面向静态手势识别的边缘序列递归模型算法

Gesture Recognition Algorithm Based on Contour Sequence Recursive Model

针对传统手势识别准确率不高、鲁棒性不强的问题,通过研究静态手势轮廓特征,从手势边缘序列角度出发提出一种基于手势边缘轮廓递归图的CK-1距离的手势识别算法CSRP.首先通过阈值分割获取手势区域图像;然后定位起始点坐标,建立随着空间位置变化的手势边缘序列;为了克服边缘序列数据的不等长问题,构造基于时空域的手势轮廓序列递归图;最后利用MPEG-1压缩算法计算手势递归图像之间的CK-1距离,完成手势识别.实验结果表明,该算法在手势发生旋转、平移、缩放时具有较高的鲁棒性,并且计算量小、效率高,手势识别的准确率高达97%.

To improve the accuracy and robustness,a novel algorithm CSRP is proposed based on compressiondistance of recurrence patterns extracted from the perspective of edge sequence.Firstly,the regionalimages of the hand with fingertips are obtained by threshold segmentation,and then,the edge sequences ofpalms are built following the change of the spatial position from the start point;In order to overcome theproblem of unequal length of the edge sequences,recurrence plots in spatial domain are established;Finally,CK-1distance is calculated by using the MPEG-1video compression algorithm for gesture recognition.Theexperimental results show that the proposed algorithm achieves high robustness in the case of rotation,translation and scaling,with97%recognition accuracy.