2D Markerless gait analysis

We present a 2D gait analysis system which is completely markerless and extracts kinematic information by analyzing video sequences obtained from an RGB video camera. These properties make the proposed approach particularly suitable in medical contexts where visual gait observation is still a recognised procedure or the invasiveness and high costs of marker-based systems can not be afforded. Markerless motion estimation literature for medical gait analysis is generally 2D oriented, since the majority of joints dysfunctions related to gait occur in the sagittal plane. Most of the approaches are based on time consuming human body models or need human-intervention. Conversely, the method we present this contribution is silhouette-based, completely automatic and uses information on the human body anthropometric proportions for the estimation of the lower limbs’ pose in the sagittal plane with good accuracy and low computational cost. Tests on a large number of synthetic and real video sequences with normal gait have been performed. Different frame rates, image resolutions and noises have been considered. The obtained results, in terms of sagittal joint angles, have been compared with the typical trends found in biomechanical studies. The performance of the proposed method is particularly encouraging for its appliance in the real medical context. Keywords— Mark