- Review Article
- Open Access
A Review of the Evolution of Vision-Based Motion Analysis and the Integration of Advanced Computer Vision Methods Towards Developing a Markerless System
Sports Medicine - Open volume 4, Article number: 24 (2018)
The study of human movement within sports biomechanics and rehabilitation settings has made considerable progress over recent decades. However, developing a motion analysis system that collects accurate kinematic data in a timely, unobtrusive and externally valid manner remains an open challenge.
This narrative review considers the evolution of methods for extracting kinematic information from images, observing how technology has progressed from laborious manual approaches to optoelectronic marker-based systems. The motion analysis systems which are currently most widely used in sports biomechanics and rehabilitation do not allow kinematic data to be collected automatically without the attachment of markers, controlled conditions and/or extensive processing times. These limitations can obstruct the routine use of motion capture in normal training or rehabilitation environments, and there is a clear desire for the development of automatic markerless systems. Such technology is emerging, often driven by the needs of the entertainment industry, and utilising many of the latest trends in computer vision and machine learning. However, the accuracy and practicality of these systems has yet to be fully scrutinised, meaning such markerless systems are not currently in widespread use within biomechanics.
This review aims to introduce the key state-of-the-art in markerless motion capture research from computer vision that is likely to have a future impact in biomechanics, while considering the challenges with accuracy and robustness that are yet to be addressed.
Biomechanists aspire to have motion analysis tools that allow movement to be accurately measured automatically and unobtrusively in applied (e.g. everyday training) situations
Innovative markerless techniques developed primarily for entertainment purposes provide a potentially promising solution, with some systems capable of measuring sagittal plane angles to within 2°–3° during walking gait. However, accuracy requirements vary across different scenarios and the validity of markerless systems has yet to be fully established across different movements in varying environments
Further collaborative work between computer vision experts and biomechanists is required to develop such techniques further to meet the unique practical and accuracy requirements of motion analysis for sports and rehabilitation applications.
Vision-based motion analysis involves extracting information from sequential images in order to describe movement. It can be traced back to the late nineteenth century and the pioneering work of Eadweard Muybridge who first developed techniques to capture image sequences of equine gait . Motion analysis has since evolved substantially in parallel with major technological advancements and the increasing demand for faster, more sophisticated techniques to capture movement in a wide range of settings ranging from clinical gait assessment  to video game animation . Within sports biomechanics and rehabilitation applications, quantitative analysis of human body kinematics is a powerful tool that has been used to understand the performance determining aspects of technique , identify injury risk factors , and facilitate recovery from injury  or trauma .
Biomechanical tools have developed considerably from manual annotation of images to marker-based optical trackers, inertial sensor-based systems and markerless systems using sophisticated human body models, computer vision and machine learning algorithms. The aim of this review is to cover some of the history of the development and use of motion analysis methods within sports and biomechanics, highlighting the limitations of existing systems. The state-of-the-art technologies from computer vision and machine learning, which have started to emerge within the biomechanics community, are introduced. This review considers how these new technologies could revolutionise the fields of sports biomechanics and rehabilitation by broadening the applications of motion analysis to include everyday training or competition environments.
General Principles and Requirements of Vision-Based Motion Analysis in Sports Biomechanics and Rehabilitation
Optical motion analysis requires the estimation of the position and orientation (pose) of an object across image sequences. Through the identification of common object features in successive images, displacement data can be “tracked” over time. However, accurate quantification of whole-body pose can be a difficult problem to solve since the human body is an extremely complex, highly articulated, self-occluding and only partially rigid entity [8,9,10]. To make this process more tractable, the structure of the human body is usually simplified as a series of rigid bodies connected by frictionless rotational joints.
Three-dimensional (3D) pose of rigid segments can be fully specified by six degrees of freedom (DOF): three relating to translation and three defining orientation. As such, even for a relatively simple 14-segment human body model, a large number of DOF (potentially as many as 84 depending on the anatomical constraints employed) need to be recovered to completely characterise 3D body configuration. From such a model, it is possible to compute joint angles, and with the incorporation of body segment inertia parameters, the whole body centre of mass location can be deduced, as in previous research in sprinting [11, 12], gymnastics  and rugby place-kicking . Moreover, kinematic and kinetic data can be combined to allow the calculation of joint moments and powers through inverse dynamics analysis . Such analyses have value across diverse areas from, for example, understanding lower-limb joint power production across fatiguing cycling efforts  to characterising joint torque profiles following anterior cruciate ligament reconstruction surgery . However, obtaining accurate body pose is an essential step before reliable joint moments (and powers) can be robustly acquired, as inaccuracies in kinematic data will propagate to larger errors in joint kinetic parameters .
In certain cases within biomechanics, two-dimensional (2D) analyses using relatively simple body models suffice. Examples include when assessing movements which are considered to occur primarily in the sagittal plane such as walking  and sprinting , or when experimental control is limited, such as when ski jumpers’ body positions were analysed during Olympic competition . Conversely, when the movement under analysis occurs in multiple planes, a multi-camera system and a more complex 3D model are required; for instance, when investigating shoulder injury risks associated with different volleyball spiking techniques . The more extensive experimental set-ups for whole-body 3D analysis typically necessitate controlled laboratory environments and a challenge is to then ensure ecological validity (that movements accurately represent reality).
The main differences between 2D and 3D analysis relate to the complexity of the calibration and coordinate reconstruction processes, and joint angle definitions. Planar (2D) analysis can be conducted with only one camera, whereas at least two different perspectives are required to triangulate 2D information into 3D real-space coordinates [22, 23]. The number of DOF that need to be recovered (and consequently the number of markers required) in order to define segment (or any rigid body) pose differs between 2D and 3D methods. In 2D analysis, it is only possible to recover three DOF and this requires a minimum of two known points on the segment. Conversely, for 3D reconstruction of a rigid body, six DOF can be specified by identifying at least three non-collinear points.
A wide range of motion analysis systems allow movement to be captured in a variety of settings, which can broadly be categorised into direct (devices affixed to the body, e.g. accelerometry) and indirect (vision-based, e.g. video or optoelectronic) techniques. Direct methods allow kinematic information to be captured in diverse environments. For example, inertial sensors have been used as tools to provide insight into the execution of various movements (walking gait , discus , dressage  and swimming ). Sensor drift, which influences the accuracy of inertial sensor data, can be reduced during processing; however, this is yet to be fully resolved and capture periods remain limited . Additionally, it has been recognised that motion analysis systems for biomechanical applications should fulfil the following criteria: they should be capable of collecting accurate kinematic information, ideally in a timely manner, without encumbering the performer or influencing their natural movement . As such, indirect techniques can be distinguished as more appropriate in many settings compared with direct methods, as data are captured remotely from the participant imparting minimal interference to their movement. Indirect methods were also the only possible approach for biomechanical analyses previously conducted during sports competition [20, 30,31,32,33]. Over the past few decades, the indirect, vision-based methods available to biomechanists have dramatically progressed towards more accurate, automated systems. However, there is yet to be a tool developed which entirely satisfies the aforementioned important attributes of motion analysis systems.
Historical Progression of Vision-Based Motion Analysis in Sports Biomechanics and Rehabilitation
Manual digitisation was the most widespread motion measurement technique for many decades, and prior to digital technologies, cine film cameras were traditionally used [32, 34,35,36]. These were well-suited to the field of movement analysis due to their high image quality and high-speed frame rates (100 Hz in the aforementioned studies). However, the practicality of this method was limited by long processing times. With the advent of video cameras (initially tape-based before the transition to digital), cine cameras have become essentially redundant in the field of biomechanics. Digital video cameras are now relatively inexpensive, have increasingly high resolutions and fast frame rates (consumer cameras are generally capable of high-definition video at greater than 120 Hz, whereas industrial cameras are significantly faster), and are associated with shorter processing times.
Regardless of the technology used to capture motion, manual digitising requires the manual localisation of several points of interest (typically representing the underlying joint centres) in each sequential image from each camera perspective. Providing a calibration trial has been performed (where several control points of known relative location are digitised in each camera view), the position of the image body points can be reconstructed into real-space coordinates, most commonly via direct linear transformation . Several software packages exist to aid this process and allow accurate localisation of points on rigid structures [37, 38]. Moreover, the repeatability of these methods has been supported by reliability analyses, for example within sprint hurdle  and cricket  research.
One of the primary advantages of manual digitising, which has allowed this method to persist as a means to collect kinematic data, is that the attachment of markers is not necessarily required. As such, manual digitisation remains a valuable tool particularly in sports biomechanics as it allows analysis of movement in normal training [12, 40, 41] and also competition [20, 30,31,32,33] environments without impeding the athlete(s). Additionally, this methodology provides a practical and affordable way of studying gait in applied therapy settings .
Unfortunately, a trade-off exists between accuracy and ecological validity when adopting field-based compared to laboratory-based optical motion analyses . Specifically, the manual digitising approach can be implemented unobtrusively in applied settings with relative ease. However, the resultant 3D vector-based joint angles are difficult to relate to anatomically relevant axes of rotation. Moreover, if angles are projected onto 2D planes (in an attempt to separate the angle into component parts) movement in one plane can be incorrectly measured as movement in another, as discussed in relation to the assessment of elbow extension legality during cricket bowling . Improvements to this early modelling approach are made by digitising external markers, such as medial and lateral condyles, which provide more accurate representations of 3D joint angles. However, certain drawbacks remain including the fact that manual digitising is a notoriously time-consuming and laborious task, and is liable to subjective error. These limitations have provided motivation for the development of automatic solutions made available by the emergence of more sophisticated technologies.
Automatic Marker-Based Systems
A large number of commercial automatic optoelectronic systems now exist for the study of human movement. The majority of these utilise multiple cameras that emit invisible infrared light, and passive markers that reflect this infrared back to the cameras and allow their 3D position to be deduced. Although the specifications of these systems differ markedly , the same underlying principles apply in the sense that several points of interest are located in sequential images, converted to real-space coordinates and used to infer 3D pose of the underlying skeleton. However, the primary difference between methodologies is that optoelectronic systems are capable of automatically locating large numbers of markers, substantially improving the time efficiency of this process. At least three non-collinear markers must be affixed to each segment to specify six DOF. If only two joint markers are used to define a segment, the same challenges exist as those outlined above in relation to manual digitisation. Increasing the number of markers attached to each segment increases the system’s redundancy. However, extensive marker sets encumber the natural movement pattern and tracking marker trajectories can become challenging if markers are clustered close to one another or become occluded .
The accuracies of several widely utilised commercial marker-based systems have been evaluated using a rigid, rotating structure with markers attached at known locations . Root mean square errors were found to be typically less than 2.0 mm for fully visible moving markers and 1.0 mm for a stationary marker (errors were scaled to a standard 3-m long volume) indicating excellent precision when markers are attached to a rigid body. However, the exact placement of markers on anatomical landmarks is difficult to realise and markers placed on the skin do not directly correspond to 3D joint positions. Various protocols exist to locate joint centres and/or define segment pose from markers placed on anatomical landmarks; however, these different conventions produce varying out-of-sagittal plane results when compared over the same gait cycles . In fact, there is also inevitable day-to-day and inter-tester variability in marker placement, which reduces the reliability of marker-based measurements, particularly for transverse plane movements [47, 48].
It is well acknowledged that the rigid body assumption (underlying marker-based motion analysis) can be violated by soft tissue movement, particularly during dynamic activities . This phenomenon has been consistently demonstrated by studies which have compared marker-derived kinematics with those using “gold standard” methods such as fluoroscopy , Roentgen photogrammetric techniques [51, 52] and intra-cortical bone pins [53,54,55]. As soft tissue movement introduces both systematic and random errors of similar frequency to the actual bone movement, this is difficult to attenuate through data smoothing . Over the last two decades, careful design of marker sets  and the use of anatomical calibration procedures  have somewhat alleviated this measurement artefact. For example, initial static calibration trials can be captured whereby joint centres and segment coordinate systems are defined relative to markers. It is then possible to remove certain markers that are liable to movement during dynamic movement, without compromising the deduction of segment pose [59,60,61]. Moreover, the placement of marker clusters (typically three or four non-collinear markers rigidly affixed to a plate) not only provides a practical method to define the segment’s six DOF, but can also be strategically positioned to reduce soft tissue artefact . The development of more sophisticated pose-estimation algorithms  and joint angle definitions  has further advanced the accuracy of marker-based analyses.
Optoelectronic systems are also relatively sensitive to the capture environment. In particular, sunlight, which includes a strong infrared component, can introduce undesirable noise into the measurements. In the past, marker-based analysis has therefore been restricted to indoor conditions (where light conditions could be stringently controlled). However, innovative active filtering features can alleviate these errors and have even allowed data to be captured during outdoor snow sports .
It is, therefore, clear that optoelectronic systems have made significant advancements in recent times within the field of biomechanics. However, even though careful methodological considerations can improve the accuracy of the data acquired, several limitations remain, including long participant preparation times, potential for erroneous marker placement or movement, and the unfeasibility of attaching markers in certain settings (e.g. sports competition). Perhaps one of the most fundamental problems is the physical and/or psychological constraints that attached markers impart on the participant, influencing movement execution. These drawbacks can limit the utility of marker-based systems within certain areas of sports biomechanics and rehabilitation, and have driven the exploration of potential markerless solutions.
Markerless Motion Analysis Systems
An attractive future advancement in motion analysis is towards a fully automatic, non-invasive, markerless approach, which would ultimately provide a major breakthrough for research and practice within sports biomechanics and rehabilitation. For example, motion could be analysed during normal training environments more readily, without the long subject preparation times associated with marker-based systems or the laborious processing required for manual methods. Moreover, it could provide a potential solution for a common dilemma faced by biomechanists, which stems from the trade-off between accuracy (laboratory-based analyses) and external validity (field-based analyses).
Markerless methods are not yet in widespread use within biomechanics, with only a small number of companies providing commercial systems (details for a selection of which are provided in Table 1). However, it remains unclear exactly what precision these systems can achieve in comparison to the other, more established motion analysis systems available on the market. Certainly, the technology is under rapid development with modern computer vision algorithms improving the robustness, flexibility and accuracy of markerless systems. A number of reviews [10, 64,65,66,67] have been previously published detailing these developments, targeting specific application areas such as security, forensics and entertainment. The aim of this section is to introduce the biomechanics community to the current state-of-the-art markerless technology from the field of computer vision, and to discuss where the technology stands in terms of accuracy.
Recent Computer Vision Approaches to Markerless Motion Capture
In marker-based motion capture, cameras and lighting are specially configured to make observation and tracking of markers simple. Where multiple markers are used, individual markers need to be identified, and measurements then taken either directly from the positions of the markers, or from inferring the configuration of a skeleton model that best fits the marker positions. Markerless systems have some similarity to this, with differences mostly induced by the significantly more difficult process of gathering information from the images.
The four major components of a markerless motion capture system are (1) the camera systems that are used, (2) the representation of the human body (the body model), (3) the image features used and (4) the algorithms used to determine the parameters (shape, pose, location) of the body model. The algorithms used to infer body pose given image data are usually categorised as either “generative” (in which model parameters can be used to “generate” a hypothesis that is evaluated against image data and then iteratively refined to determine a best possible fit) or “discriminative” (where image data is used to directly infer model parameters). In general, a markerless motion capture system will have the form shown in Fig. 1. This consists of an offline stage where prior data inform model design or training of a machine learning-based discriminative algorithm, and then image data are captured, processed and input into the algorithms that will estimate body pose and shape.
Camera Systems for Markerless Motion Capture
Two major families of camera systems are used for markerless motion capture differing by whether or not a “depth map” is produced. A depth map is an image where each pixel, instead of describing colour or brightness, describes the distance of a point in space from the camera (Fig. 2). Depth-sensing camera systems range from narrow-baseline binocular-stereo camera systems (such as the PointGrey Bumblebee or the Stereolabs Zed camera) to “active” cameras which sense depth through the projection of light into the observed scene such as Microsoft’s Kinect. Depth information can help alleviate problems that affect traditional camera systems such as shadows, imperfect lighting conditions, reflections and cluttered backgrounds. Active, depth-sensing camera systems (often termed RGB-D cameras where they capture both colour and depth) have proven effective for real-time full-body pose estimation in interactive systems and games [68, 69]. The devices most commonly use one of two technologies: structured light or time-of-flight (ToF). Structured light devices sense depth through the deformations of a known pattern projected onto the scene, while ToF devices measure the time for a pulse of light to return to the camera. The two technologies have different noise characteristics and trade-offs between depth accuracy and spatial resolution . The most well-known active cameras are Microsoft’s original structured light “Kinect”, and the replacement ToF based “Kinect For Xbox One” (often referred to as the Kinect 2), which are provided with body tracking software designed for interactive systems. The performance of this tracking system has been analysed for both versions of the cameras , but clearly falls well below the accuracy required for precision biomechanics (it could be speculated, however, that a tracking system not dedicated to interactive systems may achieve greater accuracy using these devices). Active cameras have been applied to sports biomechanics [72, 73] using bespoke software, but current hardware limitations (effective only over short range, maximum 30 Hz framerates, inoperability in bright sun light, and interference between multiple sensors) are likely to limit their application in sports biomechanics for the foreseeable future.
The body models used by markerless motion capture are generally similar to those used by traditional marker-based approaches. A skeleton is defined as a set of joints and the bones between these joints (Fig. 3). The skeleton is parameterised on the lengths of the bones and the rotation of each joint with pose being described by the joint angles. For discriminative approaches, this skeleton model can be enough, but generative approaches will also require a representation of the person’s volume.
In earlier works, the volume of the model is represented by simple geometric shapes  such as cylinders. Such models remain the state-of-the-art within computer vision in the form of a set of “spatial 3D Gaussians”  attached to the bones of a kinematic skeleton (Fig. 4). The advantage of this representation has been to enable fast, almost real-time fitting in a generative framework using passive cameras and a very simple set of image features.
In general though, the trend has been to use 3D triangle meshes common in graphics and computer games, either created by artists, as the product of high-definition 3D scans , or more recently, by specialising a generic statistical 3D shape model [77,78,79] (Fig. 5). Statistical body shape models allow a wide range of human body shapes to be represented in a relatively small number of parameters and improve how the body surface deforms under joint rotations. However, because these models focus on the external surface appearance of the model, the underlying skeleton is questionably representative of an actual human skeleton and as such, care must be taken should these models be used for biomechanical measurements.
The parameterisation of the human body used by motion capture models is always a simplification, and although capable of a realistic appearance, can also represent physically unrealistic shapes and poses. If the algorithms are not carefully constrained, these solutions can appear optimal for the available data. To ensure only physically realistic solutions are produced, the algorithms must be supported by constraints on the body model such as explicit joint limits  or probabilistic spaces of human pose and motion deduced by machine learning . In either case, there exists a balance between enforcing the constraints and trusting the observed data to achieve a solution that is both plausible and precise.
Image Features for Markerless Motion Capture
A digital image, fundamentally, is a 2D grid of numbers each representing the brightness and colour of a small region, or pixel. The process of determining how pixels relate to objects is a fundamental task in computer vision, and there have been many proposed approaches to extracting “features” from an image that are meaningful. It is the great difficulty of this task that marker-based systems have been developed to avoid.
For motion capture, the primary aim is to determine the location and extents in the image of the person being captured. The earliest and one of the most robust approaches to this task is termed chroma keying. This is where the background of the scene is painted a single specific colour, allowing the silhouette of the person (who is dressed in a suitable contrasting colour) to be easily segmented (Fig. 6). For environments where chroma keying is not possible, there are a large number  of background subtraction algorithms. However, these can all suffer from problems with shadows, lighting changes, reflections and non-salient motions of the background (such as a crowd or other athletes).
Image silhouettes are also inherently ambiguous and provide no information on whether the observed subject is facing towards or away from the camera (Fig. 6). This ambiguity can only be reduced by the use of extra cameras or more sophisticated image features. Where large numbers of cameras are available, silhouettes can be combined into a 3D representation known as a visual hull , which is an approximation of the space occupied by the observed person (Fig. 7). More sophisticated 3D reconstructions can also be carried out ; however, any added accuracy must be traded off against increased computational complexity. Improving the reconstruction does not fully resolve all fitting difficulties, however, and extra information that identifies which regions of the silhouettes correspond to which regions of the body are often needed to completely resolve all possible confusion . Nevertheless, silhouettes have formed a key aspect of many markerless motion capture works including the work of Corazza et al. , which has reported some of the most accurate results for automatic markerless body motion capture, and Liu et al. , which enables the kinematic motion analysis of multiple persons. The trend, however, has been to move away from the use of image silhouettes to improve robustness, reduce ambiguities, reduce the number of cameras and simplify capture procedures. In this regard, the work of Stoll et al.  is significant for enabling the body model to fit to the image using only a simple colour model, while the advent of deep learning  and its provision of robust and fast body-part detectors has made dramatic improvements on what can be done outside of laboratory conditions [89, 90], including recognising body pose of many people from a single uncalibrated and moving camera .
In generative motion capture approaches, the pose and shape of the person is determined by fitting the body model to information extracted from the image. For a given set of model parameters (body shape, bone lengths, joint angles), a representation of the model is generated. This representation can then be compared against the features extracted from the image and a single “error value” calculated, which represents how much the hypothesis differs from the observed data. In one possibility, the 3D triangle mesh resulting from the predicted parameters can be projected into the 2D image, and the overlap of the mesh and the silhouette of the person can be maximised . Alternatively, the 3D body model can be compared against a 3D reconstruction such as a visual hull by minimising the distances between the 3D vertices of the model, and the 3D points of the visual hull [86, 93] through a standard algorithm known as iterative closest point.
A key factor of generative approaches is the appropriate definition of the function that compares a specific hypothesis with the information available in the images. If this is not carefully considered, then the search for the optimal set of model parameters can easily fail, resulting in poor estimates or nonsense configurations where joints bend at unrealistic angles and limbs penetrate inside the body. Constructing a cost function that is robust to image noise and to unrealistic model configurations is difficult meaning generative models often need a reliable initial guess of the model parameters. In extremes this would mean forcing the person being captured to assume a specific pose at the start of tracking. If the fitting then becomes confused by occlusions, image noise or other failure, tracking will not be able to correct itself without manual intervention. Researchers have attempted to address this situation using improved searching algorithms , extra information derived from robust body part detectors  and recent pose-recognition algorithms [94,95,96,97], or by coupling generative methods with discriminative methods .
Discriminative algorithms avoid the process of iteratively tuning the parameters of a body model to fit the image and as such they are also often referred to as model-free algorithms. Compared with generative approaches, they will often have a much faster processing time, improved robustness and reduced dependence on an initial guess. However, they can have reduced precision, and they require a very large database of exemplar data (far more than is required even for constructing the statistical body shape models used by generative algorithms) from which they can learn how to infer a result.
Discriminative approaches have two major families. One approach is to discover a mapping directly from image features to a description of pose, such as by using machine learning-based regression [99, 100]. In this way, it is possible to “teach” the computer how to determine the pose of a simple skeleton model using only the image data. The most recent approaches in this family use deep learning to train a system that can identify the body parts of multiple people, the probable ownership of joints, and then quickly parse this to determine skeletons . Alternatively, a database of pose examples can be created and then searched to discover the most similar known pose given the current image, as used in previous studies [101,102,103].
The main difficulty in the use of discriminative algorithms is the creation of the exemplar data. If the available data is insufficient, then poses, physiques and even camera positions that were not suitably represented will lead to false results as the system will not be able to generalise from what it “knows” to what it “sees”. This will also affect the precision of the result because the algorithm is restricted to giving solutions close to what it “knows” about, so small variations may not be fully represented in the results. As a result, discriminative approaches are used as initial guesses for generative approaches .
Summary of Markerless Approaches
The current state-of-the-art shows the computer vision community aiming to develop solutions to markerless motion capture that are applicable and reliable outside of laboratory conditions. Although carefully calibrated silhouette-based algorithms using sophisticated subject-specific body models have shown the most accurate results to date, they have been limited to laboratory conditions using a large number of cameras . By taking advantage of modern technologies such as improved solvers , advanced image features and modern machine learning , recent works are providing solutions that reduce the required number of cameras , allow moving cameras , increase the number of people that can be tracked and provide robust detection and fitting in varied environments . The ability to do this without knowledge of camera calibration further improves the potential ease of use of future systems; however, calibration is likely to remain a necessity where precise measurement is needed such as in biomechanics.
Accuracy of Current Markerless Motion Capture Systems
There are distinct differences in the accuracy requirements between motion analysis techniques in the fields of computer vision and biomechanics, which must be taken into account when attempting to apply computer vision methods more broadly across other disciplines. For instance, accuracy within computer vision (primarily entertainment applications) is typically assessed qualitatively and is primarily evaluated based on appearance. Conversely, in biomechanical settings, it is fundamental that any motion analysis system is capable of robustly quantifying subtle differences in motion, which could be meaningful from a musculoskeletal performance or pathology perspective. Nonetheless, there is no general consensus regarding the minimum accuracy requirement of motion analysis systems and the magnitudes of the inevitable measurement errors will vary depending on the context (laboratory vs. field), the characteristics of the movement and the participant, the experimental setup and how the human body is modelled.
As previously described, marker-based approaches are currently the most widely used systems in biomechanical laboratories. However, a prominent source of measurement error in marker-based systems is skin movement artefact , which violates the rigid body assumption underlying these methods. Reports suggest that errors due to soft tissue movement can exceed 10 mm for some anatomical landmarks and 10° for some joint angles when compared to more precise, yet invasive, methods (e.g. intra-cortical bone pins) . However, these errors in joint angle measurements may be reduced to 2°–4° by using more sophisticated pose-estimation algorithms such as global optimisation (inverse kinematics) with joint constraints . As the “gold standard” methods are inappropriate in many contexts, and marker-based systems are the most frequently utilised motion analysis technique in the field, agreement between markerless and optoelectronic systems would be considered to provide evidence for the validity of markerless motion analysis techniques.
There are already studies which have attempted to evaluate the accuracy of markerless systems (summarised in Table 2) by comparing the kinematic output variables against those obtained using marker-based optoelectronic systems [10, 86, 93, 107,108,109] or manual digitisation . These validations mostly study relatively slow movements (typically walking gait), whereas to verify the utility of these approaches in sports applications, much quicker movements need to be thoroughly assessed. One clear observation from these results is that transverse plane rotations are currently difficult to extract accurately and reliably by markerless technologies [107, 110].
In the computer vision community, it is common practice to advance technology by establishing benchmark datasets against which many authors can rank their algorithm’s performance. Two such benchmarks are the widely used HumanEva dataset  and the more recent Human 3.6M dataset . These datasets provide video of people performing actions (walking, jogging, boxing etc.) while also being tracked with marker-based tracking systems. Table 3 shows a sample of published comparison results for the HumanEva dataset. These results show that precision of markerless techniques remains too low to be applicable for biomechanics analyses. However, the video data and motion capture data in the HumanEva dataset are themselves of limited quality. For example, videos are low resolution and camera placement is sub-optimal, while markers are limited and sub-optimally located (often on relatively loose clothing) with no marker clusters to aid tracking (Fig. 8). For comparison, the results of Corazza et al.  on HumanEva have a mean joint centre position error of 79 ± 12 mm, while on the authors’ own higher resolution data with better camera and marker placement, a much smaller 15 ± 10 mm error was achieved.
The discrepancies observed by Corazza et al.  between the validation results against the HumanEva benchmark and the more rigorously captured marker-based data show the difficulties of treating marker-based motion capture as the criterion method. In fact, although these benchmarks are useful for showing the general performance of different algorithms, neither the marker-based nor manual digitising methods used to validate markerless technologies can provide exact “true” body pose due to experimental artefacts that are inevitably introduced. Additionally, ensuring a close match between the body model applied to both systems is a challenge, which may necessitate “off-line” stages perhaps involving imaging, as in previous work . Adding markers to the validation images might also unduly bias the performance of a markerless system under test, if the algorithm detects the markers and uses them for its benefit or if the markers adversely affect the performance of the markerless algorithm (altering silhouette shapes, for example) . As such, alternative methods of validating the performance of markerless systems have also been considered, such as utilising force plate data to analyse centre of mass movement  and creating virtual environments (synthesised images) in which a predefined model moves with known kinematics . Although synthesised images can be invaluable for developing an algorithm (synthetic images were used for generating training images for Microsoft’s Kinect pose tracker ), the idealised image data is unlikely to capture the noise and error sources of real imagery.
Future of Markerless Approaches to Analyse Motion in Sports Biomechanics and Rehabilitation
It is clear that a broad range of markerless technologies have emerged from computer vision research over recent times, which have the potential to be applied across diverse disciplines and settings. The priorities and requirements for a markerless motion capture system will depend on the research area and the unique capture environment, and are thus non-uniform across disciplines. In sports biomechanics and rehabilitation applications, motion analysis systems must be highly accurate in order to detect subtle changes in motion, as well as being adaptable, non-invasive and unencumbering. With these system requirements in mind, the current progression of technologies suggests that the future of practical markerless motion capture will lie with techniques such as those presented by Elhayek et al. , which fuse together a discriminative approach (to get good initialisation and robustness) with a robust silhouette-free kinematic model fitting approach for precision.
A fast, approximate pose estimation system has previously been combined with a slow, more-accurate technique in order to provide basic parameters to athletics coaches and inform training in real-time . This type of system may have utility in the applied field by allowing some of the primary, “top-level” biomechanical determinants (for example, step frequency and step length in gait) to be fed-back during normal training or rehabilitation situations. Importantly, the more complex and computationally expensive kinematic variables (such as 3D joint angles, which require modelling of the body) may still be acquired. However, the likelihood is that more time-consuming, offline processing will be necessary. This two-part approach could help address the apparent disconnect between sports science research and practice  as short participant preparation times and timely feedback from the system may increase the perceived (and actual) value of such studies to those operating in the applied field. Importantly, the more complex kinematic information could still be computed and communicated to applied practitioners across longer time frames, but equally these data can be used in research studies to continually progress our scientific understanding of human movement.
It should be noted that resolution (both spatial and temporal) will affect the accuracy of markerless systems in the same way as it does for marker-based systems. However, video-based automatic systems must also consider the fact that the size of the data captured will be considerably larger and thus, markerless systems may need to compromise accuracy to make a deployable, fast system feasible. Such a system requires large amounts of video data to be handled efficiently and effectively, which is likely to necessitate the purchase of an expensive (perhaps specially engineered) video-based system (e.g. machine vision).
Vision-based motion analysis methods within sports and rehabilitation applications have evolved substantially over recent times and have allowed biomechanical research to contribute a vast amount of meaningful information to these fields. However, the most widespread kinematic data capture techniques (marker-based technologies and manual digitisation) are not without their drawbacks. Considerable developments in computer vision have sparked interest in markerless motion analysis and its possible wider applications. Although this potential is promising, it is not yet clear exactly what accuracy can be achieved and whether such systems can be effectively and routinely utilised in field-based (more externally valid) settings. Over the coming years, collaborative research between computer vision experts and biomechanists is required to further develop markerless techniques so they are able to meet the unique practical and accuracy requirements of motion analysis within sports and rehabilitation contexts.
Degrees of freedom
Root mean square
Skinned Multi-Person Linear Model
Muybridge E. Complete human and animal locomotion (all 781 plates from the 1887 animal locomotion). In: Cappozzo A, Marchetti M, Tosi V, editors. Biolocomotion: a century of research using moving pictures. Rome: Promograph; 1979. p. 69.
Cimolin V, Galli M. Summary measures for clinical gait analysis: a literature review. Gait Posture. 2014;39(4):1005–10.
Cosker D, Eisert P, Helzle V. Facial capture and animation in visual effects. In: Magnor MA, Grau O, Sorkine-Hornung O, Theobalt C, editors. Digital representations of the real world: how to capture, model, and render visual reality. Boca Raton, FL: CRC Press; 2015. p. 311–21.
Bezodis NE, Salo AIT, Trewartha G. Relationships between lower-limb kinematics and block phase performance in a cross section of sprinters. Eur J Sport Sci. 2015;15(2):118–24.
Ford KR, Myer GD, Toms HE, Hewett TE. Gender differences in the kinematics of unanticipated cutting in young athletes. Med Sci Sports Exerc. 2005;37(1):124–9.
Devita P, Hortobagyi T, Barrier J. Gait biomechanics are not normal after anterior cruciate ligament reconstruction and accelerated rehabilitation. Med Sci Sports Exerc. 1998;30(10):1481–8.
Sjödahl C, Jarnlo G-B, Söderberg B, Persson BM. Kinematic and kinetic gait analysis in the sagittal plane of trans-femoral amputees before and after special gait re-education. Prosthetics Orthot Int. 2002;26(2):101–12.
O'Rourke J, Badler NI. Model-based image analysis of human motion using constraint propagation. IEEE Trans Pattern Anal Mach Intell. 1980;2(6):522–36.
Leung MK, Yang Y-H. First sight: a human body outline labeling system. IEEE Trans Pattern Anal Mach Intell. 1995;17(4):359–77.
Mündermann L, Corazza S, Andriacchi TP. The evolution of methods for the capture of human movement leading to markerless motion capture for biomechanical applications. J Neuroeng Rehabil. 2006; https://doi.org/10.1186/1743-0003-3-6
Bezodis IN, Kerwin DG, Salo AIT. Lower-limb mechanics during the support phase of maximum-velocity sprint running. Med Sci Sports Exerc. 2008;40(4):707–15.
Churchill SM, Salo AIT, Trewartha G. The effect of the bend on technique and performance during maximal effort sprinting. Sports Biomechanics. 2015;14(1):106–21.
Hiley MJ, Yeadon MR. Achieving consistent performance in a complex whole body movement: the Tkatchev on high bar. Hum Mov Sci. 2012;31(4):834–43.
Bezodis NE, Trewartha G, Wilson C, Irwin G. Contributions of the non-kicking-side arm to rugby place-kicking technique. Sports Biomechanics. 2007;6:171–86.
Winter DA. Biomechanics and motor control of human movement. 2nd ed. New York: Wiley; 1990.
Martin JC, Brown NAT. Joint-specific power production and fatigue during maximal cycling. J Biomech. 2009;42(4):474–9.
Devita P, Hortobagyi T, Barrier J, Torry M, Glover KL, Speroni DL, et al. Gait adaptation before and after anterior cruciate ligament reconstruction surgery. Med Sci Sports Exerc. 1997;29(7):853–9.
Camomilla V, Cereatti A, Cutti AG, Fantozzi S, Stagni R, Vannozzi G. Methodological factors affecting joint moments estimation in clinical gait analysis: a systematic review. Biomed Eng Online. 2017;16(106); https://doi.org/10.1186/s12938-017-0396-x.
Matsas A, Taylor N, McBurney H. Knee joint kinematics from familiarised treadmill walking can be generalised to overground walking in young unimpaired subjects. Gait Posture. 2000;11(1):46–53.
Schmölzer B, Müller W. Individual flight styles in ski jumping: results obtained during Olympic Games competitions. J Biomech. 2005;38(5):1055–65.
Seminati E, Marzari A, Vacondio O, Minetti AE. Shoulder 3D range of motion and humerus rotation in two volleyball spike techniques: injury prevention and performance. Sports Biomechanics. 2015;14(2):216–31.
Zhang Z. Camera calibration. In: Ikeuchi K, editor. Computer vision: a reference guide. Boston: Springer US; 2014. p. 76–7.
Abdel-Aziz YI, Karara HM, Hauck M. Direct linear transformation from comparator coordinates into object space coordinates in close-range photogrammetry. Photogramm Eng Remote Sens. 2015;81(2):103–7.
Mayagoitia RE, Nene AV, Veltink PH. Accelerometer and rate gyroscope measurement of kinematics: an inexpensive alternative to optical motion analysis systems. J Biomech. 2002;35(4):537–42.
Ganter N, Krüger A, Gohla M, Witte K, Edelmann-Nusser J. Applicability of a full body inertial measurement system for kinematic analysis of the discus throw. Michigan: Proceedings of the 28th conference of the International Society of Biomechanics in Sports; 2010.
Eckardt F, Münz A, Witte K. Application of a full body inertial measurement system in dressage riding. J Equine Vet Sci. 2014;34:1294–9.
de Magalhaes FA, Vannozzi G, Gatta G, Fantozzi S. Wearable inertial sensors in swimming motion analysis: a systematic review. J Sports Sci. 2015;33(7):732–45.
Dadashi F, Crettenand F, Millet GP, Aminian K. Front-crawl instantaneous velocity estimation using a wearable inertial measurement unit. Sensors. 2012;12(10):12927–39.
Atha J. Current techniques for measuring motion. Appl Ergon. 1984;15(4):245–57.
Salo AIT, Grimshaw PN, Marar L. 3-D biomechanical analysis of sprint hurdles at different competitive levels. Med Sci Sports Exerc. 1997;29(2):231–7.
Manning ML, Irwin G, Gittoes MJR, Kerwin DG. Influence of longswing technique on the kinematics and key release parameters of the straddle Tkachev on uneven bars. Sports Biomechanics. 2011;10(3):161–73.
Lees A, Graham-Smith P, Fowler N. A biomechanical analysis of the last stride, touchdown, and takeoff characteristics of the men’s long jump. J Appl Biomech. 1994;10:61–78.
Portus MR, Rosemond CD, Rath DA. Cricket: fast bowling arm actions and the illegal delivery law in me’s high performance cricket matches. Sports Biomechanics. 2006;5(2):215–30.
Procter P, Paul JP. Ankle joint biomechanics. J Biomech. 1982;15(9):627–34.
Ericson M. On the biomechanics of cycling. A study of joint and muscle load during exercise on the bicycle ergometer. Scand J Rehabil Med. 1986;16:1–43.
Bobbert MF, Huijing PA, van Ingen Schenau GJ. A model of the human triceps surae muscle-tendon complex applied to jumping. J Biomech. 1986;19(11):887–98.
Wilson DJ, Smith BK, Gibson JK, Choe BK, Gaba BC, Voelz JT. Accuracy of digitization using automated and manual methods. Phys Ther. 1999;79(6):558–66.
Scholz JP, Millford JP. Accuracy and precision of the PEAK performance technologies motion measurement system. J Mot Behav. 1993;25(1):2–7.
Salo AIT, Grimshaw PN. An examination of kinematic variability of motion analysis in sprint hurdles. J Appl Biomech. 1998;14:211–22.
Bezodis I, Salo A, Kerwin D. A longitudinal case study of step characteristics in a world class sprint athlete. In: Kwon Y-H, Shim J, Shim JK, Shin I-S, editors. . Seoul: Proceedings of 26th international conference on biomechanics in Sports; 2008.
Bezodis NE, Salo AIT, Trewartha G. Choice of sprint start performance measure affects the performance-based ranking within a group of sprinters: which is the most appropriate measure? Sports Biomechanics. 2010;9(4):258–69.
Churchill AJG, Halligan PW, Wade DT. RIVCAM: a simple video-based kinematic analysis for clinical disorders of gait. Comput Methods Prog Biomed. 2002;69(3):197–209.
Elliott B, Alderson J. Laboratory versus field testing in cricket bowling: a review of current and past practice in modelling techniques. Sports Biomechanics. 2007;6(1):99–108.
Elliott BC, Alderson JA, Denver ER. System and modelling errors in motion analysis: implications for the measurement of the elbow angle in cricket bowling. J Biomech. 2007;40(12):2679–85.
Richards JG. The measurement of human motion: a comparison of commercially available systems. Hum Mov Sci. 1999;18(5):589–602.
Ferrari A, Benedetti MG, Pavan E, Frigo C, Bettinelli D, Rabuffetti M, et al. Quantitative comparison of five current protocols in gait analysis. Gait Posture. 2008;28:207–16.
Growney E, Meglan D, Johnson M, Cahalan T, An K-N. Repeated measures of adult normal walking using a video tracking system. Gait Posture. 1997;6(2):147–62.
Tsushima H, Morris ME, McGinley J. Test-retest reliability and inter-tester reliability of kinematic data from a three-dimensional gait analysis system. J Japan Phys Ther Assoc. 2003;6(1):9–17.
Cappozzo A, Catani F, Leardini A, Benedetti MG, Della Croce U. Position and orientation in space of bones during movement: experimental artefacts. Clin Biomech. 1996;11(2):90–100.
Stagni R, Fantozzi S, Cappello A, Leardini A. Quantification of soft tissue artefact in motion analysis by combining 3D fluoroscopy and stereophotogrammetry: a study on two subjects. Clin Biomech. 2005;20(3):320–9.
Maslen BA, Ackland TR. Radiographic study of skin displacement errors in the foot and ankle during standing. Clin Biomech. 1994;9:291–6.
Tranberg R, Karlsson D. The relative skin movement of the foot: a 2D roentgen photogrammetry study. Clin Biomech. 1998;13(1):71–6.
Fuller J, Liu L-J, Mann RW. A comparison of lower-extremity skeletal kinematics measured using skin- and pin-mounted markers. Hum Mov Sci. 1997;16(2–3):219–42.
Benoit DL, Ramsey DM, Lamontagne M, Xu L, Wretenberg P, Renström P. Effect of skin movement artifact on knee kinematics during gait and cutting motions measured in vivo. Gait Posture. 2006;24(2):152–64.
Reinschmidt C, van den Bogert AJ, Nigg BM, Lundberg A, Murphy N. Effect of skin movement on the analysis of skeletal knee joint motion during running. J Biomech. 1997;30(7):729–32.
Leardini A, Chiari L, Della Croce U, Cappozzo A. Human movement analysis using stereophotogrammetry: part 3. Soft tissue artifact assessment and compensation. Gait Posture. 2005;21(2):212–25.
Manal K, McClay I, Stanhope S, Richards J, Galinat B. Comparison of surface mounted markers and attachment methods in estimating tibial rotations during walking: an in vivo study. Gait Posture. 2000;11(1):38–45.
Cappozzo A, Catani F, Della Croce U, Leardini A. Position and orientation in space of bones during movement: anatomical frame definition and determination. Clin Biomech. 1995;10(4):171–8.
Milner CE, Ferber R, Pollard CD, Hamill J, Davis IS. Biomechanical factors associated with tibial stress fracture in female runners. Med Sci Sports Exerc. 2006;38(2):323–8.
Chmielewski TL, Rudolph KS, Fitzgerald GK, Axe MJ, Snyder-Mackler L. Biomechanical evidence supporting a differential response to acute ACL injury. Clin Biomech. 2001;16(7):586–91.
Snyder KR, Earl JE, O'Connor KM, Ebersole KT. Resistance training is accompanied by increases in hip strength and changes in lower extremity biomechanics during running. Clin Biomech. 2009;24(1):26–34.
Lu TW, O'Connor JJ. Bone position estimation for skin marker co-ordinates using global optimisation with joint constraints. J Biomech. 1999;32(2):129–34.
Nedergaard NJ, Heinen F, Sloth S, Hébert-Losier K, Holmberg H-C, Kersting UG. The effect of light reflections from the snow on kinematic data collected using stereo-photogrammetry with passive markers. Sports Eng. 2014;17:97–102.
Moeslund TB, Hilton A, Krüger V. A survey of advances in vision-based human motion capture and analysis. Comput Vis Image Underst. 2006;104:90–126.
Poppe R. Vision-based human motion analysis: an overview. Comput Vis Image Underst. 2007;108:4–18.
Holte MB, Tran C, Trivedi MM, Moeslund TB. Human pose estimation and activity recognition from multi-view videos: comparative explorations of recent developments. Select Topics Signal Process. 2012;6:538–52.
Yang SXM, Christiansen MS, Larsen PK, Alkjær T, Moeslund TB, Simonsen EB, et al. Markerless motion capture systems for tracking of persons in forensic biomechanics: an overview. Comput Methods Biomechanics Biomed Eng: Imaging Visual. 2014;2(1):46–65.
Shotton J, Fitzgibbon A, Blake A, Kipman A, Finocchio M, Moore R, et al. Real-time human pose recognition in parts from a single depth image. Colorado: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition; 2011.
Ye M, Zhang Q, Wang L, Zhu J, Yang R, Gall JA. Survey on human motion analysis from depth data. In: Grzegorzek M, Theobalt C, Koch R, Kolb A, editors. Time-of-flight and depth imaging sensors, algorithms, and applications. Berlin, Heidelberg: Springer; 2013. p. 149–87.
Sarbolandi H, Lefloch D, Kolb A. Kinect range sensing: structured-light versus time-of-flight kinect. J Comput Vision Image Understand. 2015;139:1–20.
Wang Q, Kurillo G, Ofli F, Bajcsy R. Evaluation of pose tracking accuracy in the first and second generations of Microsoft Kinect. IEEE Int Confer Healthc Inform (ICHI). 2015:380–9.
Choppin S, Wheat J. The potential of the Microsoft Kinect in sports analysis and biomechanics. Sports Technol. 2013;6(2):78–85.
Schmitz A, Ye M, Boggess G, Shapiro R, Yang R, Noehren B. The measurement of in vivo joint angles during a squat using a single camera markerless motion capture system as compared to a marker based system. Gait Posture. 2015;41:694–8.
Gavrila DM, Davis LS. 3-D model-based tracking of humans in action: a multi-view approach. San Juan, Puerto Rico: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition; 1996.
Stoll C, Hasler N, Gall J, Seidel HP, Theobalt C. Fast articulated motion tracking using a sums of Gaussians body model. Barcelona: Proceedings of the international conference on computer vision; 2011.
El-Sallam A, Bennamoun M, Sohel F, Alderson J, Lyttle A, Rossi R. A low cost 3D markerless system for the reconstruction of athletic techniques. Tampa: IEEE Workshop on Applications of Computer Vision; 2013.
Allen B, Curless B, Popović Z. The space of human body shapes: reconstruction and parameterization from range scans. ACM Trans Graph. 2003;22(3):587–94.
Anguelov D, Srinivasan P, Koller D, Thrun S, Rodgers J, Davis J. SCAPE: shape completion and animation of people. ACM Trans Graph. 2005;24(3):408–16.
Loper M, Mahmood N, Romero J, Pons-Moll G, Black MJ. SMPL: a skinned multi-person linear model. Proc of SIGGRAPH Asia. 2015;34(6):1–16.
Akhter IB, Black MJ. Pose-conditioned joint angle limits for 3D human pose reconstruction. Boston: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition; 2015.
Holden D, Saito J, Komura T, Joyce T. Learning motion manifolds with convolutional autoencoders. SIGGRAPH Asia Tech Briefs. 2015;18; https://doi.org/10.1145/2820903.2820918.
Bouwmans T. Traditional and recent approaches in background modeling for foreground detection: an overview. Comput Sci Rev. 2014;11-12:31–66.
Laurentini A. The visual hull concept for silhouette-based image understanding. IEEE Trans Pattern Anal Mach Intell. 1994;16(2):150–62.
Furukawa Y, Hernández C. Multi-view stereo: a tutorial. Found Trends Comput Graphics Vision. 2015;9(1–2):1–148.
Kanaujia A, Kittens N, Ramanathan N. Part segmentation of visual hull for 3D human pose estimation. Oregon: IEEE Conference on Computer Vision and Pattern Recognition Workshops; 2013.
Corazza S, Mündermann L, Gambaretto E, Ferrigno G, Andriacchi TP. Markerless motion capture through visual hull, articulated ICP and subject specific model generation. Int J Comput Vis. 2010;87:156–69.
Liu Y, Gall J, Stoll C, Dai Q, Seidel HP, Theobalt C. Markerless motion capture of multiple characters using multiview image segmentation. IEEE Trans Pattern Anal Mach Intell. 2013;35(11):2720–35.
LeCun Y, Bengio Y, Hinton G. Deep learning. Nature. 2015;521(7553):436–44.
Bogo F, Kanazawa A, Lassner C, Gehler P, Romero J, Black MJ. Keep it SMPL: automatic estimation of 3D human pose and shape from a single image. Amsterdam: Proceedings of the European Conference on Computer Vision; 2016.
Elhayek A, de Aguiar E, Jain A, Tompson J, Pishchulin L, Andriluka M, et al. Efficient ConvNet-based marker-less motion capture in general scenes with a low number of cameras. Boston: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition; 2015.
Cao Z, Simon T, Wei SE, Sheikh Y. Realtime multi-person 2D pose estimation using part affinity fields. Comput Vision Pattern Recogn. 2016; https://doi.org/10.1109/CVPR.2017.143.
Saini S, Zakaria N, Rambli DRA, Sulaiman S. Markerless human motion tracking using hierarchical multi-swarm cooperative particle swarm optimization. PLoS One. 2015;10(5):1–22.
Corazza S, Mündermann L, Chaudhari AM, Demattio T, Cobelli C, Andriacchi TP. A markerless motion capture system to study musculoskeletal biomechanics: visual hull and simulated annealing approach. Ann Biomed Eng. 2006;34(6):1019–29.
Amin S, Andriluka M, Rohrback M, Schiele B. Multi-view pictorial structures for 3D human pose estimation. British Machine Vision Confer. 2013; https://doi.org/10.5244/C.27.45.
Belagiannis V, Amin S, Andriluka M, Schiele B, Navab N, Ilic S. 3D pictorial structures for multiple human pose estimation. Ohio: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition; 2014.
Burenius M, Sullivan J, Carlsson S. 3D pictorial structures for multiple view articulated pose estimation. Oregon: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition; 2013.
Kazemi V, Burenius M, Azizipour H, Sullivan J. Multi-view body part recognition with random forests. Br Machine Vision Confer. 2013; https://doi.org/10.5244/C.27.48.
Rhodin H, Robertini N, Casas D, Richardt C, Seidel H-P, Theobalt C. General automatic human shape and motion capture using volumetric contour cues. In: Leibe B, Matas J, Sebe N, Welling M, editors. . Amsterdam: Proceedings of the European Conference on Computer Vision; 2016.
Agarwal A, Triggs B. Recovering 3D human pose from monocular images. Pattern Anal Machine Intelligence. 2006;28:44–58.
Hong C, Yu K, Xie Y, Chen X. Multi-view deep learning for image-based pose recovery. Hangzhou: Proceedings of the IEEE 16th International Conference on Communication Technology; 2015.
Chen C, Yang Y, Nie F, Odobez J-M. 3D human pose recovery from image by efficient visual feature selection. Comput Vis Image Underst. 2011;115(3):290–9.
Babagholami-Mohamadabadi B, Jourabloo A, Zarghami A, Kasaei S. A Bayesian framework for sparse representation-based 3-D human pose estimation. IEEE Signal Proc Lett. 2014;21(3):297–300.
Guo Y, Chen Z, Yu J. Supervised spectral embedding for human pose estimation. Suzhou: Proceedings for the 5th International Conference of Intelligence Science and Big Data Engineering Image and Video Data Engineering; 2015.
Mehta D, Sridhar S, Sotnychenko O, Rhodin H, Shafiei M, Seidel H-P, et al. VNect: real-time 3D human pose estimation with a single RGB camera. ACM Trans Graph. 2017;36(4): https://doi.org/10.1145/3072959.3073596.
Elhayek A, Stoll C, Kim KI, Theobalt C. Outdoor human motion capture by simultaneous optimization of pose and camera parameters. Computer Graphics Forum. 2015;34(6):86–98.
Peters A, Galna B, Sangeux M, Morris M, Baker R. Quantification of soft tissue artifact in lower limb human motion analysis: a systematic review. Gait Posture. 2010;31:1–8.
Ceseracciu E, Sawacha Z, Cobelli C. Comparison of markerless and marker-based motion capture technologies through simultaneous data collection during gait: proof of concept. PLoS One. 2014;9(3):e87640. https://doi.org/10.1371/journal.pone.0087640
Sandau M, Koblauch H, Moeslund TB, Aanæs H, Alkjær T, Simonsen EB. Markerless motion capture can provide reliable 3D gait kinematics in the sagittal and frontal plane. Med Eng Phys. 2014;36(9):1168–75.
Ong A, Harris IS, Hamill J. The efficacy of a video-based marker-less tracking system for gait analysis. Comput Methods Biomechanics Biomed Eng. 2017;20(10):1089–95.
Trewartha G, Yeadon MR, Knight JP. Marker-free tracking of aerial movements. In: Müller R, Gerber H, Stacoff A, editors. . Zürich: Proceedings of the 18th congress of the international society of Biomechanics; 2001.
Sigal L, Balan AO, Black MJ. HumanEva: synchronized video and motion capture dataset and baseline algorithm for evaluation of articulated human motion. Int J Comput Vis. 2010;87:4–27.
Ionescu C, Papava D, Olaru V, Sminchisescu C. Human3.6M: large scale datasets and predictive methods for 3D human sensing in natural environments. IEEE Trans Pattern Anal Mach Intell. 2014;36(7):1325–39.
Corazza S, Andriacchi TP. Posturographic analysis through markerless motion capture without ground reaction forces measurement. J Biomech. 2009;42(3):370–4.
Bishop D. An applied research model for the sport sciences. Sports Med. 2008;38(3):253–63.
This review was funded by CAMERA, the RCUK Centre for the Analysis of Motion, Entertainment Research and Applications, EP/M023281/1.
SC has a PhD and is a Post-Doctoral Research Associate in the CAMERA project (see funding above) with expertise in analysing athletes’ technique. ME has a PhD and is a Post-Doctoral Research Associate in the CAMERA project with expertise in computer vision with time spent in the computer industry before joining back to academia. DC has a PhD and is a Professor in Computer Science specialising in computer vision. He is the principal investigator and the director of the CAMERA project. AS has a PhD and is a Reader (Associate Professor) in Sports Biomechanics with expertise in analysing athletes’ technique. He is a co-investigator in the CAMERA project.
Ethics Approval and Consent to Participate
The authors Steffi L. Colyer, Murray Evans, Darren P. Cosker and Aki I. T. Salo declare that they have no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
About this article
Cite this article
Colyer, S.L., Evans, M., Cosker, D.P. et al. A Review of the Evolution of Vision-Based Motion Analysis and the Integration of Advanced Computer Vision Methods Towards Developing a Markerless System. Sports Med - Open 4, 24 (2018). https://doi.org/10.1186/s40798-018-0139-y
- Automatic analysis
- Body model
- Discriminative approaches
- Generative algorithms
- Motion capture
- Sports biomechanics