Large-scene non-common-view multi-camera calibration method

Abstract

The invention provides a large-scene non-common-view multi-camera calibration method which comprises the following steps: arranging a plurality of markers in a test area, and measuring the world coordinate system positions of the markers; observing the marker by using a pan-tilt camera, and observing the pan-tilt camera by using each to-be-calibrated camera; calculating a real three-dimensional track of the pan-tilt camera under a world coordinate system by observing the marker; solving the precise pose of the to-be-calibrated camera under the world coordinate system by utilizing the pan-tiltcamera track segment, correspondingly appearing in the view of each to-be-calibrated camera, of the timestamp; performing matrix transformation on the pose of each to-be-calibrated camera in a world coordinate system to obtain an external parameter matrix between any two cameras. The method is simple, convenient and rapid; calibration of the external parameters of the camera can be completed without specially arranging an experiment site, and visual coverage of a larger-scale scene can be achieved.

Description

technical field [0001] The invention belongs to a camera extrinsic parameter calibration method in computer vision problems, in particular to a method for calibrating external parameters among multiple cameras. Background technique [0002] The current multi-camera calibration methods for large scenes require the cameras to have a common field of view. [0003] One type of method is to obtain image pairs of the same scene captured by multiple cameras, extract feature points from the sequence images, and then perform matching between feature point pairs, so as to use the knowledge of multi-view geometry to obtain the relationship between multiple cameras. Rotation matrix and translation vector, that is, external parameters, and then use other methods to optimize the external parameters. [0004] The implementation of another type of method needs to set up notable markers from multiple angles in the field, such as photospheres, calibration plates, etc. The markers need to be ...

AI Technical Summary

Problems solved by technology

[0010] However, in the actual test, it is difficult to achieve a common field of view for all cameras in a large scene test environment

The proposal of a common field of view will significantly reduce the coverage of the camera on the test scene, and will also largely limit the camera's placement angle and placement

Images