多形状模板增强的远距离步态识别算法  

Gait Recognition at a Distance with Enhanced Multiple Shape Templates

作　　者：马勤勇[1] 聂栋栋[2] 王毅[3] 

机构地区：[1]燕山大学信息科学与工程学院,河北秦皇岛066004 [2]燕山大学理学院,河北秦皇岛066004 [3]张家口供电公司,河北张家口075000

出　　处：《小型微型计算机系统》2014年第4期922-925,共4页Journal of Chinese Computer Systems

基　　金：河北省自然科学基金项目(A2011203053)资助;秦皇岛市科学技术研究与发展计划项目(2012021A044)资助

摘　　要：利用步态的统计特征可以有效地降低远距离复杂背景下提取的步态图像中的各种噪声干扰,然而现有算法在减少噪声的同时降低了步态的信息量,从而减少了用于分类识别的有效步态特征.为了解决这个问题,提出使用多形状模板来增强有效的步态信息.基于一个对象的步态轮廓图序列,分别计算出横向方向模板、纵向方向模板、形状分布模板和运动分布模板,作为此步态对象的特征.为了减小步态对象尺度变化对分类的影响,提出一种高效的缩放比率计算方式来对多形状模板进行缩放.在USF步态数据集上的实验表明,算法的综合识别率明显高于另外两个典型算法.Statistical features of gait can be used to render the algorithm less susceptible to the noises introduced by extracting silhou- ettes from gait images taken at a distance with complex background, but current algorithms reduce gait information with noises, result in valid gait features used to classify gait are reduced. To solve the problem, multiple shape templates are proposed to enhance valid gait information. Horizontal direction template, vertical direction template, shape distribution template and motion distribution tem- plate are calculated based on the gait image sequence of an object, and the four templates are used as the objects feature. To reduce the effect of the variation of object scale to the classification result, an efficient calculation of zoom ratio is proposed to zoom multiple shape templates. Experimental result shows that, the proposed algorithm achieves higher overall recognition rate than two other algo- rithms.

关 键 词：生物特征 步态表示 步态识别 多形状模板 步态分类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]