Real-Time Hand Gesture Recognition Using Depth Information and Path Analysis
You-Jia Wu, Jian–Jiun Ding · 2023
Applications of human-computer interaction (HCI) have become common recently. Among a variety of methods, hand gestures are used in communication natural tools. Thus, hand gesture recognition has been popular in research, especially the vision-based methods with the advantage of contact-free characteristics. In this study, we developed a real-time vision-based dynamic hand gesture recognition system to continuously classify 6 different gestures: 4 directions of swipe and 2 orientations of rotation. With the development of depth sensor cameras, we segmented possible hand regions by a depth value of interest and tracked the hand locations with a combination of frame difference, connected components, and centroid calculation. In recognition, we analyzed the moving hand path by line, circle, and ellipse approximation. The proposed algorithm can be applied to the multi-view 3D synthesis display system and the defined gestures can be used for 3D model view control. The overall accuracy was 92.58% with over a 150 Hz gesture report rate. Furthermore, the swipe identification showed an accuracy of 84.5% on the 4 swipe gestures of the challenging DHG-14/28 dataset.