Research On Large-scale Scene Calibration Method Based On Local Objects

In recent years,with the development of computer vision and the popularity of digital cameras,the use of ordinary digital cameras to obtain the structural information or achieve the 3D reconstruction of the large-scale scene has been more popular.The crucial step in realizing the 3D reconstruction or measurement is camera calibration and thus achieving the projection model between the three-dimensional spatial points with the two-dimensional image points.For camera calibration in the large-scale scene,calibration precision and flexibility are two important factors that should be considered at the same time.However,the existing camera calibration methods have limitations when applied to the cameras in the large-scale scene,camera calibration methods based on calibration object have a high calibration precision while a poor flexibility and camera self-calibration method is operable but cannot be applied to most occasions requiring accuracy.The paper analysis and summarizes the characteristics of the existing camera calibration methods and then focus on the balance of calibration precision and flexibility.Main work is as follows:Firstly,a large-scale scene calibration method based on multi-chessboards is proposed,which can solve the limitations of poor flexibility of Zhang's classical calibration method when applied to the large-scale scene.In the method,multi-chessboards are placed in the large-scale scene,by computing the homograph matrix between each small chessboard plane with the image plane we can figure out the rotation and translation vectors between each small chessboard plane and the image plane,and then the rotation and translation vectors between each chessboard plane and the given reference chessboard plane can be obtained,next rotate and translate the multi-chessboards to the given reference chessboard plane to form a whole big chessboard,finally construct the reprojection error function of the image corners on the big chessboards.Take the calibration results of the reference chessboard as initial value and figure out the final camera parameters through iterative optimization of the reprojection error function.Both the simulation and real data experiments show that the proposed method has a relatively high calibration precision and the calibration flexibility has been improved at the same time.Secondly,a large-scale scene calibration method based on manual vanishing points is introduced,which can not only conquer the drawbacks of existing method requiring structural information in the large-scale scene but also extend the scope of existing calibration method by widening the calibration method based on perpendicular vanishing points to partly non-perpendicular vanishing points.In the method,two designed identical posts with circular markers are placed in the large scale scene to introduce in two groups of parallel lines,and the equation of the introduced parallel lines are fitted through the detection of the centers of the circular markers,then construct the constrained functions among camera intrinsic parameters,the angle of the introduced parallel lines and the corresponding vanishing points.Take the calibration results when the vanishing points are perpendicular as the initial values and obtain the final results through iterative optimization of the constructed function.Both the simulation and real data experiments show that the algorithm can calibration the camera in the large-scale scene well.