基于Haar-like特征多分类器集成的行人检测

Algorithm of pedestrian detection with multiple classifiers ensemble based on Haar-like features

Haar-like特征主要用于人脸等目标识别领域,然而因行人流动方向不确定,监控视频拍摄到的行人有正面、侧面和背面三种情况,因此文中提出一种根据三种特征设计的多分类器集成算法。在训练其中一面样本分类器时,将另外两面样本和背景图像作为负样本,从而得到三个分类器,分别检测识别后再筛选统计目标。实验证明该算法具有良好的检测识别效果。

Haar-like features are mainly used in the fields of faces and other targets recognition. But due to the uncertainty of pedestrian flow direction,the pedestrian heads appear three cases of front,side and back in the surveillance video. This paper presents a multiple classifiers ensemble algorithm according to the three feature cases. When one feature classifier is trained,the other two feature images and the background images are used for negative samples. Thereby it makes three classifiers to detect the pedestrian targets,and then filter and count. The experiments show that the algorithm has good effect in detection and recognition.