不变矩系数拟合的步态识别  

Gait Recognition by Fitting Moment Invariants

作　　者：袁海军[1] 文玉梅[1] 李平[1] 叶波[1] 

机构地区：[1]重庆大学光电技术及系统教育部重点实验室

出　　处：《计算机应用》2007年第4期922-924,928,共4页journal of Computer Applications

基　　金：重庆市自然科学基金资助项目(CSTC2006BB2155)

摘　　要：根据综合利用步态的静态和动态信息的思想,结合不变矩描述图像几何特性的功能,从步态序列提取不变矩作为步态特征进行识别。采用傅立叶级数描述步态图像序列人体轮廓不变矩的变化,利用遗传算法搜索傅立叶级数的系数,最后再用k近邻分类器对不变矩变化的幅度信息分类。在CMU步态数据库上进行的实验,达到了90%以上的识别率。结果表明,该方法具备很高识别性能,能较好地利用步态的静态和动态信息。According to the idea of combining static components and dynamic components from the walking way, as the moment invariants may represent geometrical traits in images, they were extracted as gait features from a subsequent gait series. The moment invariants of human silhouettes were represented by the Fourier series. A genetic algorithm was deployed to obtain the Fourier coefficients. The magnitudes of the coefficients were classified through the kNN classifier. The recognition restdts with the proposed scheme in the CMU gait database have recognition rates of more than 90% and show that it has achieved a high performance and effectively made use of two kinds of contents.

关 键 词：特征提取 步态识别 不变矩 遗传算法 KNN分类器 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]