Stereo Cameras Self-Calibration Based on SIFT

At present, a new algorithm of feature matching-SIFT has become a hot topic in the feature matching field, whose matching ability is strong, and could process the matching problems with translation, rotation and affine transformation among different images, and to a certain extent is with more stable feature matching ability for images which are captured from random different angles. In this paper, single camera is first calibrated using plane chessboard based on OpenCV, in order to overcome shortcomings in traditional and previous self-calibration methods, SIFT algorithm is proposed to calibrate stereo cameras after two cameras intrinsic parameters are calibrated. Fundamental matrix is gained through several matching points in two images using SIFT feature matching method, combined with intrinsic parameters, we can compute essential matrix. Translation matrix and the rotation matrix of stereo cameras can be resolved through SVD of essential matrix known as Huang-Faugeras´ constrains. Experiment results show that our method can calibrate relationship of stereo cameras accurately, and be able to calibrate two cameras in any circumstances. The algorithm has strong adaptability and robustness, but the expense time needs to be further improved.