基于光流空间分布的步态识别方法

摘要

Concerning the disadvantages of the traditional gait recognition method based on optical flow,such as complex system and low recognition rate.This paper proposed an improved model-free method—the spatial distribution of the optical flow to descript and recognize moving target.First,it computed dense optical flow for each image in a sequence and derive scale-independent scalar features which characterized the spatial distribution of the flow.Then,it analyzed periodic structure of these sequences of scalars.The scalar sequences for an image sequence has the same fundamental period but differ in phase.The phase feature vectors could be used to recognize individuals.Lastly,for each sample,trained the average of feature vectors as cluster centers.Sequences were classified to the nearest class based on the nearest neighbor rule.The experiment results show that,in the CASIA gait database,90％ recognition rate or higher can be reached.%针对传统基于光流法步态识别复杂、识别率不高的缺点,提出了一种非模型化的方法——光流空间分布来描述并识别运动目标.首先,计算每帧步态序列中的密度光流场,所得的与尺度无关的矩描述了光流的空间形状分布；然后,分析每一组矩的周期性结构特征,不同图像序列对应的矢量有基本相同的周期特征和不同的相位特征,利用相位特征区分不同个体步态之间的差异；最后,训练时计算各个样本特征矢量的平均值作为聚类中心,识别时计算待识别序列矢量和每个聚类中心的距离,采用最近邻法则,把序列归类到距离最近的类中.实验证明,该算法在CASIA步态数据库上最高能达到90％以上的识别率.