基于运动和外形特征的人体行为识别.pdfVIP

第36 卷 第5 期 计 算 机 工 程 2010 年3 月 Vol.36 No.5 Computer Engineering March 2010 ·人工智能及识别技术· 文章编号：1000—3428(2010)05—0193—03 文献标识码：A 中图分类号：TN911.73 基于运动与外形特征的人体行为识别 黄先锋，张 彤，莫建文，袁 华，欧阳宁 (桂林电子科技大学信息与通信学院，桂林 541004) 摘 要：多数现有特征提取方法仅采用简单的形态特征，存在走与跑识别率较低的问题。将运动速度特征与较精确分割并归一化图像大小 后的主分量分析外形特征相结合，采用支持向量机从8 个方向对跑、蹲、站、弯腰、招手、指和走7 种人体行为进行识别，结果证明走与 跑的识别率得到很大提高。 关键词：行为识别；计算机视觉；支持向量机；主分量分析 Human Behavior Recognition Based on Characteristics of Movement and Shape HUANG Xian-feng, ZHANG Tong, MO Jian-wen, YUAN Hua, OUYANG Ning (Information Communication College, Guilin University of Electronic Technology, Guilin 541004) 【Abstract 】Most of the existing characteristic extraction methods just use simple shape characteristics and exist problem of low walking and running recognition rate. This paper combines the velocity characteristics of movement and the Principal Component Analysis(PCA) shape characteristics obtained after more accurate segmentation and unifying the size of images. It uses Support Vector Machine(SVM) to recognize seven kinds of human behaviors including running, squat, standing, bending, waving, directing and walking from eight directions. Experimental results show that walking and running get higher recognition rate. 【Key words 】behavior recognition; computer vision; Support Vector Machine(SVM); Principal Component Analysis(PCA) [1] 现有人体行为识别方法一般使用隐马尔科夫模型 ，虽 1.2 特征提取 然支持向量机(Support Vector Machine, SVM)在行为识别中 分别从每个较精确的分割后样本视频中提取样本，每个