Abstract— There are increasing numbers of hand interaction system based on Microsoft Kinect for its providing the extra depth information which facilitates the hand segmentation. However, most of the recent papers focus on the stable hand recognition whose assumption can hardly be fulfilled in highly interactive HCI environment. We proposed a robust method to recognize both the moving and stable hand. We explored several features in both RGB and depth images based on a novel descriptor, and then built a feature selection model to find the robust features for recognition. Then, we exploited the hierarchical self-similarity matrix (H-SSM) to detect the gesture transformation to further enhancing the system robustness. Our method can be widely used in gesture recognition and the effectiveness has been verified through experiment.
Index Terms— Self-similarity matrix, feature selection, HOG feature, kinect
Yuhang Wu and Hui Ding are with the Information and Engineering College, Capital Normal University, Beijing, PRC (e-mail: sscommanderh@126.com, dinghui@ie.cnu.edu.cn)
Liangke Zhao is with Capital Normal University, PRC. He is now an internship student with the NLPR, Chinese Academy of Science (e-mail: bensvage1989@yahoo.com.cn)
[PDF]
Cite: Yuhang Wu, Liangke Zhao, and Hui Ding, " Robust Hand Gesture Recognition with Feature Selection and Hierarchical Temporal Self-Similarities," International Journal of Information and Electronics Engineering vol. 3, no. 5, pp. 510-515, 2013.
