基于Kinect深度图像信息的手势跟踪与识别

Research on gesture tracking and recognition based on Kinect depth data

针对基于视觉的手势识别技术对环境背景要求较高的问题,提出了一种利用深度信息进行手势提取和识别的研究方案。采用Kinect深度摄像头,通过中值滤波以及深度信息与邻域特点来分割手部区域并用Canny算子提取出手势轮廓,再以深度图像的凸缺陷指尖来完成对指尖的检测,从而实现对数字手势1到5的手势识别。该方法可快速有效地对指尖进行检测,鲁棒性和稳定性都比其他方法更好。实验结果表明,该手势识别方案的平均识别率达到92%,证明了该方法的可行性。

Aiming at the problem that gesture recognition technology based on vision requires a lot on enviroment and background, a gesture extraction and recognition scheme based on depth data is presented in this paper. By Kinect, the hand is segmented through the median filtering, the depth information and neighborhood features, and Canny operator is used to extract the hand contour. Then fingertip detection is completed by depth-based convex defects detection, implementing gesture recognition for number gesture 1 to 5. This method can quickly and effectively detect the fingertip, robustness and stability are better than other methods. Experimental results show that the average recognition rate of this scheme is 92%, which proves the feasibility of the method.