基于多特征的行人检测算法  

Pedestrian detection method based on multi-features

作　　者：胡庆新[1] 张春阳[1] 方静[1] 

机构地区：[1]合肥工业大学计算机与信息学院,合肥230009

出　　处：《计算机应用研究》2014年第10期3161-3164,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(61102154)

摘　　要：针对单特征辨识度较低的问题,提出一种基于多特征的行人检测方法。首先构造二维对数Gabor函数,利用此函数提取样本的相位一致性特征,将样本的相位一致性特征和样本的局部二值模式算子(LBP)特征相结合,得到新的行人检测方法。使用支持向量机(SVM)进行分类,并与基于HOG特征的检测方法进行比较,在INRIA行人数据库上的实验证明,基于多特征的检测方法提高了行人检测精度、降低了误检率,检测率高达99.4824%。In view of the problem of the low degree of single-feature recognizability in pedestrian detection, this paper pro- posed a new pedestrian detection method based on multi-feature. Firstly it constructed a two-dimensional log-Gabor function and extracted the phase congruency feature from images using the log-Gabor function. Secondly it combined the phase congru- ency feature with local binary pattern （LBP） feature to obtain a new pedestrian detection algorithm. The support vector ma- chine （SVM） was used to train the pedestrian classifier. Compared with the single-feature pedestrian detection method based on HOG on the INRIA dataset, the multi-feature detection method is more discriminative and can get higher detection rate. The accuracy of the new method is 99. 4824%.

关 键 词：行人检测 相位一致性特征 方向梯度直方图 局部二值模式算子 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]