Comparison of methods for hand gesture recognition based on Dynamic Time Warping algorithm

Gesture recognition may find applications in rehabilitation systems, sign language translation or smart environments. The aim of nowadays science is to improve the recognition systems' efficiency but also to allow the user to perform the gesture in a natural way. The article presents different methods (DTW - Dynamic Time Warping, DDTW - Derivative Dynamic Time Warping, PDTW - Piecewise Dynamic Time Warping) based on Dynamic Time Warping algorithm, which is commonly used for hand gesture recognition using small wearable three-axial inertial sensor. Additionally, different approaches to signal definitions and preprocessing are discussed and tested. To verify which of the methods presented is more accurate in case of gesture recognition, database of 2160 simple gestures was collected, and recognition procedure was implemented. The main goal was to compare the efficiency of each method assuming that each person should perform the movement naturally. Obtained results suggest that the most efficient method for the presented problem was the DDTW. The worst recognition performance was achieved with the PDTW method.