A Novel Multi-view Framework for 3D Multiple People Detection, Location and Labeling in Crowded Scenes
In this paper, we address the problems of multiple people detection, location and labeling under a multi-view set. We are interested in situations where the crowds are dense and occlusions are common. In this case, occlusion and lack of visibility make these problems very difficult to deal with. A novel framework is proposed, under which the problems of detection, location and labeling are all handled. The proposed framework has a robust performance against occlusion and can be implemented in real time. The core of our approach is the space field model which is a process of fast rough 3D reconstruction. By analyzing the result of the space field model, we detect and locate human in 3D space. Then we detect and label them across views in 2D image space using a detection-labeling structure.
Multiview detection location labeling
Jian Xu Xianglong Liu Xiaoli Zhang YanXu Huixing Jia
China Academy of Transportation Sciences Beijing.China ABB Reseach Center Shanghai.China Beijing Safevision Scientific Limited.Corp.Beijing,China
国际会议
太原
英文
203-209
2011-02-26（万方平台首次上网日期，不代表论文的发表时间）