A markerless gait analysis protocol based on a single RGB-Depth camera: sensitivity to background changes

This work presents a two-dimensional markerless clinical gait analysis protocol to estimate the sagittal lower limb joint kinematics based on markerless recordings from a single RGB-Depth camera. The influence of different experimental setup conditions (presence/absence of a green coloured background) on the accuracy of this proposed method was also evaluated. The proposed protocol includes a subject separation from the background, the definition of a multi-segmental model of the lower limb and the estimation of the relative joint kinematics. The segmentation algorithm performance was assessed by computing the similarity between the computer-obtained segmentations and the corresponding manual tracings (ground-truth). The estimated joint angles were compared with those obtained using a reference optoelectronic marker-based clinical protocol. The offset between the mean waveforms and the RMS value of the waveforms difference after removing their offset were computed. The segmentation accuracy is higher than 0.9 in both the experimental setup conditions. There is a little worsening (2.2%) when the segmentation is computed without the green background, but it does not affect the accuracy of the estimated kinematics since the RMSD values are lower or equal to those obtained with the green background.