摘要： 为有效抑制观察视角及鞋帽服饰等外界因素的干扰，克服目前常用整体模型步态识别算法的不足，提出将人体轮廓面积特征与支持向量机分类器相结合的识别方法。该方法在步态序列图像的人体轮廓进行提取和规格化，将轮廓图叠加后进行网格式划分，提取轮廓单元模块面积作为步态特征识别参量。使用南佛罗里达大学的步态数据库，分别采用线性、多项式和径向基内核函数对5种不同外界因素条件下的数据进行实验，该方法的正确识别率为82%～100%，且对视角及鞋帽服饰的干扰不敏感，具有更强的鲁棒性。实验表明人体轮廓面积更能反映步态特征，将该面积特征与SVM分类相结合可以获得更好的识别性能。

关键词: 生物特征识别, 步态识别, 人体轮廓, 面积特征, 支持向量机

Abstract: In order to decrease the noises (view, shoe, clothes and etc) and overcome the defects of existing recognition methods, the new technique based on area feature combining Support Vector Machine (SVM) is propseed. Body silhouette sequences of gait are extracted and normalized. The sequences are added together and images of addition are separated into segments. Area features are extracted and used as parameters for gait recognition. Three kernel functions responding to linear, polynomial and Radial Basis Function (RBF) are used in the experiments. This method is applied to USF gait data-set and achieves recognition rate of 82%～100%. SVM recognition is insensitive for noises of view and shoe. Results show robust stability and good performance. Area features present better essential characters of gait. Experimental results show that SVM can function as an efficient gait classifier for recognition.

Key words: biometrics recognition, gait recognition, body silhouette, area feature, support vector machine

中图分类号: 

• TP391