Medical Rehabilitation Assessment using Gait Analysis in Video Sequences

An automatic method to analyze human locomotion for subjects performing normal and abnormal walking in video sequences is proposed in this paper. Abnormal walking performed by patients is due to lower limb injuries or to some musculoskeletal and neurological pathologies. The method can be useful for the rehabilitation process of patients or in the detection of abnormal gait. To detect joint positions of the human body segments from video sequences the MoveNet Thunder pose estimation model is used. Having body segments detected, angles from joints on the left and right side are computed in each frame to analyze their variations. Because the video sequences differ from one to another, statistical parameters for each stride are computed and used in classification. The data extracted from the video sequences is small, therefore an augmentation was applied to increase data.