基于统计结构梯度特征的行人检测.pdfVIP

2012年第 07期，第 45卷 通 信 技 术 Vol.45，No.07,2012 总第 247期 Communications Technology No.247,Totally 基于统计结构梯度特征的行人检测 定志锋， 戴声奎， 陈美龙 （国立华侨大学信息科学与工程学院，福建 厦门 361021） 【摘 要】针对自然背景下的行人检测问题，从人体最显著的轮廓信息出发，提出一种基于统计结构 梯度（SSG, Statistical Structural Gradient）的行人检测算法。采用支持向量机（SVM, Support Vector Machine）算法对数据库中行人样本的 SSG 特征进行训练，得到行人检测分类器。提出的 SSG 算法在 sobel 图像中仅做结构统计和角度幅值计算，能够采用积分图加速行人检测速度。对 INRIA 数据库中的 288 张样 本进行测试，误检率(FPPW, False Positives Per Window)为 10-5 ，检测率等于 90.89%。 【关键词】边缘信息; 统计结构梯度; SSG; SVM 【中图分类号】TP391.41         【文献标识码】A         【文章编号】1002-0802(2012)07-0075-03  Pedestrian Detection based on Statistical Structure Gradient DING Zhi-feng， DAI Sheng-kui， CHEN Mei-long (School of Information Science and Technology, Huaqiao University, Xiamen Fujian 362021, China) 【Abstract】For the pedestrian detection under natural background, and from the bodys essential characteristics of contour information, a pedestrian detection algorithm based on statistical structure gradient (SSG) is proposed. With support vector machine (SVM) algorithm, the SSG feature of pedestrian sample in INRIA database is trained, thus to get pedestrian detection classifier. The SSG algorithm is only applied for structure statistics and angular amplitude calculation on sobel image, therefore, integral technology is employed to accelerate detection of pedestrians. Testing on 288 samples from INRIA database indicates that the false detection rate is 10-5 FPPW conditions with a detection rate of 90.89%. 【Key words】contour; statistical structure gradient; SSG