A Skeleton Object Detection-Based Dynamic Gesture Recognition Method

A skeleton-based object detection method is proposed to recognize the dynamic gesture. The Kinect depth camera is utilized to capture the skeleton of human beings in the dynamic gesture motions, all gestures of the same dynamic gestures are marked and trained. Then a modified Single Shot MultiBox Detector (SSD) network is adopted to locate the arm in the skeleton images. The dynamic gesture is recognized based on the arm skeleton movements in the motion. To balance the precision and recognition time in the identification period, a corresponding comprehensive index is constructed to find a suitable proportion of arm skeleton images for detection by the trained neural network. It can reduce the redundancy and improve detection efficiency. Moreover, it will decrease the impact of noise and enhance the recognition accuracy. At last, an experimental analysis validated the effectiveness of the proposed method.