Multi-camera Joint Calibration Based On Vehicle

In the field of intelligent transportation,multi-camera systems have been acting as the "eye" of the intelligent vehicle,which ensures the vehicle to obtain the information from the scene.In order to enlarge the field of view of the vehicle multi-camera system,we usually choose the large-viewing fisheye camera.In order to make the driver receive information directly,it is necessary to design an application that can meet the requirements of the human beings' visual habits.The precondition of designing this application is to make a joint calibration for the multi-camera system.The meaning of joint calibration is to use the calibrated parameters to correct the severe distortion of the fisheye camera.Furthermore,the position relation among cameras in the multi-camera system can be determined to be ready for the next step.In this paper,firstly we introduce the traditional target-based calibration method used in multi-camera systems,which calibrates the intrinsic parameters and the distortion coefficients of cameras by using the small size handheld calibration targets,then calibrates extrinsic parameters through the large size calibration targets which are set around the vehicle.This approach is time-consuming and complicated.Therefore,we propose two joint calibration methods to apply on vehicle multi-camera systems.1.We propose a new target-based calibration method.Around the vehicle body,large size targets are first paved and the vehicle moves some distance forward and backward,then the feature information of large size targets on the ground is used to calibrate the fisheye camera parameters and distortion coefficients.In this paper,we introduce the procedures of algorithms in detail.In this method,the homography matrix constraint and nonlinear optimization algorithm are used to calibrate the fisheye camera intrinsic parameters and distortion coefficients,and then we study how homography matrix converts images that single camera obtains to top view and LK optical flow tracking method implements the mosaic of top view by tracking feature points.Finally,the calibration of extrinsic parameters relative to the ground is achieved by the existing feature information of large size targets of top view.2.We propose a joint calibration method based on natural scenes.Geometric relationshipthat exists in natural scenes is used as the constraint,and the nonlinear optimization method is used to optimize in order to obtain the intrinsic parameters and the distortion coefficients of camera.In the calibration process of extrinsic parameters,we use the ORB-SLAM algorithm to store the scene information that cameras obtained and match each obtained BRIEF descriptor to achieve the coarse registration of the scene point cloud.And the accurate registration is achieved by using the ICP algorithm so that parameter relation among multi-camera is obtained.Through the two methods above,the shortcomings of the traditional calibration method are overcome.According to the experimental results,it can be seen that the algorithms are correct and reliable and can be used in practical applications.