基于边缘对称性和HOG的行人检测方法.pdf

文档介绍：计算机工程 2012 年 3 月
第 38 卷第 5 期
March 2012
Engineering

·人工智能及识别技术· 文章编号:1000—3428(2012)05—0179—04 文献标识码:A 中图分类号:

基于边缘对称性和 HOG 的行人检测方法
姚雪琴 a,李晓华 b,周激流 b
(四川大学 a. 电子信息学院;b. 计算机学院,成都 610065)
摘要:传统基于 HOG 特征的行人检测方法存在检测速度慢的问题。为此,提出一种基于边缘对称性和 HOG 的行人检测方法。利用对称
差分提取输入窗口的垂直边缘,根据垂直边缘的对称性快速检测出行人候选区,采用 HOG 特征和线性支持向量机对行人候选区进行验证。
实验结果表明,该方法在保持传统方法检测率的同时,能提高检测速度。
关键词:行人检测;对称差分;垂直边缘;边缘对称性;HOG 特征;线性支持向量机;十折交叉验证
Pedestrian Detection Method Based on Edge Symmetry and HOG
YAO Xue-qin a, LI Xiao-huab, ZHOU Ji-liub
(a. College of Electronics and Information Engineering; b. College puter Science, Sichuan University, Chengdu 610065, China)
【Abstract】The detection speed of traditional pedestrian detection method based on Histograms of Oriented Gradient(HOG) feature is slow. In
order to detect pedestrian faster, a pedestrian detection method based on edge symmetry and HOG is proposed. Symmetric difference is used to
extract the vertical edge of the input window. Candidate pedestrians can be detected rapidly according to the symmetry of the vertical edge.
Candidate pedestrians are validated by using HOG feature and linear Support Vector Machine(SVM). Experimental results show that the proposed
method can increase the detection speed while maintaining parable detection rate to the traditional method based on HOG feature.
【Key word