Scene representation for driver assistance by means of neural fields

In this paper a method for scene representation in driver assistance applications is proposed. To gain noise reduction and unique environmental information detection object-hypotheses are evaluated. For each object an object-dynamic is built. The Object-dynamic consists of four one-dimensional neural fields for information evaluation of relative position, size and relative velocity. The chosen formulation enables fusion and separation of object-hypotheses gained from different sensors. Due to the dynamic character of the representation a reduction of noise and a prediction over short time periods is possible. The advantages of the representation are shown by inspecting real world sensor data.