Projekt

Up to now, it is an unsolved problem, how sensor data selection and fusion shall be done in the case that multiple cameras and multiple cues from each of the cameras are available. Thus, the goal of the project is the development of a probabilistic cue integration mechanism for object tracking using multiple cameras. The basis is given by the approach on cue integration by Democratic Integration (DI) and probabilistic methods for state estimation, and optimal sensor data selection. The key idea is to first integrate the DI priciple in a probabilistic framework. Then, the method is extented from single sensors to multiple sensors by exploiting techniques from projective geometry. As a result one gets global fusion maps from the local maps of the different sensors. The approach is applied and verified in real-time object tracking using multiple cameras in two selected applications: surveillance tasks and moble robot navigation.