基于AdaBoost-SVM级联分类器的行人检测

Pedestrian detection based on cascade AdaBoost-SVM

针对实时行人检测中AdaBoost级联分类算法存在的问题,改进AdaBoost级联分类器的训练算法,提出了Ada-Boost-SVM级联分类算法,它结合了AdaBoost和SVM两种算法的优点。对自定义样本集和PET图像库进行行人检测实验,实验中选择固定大小的窗口作为候选区域并利用类Haar矩形特征进行特征提取,通过AdaBoost-SVM级联分类器进行分类。实验结果表明AdaBoost-SVM级联分类器的分类器准确率达到99.5%,误报率低于0.05%,优于AdaBoost级联分类器,训练时间要远远小于SVM分类器。

For the problem of the cascade AdaBoost classification algorithm which is used in the real time pedestrian detection, the training algorithm of cascade AdaBoost classifier is improved and a AdaBoost-SVM cascade classification algorithm is proposed, which combined the advantages of the AdaBoost algorithm with that of SVM. The pedestrian detection experiment with the database of captured samples and PET database is conducted. The candidate areas with a window of fixed size are captured for Haar-like rectangle feature extracting in the experiment, and then two classes classification is performed by using the proposed cascade AdaBoost-SVM classifier. The experimental result shows that the cascade classifier proposed by us can get better performance than cascade AdaBoost classifier and its accuracy can reach 99.5o//00 and the false alarm rate is less than 0. 05%. Compared with SVM classifier, the training time is greatly reduced.