为了满足智能监控系统自动、准确、实时识别行人的要求，提出了一种表示方法简单、计算复杂度低的步态识别算法。首先建立环境的高斯背景模型，从步态视频序列中提取轮廓图像，计算质心以及轮廓线上的点到质心的欧氏距离。再将轮廓线以最高点为起点顺时针展开，将2维轮廓线特征转换为1维距离特征并标准化，建立标准步态模型。然后计算训练序列与标准步态模型之间的动态时间规整距离，确定阈值。最后，输入测试序列，计算动态时间规整距离并与阈值比较，识别人体的步态。与常用步态识别方法相比，该方法兼顾了计算复杂度和识别率，符合智能监控系统的性能要求。

This paper presents a new approach to gait identification and authentication with simple representation and lower computational complexity, which can meet intelligent surveillance’s need in precision and response. It creates Gaussian Mixture Model for each scenario, and contour of gait is extracted from binary silhouette for Euclidean distance between the centroid and any pixel on it. Contour is unfolded clockwise by the distance from the uppermost pixel, and then 2D features are transformed into 1D and normalized according to a standard model of gait. Thresholds are determined by dynamic time warping (DTW) distance between training sequences and standard model. Finally, gait recognition is performed by comparing DTW distance of testing sequences with predetermined threshold. Compared with other methods, it balances both computational cost and recognition rate, and achieves performance of intelligent surveillance.