Calibration. Multi-view Camera Calibration And Multi-view Three-dimensional Recovery,

Camera calibration is the essential step of obtaining 3D information from views in the field of computer vision, which is widely used in the area of 3D reconstruction, navigation, visual supervision, etc. Therefore, the associative theoretic research has been put more and more attention in this field. The intrinsic parameters of camera are found from the feature-point -set of views by means of any one of camera self-calibration techniques. And the traditional camera calibration techniques are off-line, they first do the self-calibration off-line, then, they do the 3D reconstruction on-line, so in the progress of 3D reconstruction, the intrinsic parameters should not changed, which has limited its range of application greatly, especially can not applies to such new technologies of the vision as attention concentrating , initiative vision ,etc. Among these new technologies, intrinsic parameters of camera are changing as required, so we must develop camera self-calibration technologies. At present, there are little successful multi-view camera self-calibration technologies, and, existing technologies are limited to specific camera motion modes, so they have theory value and practical value in developing technologies of multi-view camera self-calibration with 3D reconstruction without camera motion modes restricts. These technologies not only can utilize whole information, but also the restrictions on the developed algorithms are so less that it can be widely applied to many practical situations and promote the appearance and development of the new technologies of the vision.This paper introduces two kinds of techniques of multi-view camera self-calibration with 3D reconstruction, one is linear multi-view camera self-calibration with 3D reconstruction, the other is non-linear multi-view camera self-calibration with 3D reconstruction. The paper mainly includes three parts. The first part presents linear multi-view camera self-calibration with 3D reconstruction technique and the performance in detail. The second part provides non-linear multi-view camera self-calibration with 3D reconstruction technique and the performance in detail. The third part gives the comparison of linear multi-view camera self-calibration with 3D reconstruction technique and non-linear multi-view camera self-calibration with 3D reconstruction technique.In this paper, the classical pinhole imaging is adopted as the model ofcamera, which means the matrix of camera intrinsic parameters has five independent arguments. In linear multi-view camera self-calibration with 3D reconstruction technique the epipoles and fundamental matrixes are firstly estimated from the views correspondences and, then, the intrinsic parameters are computed from them. Finally, the camera self-calibration with 3D reconstruction are completed from all the above information. The non-linear multi-view camera self-calibration with 3D reconstruction technique uses the restriction of rank 3 of centralized multi-view figure matrix to compute the intrinsic parameters and then complete camera self-calibration with 3D reconstruction. Through comparison, we can get the conclusions as follows: The non-linear multi-view camera self-calibration with 3D reconstruction technique is better than linear multi-view camera self-calibration with 3D reconstruction technique. Furthermore, the influence on the calculation errors in simulation is less. But the linear multi-view camera self-calibration with 3D reconstruction technique is more easily done in less time. At the same time, the two techniques are used to do real image experiments and the experimental results demonstrate that the non-linear multi-view camera self-calibration with 3D reconstruction technique is indeed better.The paper introduces two kinds of technique of multi-view camera self-calibration with 3D reconstruction. The results of experiments demonstrate that the non-linear algorithms and linear algorithms are both very good, so it have valuable in theory and in practice. But there are some limitations in the theory, such as the future-points run out of the views, which are worth studying further.