Triple Lines Approach to Calibration of Camera Mounted in Intelligent Vehicle
Q Li,NN Zheng,XT Zhang
DOI: https://doi.org/10.1117/12.541996
2004-01-01
Abstract:In the vision based navigation of intelligent vehicle(1,2), it is necessary to reconstruct or recognize objects according to images acquired by camera. So calibrating camera is preliminary process for the correctness of the results of all the functionalities based on vision system, including lane detection, obstacle detection, vehicle detection, pedestrian detection and navigation(3). There are many solutions on calibration of camera on-board. In ARGO system, the intrinsic parameters are assumed known. Some grids whose size is prefixed are drawn on the ground and interaction points are used to calibrate extrinsic parameters(4). Additionally, a so-called self-calibration method is adopted in the system where some markers on hook are used to calibrating object and their position relative to vehicle can be measured(5). In GRASP Laboratory, University of Pennsylvania, for intrinsic calibration, they use a calibration toolbox developed at the University of Southern California, and for extrinsic calibration, they use parallel road edges, but only calibrate two parameters, the height and pitch of the camera(6). In Tsinghua University, researchers use grids drawn on ground to calibrate camera(7). Although the methods mentioned above can calibrate camera well, they have some shortcomings to a certain extent, for example, special field being needed and inconvenient operations existing(4,7), only few parameters being calibrated(6). For solving these problems, a simple and practical method to calibrate extrinsic parameters of camera mounted in intelligent vehicle is presented. Because mapping of three parallel lines into image plane have same vanishing point and different slope, according to perspective projection principle, an inherent relationship exists among camera external parameters and vanishing point and slope, so analytic solutions of camera external parameters can be established by mathematical derivation, without iterative and optimizing calculation. After positioning intersection point and slope of imaging lines by hand or automatically, external parameters can be calculated directly. The method doesn't need specific field, even can be implemented when the vehicle is running.