MARKERLESS GESTURE RECOGNITION ACCORDING TO BIOMECHANICAL CONVENTION

During rehabilitation it is important to perform determined exercises according to joint biomechanics in order to achieve the right improvements. In order to rehabilitation systems and applications to be able of recognizing movements according to these necessities, this paper proposes a gesture recognition method developed considering International Society of Biomechanics standards for upper and lower limbs. Joints positions were obtained through Kinect sensor and the angles between consecutive joints associated with normal vectors of the anatomic planes were computed based on body references. Movements of upper and lower limbs were classified according to the biomechanical standards. All movements were tested by a specialized physiotherapist. Each movement was performed 60 times: 20 correct executions at normal velocity, 20 fast and 20 wrong performances. These movements were submitted to different movement tolerance margins in order to analyze the most adequate for applications. The method proposed presented good capability in classifying biomechanical exercises, and therefore, it is indicated as recognition method for rehabilitation applications.