基于深度信息的实时手势识别和虚拟书写系统

鉴于无接触体感交互技术在人机交互领域的成功应用，提出了一种基于Kinect深度相机的实时隔空虚拟书写方法。结合颜色和深度数据检测和分割出手掌区域；进一步，通过修改的圆扫描转换算法获得手指的个数，以识别不同的手势指令；根据指尖检测从指尖的运动轨迹分割出独立的字符或汉字运动轨迹，并采用随机森林算法识别该字符或汉字。这种基于深度信息的手势检测和虚拟书写方法可以克服光照和肤色重叠的影响，可靠实时地检测和识别手势和隔空书写的文字，其识别率达到93.25%，识别速度达到25 frame/s。

Depth-based real-time hand gesture recognition and virtual writing systems

Inspired by the successful application of the contactless somatosensory interaction technology in the field of human-
computer interaction, this paper presents a real-time Kinect depth camera based contactless virtual hand writing technique. In the technique, hand regions are detected and segmented utilizing both color and depth information. Hand fingers are rapidly recognized using a modified scan conversion method of a circle in order to recognize different hand gestures. It uses tracking by detection of a fingertip to obtain the fingertip trajectory and segment out individual trajectories for English and Chinese characters. Further a random forests algorithm is used to identify the correct characters from these individual trajectories. This depth information based hand gesture recognition and virtual hand writing technique can overcome the effects of light and color overlap and make the real-time detection and identification of gestures and hand writing more robust and reliable. The recognition accuracy?can arrive at 93.25% and the speed of the recognition is about 25 frame/s.