Simultaneous localization and tracking of persons in a cluttered scene with a single camera
Holger Fillbrandt, Karl–Friedrich Kraiss · 2005
Abstract: Visual surveillance systems often require exact localization of tracked persons on the ground plane of the monitored scene. This becomes a challenging task in a cluttered environment, where persons not only overlap each other in the image plane but are also at times partially occluded by furniture or similar objects. In contrast to many multi-camera or stereo approaches to overcome these difficulties, we present a framework to solve the surveillance task using only one stationary monocular color camera. Our approach is to reconstruct the full silhouette from the visible parts, to merge several position assumptions and to incorporate relevant a priori knowledge about the monitored scene into the system. The paper will discuss this approach and present special algorithms developed for this task.