Multi level fusion of competitive sensors for automotive environment perception

A reference sensor system, consisting of a multilayer laser scanner and a stereo camera system, is used for detecting vehicle surroundings. Via a novel multi level multi sensor fusion framework the heterogeneous sensor information can be fused on three succeeding processing levels (low, mid and high level). Here the low level fusion achieved the highest accuracy in the description of the object hypotheses. Detection and processing faults can be recognized and reduced by competing sensor information within higher fusion levels. To combine the individual advantages and compensate for the disadvantages of the each level the different levels´ results are merged in a V-shaped descent and ascent in the process chain. In the following the framework and the various methods for data fusion are presented and finally validated using real and simulated scenarios.