Application Research Of Hand Recognition Based On Monocular Vision

With the rapid development of human-computer interaction, new interactive way has become a research hotspot in recent years. As a natural and intuitive way of communication, gestures play an important role in human-computer interaction. Specifically, vision-based gesture recognition has always been a very active and challenging research topic in the field of computer vision and is attracted widespread attention at home and abroad.Based on inductive analysis in gesture recognition technology at home and abroad, we choose to do some research in monocular visual gesture recognition. The reason is that binocular vision-based gesture recognition can achieve better recognition effects, but the hardware requirements is high and the promotion is not strong at the same time. However, the gesture recognition based on monocular vision is low-cost and very versatile although it is vulnerable to the constraints of reality. Therefore, we study the algorithm correspond to hand detection, hand tracking and hand recognition from the perspective of practical applications. The major work and innovation are as follows:In hand detection, the cascade classifier object detection framework proposed by Viola and Jones is utilized against the complex background and lighting conditions in real environment. This method can effectively overcome the effects of light on the background and effectively reduced false detection rate in Viola-Jones detector by combining with the color information. The detection results can be target initialization in hand automatic tracking.In hand tracking, in order to solve the problem that traditional target tracking methods are easy to lose track in the case of skin color background, excessive movement and gesture shape changing, we research and analyze a method named Local-Global Tracking (LGT). LGT tracking method uses two-layer model for mutual restraint and remains dynamic update in color, motion and shape features while tracking, suitable for tracking significant appearance changed target, but the time of color update is long when use this method in hand tracking. By introducing new color probability, we propose an improved LGT method which shortens the tracking time and ensures the tracking results. This method admirably solved tracking lost problem in gesture tracking.In hand recognition, the binary images of gestures are obtained by color segmentation after positioning gestures, then we extract the Fourier descriptors and Hu moment features of hand shape. At last, BP neural network is used as a classifier and we implement a robot hand gesture recognition system.After thorough study of the algorithm principle in this article and experimental results show that the adopted gesture detection, gesture tracking and gesture recognition program has higher accuracy and robustness in real environments.