基于时空变化的步态识别算法研究

Research on Motion Change-Based Gait Recognition Algorithm

本文提出了一种八椭圆人体模型,并在此基础上提出了基于变化信息的步态识别算法。对每个视频序列,采用基于贝叶斯规则的检测算法检测出目标人体区域;将目标人体区域按比例划分为八个区域并分别用椭圆拟合,建立人体椭圆模型;用人体姿势的时空变化——相邻帧间的模型参数的差值作为特征,用归一化后的Mahalanobis距离和Euclidean距离分别进行相似性度量,NN和KNN技术用于最终的分类。实验结果表明,该算法拥有较高的识别率和较低的计算代价。

This paper proposes a 8-ellipse human body model, and based on this model a motion change-based gait recognition for human identification is proposed. For any given video sequence, the object human body region will be detected precisely by a method based on Bayes rules and then the body region is divided into eight parts which are fitted by different ellipses, then the human body model can be established. Using the amount of motion change as the feature, which can be gained by subtracting the values of model parameters of the prior frame from the current frame, the comparability measure is achieved by the normalizedMahalanobis Distance or Euclidean Distance. NN and KNN classifiers are used for the human identification. Experimental results demonstrate the higher performance and lower computational complexity.