Abstract:

Camera calibration is an essential step of 3D reconstruction. Conventional calibration methods require high precision equipment and sophisticated operations. Compared with them, camera self-calibration is simple but has low precision, which leads to significant performance degradation of 3D reconstruction. Therefore, there is a growing need for a simple and accurate high precision self-calibration method. By using the bundle adjustment algorithm and SIFT points matching relationship, we propose a local-global hybrid iterative optimization method. As for the large number of matching features, we propose a neighborhood image matching method, which can significantly reduce the matching time under the premise of maintaining accuracy. Experimental results show that the proposed method is effective  and accurate, and it can reduce the time consumption of image matching. Based on the relationship between corresponding  matching points in multi view images, our method makes full use of the local-global hybrid idea to compute the parameters of the camera. Compared with other existing methods, it is more robust with higher precision.

Key words: camera self-calibration, multiple view images, image matching, bundle adjustment algorithm