A High Accuracy Calibration Method for Wide-Area Distributed Multi-viewpoint Camera Group

摘要

Camera calibration, playing a key role in three-dimensional reconstruction, has been studied extensively in computer vision and photogrammetry for many years. With the trend of enlarging the scene and increasing the number of cameras, more calibration points need to be matched and besides, some calibration points may be partially captured by a camera, so that calibrating cameras accurately in a wide-area scene becomes a challenge. In order to solve these problems, a high accuracy calibration method for wide-area distributed multi-viewpoint camera group is put forward in this paper. On one hand, homography matrix is calculated iteratively based on the previous result in a proliferation manner to create conjugate pairs of feature points automatically. Meanwhile, a pre-prepared black and white chessboard template is used as the calibration object to improve the stability of the matching process. On the other hand, a multi-hierarchical tree-structural grouping strategy is presented, in which inner-group calibration is conducted on the lowest hierarchy first, after which the calibration results are combined to a unified coordination system in a parallel way through a series of iterative transformation of coordinates and optimization. Experimental results show that the iterative feature points matching algorithm makes feature points matching accurate and automatic even in the condition of a large calibration plane. Moreover, with the multi-hierarchical grouping strategy, the error of extrinsic parameters is tiny even in a wide-area scene with many cameras. In a word, the method works stably and accurately in calibrating distributed multi-viewpoint cameras in a wide-area scene.