Abstract: This paper proposes a method for gait representation based on body’s centroid and outline’s key points. Firstly a three-frame difference method is given to detect moving targets. Then the body centroid, the edge points of feet and head are found, extracting four parameters of each frame：the height of centroid, the distance between body centroid and the edge point of head, the distance between body centroid and the edge point of left foot, the distance between body centroid and the edge point of right foot. It selects a cycle of gait sequence, using automatic warping algorithm Dynamic Time Warping（DTW） for recognition. In terms of computational cost, less dimension parameters reduce the complexity of the algorithm. Unlike BP, HMM trains the model parameters first, this algorithm of gait recognition uses template matching directly, guarantees the real-time performance of gait recognition. Experimental results show that the algorithm not only wins encouraging recognition performance, but also proves robust for objects’ clothes to a certain extent.

Key words: biometrics feature recognition, gait recognition, body centroid, key point of silhouette, Dynamic Time Warping（DTW）

摘要： 提出了一种基于人体的质心和轮廓关键点的步态表示方法，用三帧差分法对运动目标进行检测，然后确定人体的质心和外轮廓上头顶、双足的边缘点的位置，对每一帧图像提取4个特征参数：质心高度、质心到头顶边缘点的距离、质心到左足边缘点的距离、质心到右足边缘点的距离。选取一个周期的步态特征用自动规整算法（Dynamic Time Warping，DTW）进行识别。就计算代价而言，步态特征参数维数少，降低了算法复杂度，步态识别阶段所用算法不像BP、HMM等算法要进行模型参数的训练，而是直接针对提取的特征进行模板匹配，可以保证步态识别的实时性。实验结果表明该算法不仅获得了令人鼓舞的识别性能，并且对衣着具有一定的鲁棒性。

关键词: 生物特征识别, 步态识别, 质心, 轮廓关键点, 动态时间规整（DTW）