Extrinsic camera calibration in vehicles with explicit ground estimation

The use of multiple cameras in vehicles becomes more and more attractive as hardware prices decrease rapidly. Multiple camera sensors can be used to cover larger areas of the environment of a vehicle and for 3D scene reconstruction using stereo or structure from motion techniques. To be able to merge the sensor data in a common coordinate frame, it is necessary to know the relative positions and orientations of the cameras. However, as camera configurations may have non-overlapping fields of view due to cost or design reasons, no point correspondences between the camera images can be used. Instead, we apply a motionbased technique that allows general, in especially nonoverlapping camera configurations. By estimating the ground parameters in an intermediate step, we overcome the typical problems of purely planar motion. Finally we are able to estimate all 6 Euclidean calibration parameters between each camera pair. This contribution outlines a concept to perform an online calibration of multiple cameras on a mobile platform with non-overlapping fields of view. Results with simulated data are presented.