Gait Segmentation in Video Sequences Analysis for Medical Rehabilitation Assessment

An automatic method for gait segmentation in video sequences is proposed. The recovery level of the patients with lower limb injuries and also neurological pathologies can be automatically tracked, analyzed and evaluated using video recordings with gait sequences. Since the analysis is performed at the level of each stride, it is necessary to identify the starting frames of each of them. This task is more difficult when the camera is moving to follow the subject. The proposed approach is based mainly on the subject’s pose estimation. The MoveNet model is used to detect the joints of body segments in all frames. The first frame in a sequence in which the subject’s ankle remains fixed is considered the beginning of a step. The Lucas-Kanade Optical Flow algorithm is used to track fixed features in the scene in order to update the position of the subject according to the movement of the camera. This approach is already included in the medical rehabilitation assessment system developed by authors.