A sensor independent probabilistic fusion system for driver assistance systems

In this contribution we present a probabilistic fusion framework for implementing a sensor independent measurement fusion. All interfaces are using probabilistic descriptions of measurement and existence uncertainties. We introduce several extensions to already existing algorithms: the support for association of multiple measurements to the same object is introduced, which reduces the effects of split segments in the data preprocessing step of high-resolution sensors like laser scanners. Furthermore, we present an approach for integrating explicit object birth models. We also developed extensions to speed up the algorithm which lead to real-time performance with fragmented data. We show the application of the framework in an automotive multi-target multi-sensor environment by fusing laser scanner and video. The algorithms were evaluated using real-world data in our research vehicle.