Model-based improvement of motion vector fields for driver assistance systems

This paper presents a model-based approach for motion vector field optimization for driver assistance systems. In order to optimize the vector quality a predictive motion estimator which computes motion vector fields out of video signals which are captured by a front camera inside a car, is enhanced by synthetically generated motion vectors. These synthetic motion vectors are adapted to typical driving situations. Then, these vectors are recursively fed back into the motion estimator loop and are serving as additional candidate motion vectors. The proposed approach provides an improvement in terms of objective and subjective performance while requiring only very low processing complexity.