基于多参数模型结合的步态特征提取新方法  

A New Technique of Gait Feature Extraction Based on Multi-parameter Model

作　　者：薛召军[1] 万柏坤[1] 刘晓辉[1] 明东[1] 靳世久[1] 

机构地区：[1]天津大学精密仪器与光电子工程学院,天津300072

出　　处：《中国生物医学工程学报》2009年第1期22-26,共5页Chinese Journal of Biomedical Engineering

基　　金：国家自然科学基金资助项目(60501005);"十一五"国家高技术研究发展计划(2007AA04Z236);天津市科技支撑计划重点项目(07ZCKFSF01300)

摘　　要：为弥补单一模型识别能力的不足,削弱因步速、衣着、光照等变化的影响,提出小波分解(WD)、不变矩(IM)并结合骨架理论(ST)提取步态特征参数的新方法。其技术流程为:先分割出人体目标轮廓,并将其规格化叠加处理,获取步态特征图;再将小波分解与不变矩结合,提取含人体整体模型信息的矩参数;同时将步态特征图骨架化,提取含人体简化模型信息的骨架特征参数;最后将矩参数与骨架参数作为识别参量,输入支持向量机(SVM)进行步态识别。使用自建的天津大学红外步态数据库(TIGD)进行试验,其正确识别率为84%～92%。表明多参数模型相结合的方法有利于提取步态的本质结构特征。In order to improve the recognition capability with single model and restrain the impact of noise （walking speed, clothing, illumination, etc） in gait recognition, a novel technique of feature extraction was presented for gait parameters in this paper. This method was based on wavelet decomposition（WD）, invariant moments （IM） and skeleton theory（ST）. Body silhouette sequences were extracted and normalized. The sequences were added together and the gait feature image could be achieved. The moment parameters with information of integral model were obtained by using wavelet decomposition and invariant moments. The skeleton was extracted from the gait feature image. Parameters of skeleton involving simplified model were extracted. These parameters, including invariant moments and skeleton, were given to support vector machines （SVM） for gait recognition. This method was applied to Tianjin University Infrared Gait Data-set （TIGD） and achieved recognition rate of 84 - 92%. Results proved that this method would benefit extracting the gait essential feature.

关 键 词：步态识别 小波分解 不变矩 骨架理论 支持向量机 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]