说明：  基于梯度方向直方图( H OG) 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提高、向量维数大的问题, 提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 提高检测精度。
(H istog r am o f or iented g radient( H OG) based on pedestr ian detect ion is the popular method w ith t he hig hest detection rat e curr ent ly. To further im pr ove its detection r ate and decrease its larg e dimensions o f feat ur es, this paper in􀀁 tegr ates HOG w ith color frequent and skin colo r feature to descr ibe pedestr ian, w e select some blo ck features w ith better classify ing ability as t he final featur e, then use linear SVM as our classifier. T he ex per iment o n INRIA demo nstr ates t he hig h detection r ate of our met ho d.)