Research On Intelligent Calibration Method Of Lidar And Multi-Cameras Fusion

The technology of driverless car is a hot topic at the moment,and the perception components is a very important part,which plays a role in linking up and down.The most important thing in the perception components is the data received from the sensor.At present,the fusion of various sensors is the trend of times.The experimental platform in this subject is the two sensors applied to the perception components.It consists of a LIDAR and five cameras.The work of our project is to complete the extrinsic calibration of the LIDAR and five cameras before the platform is used,as well as the intrinsic calibration of the camera.The main research work of this paper are as follows:(1)The construction of the experimental platform,the collection,synchronization and screening of calibration data.First,design the mechanical structure of the platform,then use the generated random noise pattern to move around the multi-camera platform.And use the checkerboard calibration pattern to capture 90 pose images at nine positions in the camera's field of view.Then use the message_filters in the ROS system to synchronize the data and eliminate the invalid images in the calibration image data.(2)Research on calibration methods of multi-camera platform's extrinsic and intrinsic.Firstly,several methods of image feature points detection are compared.And the SURF detection method which is most suitable for calibration is selected.The pattern image is associated with the generated calibration image by feature points detection and matching.And the homography matrix is calculated by matching the feature points with the homography of the spatial corresponding plane.Decompose the homography matrix to get the pose of the pattern in the camera coordinate system and the intrinsic of the camera.Optimize the pose of the camera intrinsic and pattern using minimum reprojection error.Finally,the extrinsic is completed by the graph optimization by using the pose of the pattern by the adjacent cameras.(3)Research on LIDAR and camera extrinsic calibration methods.Intercept the point cloud of each pose and the pattern of the corresponding frame.Select the point on the pattern in the point cloud,calculate the pose of the pattern in the LIDAR coordinate system by least squares algorithm and its normal vectors.In the image,the pixel coordinates of the inner corner points are detected by the method of corner detection,and the pose of the pattern in the camera coordinate system and its normal vector are calculated by the corresponding corner points.The rotation matrix is calculated by the normal vector in the two coordinate systems,and the translation vector is calculated by minimizing the distance from the point to the pattern.Finally,the extrinsic of the LIDAR and the camera is completed.(4)The extrinsic between the multi-cameras and LIDAR are combined to complete the calibration of the entire system...