Camera Self-Calibration with Planar Pattern Using Genetic Algorithm

Abstract

In the paper, a stereo camera self-calibration method with Genetic Algorithm (GA) applied to the navigation of autonomous land vehicle (ALV) in a natural environment is proposed. The proposed method does not require specific object as a calibration pattern, e.g. checkerboard; conversely, it exploits common feature, for example: planes among natural scenes. In the evaluating process of GA, the coplanar condition of 3D points is employed as a fitness function to inspect the camera parameters. In addition, real valued GA is used because it does not only decrease the complexity of encoding and decoding process, but also increase the precision of solution. Comparing to conventional optimization methods, the camera self-calibration method based on GA can avoid being trapped in local minimum and does not need initial value or gradient information. Several experiments of the camera calibration with the stereo vision show that the proposed method can find approximate optimum solution.