The effects of markerless inconsistencies are at least as large as the effects of the marker-based soft tissue artefact

The soft tissue artefact is a well-known issue for marker-based motion analysis and markerless motion analysis is by definition free from this artefact. The goal of this study is to compare the limb skeletal inconsistencies generated by the neural networks in markerless motion capture and generated by the soft tissue artefact in marker-based motion capture using retrospective data. Sixteen volunteers were included and were asked to perform four motor tasks (walk, sit-to-stand, stand-to-sit, countermovement jump) acquired with ten optoelectronic cameras and ten video cameras. Keypoint identification was performed in videos using Openpose. Triangulation and data augmentation algorithms were used to get an extension of anatomical landmarks. Then, lower limb skeletal inconsistencies (length variations and apparent joint dislocations) for both marker-based and markerless data were analyzed. The length variation of the lower limbs was generally larger with markerless data (triangulated keypoints and augmented anatomical landmarks) as found with marker-based data. Mean dislocations were found smaller for the markerless data than for the marker-based data for the hip only. The effect of the markerless inconsistencies are at least as large as the effect of the soft tissue artefact except for the hip dislocation, probably due to the soft tissue artefact that is main at the pelvis level. These inconsistencies are related to different phenomena than skin sliding as there are no correlation with joint flexion-extension angles. Thus, compensation methods proposed for soft tissue artefact are not all applicable.