Can Markerless Pose Estimation Algorithms Estimate 3D Mass Centre Positions and Velocities during Linear Sprinting Activities?

Abstract

The ability to accurately and non-invasively measure 3D mass centre positions and their derivatives can provide rich insight into the physical demands of sports training and competition. This study examines a method for non-invasively measuring mass centre velocities using markerless human pose estimation and Kalman smoothing. Marker (Qualysis) and markerless (OpenPose) motion capture data were captured synchronously for sprinting and skeleton push starts. Mass centre positions and velocities derived from raw markerless pose estimation data contained large errors for both sprinting and skeleton pushing (mean ± SD = 0.127 ± 0.943 and −0.197 ± 1.549 m·s−1, respectively). Signal processing methods such as Kalman smoothing substantially reduced the mean error (±SD) in horizontal mass centre velocities (0.041 ± 0.257 m·s−1) during sprinting but the precision remained poor. Applying pose estimation to activities which exhibit unusual body poses (e.g., skeleton pushing) appears to elicit more erroneous results due to poor performance of the pose estimation algorithm. Researchers and practitioners should apply these methods with caution to activities beyond sprinting as pose estimation algorithms may not generalise well to the activity of interest. Retraining the model using activity specific data to produce more specialised networks is therefore recommended.

The accuracy of several pose estimation methods for 3D joint centre localisation

“…Evidence of activity specific pose estimation performance was observed, with the largest mean differences and random error measured during running and the smallest mean differences and random error measured during jumping. Lower pose estimation performance during running may be attributed to the greater range of observed limb configurations and segment velocities when compared to jumping and aligns with previous evidence that pose estimation performance is highly task specific 35 —an important consideration for researchers. Furthermore, greater limb velocities may introduce image noise in the form of motion blur, making it harder for pose estimation methods to detect image features that represent a given joint centre.…”

“…The final 3D joint centre solution was taken as the intersection of the remaining inlier rays. For a more detailed description of this process see 35 . The 3D joint centre reconstructions were filtered using a bi-directional Kalman filter 36 before being written to C3D file format.…”

A Systematic Review of the Application of Camera-Based Human Pose Estimation in the Field of Sport and Physical Exercise

“…Regarding the used data for the development and testing of the systems, on the one hand, several papers such as [ 12 , 13 , 14 , 15 , 17 , 18 , 20 , 37 , 38 , 42 ], developed their own datasets using manual annotations, MoCap systems, or other ground truth generation methods, but did not make them publicly available. Other papers created and published their dataset to contribute to the research community, such as [ 22 , 30 ].…”

“…Other metrics are used apart from the estimation of the joints of the body, such as in [ 17 , 18 ] focused on the estimation of the Center of Posture (CoP) or Center of Mass (CoM). In those cases, the error is calculated in relation to the ground truth location.…”

“… RGB images Output: 2D (+ gait output category) Own not public dataset of RGB images of people walking in different situations using markers for the hip, knee, and ankle (no HPE data is specified as ground truth, the walking category is labeled by hand) Can Markerless Pose Estimation Algorithms Estimate 3D Mass Centre Positions and Velocities during Linear Sprinting Activities? [ 17 ] Test the capacity of estimating the 3D mass center positions and velocities during linear sprinting activities using 3D HPE. (in such actions in which skeleton is pushing, current HPE methods show quite high error for the objective of the paper, at least for the proposed method) RGB images + depth Output: 3D Own not public dataset created using maker-based MoCap system Qualysis and markerless OpenPose system to record sprints.…”