Hand Gesture Recognition Based On Depth Image And Dexterous Hand Interaction

The research of human-computer interaction has developed into multi-modalway of user-interface from simple-modal way with the progress of computertechnology. As one of technologies of human-computer interaction, hand gestureanalysis and recognition based on computer vision have been the focus of researchers.The problems of how to achieve non-contact gesture recognition based on computervision, slip the leash of mouse and keyboard, and make the human-computerinteraction more intuitionistic are especially concerned.Comprehensive analysis and description of the real-time system forhuman-computer interaction through gesture recognition was proposed in this thesis.The process of hand gesture recognition consists of four parts: collection of handgesture datum, segmentation of gesture region, characteristic parameter extraction,and hand gestures classification.The main contents of this paper include as follows:(1) An algorithm, which can be used in human-computer interaction based ondepth image for gesture recognition, was proposed according to the problem ofrecognition caused by environmental disturbance which gesture recognition issusceptible to at the condition of complicated background through analyzing theapparent characteristics of gesture.(2) In the part of segmentation, the algorithm achieved the segmentation ofgesture region based on depth image. Comparing with the traditional segmentationalgorithms based on complexion and movement, it has a better robustness andsegmentation, and is almost insusceptible to the influence of light and environment.(3) In the part of characteristic parameter extraction, the apparent features ofradian between fingers, number of radian and finger radian with invariance whenrotating and scaling are extracted in the paper which effectively improve the rate ofrecognition and make sure the reliability and practicability of experiment. (4) In the part of gestures classification, template matching with the shortestdistance was used in the system to recognize gesture. The way of template matchinghas the characteristics of less calculation and easier work. The real-time system forhuman-computer interaction through gesture recognition achieved at last also has abetter robustness.(5) The algorithm based on depth image for gesture recognition was applied forthe platform of dexterous hand with three fingers. And the human-computerinteraction for dexterous hand was designed in this paper, which the process can beseparate into two parts: the feature extraction and recognition of gesture and theconverting from hand gestures to control commands. The whole process is that theresult of recognition based on depth image was converted to the control commands tocontrol the hand move. This system could ideally do some simple interactions withobjects and achieved satisfying results according to the experiments.