模板匹配的三维手势识别算法

Template Matching Algorithm for 3D Gesture Recognition

针对三维空间的手势识别,基于轻量级的二维平面手势识别算法进行了扩展及改进,将其扩展到了三维空间,提出了基于投影的单次旋转模板匹配识别方法.将手势模板集划分为空间平面手势和空间立体手势2类模板;对于空间平面手势模板,输入手势轨迹需要先投影到相应的坐标平面,再进入模板匹配流程;基于矩阵奇异值分解,计算归一化手势轨迹与模板之间的最优旋转矩阵,经过一次刚体转置即可完成与模板的匹配度计算.实验结果表明,该方法对三维手势具有较高的识别率,且对于空间平面手势、空间立体手势的识别能力较为均衡.

This paper extended and improved the algorithm, a lightweight planar 2D hand gesture recognition algorithm, and realized a 3D gesture recognition algorithm based on projection and single rotation. The gesture template library was divided into two types： spatial planar gesture and spatial stereo gesture. For the templates of the spatial planar gesture, the input gesture trajectory needed to be projected onto the corresponding coordinate plane before the template matching. Based on the matrix singular value decomposition method, the optimal transposing rotation matrix between the normalized input gesture and the templates was calculated. With the transposing matrix, only a rigid transpose operation was needed for matching degree calculation of a template. Experiments show that this method has a high recognition rate for 3D gesture recognition, and the recognition ability is more balanced for 3D plane and stereo gestures.