardized forecast errors of the model. According to the respective V-mask parameters, alarms are given for each of the three states (oestrus, lameness, others) when the deviations between model predicted values and observations exceed some defined parameters. External information is incorporated into the model to limit the number of false alarms when a subgroup of sows enters and exits a group or both. The detection method was implemented on data collected within three production herds over 12 months. Visual recordings were performed to identify sows in oestrus or with health disorders. The detection method showed a high specificity. For oestrus detection, there was a sensitivity of 59%, 70% and 75% for the three herds as compared to 9% (herd 1) and 20% (herd 2) using lists of sows as alarms. Monitoring lameness results in a sensitivity of 56%, 70% and 41%, vs. 39%, 32% and 22% using the lists; monitoring other health disorders resulted in a sensitivity of 0%, 75% and 39% for the three respective herds, vs. 34% and 16% for herds 2 and 3 using the lists. To limit the number of false alarms, it is suggested to expand the model by including daily feed intake or body activity as other response variables.
Article
Full-text available
The aim of this study was to test the hypothesis that valuable on-farm outcome measures of broiler (meat) chicken welfare can be derived from optical flow statistics of flock movements recorded on video or CCTV inside commercial broiler houses. ‘Optical flow’ describes the velocity of image motion across an eye or camera and statistical patterns can be derived automatically and continuously throughout the life of a flock. We provide descriptive statistics (mean, variance, skewness and kurtosis) of optical flow of 10 intensively housed commercial broiler flocks between the ages of 32 and 35 days. There were no significant correlations between any of these measures and flock mortality. However, all four measures were correlated significantly with the % of birds in a house showing poor walking (high gait scores). Furthermore, these gait scores were highly negatively correlated with the % of time chickens spent walking and with their stride rate (no. of strides/min), as measured by focal behaviour analysis of individual birds from the same video records. The results suggest that optical flow measures have the potential to be used as an adjunct or even a substitute for gait scoring on commercial farms with the added advantage that the measurements could be made continuously throughout the life of a flock, are fully automated, completely non-invasive and non-intrusive and do not involve the biosecurity risk of having people visiting different farms to carry out gait scoring. The correlations between gait scores and optical flow also suggest that gait scoring itself has an objective basis.
Article
Full-text available
This article suggests a method for classifying sows’ activity types performed in farrowing house. Five types of activity are modeled using multivariate dynamic linear models: high active (HA), medium active (MA), lying laterally on one side (L1), lying laterally on the other side (L2) and lying sternally (LS). The classification method is based on a Multi-Process Kalman Filter (MPKF) of class I. The performance of the method is validated using a Test data set. Results of activity classification appear satisfying: 75–100% of series are correctly classified within their activity type. When collapsing activity types into active (HA and MA) vs. passive (L1, L2, LS) categories, results range from 96 to 100%. In a second step, the suggested method is applied on series collected for 19 sows around the onset of farrowing, including 9 sows that received bedding materials (57 sow days in total) and 10 sows that received no bedding material (61 sow days in total). Results indicate that there is a marked (i) increase of active behaviours (HA and MA, p
Article
The aim of this study was to explore the possibility of capturing cow locomotion activity by computer vision techniques and to calculate the correlation between step overlap and manually measured locomotion scores. In two experiments, a total of 208 video recordings of 85 individual lactating cows were gait scored visually by an observer. The side-view videos were recorded when cows were freely passing the experimental setup. After image processing, the imprint location, step overlap, body size, and relative step overlap were calculated. The values of automatically measured step overlap showed a high correlation with the manually measured step overlap (R2 = 0.739, p < 0.001; R2 = 0.809, p < 0.001). The maximal step overlap allowed differentiation between gait scores 1 and 3 (p = 0.032) and between gait scores 2 and 3(p = 0.039). The difference between gait scores 1 and 2 was not significant (p = 0.079). There was a large variation between individual cows, in both the progress of lameness and the influence on step overlap. Changes in step overlap were also seen that were not matched by changes in gait score. Step overlap is a variable that shows a relationship with manual gait scores, but it is not strong enough to be used as a single classifier for lameness in all cows. ©2010 American Society of Agricultural and Biological Engineers ISSN 2151-0032.
Article
The inflammatory response evokes changes in behaviour including increased thermoregulatory activities and sleep, reduced social exploration and appetite, and altered food preferences. This sickness response also includes feelings of lethargy, depression, and pain, collectively referred to as 'malaise'. Recent experiments involving laboratory rodents reveal information about proximate mechanisms of sickness behaviour, but scant information exists about how sickness behaviour is expressed by farmed species or within social environments. The behavioural needs of ill individuals differ from those of conspecifics, and failure to accommodate the needs of ill individuals may exacerbate suffering. Policy makers, industry and animal welfare certification programs recommend hospital pens to address the housing and handling needs of ill livestock and to reduce risks of disease transmission. However, a survey of swine farms in Ontario, Canada revealed deficiencies in the use of hospital pens and gaps in knowledge about best management practices for this vulnerable population. There is considerable scope to improve the welfare and husbandry of ill and at risk animals through effective use of hospital pens and supportive therapies.
Article
Cattle lameness causes considerable animal welfare problems and negatively affects the farm economy. Gait scoring techniques and claw health reports are commonly used for research and surveys, but few daily management solutions exist to monitor gait parameters from individual cows within a herd. A tool to automate and process the measurement of spatiotemporal kinematic and force variables was developed using a pressure sensitive walkway, commercial farm infrastructure and management tools. A fully automatic setup on ILVO's experimental farm measures and analyses the gait of each cow that exits the milking parlour by registration of 20 basic kinematic gait variables. Based on this variable set, a wide range of typical gait parameters such as triple support time, abduction, etc. can be calculated. This paper presents some practical results and considerations related to this system of automated gait analysis. A first validation of the acquired variables shows that the Gaitwise system is capable of clustering observations in relation to the observer score with an overall sensitivity of 76-90% and specificity of 86-100%. Variables of asymmetry and speed seem most promising for further research on the detection of lameness. For future reference, the resulting cow gait variables will be stored for time series analysis to detect changes in individual cow walking behaviour in relation to any source (claw lesions, gestation stage, etc.). Such measurements could provide valuable information for management, veterinary check-ups and further research for automatic lameness detection in cattle.
Article
The aim of the present study was to clarify whether wavelet analysis could identify water intake variation due to health problems and to differentiate between healthy and treated sows. A total of 85 sows (Large White, German Landrace and their crossbreeds) in parities 1–6 were observed. Data were measured in the farrowing unit, beginning at the day of farrowing until the day 10 of lactation. During the week of farrowing the body temperature was measured rectally. If a sow had a body temperature higher than 39.5°C, the animal was treated. The individual water intake was recorded on a daily basis. The present study used the wavelet analysis to investigate individual water intake of lactating sows. After evaluation of the best wavelet transform, daily water intake of each sow was decomposed into coefficients of approximation and details. The analysis calculated five coefficients of approximation. The differences between the second and first (C21), between the third and second (C32), between fourth and third (C43) and between the fifth and fourth coefficients (C54) were calculated. The smallest value of C21 and C32 was the minimal difference of coefficients (MDC1) of decision criterion one and the smallest value of C21, C32, C43 and C54 decision criterion two (MDC2). These criteria were used to differentiate between healthy and treated sows. The mean MDC1 of healthy sows was significantly higher than the mean MDC1 of treated sows. No distinction could be derived between healthy and treated sows using the mean MDC2. The classification parameters sensitivity, specificity and error rate were calculated depending on different thresholds. Sensitivity ranged from 34% to 83% and specificity from 32% to 92%. In conclusion wavelet analysis in combination with criteria MDC1 could be used to analyse water intake of treated and healthy lactating sows.