Hand Gestures Recognition Using Time Delay Networks

This paper proposes a system for body gestures recognition using the coordinates of the body skeletal returned by a Kinect sensor who are processed in order to compute a set of angles. Gesture recognition is achieved using a Time Delay Neural Network, implemented in two ways: with delay layer, and with delay synapse. The resulting system was trained on a set of gestures, such as: hands up and elbows bent, lifting arms, lowering arms, round, greeting gesture and pointing assertion. Each gesture was repeated at least 5 times with different speed. The accuracy of the proposed method is approximately 80%.