Calibration of non-overlapping cameras in vehicles

Cameras are getting smaller and cheaper. So a cost-effective usage of multi-camera systems in vehicles gets more and more attractive. In many cases it is desirable to cover the whole environment all around the vehicle. But often design restrictions and energy consumption do not allow constellations of cameras with overlapping field of views. However, for a common and geometric usage of the extracted information, e. g. in structure from motion tasks, it is necessary to know the relative alignment of the cameras, which are the extrinsic calibration parameters. This paper addresses the extrinsic calibration of a multi camera rig with non-overlapping field of views on a mobile platform. As the field of views do not necessarily overlap, common calibration methods based on corresponding image points between the camera views will fail. This problem can be overcome by using the mobility of the platform. A pattern-based method for extrinsic calibration of the camera rig on a mobile platform is presented.