Research On Camera Calibration For Videometrics With Camera Network And With Moving Platform

Videometrics with camera network and with moving platform have a broad range of applications in both military and civilian fields.Typically,videometrics with camera series network connects multiple cameras in a relayed fashion for the deformation measurement of large scale structures,such as railways and bridge.It is important for insuring service quality and safety.In the application of videometrics with moving platform,the three-dimensional position of the ground target can be measured based on camera and inertial measurement unit(IMU)which are mounted on the platform,such as an unmanned aerial vehicle(UAV).It is a key technology to achieve precision strike in modern warfare.Camera calibration is a fundamental technical issue in videometrics.Due to the limitations of the experimental scene,camera calibration still faces many challenges.This dissertation focuses on camera calibration in the application of videometrics with camera network and with moving platform,including stereo self-calibration of fisheye cameras,calibration of large field cameras and non-overlapping cameras,rotational alignment of IMU-camera systems and relative pose estimation with known orientation angles.The main contents of this dissertation are as follows:(1)Two stereo self-calibration methods are proposed for fisheye cameras with serious radial distortion.According to the experimental scene as a planar scene or a 3D scene,the intrinsic and extrinsic parameters of stereo cameras are calibrated based on homography constraint or epipolar constraint.The traditional self-calibration methods estimate the stereo camera geometry from image correspondences corrupted by radial distortions and only deal with radial distortions by modeling them in the final optimization stage.The optimization is more likely to prone to ending up in local minima.Thus it is necessary to consider the radial distortions already when searching for image correspondences at the RANSAC stage.Our methods estimate the stereo camera geometry and two radial distortion parameters simultaneously.The RANSAC scheme is used to cope with outliers of feature matches.Then,the intrinsic and extrinsic parameters are optimized using the inliers of stereo image pair.Our methods are both efficient and accurate for the stereo self-calibration of fisheye cameras.(2)Three camera calibration methods are proposed for calibration of large field cameras and non-overlapping cameras considering the limitations of the experimental scene.Firstly,a camera calibration method is presented for large field camera pointing at sky on sway platform.All the UAV's positions in the entire flight are used as control points for camera calibration.Secondly,a flexible calibration method is presented for non-overlapping cameras based on the fixation constraint of camera rig.The proposed algorithm gives the solutions of the cameras parameters and the relative poses simultaneously using two checkerboards.Finally,a calibration method for non-overlapping cameras with double-sided telecentric lenses is proposed.The proposed method connects the double-sided telecentric cameras using the laser plane which goes through the field of views of cameras.The laser planes are generated by an ordinary line laser projector.The rotation relationship of the telecentric cameras is solved using the coplanarity of laser plane.(3)The rotational alignment of IMU-camera systems is proposed based on homography constraints.Firstly,the minimal case solutions are proposed for the rotational alignment of IMU-camera systems.These solutions depend on the calibration case with respect to camera motion(general motion case or pure rotation case)and camera parameters(calibrated camera or partially uncalibrated camera).The minimal case solutions are useful to reduce the computation time and increase the calibration robustness when using RANSAC on the point correspondences between two images.Furthermore,a non-linear parameter optimization over all image pairs is performed.Our methods do not require a known calibration device or any special hardware.The proposed methods directly minimize the image transfer residuals based on homography constraints,rather than conduct an algebraic minimization of transformation matrices between the IMU and the camera.The objective function based on the image measurements is a geometrically more meaningful criterion.(4)Several methods are proposed for the relative pose estimation of the camera with known orientation angles according to different application scenarios.Firstly,a relative translation estimation method using two matched points is proposed in the UAV application scenario.This method uses three rotation angles of camera provided by IMU.Secondly,a relative pose estimation method based on homography is proposed in the UAV application scenario.This method needs two matched points on the ground plane but only uses two known rotation angles of camera: roll angle and pitch angle.Finally,a series of relative pose estimation methods is proposed in the automatic driving application scenario.The roll angle and pitch angle of camera are also provided by IMU.These methods are derived using a homography formulation with decoupled rotation and translation estimation.A variety of experiments with synthetic and real data demonstrate that our methods improve the accuracy and robustness of the relative pose estimation of the camera.