Resolving partial occlusions in crowded environments utilizing range data and video cameras

Video surveillance systems are omnipresent in our daily life, but still suffer from some drawbacks, which hardens the integration of fully automated systems. Currently standards CCD sensors are used to monitor public and private spaces. These are not yet able to resolve revere occlusions in narrow environments. Therefore we suggest the integration of 3D sensors, in particular a photonic mixture device, into current frameworks, in order to support the reliable detection and segmentation in dense situations. We propose the use of basic techniques to segment persons in range data, to guarantee real-time processing capabilities. With a reliable foreground segmentation and the computation of depth gradients the segmentation performance will drastically rise.