Automatic Gait Analysis and Classification in Video Sequences

Gait abnormalities can be related to locomotion injuries or to various musculoskeletal and neurological pathologies. In this paper an automatic method for gait analysis and evaluation is proposed. It can be used for rehabilitation assessment but also to detect the gait abnormalities. The proposed system is noninvasive and it is based on the analysis of the video sequences captured in the sagittal plane. The MoveNet pose estimation model is used to detect the joints positions of the body segments. Then the knee and thigh-trunk angles on both left and right side are computed in each frame of the video sequence to study their variation. The video sequences can have different lengths and frame rates, therefore the analysis will be done at stride level by using statistical parameters. Experiments were made on a reduced number of video sequences that contain pathological gait and normal gait. Data augmentation was applied to increase the volume of data which is then used for gait classification.