Dynamic Hand Gesture Recognition Based On Depth Information

Abstract

Dynamic hand gesture is consisted by hand movement trajectory and the changes of hand shape. However, some existing methods only focus on the trajectory, and those methods can not accurately recognize the gesture that has the similar trajectory but different hand shape changes. For this problem, a dynamic hand gesture recognition method that combines the trajectory with the hand shape is proposed in this paper. First, we use depth images to determine the hand region and extract the location of palm center, avoiding the effect of lighting condition and complex environment. The absolute position and relative position of the palm center is adopted to represent the trajectory. Next, we present a method which combines convex hull with k-curvature to detect the fingertips contour, which can be a better representation of the hand shape change in dynamic gestures. Then we solve the image blurring problem by voting strategy. Besides, the Temporal Pyramid algorithm is applied to process the extracted features, since it can express temporal features more delicately and unify different feature dimensions. Finally, SVM algorithm is utilized to classify the dynamic hand gesture. The experimental results show that our method has higher recognition rate with less time consuming than the compared methods.