基于HOG和颜色特征行人检测_曲永宇.pdf

33 　4 武　汉　理　工　大　学　学　报 Vol.33　No.4 2011 4 JOURNAL OF WUHAN UNIVERSITY OF TECHNOLOGY 　Apr.2011 DOI:10.3963/j.issn.1671-4431.2011.04.030 HOG 曲永宇,刘　清,郭建明,周生辉 (, 430070) 　:　基于梯度方向直方图(HOG)特征的行人检测是目前检测精度较高的主流方法。针对基于梯度直方图特征的 行人检测存在检测精度还有待 高、向量维数大的问题, 出使用梯度直方图统计特征加颜色频率和肤色特征描述行 人, 选取 一些分类能力较强的block 作为最后的特征, 使用线性SVM 分类。在INRIA 库上的实验证明, 该方法能有效地 高检测精度。 :　行人检测;　梯度直方图(HOG);　颜色特征;　block 选择 :　T 751.1 :　A :1671-4431(2011)04-0134-05 HOG and Color Based Pedestrian Dete tion QU Yong -y u, L I U Qing , GUO J ian -m ing , Z H OU S heng -hu i (School of Automation, Wuhan University of Technology, Wuhan 430070, China) Abstra t:　Histogram of oriented gradient(H OG)based on pedestrian detection is the popular method with the highest detection rate currently.To further im prove its detection rate and decrease its large dimensions of features, this paper in- tegrates HOG w ith color frequent and skin color feature to describe pedestrian, w e select some block features w ith better classifying ability as the final feature, then use linear SVM as our classifier.The experiment on IN RIA demonstrates the high detection rate of our method. Key words:　pedestrian detection;　histogram of oriented gradient(HOG);　color feature;　block selection [1] 、、, 。、、、, , 。、 , 。 [2-3] 3 :1 , ; [4-6] 2 ,; 3 ,,。 2005 DALA L (Histogram of Oriented Gradient, HOG) :2010-08-15. :(2009CDB403)(2010-Va-012). :(1987-), , .E-mail:quyongyu163@163.com 33 　4 　　　　　　, 　, , : HOG 135 [7] , HOG , ,, 。 , , 。DA LA L , 。:2006 Zhu Viola