Multi-camera Calibration Technique Research And Its Application

Camera calibration is a key problem in computer vision research and is more andmore valued comprehensively lately. Because it is indispensable to obtain3Dgeometric information from2D images, the technology has been widely used indomains such as Virtual Reality, medical image reconstruction,3D measurement,human motion capture and so on.The research range in this paper mainly involves non-parallel multi-cameracalibration and its application aiming at the3D reconstruction of human body joints.On the basis of elaborating the fundamental theory of camera calibration, this papersummarizes a few multi-camera calibration methods and the detailed calibrationsteps for all. Then the multi-camera calibration methods are proposed and areapplied in3D human motionreconstruction. The main work and achievements havebeen discussed as follows：(1)The basic principles of camera calibration are briefly introduced, and cameracalibration methods including Direct Linear Transformation(DLT),Zhang’scalibration method and Tsai’s two-step calibration method are surveyed． Thefrequently-used nonlinear optimization algorithms are proposed.(2)According to the calibration object dimension, several foreign and domesticmulti-camera calibration methods are exhaustively overviewed aiming at this project.On the basisof it, the two calibration methods in this paper are proposed.(3)Two different multi-camera calibration methods are proposed.They aremulti-camera calibration method based on OpenCV and multi-camera calibrationmethod using planar patterns respectively. For the camera intrinsic parameters, thetwo all use Zhang’s calibration method.For the extrinsic parameters,the former usesmany cameras to photograph a regular region in the center location to obtainrespectively the correspondence between the specified points on this region and eachimage points. Using OpenCV method the extrinsic parameters of this region withrespect to each camera coordinate system are computed firstly, further the relationsof each camera with respect to this region are solved using reverse operations,ultimately the parameters of each camera are obtained. In the latter method,constrained optimization method of VR algorithm is applied based on ICP algorithmand the position relationship between two adjacent cameras is computed by placing a model plane at a few different locations at first, each camera is unified to a worldcoordinate system ultimately.(4)On the basis of camera calibration technique, the3D human joints arereconstructed through the marker ID matching and the strategy of using epipolarconstraint to reduce false matching points. The experiment results indicate that themethod can satisfy the precision need of subsequent3D reconstruction and areeffective methods.