Comparison of video-based algorithms for 2D human kinematics estimation: a preliminary study

Abstract Many research efforts have been spent developing robust video-based algorithms for human pose estimation. Our goal was to compare estimation gait analysis. We conducted an experiment with a healthy subject performing walking sessions on treadmill at three different speeds: slow (3.6 km/h), medium (5 and high (7 km/h). An RGB 4k camera placed laterally the sagittal plane. Four were compared: (i) colour threshold filtering blob-analysis, Deep Learning-based markerless (ii) TC-Former, (iii) FastPose (iv) Blazepose. For blob-analysis algorithm, six magenta passive markers over joint centres of subject’s lower limb. All selected deep learning-based are supported by various open-source toolboxes pre-trained several whole-body keypoint datasets. The 2D trajectories compared considering root mean square error Pearson’s coefficient. Preliminary results showed correlations between marker all speeds. TC-Former generally performed better below 35 mm did not suffer from self-occlusion issues.