Camera high-precision automatic calibration method and device for full-automatic parking

Abstract

The invention discloses a camera high-precision automatic calibration method and device for full-automatic parking, and the method comprises the steps: randomly selecting a fixed number of cameras from batch cameras as sample cameras; for the sample cameras, collecting a plurality of original images, calibrating the cameras, and constructing an internal reference set by using the internal parameters of all the sample cameras; acquiring angular points of the calibration cloth on the original image; traversing the internal parameter set, carrying out camera external parameter calibration, selecting an angular point back projection error as an index, and taking a group of internal and external parameters with minimum errors as camera internal and external parameter initial values; convertingthe fisheye camera image into a top view based on the internal and external parameter initial values, extracting the initial values of the angular points in the top view, converting the initial valuesof the angular points to the fisheye image, and performing sub-pixel-level angular point extraction to obtain high-precision angular point information; and optimizing the internal and external parameter initial values by using high-precision angular point information to obtain an optimized calibration result. According to the invention, high-precision automatic calibration can be effectively carried out on the camera.

Description

technical field [0001] The invention relates to the technical field of fully automatic parking, in particular to a camera high-precision automatic calibration method and device for fully automatic parking. Background technique [0002] At present, for the calibration of batch cameras, because internal reference calibration is time-consuming, fixed camera internal references (including optical center, focal ratio, and distortion coefficient) are used, and only external reference calibration is performed. For corner point extraction, the initial coordinates of corner points are usually roughly extracted directly on the fisheye image, and then sub-pixel level corner point extraction is performed. [0003] It can be seen from this that although there is not much difference in the process of batch cameras, the internal parameters of the cameras cannot be exactly the same. If the same set of parameters is used for batch cameras, it will obviously introduce a large error, and autom...

AI Technical Summary

Problems solved by technology

[0003] It can be seen from this that although there is not much difference in the process of batch cameras, the internal parameters of the cameras cannot be completely the same. If the same set of parameters is used for batch cameras, it will obviously introduce a large error, and automatic parking requires relatively high accuracy of camera parameters.

In addition, due to the large distortion of the fisheye camera, it is difficult to accurately extract the initial value of the corner point in the non-middle position of the image directly from the original image of the fisheye camera

Images