The Study On Multi-camrea Calibration Using Multi-plane 3D Target

With the improvement of the hardware equipment and the operational speed of computers,machine vision technology is widely applied in artificial intelligence,vision measurement,robotics and image processing etc.Cameras are the central sensors in machine vision technology.So camera calibration is the crucial step of this technique which concerns rapidly building accuracy geometric relationship among different coordinates by calibrating cameras in object tracking,dimension measurement and surface reconstruction etc.The accuracy of the calibration is directly related to the reliability of these measurement results.Therefore,the core technology of vision measurement is to design an appropriate calibration method which is measured with different object and flexible movement.By studying existed calibration technology,it is found that rapidity,robustness and accuracy are still three major challenges for the camera calibration methods.A calibration technology which can satisfy both three indicators and calibrate all the parameters of the camera at the same time is still essential in vision measurement technology.This thesis proposes a camera calibration method using a multi-plane 3D target.The designed calibration target consisting of four fixed planes attached with coded patterns and feature circles is designed into the rectangular pyramid.The encoding pattern in the junction of each target plane is used for recognizing different target plane automatically.Only one captured image of the proposed method can identify the intrinsic and extrinsic parameters of camera systems.The experimental results confirm that the proposed method has the better accuracy and narrowest range of errors than Zhang's classical calibration method.In order to guarantee the single solution for the intrinsic matrix,the four non-coplanar planes provide four different homography matrixs at one time.The external parameters of the cameras are calculated by the geometric constrains of the feature circles on each plane.The proposed target with fixed structure avoids the errors from the wrong poses of the plane target.In this thesis,single-plane-circle target,biplane-circle target,multi-plane 3D target and a checkerboard target are used to calibrate the same camera system respectively under the same experimental conditions and the results indicate that themulti-plane 3D target owns the better robustness and the better accuracy than other targets through comparing the calibration data of these four targets.Then,ten groups of multi-plane 3D target with different dihedral angle are selected for calibrating camera,and the impact of the angle on the calibration results is further analyzed.The dihedral angle of the target must have influences on the calibration results.If the dihedral angle was 180?,the proposed target would reduce to the one plane target.The results indicate that the accuracy and the robustness of the calibration parameters are improved with the dihedral angle in the range from 80? to 140? and 220? to 280?.In order to study the effects of the size of feature circles on calibration results for the proposed method,comparison experiments are carried out.The feature circle with diameter larger than 12 mm can improve the calibration results.Finally,the proposed target is used to calibrate a camera system consisting of three cameras.And the reconstruct results of standard quantity of high precision glass checkerboard are used for error analysis.Finally,the proposed method calibration is to a case of visual measurement for tracking the motion of human eyes iris.Firstly,a picture of the human face in sinusoid way is taken on the vibrating table to extract the iris center and calculate the three-dimensional trajectory of the center.The measurement result compares with setting sinusoidal signal to verify the accuracy of the measurement scheme.Then,the actual human eyes iris movement measurement based on the experiment on the vibrating table is carried out to measure the motion of the model which is on the regular movement.The results which have the same regularity with the motion of the model indicate that the proposed method has practical application value.The measurement results provide the reference value for the analysis of the asthenopia caused by the relative movement between human eyes and the car display screen in the course of a vehicle.