Abstract: The drawbacks of pedestrian detection method based on Histograms of Oriented Gradient(HOG) are larger dimensions of features and slow detection speed. Aiming at this point, a method based on HOG of Region Of Interest(ROI) is proposed. HOG is calculated in four important regions which locate in head and limbs’ regions respectively. Through this method, dimensions of features are decreased effectively. Experimental results show this method speeds up detection process while maintaining comparably accuracy to the method based on HOG.

Key words: pedestrian detection, Histograms of Oriented Gradient(HOG), Region Of Interest(ROI), Support Vector Machine(SVM)

摘要： 针对以梯度方向直方图作为人体特征的行人检测存在向量维数较大、检测时间较长的问题，提出基于感兴趣区梯度方向直方图的行人检测方法，分别在头部及四肢等重点区域计算梯度方向直方图，有效地减少了向量维数。实验结果表明，该方法在检测率基本不变的情况下提高了检测速度。

关键词: 行人检测, 梯度方向直方图, 感兴趣区, 支持向量机

CLC Number: 

• TP391