A real-time identification for hand-based movements using Recurrent Complex-Valued Neural Networks

This paper presents an application for hand-based movements using two Recurrent Complex-Valued Neural Networks (RCVNN) in real-time. The proposed system identifies hand-based movements using two angles of human arm model acquired by the infrared vision time of flight depth system integrated in Kinect v2. The results of the experiments compare the performance of the RCVNN with the inverse kinematic. Finally, this topology helps us to identify hand-based movements avoiding singularities.