Research Of Static Gesture Recognition Based On Kinect Depth Data And Combinaional Feature

Hand gesture, as a natural way of human communications, is one kind of latest human-computer interaction way. The operator only needs to form different gestures in front of the equipment to interact with the computer. Although automatic hand gesture recognition has been studied extensively for about a decade, the related theories and technologies are not mature enough. Kinect, one of the most popular motion sensing input devices, is capable of capturing depth data along with RGB frames. In this paper, we employ the new Kinect version 2 as the provider for gesture depth frame, then we use digital image processing approaches to process the depth frame, the result is offered to data feature extraction. At last, we use classifier to perform static gesture recognition.We firstly discuss how to acquire depth frame from Kinect sensor in great detail. After acquisition of depth data, we utilize image segmentation algorithm based on lower-upper threshold to obtain a binary image which contains only the gesture. In the stage of feature extraction, we calculate some important features such as contour, convex hull, convex defect and gravity center. By combining different simple features, we propose our way of describing the shape of convex hull and the relative position of convex defects. These two descriptor altogether establish the combinational feature descriptor of hand gesture. Finally, we tested our approach by two different supervised classification methods, and carry out experiments in different lighting conditions.In our 12 different hand gesture patterns database, experiment result shows our descriptor can train the classifier effectively, the mean accuracy rate reaches 91.7%. Moreover, the experiments in different lighting conditions show our approach is almost not influenced by lighting condition. This approach is easy to implement and it performs stably.