Self-Calibration System for the Orientation of a Vehicle Camera

Abstract

The process of calibration is a prerequisite for each computer vision system. Calibration involves calculating both intrinsic and extrinsic parameters of the camera. While intrinsic parameters (focal length, principal point, etc.) are usually fixed, extrinsic ones (position and angles of the camera) have to be determined when the camera moves in relation to world coordinates. The calibration of the extrinsic parameters is usually performed with help of some reference objects or known measured points (GCP’s) in the scene. In the case of a vehicle camera, where the coordinates refer to vehicle coordinates, not the extrinsic calibration but the alignment of the camera to the Inertial Measurement Unit (IMU) is necessary. This paper proposes a solution for determining the orientation of a vehicle camera in relation to the vehicle. This novel approach escapes from tedious laboratory setups and reference measurements. It benefits from a known property of the road’s infrastructure, namely the parallelism of the road markers. For this reason lane markers are detected and transformed through a fast perspective removal (FPR) to an orthographic perspective. Newton’s Method is used for searching an optimal parameter set for this transformation. The algorithm works under the assumptions that the calibration is performed when driving on a straight and flat segment and the lane markers are visible. It reaches very good performance (via parametrical instead of image transformations) and good accuracy for lateral detection of features in automotive applications (for depth information, the algorithm must be improved).