People Detection and Tracking with World-Z Map from a Single Stereo Camera

In this paper we propose a new people tracking system that uses a single stereo camera fixed at a high position and observing the scene at an oblique angle. We introduce the notion of world-Z map and show that 3D people detection and segmentation can be performed efficiently in this map and outperforms significantly the methods based on depth or plan-view. Detection and tracking play complementary roles, we derive a probabilistic framework for tracking based on motion. The system successfully deals with very complex situations without loosing track of people in highly cluttered scenes for long observation periods and achieves around 10fps on a single PC.