Conference PaperPDF Available

The practical design of in-vehicle telematics device with GPS and MEMS accelerometers


Abstract and Figures

Latest generation of vehicle tracking devices relies not only on Global Positioning System (GPS) but also uses a low-cost Micro-Electro-Mechanical Systems (MEMS) accelerometers. This combination supports new services as driving style characterization and Automatic Crash Notification (ACN). Focus will be on practical considerations of such telematics unit. Work will consider boundaries of allowed errors and minimal requirements for sensors and mounting requirements. Sensor range for crash detection and impact angle estimation was tested on field trials with two units containing accelerometers range of 18g and 2g. Kinematic orientation of vehicle is evaluated in series of field trials with resulting standard deviation of estimation of 1.67°. The second run of experiments considers dynamic range and sampling rate of sensors during collision. Sensor range of 8g (typical for current telematics devices) can be used to detect crash without accurate knowing of impact angle.
Content may be subject to copyright.
128 Telfor Journal, Vol. 4, No. 2, 2012.
Abstract The latest generation of vehicle tracking
devices relies not only on Global Positioning System (GPS)
but also uses low-cost Micro-Electro-Mechanical Systems
(MEMS) accelerometers. This combination supports new
services such as driving style characterization and Automatic
Crash Notification (ACN). Our focus will be on practical
considerations of such a telematics unit. The paper will
consider the boundaries of allowed errors and minimal
requirements for sensors and mounting requirements. Sensor
range for crash detection and impact angle estimation was
tested on field trials with two units containing accelerometers
range of 18g and 2g. The kinematic orientation of vehicle is
evaluated in a series of field trials with a resulting standard
deviation of estimation of 1.67˚. The second run of
experiments considers the dynamic range and sampling rate
of sensors during collision. A sensor range of 8g (typical for
present-day telematics devices) can be used to detect crash
without accurate knowledge of impact angle.
Keywords — Accelerometer, advanced vehicle tracking,
automatic crash notification, GPS, telematics.
PS tracking devices and event data recorders (EDR)
are common equipment in modern vehicles, either as
a part of embedded car electronics or as “aftermarket”
devices installed to support various telematics services.
Evolution of modern fleet management tools brings a
wider span of supported services. Common examples are
characterization of driver behavior and automatic crash
notification. This is followed by an increase of processing
power available on board and introduction of MEMS
motion sensors. Telematics services today are commonly
based on “aftermarket” tracking units that are built-in after
purchase of the vehicle. This is partly caused by flexibility
reasons. The backbone of information infrastructure are
GPRS links and “cloud computing”. GPS and motion
sensors are often combined with data coming from a
vehicle On-Board-Diagnostics (OBD) bus. This
redundancy makes speed information more reliable.
Current trends show a need for a deeper analysis of driver
behavior and recognition of undesired driving patterns.
Devices are becoming a step closer to GPS/INS navigation
Being “aftermarket” is usually a drawback for devices
with accelerometers. Such devices always need rigid and
Milan B. Vukajlović, Bitgear, Generala Mihajla Nedeljkovića 12,
11000 Belgrade, Serbia (e-mail:
Srdjan Tadić, Bitgear, Generala Mihajla Nedeljkovića 12, 11000
Belgrade, Serbia (e-mail:
Dejan M. Dramićanin, Bitgear, Generala Mihajla Nedeljkovića 12,
11000 Belgrade, Serbia (e-mail:
precise mounting. Vibrations or improper axis alignment
affects data integrity. Hence, device installation is a
critical operation considering reliability/repeatability, time
consumption and cost. Optimization of this process would
bring large benefits to service suppliers.
Complex and fatal crashes are often a matter of dispute
in forensic investigation because existing tools provide
only lines of march and demand a high level of expertise
to resolve ambiguities. This paper will consider technical
requirements for measuring specific forces in a vehicle
during crash that would provide accurate characterization
of crash severity – such as sensor dynamic range,
resolution and sampling rate regarding real-vehicle physics
validated on field trials.
A. Device Leveling
The accelerometer measures specific forces in a sensor
“body” coordinate frame, which in general differs from a
vehicle coordinate frame. An application like this usually
requires a simple horizontal leveling procedure. One
option is to measure gravity components while a vehicle is
stationary and o a flat surface [1]. This provides two out of
three Euler angles required to estimate the initial
Transformation Cosine Matrix. The Transformation
Cosine Matrix can be defined considering three successive
rotations of angles φ (roll), θ (pitch) and ψ (yaw) around
the x, y and z axis respectively (Fig. 1). It is a link between
acceleration in “vehicle” and “body” coordinate frame (1).
Fig. 1. Euler angles.
The lower boundary of accelerometer resolution
(sensitivity) depends on allowed leveling and orientation
error. Rough installation introduces leakage of specific
forces to other axis which increases a chance of false-
alarming. Undesired components should be kept at least
one order of magnitude lower than forces typical for listed
driving events. The rule-of-thumb for this application is to
The Practical Design of In-vehicle Telematics
Device with GPS and MEMS Accelerometers
Milan B. Vukajlović, Srdjan Tadić, and Dejan M. Dramićanin
Vukajlović et al.: The Practical Design of In-vehicle Telematics Device with GPS and MEMS Accelerometers 129
enable leveling within 1-2 degrees that should be
achievable with sensor sensitivity in the order of 10-20mg.
At consumer/automotive grades of sensors there is also
a significant influence of imperfections of MEMS
technology on leveling accuracy [2]. Dominant is the
effect of zero-G bias, which at this grade might be on a
level of 20mg. Sensor temperature variation is usually
almost one order of magnitude lower over the whole
temperature span of interest. A significant source of error
might also be the physical non-orthogonality of sensor axis
which is above 1% for this type of IC’s. Leveling is
usually less sensitive to other usually modeled errors such
as a scale factor and nonlinearity. These effects come to
importance in high dynamic environment and will more
affect reliability while extracting crash angles. A
cumulative effect might degrade leveling accuracy.
Though, forces present at vehicle during typical driving
provide a certain level of robustness when using such
devices for driving behavior analysis. As a consequence,
in practice, it is not usually necessary to do specific
calibration. Device leveling is done in a real time. Angles
are recalculated only if vehicle is stationary.
B. Orientation Algorithm
The GPS provides vehicle heading at 1s time instants.
In general this orientation is not the same as for device.
One approach to override rigid and time consuming
measuring of device orientation is to incorporate a filter
for kinematic alignment [3]-[5]. Kinematic alignment
relies on external speed and acceleration sources such as
GPS or odometer, and external heading info as from GPS
or a magnetic sensor. Kinematic alignment algorithms use
external heading information to estimate yaw angle. GPS
heading is used in the algorithm instead of magnetic
sensor heading because of indispensable magnetic sensor
calibration and the influence of uncorrelated magnetic
disturbances caused in the proximity of metal and magnets
on sensor output [6], [7]. Hence, the kinematic alignment
method with GPS heading, odometer data and three-axis
accelerometer is tested. Device orientation relative to
vehicle is recursively estimated in real time while vehicle
is accelerating or braking with a constant direction. A
constant direction is detected according to GPS heading
data (2), where VE and VN are speed components in east
and north directions. GPS receiver uses knowledge of the
components of speed directed to north and east to find
heading. By elementary math, heading is less accurate at
low speeds or while driving along North or East. These
cases should be filtered out as outliers. Another fact is that
GPS data are always delayed regarding inertial data. Delay
is usually greater than half a second. When vehicle is
turning the effect is harmful. A simple filter would be to
search only for intervals with a constant GPS heading for
at least 3 seconds. This excludes potential start of
cornering in latter second. Accelerations ay and ax (3) are
mean values from a previous second. Precise device
orientation is possible only if these components have a
notable value. Hence, two conditions should be met for
kinematic orientation in this simple way – a certain level
of acceleration and a constant heading.
Algorithm 1 - Simple Kinematic Alignment
while 1 do
if (|VN| > THR_SPEED && |VE| > THR_SPEED)
if heading_gps(t) = heading_gps(t-1s)
if ((cnt>3) and (speed(t-1s)– speed(t-2s)>THR))
heading = -atan(ay(t-1s)/ax(t-1s));
cnt = 0;
cnt = 0;
end while
Harctan (2)
C. Field Trials
A passenger car was equipped with a GPS tracking unit
with consumer grade MEMS accelerometers. Two 20-
minute rides were logged under regular driving conditions
in urban environment (Fig. 2). Initial roll (-0.4˚) and pitch
(4.1˚) angles were calculated after mounting. According to
the photo of mounted device (Fig. 3), a relative angle
between device and vehicle longitudinal axes (“box
orientation”) was approximately 68˚.
0200 400 600 800 1000 1200
Time (s)
Speed (km/h) / Heading (degrees)
Fig. 2. Speed, heading and route of the first of test drives.
The emphasis of tests was on device orientation. By the
proposed method, orientation angle should slowly
converge to an acceptable value. In reality, convergence
time might be even measured in days after device
installation. Calculation is affected by misalignment
between the sensor, printed circuit board and enclosure, as
well as with the built-in non-orthogonality of sensor axes.
These effects could not be measured by photo or
mechanical tools. As after effect, the overall deviation of
130 Telfor Journal, Vol. 4, No. 2, 2012.
“true” orientation from assumed value might be roughly
up to ±5˚.
Fig. 4 shows results of the first trial. Speed difference
threshold in this run was set to 1.9m/s, and calculated yaw
angle was 67.33˚. The standard deviation of 18 obtained
results was 1.67˚. By lowering the threshold which is dual
to increasing the filter bandwidth, the number of available
estimations will rise. Experiments show that an increased
number of measurements will not compensate a higher
level of noise introduced to algorithm, resulting in
worsening standard deviation.
Fig. 3. Car-mounted telematics unit.
Fig. 5 illustrates results obtained on the same run with
lower thresholds of 1.5m/s. Now, the number of yaw angle
estimations rises to 25, but this benefit was poorly traded
with an increased standard deviation of 4.6˚.
Suitable calculation intervals depend on individual
driving patterns. Still, as a general rule, it has been
identified that acceleration sections which satisfy all
conditions were much less frequent. These sections were
also less noisy and the standard deviation of estimations
done on these intervals was significantly lower than on
braking sections (2.44˚ compared to 1˚ on some tests).
The second trial illustrates a possible drawback of this
simple approach. Specific traffic conditions allowed only
one estimate of yaw angle (72˚) in a 20-minute period.
Road-traffic jam limited car dynamics and recorded
accelerations/brakings exceeded pre-defined thresholds at
only one point of journey. Other tests in regular driving
conditions and on various terrains and road inclinations
confirmed a very similar level of standard deviation of
The aim of the next experiment was to answer if this
level of orientation error is acceptable for driving style
characterization. The answer depends on the magnitude of
specific forces typically measured during targeted driving
events. As an example, similar setup recorded the lateral
acceleration of car while cornering at high speed with a
small radius of curvature (Fig. 6). Recorded data were
post-processed in simulation to add the effect of yaw angle
error of 5˚. Simulation has verified that error is one order
of magnitude below specific forces of interest. The lateral
acceleration threshold was preset to 6m/s2, while
acceleration error (Fig. 7) was below 0.4m/s2. It can be
assumed that influence on false detection of a cornering
event is limited and acceptable.
Fig. 6. Real lateral acceleration and simulated contribution
of yaw angle error of 5˚.
Fig. 7. Difference between real lateral acceleration and
simulated contribution of yaw angle error of 5˚.
A. Required Sensor Parameters
ACN is an important component of modern and future
telematics services. It is important to consider minimal
technical requirements of GPS and sensors.
The duration of “typical” vehicle crash almost never
exceeds 300ms [8]. The highest level of energy exchange
during a typical crash is usually within starting 50ms.
Following this, the accelerometer sampling rate roughly
should be at least 50-100Hz. This exceeds the rate required
for driving style characterization. It also makes GPS useful
010 20 30 40 50
Acceleration (m/s
Time (s)
010 20 30 40 50
Acceleratio n (m /s
Time (s)
Fig. 4. Yaw angle estimations – rigid thresholds.
Fig. 5. Yaw angle estimations – lower thresholds.
Yaw an gle (d egrees)
Estimated yaw angle
Mean yaw angle
Measured yaw angle
Yaw ang le (deg rees)
Estimated yaw angle
Mean yaw angle
Measure d yaw angle
Vukajlović et al.: The Practical Design of In-vehicle Telematics Device with GPS and MEMS Accelerometers 131
only during pre- and post-crash intervals.
Severe crashes may reach levels of up to 50-100g‘s
depending on conditions [8]. In the USA, federal
regulations [9] define minimal specifications of EDR
sensor if it was embedded in a vehicle in production and
for the purpose of crash recording (Table 1). These
standards are not obligatory for aftermarket devices. Most
of the aftermarket devices that are currently present in
vehicles do not match these criteria. Usually, they cover a
range of 4-10g. In case of crash, output is saturated.
Override of this problem could enable ACN on many
already produced devices.
Acceleration Min.
Sensitivity Sampling
Lateral ±5 g 0.5 g 100 Hz
Longitudinal ±50 g 0.5 g 100 Hz
Normal ±5 g 0.5 g 100 Hz
B. Crash Test Setup
Crash is a complex and non-linear event. One possible
approach to analysis is field testing. Concerning practical
reasons, a 1:5 scaled down and reinforced model car was
used for a crash test. The test was prepared considering
mechanical aspects such as weight, ratio of tire size and
surface fairness, engine vibrations and suspension. Two
inertial measurement units were used. The first unit
covered a 2g dynamic range at a rate of 100Hz with
18mg/LSB resolution. This device represented consumer
grade technology that is common for already deployed
devices. The second, referent, device covered a wider
dynamic range of 18g, at a sampling rate of 200Hz and
with a sensitivity of 3.33mg/LSB. A referent device was
industry grade IMU but this fact is not prevailing for this
type of crash analysis due to high dynamics and short
intervals of interest. Data were recorded in a flash memory
and post-processed in MATLAB as shown on a block
diagram in Fig. 8.
Fig. 8. Block diagram of crash-test setup.
C. Test Scenario
The device recorded a full frontal crash of a compact
car into a barrier. A cardboard box with partially
homogeneous content has served as a barrier. The box
weighed app. 10kg that was similar to the weight of a
model car. A barrier was placed on a concrete surface with
low friction in between. A barrier was supported by a
metal bar placed on lower back side.
Impact speed was approximately 30kmph (“event 1”). A
hit caused box rotation around the axis that coincides with
a metal bar. Moderate deformation of box combined with
its rotation brought to the rotation of model car around a
mirrored axis positioned on car lower back side. Car
reared to an angle between 45˚ and 90˚, and returned to a
previous horizontal position (“event 2”). Movement
continued until the final impact to a metal bar (full frontal)
at low speed (“event 3”). Fig. 9 shows video snapshots of
this event.
This simple test comprised multiple crash events with
various obstacles. Scope was on crash detection and angle
of impact, commonly known as Principal Direction of
Force (PDOF). PDOF is very valuable to an expert-
Referent and tested device outputs are shown in Fig. 10
and Fig. 11 respectively. A change of speed vector is
normalized by predefined law.
Fig. 9. Snapshots of crash test video.
A referent unit successfully detected crash events while
a tested unit detected only the strongest one. According to
the value of normalized delta-V, all three axes on a tested
device should be saturated to detect a crash of this
severity. This means that crash events aligned with any
device axis could not be detected.
Crash PDOF should be calculated at the moment of
detection. The peak value of delta-V thus does not have to
be crucial for its calculation. The next question is whether
it is possible to get a fair estimate of PDOF if sensor were
saturated. The assumed “true” value of PDOF was app.
10˚ in a horizontal and -5˚ in a vertical plane. A referent
device calculated (17˚h, -8˚v). Tested device PDOF output
was (47˚h, 23˚v) which is only sub-quadrant precision.
Referent sensor data were degraded in MATLAB
environment to match tested data and obtained results are
shown in Table 2. Test verifies that PDOF accuracy is
severely degraded even if sensors are being saturated at
much higher values or while operating on a higher rate.
One possible approach to partly override saturation is to
use polynomial interpolation techniques. Cubic spline
approximation is used for data reconstruction (4) – (8)
with h as duration of sensor clipping.
Approximation enabled detection of all events, but was
not able to recover PDOF (43˚h, -61˚v) (Fig. 12).
Polynomial interpolation effects maximal acceleration on
low sampling rates.
132 Telfor Journal, Vol. 4, No. 2, 2012.
Data Dynamic
18 g 200 Hz 17˚ -8˚
8 g 200 Hz 29˚ 15˚
2 g 200 Hz 45˚ 5˚
8 g 100 Hz 37˚/23˚ 27˚/7˚
2 g 100 Hz 43˚/42˚ 22˚/4˚
Tested 2 g 100 Hz 47˚ 23˚
DCxBxAxxy +++= 23 (4)
() ( )
= (5)
B= (6)
() ()
ii xyxy
''1 *
+ (7)
)( i
xyD = (8)
Proposed device orientation estimation is shown to be
good enough for driving style characterization. A standard
deviation of 1.67˚ is acceptable in driving style
characterization and crash detection. Regular vehicle
dynamics is usually within a range of 2g and it is not
sufficient for proper crash detection. It is found that 8g is a
sufficient accelerometer dynamic range for crash
detection. Cubic interpolation of saturated sensor data
further improved crash detection. PDOF information is
poorly recovered if crash is not recorded in full range no
matter if the sampling rate is increased.
For proper driving style characterization and ACN
detection with PDOF estimation with one device it is
recommended to use two accelerometer sensors. A low
dynamic range accelerometer with high sensitivity should
be used for driving style characterization and device
kinematic alignment and a high dynamic range
accelerometer should be used for ACN detection and
PDOF estimation.
[1] R. M. Rogers, “Applied mathematics in integrated navigation
systems,” 2nd ed., Rogers Engineering and Associates, AIAA,
pp. 18–19, 2003.
[2] M. Park and Y. Gao, “Error and Performance Analysis of MEMS-
based Inertial Sensors with a Low-cost GPS Receiver,” Sensors
2008, pp. 2240-2261, March 29, 2008.
[3] G. Dissanayake, S. Sukkarieh, E. Nebot, and H. Durrant-Whyte,
“The aiding of a low-cost strapdown inertial measurement unit
using vehicle model constraints for land vehicle applications,”
IEEE Trans. On Robotics and Automation, vol. 17, pp. 733, Oct.
[4] A. O. Salycheva and M. E. Cannon, “Kinematics Azimuth
Alignment of INS using GPS Velocity Information,” Proc. of NTM
Conf., San Diego, CA, pp. 3, January 26-28, 2004.
[5] P. Davidson, M. A. Vazquez, and R. Piche, “Uninterrupted portable
car navigation system using GPS, map and inertial sensors data,”
IEEE 13th Int. Symp. on Consumer Electronics, pp. 2, 2009.
[6] D. Gebre-Egziabher, G. H. Elkaim, J. D. Powell, and B. W.
Parkinson, “A Non-Linear, Two-Step Estimation Algorithm for
Calibrating Solid-State Strapdown Magnetometers,” Proceedings of
the International Conference on Integrated Navigation Systems,
St. Petersburg, Russia, May 2001.
[7] V. Renaudin, M. H. Afzal, and G. Lachapelle, “Complete Triaxis
Magnetometer Calibration in the Magnetic Domain,” Journal of
Sensors, vol. 2010, Article ID 967245, Oct. 26, 2010.
[8] M. Huang, Vehicle Crash Mechanics, CRC Press, New York, 2002.
[9] U.S. Fed. Register, vol. 76, no. 151, Aug. 5, 2011.
Fig. 10. Referent device normalized delta-V.
Fig. 11. Tested device normalized delta-V.
0 1 2 3 4 5
DeltaV (norm)
0 1 2 3 4 5
Time (s)
DeltaV (norm)
Fig. 12. Cubic approximation partly recovers delta-V.
0 1 2 3 4 5
Time (s)
DeltaV (norm)
... Similarly, non motorized movements such as walking, posture transition, gentle motion, standing, sitting and lying, are detected in [16]. Once these patterns have been identified and classified, several applications can be built on top, like the automatic crash notification system [14] or the drunk driver detection system [6]. ...
Conference Paper
Full-text available
Traditional radio frequency based wireless communication/ localization suffers from environmental effects such as multipath propagation. We introduce a robust wireless communication methodology which utilizes the physical world itself to convey information. Our proposed intelligent pavement signaling system (iPASS) places irregularities on pavement in order to encode different messages that can be deciphered in real time by an application running on e.g., a phone. iPASS utilizes available accelerometers, e.g., in today's smartphones, that can detect even small vibrations from the inside of a car that are produced by driving over irregularities on the road. Our feasibility study illustrates that iPASS can successfully decode a 5 bit message using a conventional smartphone when cruising at various speeds.
Complex and fatal road vehicle crashes are often a matter of dispute in forensic investigation. The proposed algorithms and methods, utilizing data from inertial measurement units and satellite navigation, enable road-vehicle trajectory reconstruction in 3-D suitable for postaccident forensic analysis. This paper provides an estimate of the reliability of this approach and its relation to component specifications. We start by describing an appropriate type of the strap-down inertial navigation system before providing a detailed performance analysis when the system is integrated with global navigation satellite systems. In addition, we present custom hardware and embedded software solutions, as well as experimental evaluation in real-world case study, capturing typical crash scenario performed at an approved safety performance assessment laboratory. In our experiments, strap-down inertial mechanization was used for short-term vehicle trajectory reconstruction using sensor readings and compared with ground-truth position captured on video. We have used reference equipment to validate the approach and the experiment resulted in a positioning accuracy on a decimeter level.
Full-text available
This paper presents a new algorithm for calibrating modern, solid-state strapdown magnetometers. The traditional method for cali-brating these sensors is a cumbersome procedure called "swinging" which involves leveling and rotating the vehicle containing the mag-netometer through a series of known headings. The calibration procedure presented in this paper does not require any external reference. Furthermore, unlike the method of "swinging", this calibration method performs the calibration in the magnetic field domain and not the heading domain. This makes the algorithm applicable for cases where the magnetometer is used in applications other than heading determination. The calibration algorithm uses an estimator where the states are the various sensor errors. These states in the estimator are hard iron biases, soft iron biases, and scale factor errors. This calibration procedure is implemented using a non-linear two step estimator where the first step problem is solved by using standard batch least squares linear estimation techniques and the second step is solved algebraically. This two step estimator outperforms an Extended Kalman Filter (EKF) formulation. A low-cost magnetometer triad was used to collect data and experimentally validate the calibration algorithm. The non-linear two step estimator results in magnetometer calibration residual less than 0.005 Gauss which translates to heading errors on the order of 1 degree.
Conference Paper
Full-text available
This paper presents the development of car navigation system for portable navigation devices and car telematics applications. The objective was to develop a system that can provide uninterrupted reliable navigation even when GPS signals are not available. The approach uses digital maps, 3D accelerometer and one gyro for directional measurements to improve positioning availability and reliability in weak signal environment and during short GPS signal outages. This system does not require vehicle installation and can be easily transferred between vehicles. Loosely coupled extended Kalman filter and probabilistic map-matching algorithm provide optimally tuned navigation solution and continuous auto calibration of inertial sensors. A real-time prototype was built. The system accuracy performance was investigated using field tests in an urban environment.
Full-text available
This paper presents an algorithm for calibrating erroneous tri-axis magnetometers in the magnetic field domain. Unlike existing algorithms, no simplification is made on the nature of errors to ease the estimation. A complete error model, including instrumentation errors (scale factors, nonorthogonality, and offsets) and magnetic deviations (soft and hard iron) on the host platform, is elaborated. An adaptive least squares estimator provides a consistent solution to the ellipsoid fitting problem and the magnetometer's calibration parameters are derived. The calibration is experimentally assessed with two artificial magnetic perturbations introduced close to the sensor on the host platform and without additional perturbation. In all configurations, the algorithm successfully converges to a good estimate of the said errors. Comparing the magnetically derived headings with a GNSS/INS reference, the results show a major improvement in terms of heading accuracy after the calibration.
Full-text available
Global Navigation Satellite Systems (GNSS), such as the Global Positioning System (GPS), have been widely utilized and their applications are becoming popular, not only in military or commercial applications, but also for everyday life. Although GPS measurements are the essential information for currently developed land vehicle navigation systems (LVNS), GPS signals are often unavailable or unreliable due to signal blockages under certain environments such as urban canyons. This situation must be compensated in order to provide continuous navigation solutions. To overcome the problems of unavailability and unreliability using GPS and to be cost and size effective as well, Micro Electro Mechanical Systems (MEMS) based inertial sensor technology has been pushing for the development of low-cost integrated navigation systems for land vehicle navigation and guidance applications. This paper will analyze the characterization of MEMS based inertial sensors and the performance of an integrated system prototype of MEMS based inertial sensors, a low-cost GPS receiver and a digital compass. The influence of the stochastic variation of sensors will be assessed and modeled by two different methods, namely Gauss-Markov (GM) and AutoRegressive (AR) models, with GPS signal blockage of different lengths. Numerical results from kinematic testing have been used to assess the performance of different modeling schemes.
Full-text available
This paper presents a new method for improving the accuracy of inertial measurement units (IMUs) mounted on land vehicles. The algorithm exploits nonholonomic constraints that govern the motion of a vehicle on a surface to obtain velocity observation measurements which aid in the estimation of the alignment of the IMU as well as the forward velocity of the vehicle. It is shown that this can be achieved without any external sensing provided that certain observability conditions are met. A theoretical analysis is provided together with a comparison of experimental results between a nonlinear implementation of the algorithm and an IMU/GPS navigation system. This comparison demonstrates the effectiveness of the algorithm. The real time implementation is also addressed through a multiple observation inertial aiding algorithm based on the information filter. The results show that the use of these constraints and vehicle speed guarantees the observability of the velocity and the attitude of the inertial unit, and hence bounds the errors associated with these states. The strategies proposed provides a tighter navigation loop which can sustain outages of GPS for a greater amount of time as compared to when the inertial unit is used with standard integration algorithms
The combination of the Global Positioning System (GPS) and an Inertial Navigation System (INS) can deliver a system performance that is superior to either one in stand- alone mode, since each system compensates for the other's shortcomings. Due to large errors of the inertial sensors, error estimation and further compensation have to be performed using external information from GPS. The azimuth misalignment angle, being one of the largest non-stationary INS error components, has a significant impact on the INS accuracy; therefore this paper is focused on the development and testing of a special procedure for the azimuth error estimation. The method presented is based on a traditional Kalman filter and uses GPS velocity and heading information for measurement updates. The filter is divided into two stages, defined by a different degree of observability of the error components in the state vector. The estimation of the azimuth misalignment is performed only when it is strongly observable, in other words, during high vehicle dynamics. To reduce the transition period of its estimation, the problem statement is reformulated so that the azimuth misalignment error becomes a directly measured component. The discussed algorithm is tested for land navigation applications. The results show that precise azimuth error estimation using the proposed approach can significantly improve the INS accuracy in general and the prediction accuracy in particular, when the GPS measurements are not available.
CRASH PULSE AND KINEMATICS Introduction Vehicle Impact Modes and Crash Data Recording Digital Filtering Practice per SAE J211 and ISO 6487 Basic Kinematic Relationships Impact and Excitation: Vehicle and Sled Test Kinematics Vehicle and Occupant Kinematics in Fixed Object Impact Kinematic Variables Case Study: Single Vehicle-Tree Impact Accident Restraint Coupling Occupant Ridedown Analysis and Energy Management References CRASH PULSE CHARACTERIZATION Introduction Moment-Area Method Pulse Approximations with Non-Zero Initial Deceleration Pulse Approximations with Zero Initial Deceleration Fourier Analysis of Crash Pulse References CRASH PULSE PREDICTION--THE CONVOLUTION METHOD Introduction Transfer Function via Convolution Integral Transfer Function and a Spring-Damper Model Belted and Unbelted Occupant Performance with Air Bag Body Mount and Torso Restraint Transfer Functions Effect of Sled and Barrier Pulses on Occupant Response Other Applications Response Inverse Filtering References BASICS OF IMPACT AND EXCITATION MODELING Introduction Impact and Excitation--Rigid Barrier and Hyge Sled Tests Ridedown Existence Criteria and Efficiency Basics of Spring and Damper Dynamic Modeling Two-Mass and Effective Mass Systems Vehicle-to-Barrier Impact: Spring-Mass Model Spring-Mass Occupant Model Subjected to Excitation Vehicle-to-Vehicle Impact: Spring-Mass Model A Maxwell Model Impact on Kelvin Model--Vehicle or Component Damping Factor and Natural Frequency from Tests Excitation on the Kelvin Model--Occupant and Restraint References RESPONSE PREDICTION BY NUMERICAL METHODS Introduction Hybrid Model--A Standard Solid Model Two Mass-Spring-Damper Model Natural Frequencies in Two-Mass System Numerical Searching Techniques Loading and Unloading Simulation A Lumped-Parameter Model--CRUSH II Side-Impact and Frontal-Offset Models References IMPULSE, MOMENTUM, AND ENERGY Introduction Background Center of Gravity and Motion Theorem Impulse and Circle of Constant Acceleration Principle of Work and Energy Vehicle Inertia Properties and Critical Sliding Velocity Rollover Crashes Eccentric Loading on Vehicle Rollover References CRASH SEVERITY AND RECONSTRUCTION Introduction Occupant Motion Under Impact and Excitation Preloading on an Occupant Central Collisions Non-Central Collisions Use of DV and BEV of Vehicles in Crash Severity Assessment Vehicle Acceleration and Crash Severity Velocity and Energy Distributions in Two-Vehicle Impact Computation of Barrier Equivalent Velocity Intermediate Mass Effect Modeling the Vehicle-to-Vehicle Compatibility Test Accident Reconstruction Methodology References