基于指尖角度集核密度估计手势特征提取

GESTURE FEATURE EXTRACTION BASED ON KERNEL DENSITY ESTIMATION OF FINGERTIP ANGLE SET

针对手势识别实时性和鲁棒性不足的问题,提出基于指尖角度集核密度估计的特征提取方法。通过对一般手势定义指尖角度集并提取指尖角度集核密度估计特征。为解决形状匹配的相位漂移问题和进一步提高实时性,对该特征集有效区间归一化和均匀采样得到指尖角度集核密度估计序列。基于互相关系数形状匹配算法进行手势识别。实验分析表明,采用该特征提取方法的任意手势识别实时性和鲁棒性比现有方法有显著提高。

We propose the hand gesture feature extraction method, which is based on kernel density estimation of fingertip angle set, for the problems of gesture recognition in lacking robustness and real-time property. For common gestures, the method extracts kernel density estimation of fingertip angle set by defining fingertip angle set. In order to solve the phase shift of shape matching and to further increase the real-time property, the method normalises the effective range of feature set and samples uniformly so as to obtain the estimation sequence of kernel density of fingertip angle set. The hand gesture recognition is achieved based on the shape matching of cross-correlation coefficient. It is demonstrate by the experiments that the real-time performance and robustness in arbitrary gesture recognition using the proposed feature extraction method are significantly improved compared with current method.