Towards surround stereo vision: Analysis of a new surround view camera configuration for driving assistance applications

We propose a new camera configuration for automotive surround view systems where the criteria for which the camera mountings are optimized is derived not from visualization, but optimal depth measurements. We show that based on special algorithmic processing [1] it is feasible to achieve 3D reconstruction in almost the complete surrounding of the car. In order to accomplish this, more than 4 cameras are required and the positions have to be carefully selected. An analysis of the proposed configuration is performed with regard to the field of view overlaps and we show that the estimated depth is good compared to reference measurements conducted with a lidar sensor.