Dynamic Gesture Recognition Algorithm Based On Motion Tracking

Vision-based gesture recognition is a main research direction in human-computerinteraction field. First, a camera captures the user’s gesture image sequence, and then theimage sequence will be processed and analyzed. Finally, the user’s gestures will bechanged into mouse or keyboard events which are comprehensible to the applicationprogram. Compared with the traditional human-computer interaction mode, the newinteraction mode to control computer by using hand gestures can bring better userexperience.This paper studies and realizes dynamic gesture recognition algorithm which is basedon motion tracking. In this paper, the writer proposes an adaptive color segmentationalgorithm. Specifically, at first it obtains natural complexion sample set and complexionsample set under illumination change through face detection. Then the color distributionhistogram of these samples will be calculated and merged. Meanwhile, on the basis of thehistogram distribution, the range of color segmentation will be shown. The writer proposesa real-time compression gesture-tracking algorithm, it gives the corresponding weight tothe characteristic rectangle frames according to their regional position. By using Kalmanfilter, it can predict the position of tracking targets in next frame, and then search for thetracking targets around the predicted position. Some candidate samples will then befiltered based on complexion coverage area in the rectangle frame. Finally, the mapping ofboth operation space and display space are implemented, which makes it possible tocontrol mouse movement by using gestures motion, and it also implements the recognitionof some usual dynamic gestures based on an idea of mouse touch.The experimental result shows that the selfadaptive skin clolr segmentaiton algorithmin this thesis is more adaptable to different races or illumination than other existingmethods; the modified tracing algorithm is more precise and more robust than the originalalgorithm in the gesture tracing process. As a result, the gesture recognition system basedon the above methods has higher recognition rate for dynamic gesture recognition.