In this paper,we present a robust and efficient method to multiple humans tracking based on head detection.Methods of extracting background are failed due to the high density population and the serious problem of shade.The head in front of the video is detected by Viola and Jones AdaBoost cascade classifier based on Haar-like features,the back of the head is detected by Logistic regression based on head profile feature.After determining the pedestrian's head position,the appearance color histograms for the head are modeled,finally we use a particle filter algorithm to track the pedestrian's head.The experimental results demonstrate that the proposed method is capable of tracking humans effectively in high density crowds.%针对日常户外高密度遮挡人群,本文提出一种基于行人头部检测的高效、鲁棒的多人跟踪方法.由于高密度人群的遮挡问题严重,因此提取背景的方法不可行.通过基于Haar-like特征的Viola-Jones分类器对视频中行人的头部正面进行检测,同时通过基于头部轮廓特征的Logistic回归对视频中行人的头部背面进行检测.确定行人的头部位置后,提取基于颜色直方图的头部特征,最后使用粒子滤波跟踪行人的头部.实验表明本方法能够高效地跟踪高密度遮挡的人群.
展开▼
