A fast and accurate calibration method for the mapping relationship between laser point cloud and visual image

Abstract

The invention relates to a quick and precise calibrating method of the mapping relation of a laser point cloud and a visual image. The method comprises the following steps of 1, setting a checkerboardcalibrating plate with square holes, placing the calibrating plate into the fields of a laser radar and a camera simultaneously and obtaining n corresponding characteristic points by extracting characteristic points of the laser point cloud and the visual image; 2, conducting homography matrix initial calculation; 3, estimating homography matrix maximum likelihood; 4, estimating camera distortionparameter maximum likelihood; 5, estimating maximum likelihood of all mapping parameters in the mapping relation of the laser point cloud and the visual image. Based on a homography matrix, the direct mapping relation among the three-dimensional point cloud and visual image pixels is directly built, and a camera internal parameter matrix and a sensor external parameter matrix do not need to be calibrated. By means of the calibrating method, the calibrating steps are reduced; meanwhile, since the mapping result is directly optimized, transferring of a calibrating error cannot be caused, and the calibrating precision is higher.

Description

Technical field [0001] The invention relates to a method for quickly and accurately calibrating the mapping relationship between a laser point cloud and a visual image, and belongs to the field of intelligent networked automobile environment perception. Background technique [0002] Lidar can directly measure the distance information of the surrounding environment, with accurate measurement accuracy and a long measurement range, especially multi-line Lidar has ideal three-dimensional modeling capabilities. However, because it cannot obtain rich color information, it is difficult to use a three-dimensional point cloud to understand the semantics of the surrounding environment. The camera can obtain rich color information of the surrounding environment, and the current semantic segmentation algorithm for images is relatively mature. However, due to the loss of depth information in visual images, it is difficult to accurately express the surrounding environment in three dimensions....

AI Technical Summary

Problems solved by technology

However, by calibrating all physical parameters, and then obtaining the mapping relationship between 3D point clouds and pixels, errors will accumulate, and it is difficult to obtain the global optimum of the calibration process, and multiple calibrations of different parameters will also cause cumbersome calibration processes.

Therefore, the calibration process complexity and calibration accuracy of existing calibration methods need to be improved

Images