基于Kinect的空中手写系统的手指检测及手势识别

We proposed a virtual writing system called writing in the air system. Specifically, we introduce a new fingertip detection algorithm, and a set of simple but effective gestures and their recognition method for the intelligent input system. By integrating depth data and hand physical model, we propose a two-mode switching fingertip detection algorithm. The mode switch is determined by the finger-palm-arm position and angles and the fingertip position is detected accordingly. Additionally, hand gestures are similar to body language and easy to be understand and applied for the virtual writing system interaction. They are converted to commands like "delete" and "select". Compared with traditional gesture recognition methods, our method is more natural. Meanwhile, it does not have to segment the start and end point of gesture manually as the preprocessing step, and the motion of gesture will not be confused with the writing trajectory. Experiment results indicate that the fingertip detection rate improves more than 5% comparing with recent methods, and the average recognition rate of all hand gestures is 94.17%. Generally speaking, it can sufficiently meet the demand of daily use.