融合步态和人脸特征的远距离身份识别研究 .pdfVIP

学兔兔 第32卷 第2期 仪 器 仪 表 学 报 Vol_32 No．2 2011年2月 Chinese Journal of Scientific Instrument Feb．201l 融合步态和人脸特征的远距离身份识别研究术 李 轶，明 东，王 璐，綦宏志，万柏坤 (天津大学精密仪器与光电子工程学院 天津 300072) 摘 要：在单一摄像头获得的步态图像序列中将步态信息和人脸信息相融合进行实时的远距离身份识别。在步态图像序列中 自动提取侧面人脸图像。采用基于傅里叶描绘子和关键点特征的方法提取步态特征，采用傅里叶变换和奇异值分解的人脸识 别方法对步态图像序列中的侧面人脸进行特征提取。利用欧氏距离作为度量建立匹配函数。在中科院自动化研究所CASIA 步态数据库中进行实验，通过对单独利用步态特征和人脸特征进行识别的识别率和通过最大法则、加法法则和乘法法则融合后 进行识别的识别率进行比较，实验表明将步态和人脸特征相融合可有效地提高了识别率，识别率可达95．00％。 关键词：步态识别；人脸识别；傅里叶变换；奇异值分解；融合算法 中图分类号：TP391 文献标识码：A 国家标准学科分类代码：520．2040 Gait and face profile fusion based human identification at a distance in video IJi Yi，Ming Dong，Wang Lu，Qi Hongzhi，Wan Baikun (College ofPrecision lmt~ments and Optoelectronics Engineering，rianjin University，Tianfin 300072，China) Abstract：A method of recognizing individuals at a distance through fusing gait and face profile cues from a gait im— age sequence is presented in this paper．Face profile images are extracted automatically from the gait image se- quence．A Fourier transform based algorithm is adopted for gait feature extraction．For face feature extraction，an al— gorithm based on Fourier transform and singular value decomposition(SVD)is used．Euclidean distance is adopted for classification．The fusion methods of sum rule，product rule and max rule were tested in the CASIA gait data— base．Comparison is carried out for the recognition rates obtained using gait or face profile feature only and using the fusion feature，and experiment results show that the recognition performance is improved by fusing gait and face pro— file．The recognition rate reaches 95．00％． Key words：gait recognition；face r