Pose2Sim: An End-to-End Workflow for 3D Markerless Sports Kinematics—Part 2: Accuracy

Abstract

Two-dimensional deep-learning pose estimation algorithms can suffer from biases in joint pose localizations, which are reflected in triangulated coordinates, and then in 3D joint angle estimation. Pose2Sim, our robust markerless kinematics workflow, comes with a physically consistent OpenSim skeletal model, meant to mitigate these errors. Its accuracy was concurrently validated against a reference marker-based method. Lower-limb joint angles were estimated over three tasks (walking, running, and cycling) performed multiple times by one participant. When averaged over all joint angles, the coefficient of multiple correlation (CMC) remained above 0.9 in the sagittal plane, except for the hip in running, which suffered from a systematic 15° offset (CMC = 0.65), and for the ankle in cycling, which was partially occluded (CMC = 0.75). When averaged over all joint angles and all degrees of freedom, mean errors were 3.0°, 4.1°, and 4.0°, in walking, running, and cycling, respectively; and range of motion errors were 2.7°, 2.3°, and 4.3°, respectively. Given the magnitude of error traditionally reported in joint angles computed from a marker-based optoelectronic system, Pose2Sim is deemed accurate enough for the analysis of lower-body kinematics in walking, cycling, and running.

OpenCap: 3D human movement dynamics from smartphone videos

“…The accuracy of OpenCap’s kinematic and kinetic estimates is similar to state-of-the art markerless motion capture solutions. OpenCap’s kinematic error (range of root mean squared error [RMSE] across lower-extremity degrees of freedom: 2.0–10.2°) is similar to errors reported for inertial-measurement-unit-based approaches (RMSE: 2.0–12° for walking, running, and daily living activities 18,49–55 ) and commercial and academic video-based systems with eight cameras (RMSE: 2.6–11° for walking, running, and cycling activities 30,56 ). Furthermore, in contrast with most inertial-measurement-unit-based approaches, OpenCap estimates global translations (e.g., pelvis displacement), enabling estimation of whole-body measures like center-of-mass trajectory.…”

“…OpenCap triangulates the synchronized 2D video keypoint positions to compute 3D positions. OpenCap uses a Direct Linear Transformation algorithm for triangulation 66 , and weights the contribution of individual cameras in the least-squares problem with the corresponding keypoint confidence score 56 . There are two major limitations of using 3D keypoint positions triangulated from video for biomechanical analysis.…”

Accuracy of a markerless motion capture system in estimating upper extremity kinematics during boxing

“…As these tools are primarily intended for 2D pose estimation, some studies leveraged its potential in estimating 2D kinematics of the lower limb (hip, knee, and ankle joint) during gait (Stenum et al, 2021 ), vertical jump (Drazan et al, 2021 ), and under water running (Cronin et al, 2019 ). Progressing further, others focused on estimating 3D poses from 2D images of multiple calibrated cameras using triangulation during walking (Nakano et al, 2020 ; Needham et al, 2021 ; Pagnon et al, 2022 ), jumping (Nakano et al, 2020 ; Needham et al, 2021 ), running (Needham et al, 2021 ; Pagnon et al, 2022 ), cycling (Pagnon et al, 2022 ), and throwing (Nakano et al, 2020 ). While these studies demonstrated the potential of openly accessible pose estimation tools in estimating 3D joint kinematics, most of them primarily evaluated the lower extremity.…”

Pose2Sim: An open-source Python package for multiview markerless kinematics

“…The whole workflow runs from any video cameras, on any computer, equipped with any operating system (although OpenSim has to be compiled from source on Linux.) Pose2Sim has already been used and tested in a number of situations (walking, running, cycling, dancing, balancing, swimming, boxing), and published in peerreviewed scientific publications assessing its robustness (Pagnon et al, 2021) and accuracy (Pagnon et al, 2022). Its results for inverse kinematics were deemed good when compared to marker-based ones, with errors generally below 4.0°across several activities, on both lower and on upper limbs.…”