The development of a robust and reliable 3D modelling approach to accurately represent human motion, especially highly dynamic sports motion, has spanned decades and seen multiple research teams across the globe address some of the key challenges. From a kinematic modelling perspective, initial research efforts were focused on addressing two primary challenges:

• the effect of incorrect anatomical landmark identification (mislocation), and
• the influence of soft tissue artefact (e.g. muscle, tendon, adipose movement).

Early efforts to address these were undertaken by biomechanics groups most often based in clinical gait environments (e.g., hospital based clinical gait laboratories). Where appropriate and possible, the UWA Biomechanics Group set about transferring and translating some of the leading approaches developed by clinical and methodological driven biomechanics teams across to sport specific applications. This means that you as a practitioner, have access to a model that has been informed by rigorous sport specific testing, whilst being reassured that model and associated protocols are founded on fundamental methodological work dedicated to improving motion capture accuracy and reliability.

It is important to note that due to hardware and software limitations during early development of clinical gait models (CGM) there are acknowledged inherent limitations in the marker sets and models used for clinical gait use (VCM, PIG, CGM2).

The CGM2 project aims to preserve the strengths of the previous clinical gait models whilst addressing some of these limitations. Importantly for those working in applied sports biomechanics, when developing the PiG model for clinical gait settings, Vicon added an upper body model, primarily for visualisation purposes. This upper body model has never been published (other than in proprietary literature) or validated, such that it is not recommended for upper body dynamic sporting motion analysis in research settings (Baker et al., 2017).

A great starting point to understand the historical developments and limitations of 3D motion capture are the following four review papers by Aurelio Cappozzo, Ugo Della Croce, Alberto Leardini and Lorenze Chiara published in 2005. These papers clearly outline much of the methodological considerations underpinning the foundations of the UWA model and are provided as background to the principles that underpin the UWA model development.

• Cappozzo, A., Della Croce, U., Leardini, A., & Chiari, L. (2005). Human movement analysis using stereophotogrammetry: Part 1: theoretical background. Gait & posture, 21(2), 186-196. https://doi.org/10.1016/j.gaitpost.2004.01.010
• Chiari, L., Della Croce, U., Leardini, A., & Cappozzo, A. (2005). Human movement analysis using stereophotogrammetry: Part 2: Instrumental errors. Gait & posture, 21(2), 197-211. https://doi.org/10.1016/j.gaitpost.2004.04.004
• Leardini, A., Chiari, L., Della Croce, U., & Cappozzo, A. (2005). Human movement analysis using stereophotogrammetry: Part 3. Soft tissue artifact assessment and compensation. Gait & posture, 21(2), 212-225. https://doi.org/10.1016/j.gaitpost.2004.05.002
• Della Croce, U., Leardini, A., Chiari, L., & Cappozzo, A. (2005). Human movement analysis using stereophotogrammetry: Part 4: assessment of anatomical landmark misplacement and its effects on joint kinematics. Gait & posture, 21(2), 226-237. https://doi.org/10.1016/j.gaitpost.2004.05.003

Modelling 3D human motion using motion capture relies on the accurate and reliable definition of assumed rigid body segments to associate a system of axes. This in turn relies on the identification of external palpable subcutaneous anatomical landmarks, poor identification of which may introduce errors into the rigid body axis definition and siubsequently the associated full body model. There are three main reasons these mislocation errors arise (Della Croce et al., 2005):

1. The anatomical landmarks do not present as a single point to identify, instead they are large or irregular surfaces (e.g. lateral femoral condyle, iliac crest).
2. A layer of soft tissue, of varying thickness depending on individuals, covers the anatomical landmark (e.g. anterior/posterior superior iliac spines).
3. Accuracy of identification can vary depending on the palpation procedure used (e.g. fingers, palm of hand, external device).

A specific study that highlights how the propagation of these errors affects motion analysis was conducted by Gorton and colleagues (2009), where 24 examiners placed markers on the same individual participant across twelve motion analysis laboratories. The results showed that marker placement among examiners as the largest source of variability in clinical gait outputs. The average maximum difference of all lower body joint angles was 14.8°. The authors then followed a standardisation protocol previously reported by Kadaba and colleagues (1989) which decreased the standard deviation for 7 of the 9 parameters by an average of 22% (Gorton et al., 2009).

More recent work has reinforced similar findings -- that variations in anatomical landmark identification and variations in protocols contribute the largest sources of error in motion analysis. (Kratzenstein et al., 2012; DiCesare et al., 2015; Rast et al., 2016; Kaufman et al., 2016; Donnelly et al., 2020; Fischer et al., 2020).

Another study highlighting issues associated with internal anatomical landmark mislocation showed that these errors primarily affect the estimation of the geometric centres of the femoral head and acetabulum, and the hip and knee joint centres (Della Croce et al., 1999). An anterior-posterior hip joint centre mislocation of ± 30 mm will result in an average 22% error in the knee flexion/extension moment. This level of error propagation can have serious implications for appropriate interpretation of joint kinetics, whereby a practitioner may incorrectly infer, due to the model output, a dominant quadriceps moment when in fact there was a dominant hamstring moment. There is no better example of the importance of biomechanical model veracity than that applied to illegal action testing in cricket, where mislocating the glenohumeral joint centre by as little as 15 mm can result in the elbow joint angle showing as extending when it is in fact flexing – a finding that has the potential to end a cricket bowler’s career (Alderson et al., 2008).

In order to minimise the influence of the tester to correctly and repeatably identify anatomical landmarks the UWA model implements functional (numerical) methods to aid in identifying lower body (and some upper body) joint centres and joint axes. For the hip joint centre functional techniques, where the participant moves their leg through a functional range of motion, methods previously established (Cappozzo, 1984; Leardini et al., 1999; Piazza et al., 2001) are implemented in the UWA model (Besier et al., 2003). Methods to determine ‘optimal’ or mean helical axes of rotations for the knee (Spoor and Veldpaus, 1980; Reinschmidt and van den Bogert, 1997) and the elbow (Stokdijk et al., 1999)  informed the UWA knee and elbow joint centre and axes modelling (Besier et al., 2003; Alderson, 2005), as these are both modelled as hinge joints.

Extensive research has examined methods to quantify the influence of soft tissue artefact on biomechanical model outputs, with techniques such as intra-cortical pins, external fixators, percutaneous skeletal trackers and Roentgen photogrammetry used to highlight the magnitudes that soft tissue artefact (STA) can affect human movement analysis when using skin mounted stereophotogrammetric systems (Leardini et al., 2005).

There are many articles that highlight the magnitude of STA such as the work by Matsui et al. (2006) that reported upper limb STA, specifically on the scapula, can be as high as 8.7 cm. Other work investigating the effect of STA in the lower body by Hara et al. (2014) found that ASIS markers were more susceptible to relative displacement than PSIS markers, with displacement particularly evident in positions where the hip was flexed, resulting in STA up to 17mm. This propagated to inter-ASIS distance and hip joint centre location error by up to 20mm and 10mm. Akbarshahi et al. (2012) quantified STA in knee flexion with an average RMSE of 4.45°. More recently the work of Fiorentino et al. (2017) reported mean marker-based estimates of hip angles differ from those observed with x-ray fluoroscopy by 1.9°, 0.6° and 5.8° in the sagittal, frontal and transverse planes, respectively. Given the errors that soft tissue artefact can introduce a number of authors have attempted to implement solutions to reduce or minimise the effect. A synthesised account of these techniques was completed by Leardini and colleagues (2005) with some key developments including:

1. Solidification procedure – in each frame of movement the least disrupted triangle is identified which is defined from the markers. A ‘solid’ shape is fitted to the variations over time. The optimisation finds the best fit through an iterative process -- calculated at each step of the deformation frames (Cheze et al., 1995).
2. Pliant surface modelling – introduces a novel non-rigid 12 degrees of freedom model which quantifies the standard rigid rotations and translations while also quantifying scales and shear motion due to the deformation of the marker cluster associated with the skin – ‘pliant’ motion (Ball & Pierrynowski 1998). This optimisation technique leverages methods traditionally used in computer graphics to model the non-rigid motion of the segment surface (Leardini et al., 2005).
3. Dynamic optimisation approach – is a functional based method that combined an experimental and analytical procedure by means of a dynamic model of the marker cluster technical frame to anatomical landmark relationship (Lucchetti et al., 1998). This experimental approach requires the participant to perform extra dynamic motion trials such as locking the knee in hyperextension and contracting the quadricep muscle group – with the aim to reproduce similar soft tissue artefact in these “dynamic target” activities as the artefact present in the motion trials.  These dynamic activities then go through analytical methods to determine the relationship of marker clusters and anatomical landmarks which are stored in an artefact table. The final local positions of the anatomical landmarks were corrected by subtracting the corresponding artefact component (Leardini et al., 2005).
4. Pointer cluster technique – aimed to minimise the eigenvalue changes by adjusting the mass of each marker in a cluster (Andriacchi et al., 1998). The variation in the mass distribution allows for an estimated centre of mass position and orientation of the references system. There are no requirements for collection extra calibration tasks however it does require a significant number of additional markers for each segment (Leardini et al., 2005).
5. Global optimisation – includes consideration of joint constraints and global error minimisations (Lu & O’Connor 1999;2000). This process is computed by minimising the weighted sum of squared distances between simulated and model determined marker positions - with an additional joint constraint where joints are taken as perfect ball-and-socket joints (Leardini et al., 2005).

More recent efforts have focused on the two components of STA:

1. The rigid component – that is the translations and rotations undergone by the markers or clusters (Benoit et al., 2015; Bonci et al., 2015; Camomilla et al., 2015; De Rosario et al., 2013; Dumas et al., 2015).
2. The non-rigid component – this is the geometrical transformations (changes in size and shape) that have been assessed with, predominately, principal component analysis or modal analysis (Andersen et al., 2012; Benoit et al., 2015; Bonci et al., 2015; Dumas et al., 2014; Grimpampi et al., 2013).

Barre and colleagues in 2017 explored whether a larger number of markers (80) placed on the thigh and shank would produce comparable levels of STA as reported in the literature. The results produced more complex patterns of STA than previously reported with marker clusters made of only 4-6 markers. One of the most interesting findings from this work was that translation and rotation components remain the main STA mode – which supports the current re-orientation in the literature of STA compensation methods from bone estimators which typically address the non-rigid components to kinematic driven rigid components of STA reduction models (Barre et al., 2017).

A key benefit for using the UWA model is that it contains an extensively validated upper body model. This model has been used to help peak sporting bodies assess upper body movements, such as the International Cricket Council for bowling assessments of international level players. Aside from the UWA upper body model there are currently no published validations of the plug-in-gait upper body model (Baker et al., 2017). A significant amount of research has been dedicated to developing the UWA upper limb modelling protocols with dedicated research streams for the elbow (Lloyd et al., 2000; Elliott et al., 2007; Middleton et al., 2009; Chin et al., 2010; Wells et al., 2012; Wells at al., 2015; Wells et al., 2016), the shoulder (Alderson et al., 2008; Campbell et al., 2008; Campbell et al., 2009; Zhang et al., 2011) and general validation of the upper body specific modelling methods (Elliott et al., 2005; Chin et al., 2009; Reid et al., 2010; Wells et al., 2013).

The specifics of these developments are outlined in the UWA development timeline and deep dive section. Work on shoulder modelling aimed to develop a more robust and accurate regression model for the glenohumeral joint centre of rotation, create optimisation procedures to reduce the affect of soft tissue artefact and investigations into the affects of different clusters and coordinate systems (Alderson et al., 2008; Campbell et al., 2008; Campbell et al., 2009; Zhang et al., 2011). The group’s elbow modelling developments recommended optimal helical endpoint technical reference frame referencing, established a marker-based mean finite helical axis model, the repeatability of anatomical versus cluster/functional marker sets and the improvement of inverse kinematic estimations of the elbow due to joint co-ordinate prescription (Middleton et al., 2009; Chin et al., 2010; Wells et al., 2013; Wells et al.,2017). A key takeaway for practitioners is the group’s work has resulted in one of the most robust validation efforts of an upper limb model that can be implemented in a sporting context to withstand the high speeds and abnormalities of upper limb sporting movements.

One drawback for minimisation techniques is that they are complex and are often not practicable to implement in applied settings, such as clinical or sports environments. Three minimisation strategies that have been included into the UWA model are:

Functional calibration

First introduced into the UWA model by Besier and colleagues (2003) to define functional hip joint centres and mean helical knee axes. The move to adopting functional approaches to define joint centres and axes has been shown to reduce the effect of soft tissue artefact and traditional generic regression approaches (Taylor et al., 2010). Since approximately 2014, the UWA team have ceased implementing functional methods in a Matlab PECS addin in the Vicon processing pipeline in preference to adopting the SCoRE and SARA functional methods of Vicon. In house testing at UWA showed no significant differences between their custom approach and the SCoRE and SARA outputs, with the latter being a seamless use of functional methods in the modelling pipeline.

CAST technique

The calibrated anatomical system technique (CAST) introduced by Cappozzo and colleagues (1995) proposed a method with cluster marker sets and anatomical landmark calibration procedures – typically completed with a pointer but not limited to the use. The marker cluster are affixed to the segment, such as the thigh, and technical co-ordinate systems (TCS) are established for each cluster. Static calibration trials, either an A-pose or static pointer trials, establish the location of key anatomical landmarks in their relevant TCS. This allows for the position of the landmark to be reconstructed relative to the TCS during the movement trials. This technique was introduced into the lower body developments of the UWA model in 2003, extended in 2005 (Alderson 2005), and further implemented in the upper body model (Chin et al., 2009; Chin et al., 2010).

Recently Picerno and colleagues developed a method to use the CAST pointer method for the anatomical calibration of wearable magnetic and inertial measurement units. The study proposed a new calibration procedure, that leveraged previously used 3D motion capture techniques, to provide estimates of 3D shoulder/elbow angular kinematics and the linear trajectory of the wrist (Picerno et al., 2019).

Rigid marker clusters and their placement

Studies have shown that implementing rigid cluster markers can reduce soft tissue artefact (Baker et al., 2017). At the time of compilation of these resources the UWA marker set composes of rigid marker clusters for the following segments: thigh, tibia/shank, upper arm, and forearm. More recent work has looked at alternative methods to rigid marker clusters whereby larger numbers of individual markers are placed on segments and the affect on inverse kinematics was assessed compared to a cluster-based approach (Bakke & Besier, 2022). In this work Bakke and Besier assessed the differences in joint angles when using a contemporary cluster-based approach versus one without thigh markers. The no thigh marker joint angle results differed by a median of 1.2° compared to the cluster-based approach, with most average differences smaller than previously estimated STA error. Mean no thigh hip rotation, RMSE of 2.18°, was over 3.5 degrees smaller than STA error magnitude. The average reported RMSE of 1.1° for knee flexion falls below STA-based knee flexion error previously measured at or above 4.45° as reported by Akbarshahi and colleages, 2012 (Bakke & Besier, 2022). These developments are limited in the applied sporting context as they require more markers and a longer capture time – which is not feasible in most applied contexts.

Marker cluster placement should not be overlooked as clusters are still affected by STA. Barre and colleagues in 2015 investigated the spatial distribution of STA between multiple participants and the affect of varying cluster placements on the thigh and shank during gait. Overall, they reported maximum marker displacement of 24.9mm and 15.3mm in the proximal areas of the thigh and shank. STA rigid motion was larger on the thigh with RMS error in cluster orientations between 1.2° and 8.1° highlighting that a good selection of marker cluster placement is crucial in STA compensation (Barre et al., 2015). What that means for you in an applied setting is that you should always be wary of placing clusters on areas with notable soft tissue artefact, such as the front of the thigh or too high up on the thigh towards the glutes.

The development of biomechanical protocols and models is a dynamic and evolving process in which new work builds on the learnings of others.  Before providing a broad overview of the 25 year developmental history of the data collection and modelling pipeline it is important to acknowledge that these practices and protocols would not exist without benefit of a large and comprehensive body of work completed over many years by staff and students from the University of Western Australia. We would like to acknowledge the following individuals’ critical contribution to the rigorous development, testing and ongoing iterations of the upper and lower body model collection protocols and model code:

Prof. Thor F. Besier, Assoc. Prof. Jacqueline Alderson, Prof. David G. Lloyd, Dr Siobhan Reid, Assoc. Prof. Amity Campbell, Dr Aaron Chin, Dr Denny Wells, Dr Kane Middleton, Dr Helen Bayne, Dr Daniel Cottam, Emeritus Prof. Bruce Elliott, Dr Adam Hunter (AIS multiple force plate integration).

This list is by no means exhaustive. The integral work of many other researchers, including honours students and postdoctoral fellows affiliated with the UWA Biomechanics Group have made the development and validation of these protocols possible. We also acknowledge and thank them for their contribution.

Below is a timeline of major UWA model developments with key development articles since work began on the model in the late 1990s. If you are interested in finding out about the specific research and the outcomes on the UWA marker set and model you can find it in the following pdf under the parent headers of: lower limb, shoulder, elbow and general.