基于目标分割与SVM的人数统计

People Counting Based on Object Segmentation and SVM

该文实现了垂直摄像头下的实时人数统计。为减少场景中伪目标区域对人头检测的影响,首先利用背景建模法提取前景目标,并根据发色信息筛选前景目标。然后采用线性SVM训练得到的头部分类器识别头部,并将头部中心点作为运动人体的特征点,利用最近邻匹配法进行数据关联,完成行人的跟踪计数。不同场景下的视频测试结果表明,该方法能较准确地实现人数统计。

This paper achieves real-time people counting in video sequences captured with vertical camera . To reduce the impact for head detection by spurious objects , firstly, using background modeling method to extract foreground objects , and for further select based on hair color information , then using linear support vector machine（SVM） training a classifier to recognize the head , and the head center as a feature point of the moving for data association by the nearest neighbor matching method , to complete pedestrians track and count . Testing in different scenes , the results show that this method can achieve people counting accurately .