Research On Hand Gesture Tracking And Recognition Based On Monocular Vision

The video image of hand gesture tracking and recognition technology receives much concern in the field of computer vision in recent years. As one of the most natural and common mode, hand gesture can convey information with more vivid and intuitive manner. Vision-based hand gesture recognition has a wide range of applications, such as human computer interaction, virtual reality, medical diagnosis. This paper mainly studies hand gesture tracking and recognition technology based on monocular camera. The content is as follows:Firstly, we adopt method of fixed background subtraction and mixed Gaussian skin model to segment out hand regions. According to false detection and miss detection caused by light, shadow and skin color liked noise, we proposed an improved algorithm of hand region growth. The method uses a special color space in which the brightness and color of skin color are greatly independent to fulfill the leak that caused by light and shadow, and judges the area size to reduce the influence of the skin liked noise. The experimental result shows that the improved method can effectively segment out the hand region of complete and clean.Secondly, we proposed a fast method to extraction hand contour based on corner detection and gradient vector flow, which through corner detection to initial contour to and calculate the vector flow only in the motion region. Experimental results show that the improved method can extract hand contour more robust and faster.Finally, we use sparse optical flow method to tack hand contour with corner features, and update corner features by gradient vector flow snake model to capture the real contour and detect the new corners on the curve. Through the three tracking algorithm contrast experiments showed that the proposed contour is more robust and accuracy. We define four gestures contour mode then give the detail progress of the hand gesture recognition algorithm.