Inertial-based motion capture (IMC) has been suggested to overcome many of the limitations of traditional motion capture systems. The validity of IMC is, however, suggested to be dependent on the methodologies used to process the raw data collected by the inertial device. The aim of this technical summary is to provide researchers and developers with a starting point from which to further develop the current IMC data processing methodologies used to estimate human spatiotemporal and kinematic measures. The main workflow pertaining to the estimation of spatiotemporal and kinematic measures was presented, and a general overview of previous methodologies used for each stage of data processing was provided. For the estimation of spatiotemporal measures, which includes stride length, stride rate, and stance/swing duration, measurement thresholding and zero-velocity update approaches were discussed as the most common methodologies used to estimate such measures. The methodologies used for the estimation of joint kinematics were found to be broad, with the combination of Kalman filtering or complimentary filtering and various sensor to segment alignment techniques including anatomical alignment, static calibration, and functional calibration methods identified as being most common. The effect of soft tissue artefacts, device placement, biomechanical modelling methods, and ferromagnetic interference within the environment, on the accuracy and validity of IMC, was also discussed. Where a range of methods have previously been used to estimate human spatiotemporal and kinematic measures, further development is required to reduce estimation errors, improve the validity of spatiotemporal and kinematic estimations, and standardize data processing practices. It is anticipated that this technical summary will reduce the time researchers and developers require to establish the fundamental methodological components of IMC prior to commencing further development of IMC methodologies, thus increasing the rate of development and utilisation of IMC.
1. Introduction
Motion capture systems have been used extensively in biomechanics research to capture spatiotemporal measures of stride length, stride rate, contact time, and swing time and angular kinematic measures of joint angles. Such measures are commonly used in disease/condition diagnosis, injury prevention, and sport performance analysis [1–7]. The most common technologies used to collect human spatiotemporal and kinematic measures are three-dimensional (3D) optical, two-dimensional (2D) video, and electromagnetic based systems [8]. When motion capture data is collected in conjunction with data from force platforms, angular kinetics may also be modelled.
Three-dimensional optical motion capture (OMC) systems are often considered to be the gold standard method of motion capture; however, these systems are expensive and typically confined to a small capture volume within a laboratory environment [9, 10]. For a full body motion analysis, researchers are required to place up to 50 markers at anatomically specific locations, and a line of sight to each marker must be maintained by at least two cameras for each data frame throughout the movement [9]. Maintaining a line of sight to each marker throughout the movement is a major challenge when using 3D OMC as markers often become displaced and/or occluded when implements (such as boxes for manual handling assessments and bats, balls, or barbells for sporting assessments) are included in the movement analysis [9]. The displacement and/or occlusion of markers results in loss of data, increased measurement error, increased tracking time, and sometimes the inability to analyse a captured movement.
Two-dimensional (2D) video motion capture is a more affordable alternative to 3D OMC, requiring one or more video cameras with sufficient frame rate and video processing software such as the freely available software Kinovea (http://Kinovea.org, France) or Tracker (Open Source Physics). A number of drawbacks exist for 2D video motion capture. Multiple video cameras may be required for a full motional analysis. For example, for a running gait motion analysis, cameras may be required with views of the frontal and sagittal plane to capture joint varus/vulgus rotation and joint flexion/extension, stride length, stance duration, and swing duration, respectively. The high frame rate required to ensure accuracy when capturing fast movements (particularly sporting movements) result in large file sizes and extensive processing time. Both marker-based and marker-less 2D video motion capture rely on a line of sight of the participant throughout the movement and as such see similar occlusion limitations to 3D OMC [9]. Parallax error caused by the participant performing the movement at a nonperpendicular angle (out of plane) to the camera and perspective error caused by the participant moving toward or away from the camera are additional sources of error when using 2D video motion capture [11, 12].
Electromagnetic motion capture requires the participant to wear a specially designed suit of electromagnetic receiver sensors which receive electromagnetic waves from a base station transmitter located within the vicinity of where the movement is to be performed [8]. The receiver/transmitter network allows the position and orientation of the body to which the receiver sensors are attached to be determined within space [8]. Electromagnetic motion capture systems do not rely on line of sight measurements and thus do not encounter the problems of marker displacement and/or occlusion when implements are included in the motion analysis [8]. Low sampling rates currently make electromagnetic motion capture systems unsuitable for fast movements [8]. Motion capture often takes place at laboratory, clinical, or sporting facilities where equipment in the environment emit electromagnetic disturbance. Electromagnetic motion capture systems are susceptible to electromagnetic interference from the surrounding environment, causing potentially large errors in orientation estimations [8].
While each of these traditional motion capture methodologies have their own advantages and disadvantages, no single method is appropriate for all applications. Recent developments in inertial measurement unit (IMU) and magnetic, angular rate, and gravity (MARG) sensor technologies have resulted in researchers proposing the use of such devices to overcome many of the limitations of traditional motion capture systems, particularly when data needs to be collected outside of a laboratory.
Inertial devices have been used for human motion capture in the areas of athlete external load monitoring [13–15], activity classification [16–20], and spatiotemporal and kinematic analysis [4–6, 21]. The methodology of external load monitoring using inertial devices uses the raw output data of the IMU/MARG device (often accelerations) and thresholding techniques to determine the amount of exposure an athlete may have to various magnitudes of acceleration (external load) over the course of a training session, game/competition, or other relevant period of time such as a week, month, or year [15]. Such data is typically used to provide some insight into athlete performance, training adaptation, fatigue, and risk of injury [15]. Activity classification is used to identify movement patterns such as walking, running, stair ascent/descent, and lying in various positions over an extended period of time (hours or days). Machine learning techniques such as K-nearest neighbour, decision trees, support vector machine, logistic regression, and discriminant analysis are often used to classify these common activities of everyday living [17, 22]. Activity classification can provide clinicians with valuable information about the decline in health or independence of elderly living at home, the activity levels of persons living with conditions or diseases, or the detection of falls or accidents [20].
Inertial-based human spatiotemporal and kinematic analysis requires complex sensor fusion and pose estimation methodologies to process raw MARG data. Numerous studies have demonstrated good agreement when comparing spatiotemporal and kinematic measures derived from IMU and MARG based motion capture systems with gold standard 3D OMC systems in clinical, ergonomic, and sporting applications [4, 23–27]. Similar to traditional motion capture methods, researchers have suggested the accuracy of IMU and MARG based motion capture to be dependent on the algorithms and methodologies used to process the raw data captured by the device [28, 29].
Previous research and reviews have primarily focussed on either the overall validity of inertial-based motion capture (IMC) (excluding methodology considerations) [4, 8, 30, 31], sensor fusion methodologies [32, 33], or position and orientation estimation (pose) methodologies [34–38], making it difficult and time consuming for researchers and developers to piece together all essential methodological components. Two reviews have attempted to summarise the methodological components of IMC; however, these reviews have limited detail around critical considerations such as sensor fusion, pose estimation, soft tissue artifacts (STA), sensor placement, biomechanical modelling, and magnetic calibration, which should be made when developing an IMC solution [39, 40]. The following technical summary is aimed at providing background and reference on all methodological components which must be considered when implementing an IMC solution for a given application (Figure 1). Such a summary will reduce the time spent by researchers and developers establishing the fundamental methodological components of IMC prior to further developing current techniques and enhancing the rate of development and utilisation of IMC.