基于俯仰角修正的HOG特征快速行人检测算法  

Pedestrian quick detection algorithm based on with pitch angle corrected HOG features

作　　者：曹巧慧 葛万成[1] 

机构地区：[1]同济大学电子与信息工程学院,上海200092

出　　处：《现代电子技术》2014年第21期62-65,共4页Modern Electronics Technique

基　　金：德国罗德与施瓦茨公司资助

摘　　要：传统的HOG特征对正视或侧视行人有较好的识别率,但是对俯视行人的识别率仍有所欠缺。对检测图像的HOG特征根据不同的俯仰角进行了转换,同时优化了SVM分类器训练过程,提出了一种改进的快速行人检测算法。测试结果表明,该算法优于基于传统HOG特征的检测方法,有效提高了不同俯仰角视频中行人检测的准确性。Traditional HOG feature detection method has a preferable recognition rate for envisaging and side-looking pedestrians,but has a low recognition rate for overlooking pedestrians. An improved pedestrian quick detection algorithm is proposed in this paper,which is realized by converting HOG features in the detected images according to pitch angle and optimizing the SVM classifier training process. The testing result shows that the algorithm is better than the detection methods based on the traditional HOG features and has improved the detection veracity of pedestrians in the different pitch angle video images.

关 键 词：快速行人检测算法 HOG特征 俯仰角修正 SVM 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]