Research On Hand Gesture Recognition Methods For Natural Interaction

While human-computer interaction requires friendly, convenient and efficient interfaces, hand gesture recognition becomes a key technology and research spot of the human-computer interaction for its operating through intuitive actions without the need of assistant devices. In this thesis, research on Kinect based hand gesture recognition is conducted, which includes image segmentation methods, hand gesture recognition algorithms and applications in rehabilitation training.The principle of vision-based hand gesture recognition interaction system is introduced. Algorithms for three key processes of vision-based hand gesture recognition, i.e., hand segmentation, hand tracking and hand gesture classification, are reviewed. And the advantages and disadvantages of the aboved algorithm is given.Accroding to the principle analysis of hand gesture recognition, the thesis proposed a novel depth image and hand skin color based hand segmentation method to improve the segmentation accurancy. The proposed approach used depth threshold to obtain pixels of hand in depth image space and then mapped them to color image space, where skin color segmentation was futher used to optimize the result of hand segmentation. Then static hand pose recognition system is realized through fingertips detection method. To improve the real-time and recognition rate of hand gesture recognition, the thesis proposed a constraints pre-processing based dynamic time warping method, which used the location information of hand, wrist and elbow acquired by Kinect sensor as feature. To verify the effectiviness and reliability of the proposed methods, we conduct hand pose recognition and hand gesture recognition experiments, where 10 normal subjects are involved. Experiment results show that an average recognition rate of 88% for 5 hand poses and an average recognition rate of 92.5% ARR for 4 typical hand gestures are achieved.Based on the proposed hand gesture recognition system, we futher develop several hand guesture-based rehabilitation training systems and conduct a series of experiments, including the flexibility comparison between left and right hand experiment and the rehabilitation extension evaluation experiment. Experiment results prove that the system described in the thesis can be successfully utilized for rehabilitation training.The thesis carried out the study of hand gesture recognition in natural interaction,and lays an important foundation for the further devolpments of hand gusture recognition based natural huamna-computer interaction and its applications in rehabilitation training.