Gesture Detection and Recognition Fused with Multi-feature Based on Kinect

Abstract

Aiming at the high demand of the background environment and the user, this paper designs a real-time human static gesture recognition algorithm based on the depth information of Kinect. The localization of the joint points of the hand is realized by using the Kinect skeleton. The depth image is acquired by the depth sensor, and the joint points of the hand are tracked continuously; After locating the position of the hand, the region of interest is intercepted, and the depth threshold is set up to segment the hand from the depth image; The segmentation image is processed by morphology, and the circular rate, filling rate, perimeter rate, convex hull, convex defect, Hu moments of the hand contour are 9 kinds of features; Six kinds (0 to 5) of gesture are recognized using SVM method. Recognition rate and robustness of gesture recognition experiments are conducted in static and dynamic environment respectively. The experimental results show that the proposed algorithm can achieve better recognition result in a variety of environments.