基于人体手臂关节信息的非接触式手势识别方法

为了实现在复杂环境中对连续动态手势的识别,以人体固有的手臂关节之间的约束关系及特定手势在三维空间的运动轨迹为特征,提出了一种非接触式手势识别方法. 首先,通过Kinect传感器获取人体手臂关节的三维数据;然后,对手势轨迹进行分割,并将具有三维空间特征的手势轨迹转化为一维的手势轨迹;最后,将手势预判断过程与改进的动态时间规整( dynamic time warp, DTW)算法相结合,实现对动态手势的快速高效识别. 实验结果表明:该方法对具有时空连续特征的动态手势识别率较高,在复杂背景和不同光照环境中都有较强的鲁棒性.

Non-contact Gesture Recognition Method Based on the Information of the Human Arm Joint

To realize the continuous and dynamic gesture recognition in complex environment, an approach for non-contact hand gesture recognition method was proposed by using the inherent constraint relationship in the arm joints and the specific gesture track characteristics in three-dimensional space. First, a three-dimensional data of human arm joints from the Kinect sensor was obtained. Then, the gesture track with the three-dimensional space characteristics was segmented and converted to one-dimensional gesture track. Finally, gesture to preliminary judgment process with the improved dynamic time warp( DTW) algorithm was combined to achieve fast and efficient recognition of dynamic gestures. The experimental results show that the method of gesture track with the space-time characteristics has a high recognition rate and strong robustness under different light and complicated backgrounds.