Enhanced motion estimation for driver assistance systems - Integration of a curved road model

Gregor Schewior, Holger Blume

Abstract: This paper presents an extended efficient modelbased approach for the improvement of motion vector fields for driver assistance systems. Through the integration of a curved road model the quality of motion vector fields, estimated by a predictive block-based motion estimator working on video signals which are captured by a front camera inside a car, is significantly increased. The optimization step is performed by modeling synthetic motion vector fields of typical driving scenes acting in an iterative step as additional motion vector candidates. The proposed enhanced approach which is effective for all kinds of driving situations provides a significant improvement in terms of objective and subjective performance while requiring only low computational overhead.