Research On Camera Network Calibration And Measurement Calculation Optimization Technique

Camera is the essential sensor in the Photogrammetry and Computer Vision measurement tasks;it can capture the object's three dimension information in real world and record on the image by imaging.And what we,as the member of Photogrammetry and Computer Vision community,can do is try our best to reconstruct the real 3D structure model of the object precisely with the information recorded by the 2D image plane.Before the reconstruction step,one essential and inevitable work,the calibration of camera system,must be done well and high precision calibration of the vision system's parameters(camera intrinsic parameters and the camera pose)is necessary for the accurate reconstruction.In the meantime,the camera network that consists of multi cameras with the fixed restricted could be used instead of single camera.In addition,we cannot ensure that the high precision camera calibration can be acquired in any situation and only an inaccurate initial calibration result can be obtained.Fortunately,bundle adjustment is born to be the solution for getting high accuracy of calibration or reconstruction by jointly optimal the 3D structure and viewing parameters.In this thesis,two basic but important aspects,camera calibration and bundle adjustment,of the reconstruction are studied and the thesis' s main works and technique contribution are as follow:1.A review of camera calibration is presented,which include the classification of calibration methods and the investigation of distortion coefficient calculation technique base on the unknown line.In addition,the research state of Sparse Bundle Adjustment(SBA)which decoupled 3D structure and viewing parameters is investigated.And then summarize the advantage and disadvantage of several numerical optimal methods.2.A new camera calibration technique based on the nonplanar lines is proposed.Its practicability and feasibility is verified by the simulation and physical experiment.And the result of physical experiment indicates that the proposed technique on par with Zhang's method in term of accuracy.3.Research on SBA with the multiple fixed-cameras and additional constraints.Based on the mathematical and computational model of single camera SBA,the model of multiple fixed-cameras SBA attach with additional constraints is deuced.The algorithm flow is presented and its practicability and feasibility is verified by the physical experiments.In addition,the experiment verify the fact that taking into account priori information of objects can improving the accuracy.4.The core part of software for multiple fixed-cameras SBA is finished based on the accomplishment of part 3.It can be suitable for the configuration that include single camera,camera network,additional constraints or the combine of them.Furthermore,it can be configured with any parameter to be optimized and can be apply for camera network calibration or the last step of 3D reconstruction.