Real-Time Hand Gesture Recognition for Robot Hand Interface

In this paper, a highly accurate real-time hand gesture recognition system is proposed and implemented. This system can drive a dexterous robot hand behave as humans do without motional sensors attached to humans. The gesture region is segmented from complicated background based on the depth image which is obtained from Kinect. The features, such as the number of fingers, the radians between fingers, are extracted to improve the rate of recognition effectively. Then template matching with the shortest distance was used to recognize the gesture. The result of recognition is sent to the lower computer via RS232. Finally, the five fingers dexterous hand can behave as what the humans do. It can be seen from the experiments that our system can track humans' hand gesture robustly and recognize more than 90 percent of the hand gestures of our depth image database.