用时变模板自动识别行人的步态

提出了一种以行人轮廓随时间变化的灰度图像为模板的步态识别方法．首先提取行人二值轮廓序列;然后针对轮廓的关键运动区域分析边缘点分布直方图的变化，检测出包含两个单步的步态周期;随后对单步范围内的轮廓序列，经帧间轮廓前向运动区域叠加产生单步运动历史图像，从而将三维信息表示到二维图像上;继而用一组同心矩形分割两个单频运动历史图像，提取出局部性的矩统计量作为步态特征向量，最终实现了步态识别．在Soton数据库上进行了实验，这种算法的正确识别率可达85. 57％，与相关文献的对比表明该算法优于现有算法．

Automatic human gait recognition using temporal templates

A novel temporal templates representation method for gait analysis and recognition applications is introduced.The method includes following steps: first,silhouette extraction is performed to the images of the video sequence;secondly,the gait cycle including two steps is detected by analysizing the contour points distribution of silhouette sequence; thirdly,by cumulating interframes forwards sihouette differences,two single step history images(SSHI) are obtained to represent how and where human walking is evolved.The SSHI image is then segmented by a family of rectangles,from which a set of local moment invariant vectors are abstracted to be the feature vectors of the SSHIs.These feature vectors are finally used to achieve gait recognition.Recognition capability is illustrated by an 85.57% CCR on Soton database and the results show that the method outperforms the existing methods.