Automatic driving single-target calibration algorithm evaluation method, device and equipment

Abstract

The invention provides a single-target calibration algorithm evaluation method and device, electronic equipment, a storage medium and a product, and relates to the field of computer vision and the technical field of automatic driving. According to the specific implementation scheme, the method comprises the following steps: acquiring N internal reference calibration results of two monocular cameras determined by adopting N monocular calibration algorithms; determining external parameter calibration results respectively corresponding to the N internal parameter calibration results of the two monocular cameras based on a double-target calibration algorithm and the N internal parameter calibration results; based on the N internal reference calibration results and the external reference calibration results corresponding to the N internal reference calibration results respectively, correcting binocular images which are shot by the two monocular cameras and contain the target scene respectively, wherein binocular images corresponding to the N internal reference calibration results respectively are obtained; based on the binocular images corresponding to the N internal reference calibration results and the true value image of the target scene, determining evaluation results of the single target calibration algorithms corresponding to the N internal reference calibration results. According to the technical scheme of the invention, the method can improve the evaluation accuracy of the single-target calibration algorithm.

Description

technical field [0001] The present disclosure relates to the field of computer vision, in particular to the technical field of automatic driving. Background technique [0002] If the monocular camera is combined with lidar and integrated navigation, the internal reference calibration of the monocular camera is required. Monocular camera internal reference calibration itself, if not combined with other sensors, can only be used for monocular image de-distortion, it is difficult to determine the quality of monocular calibration with the naked eye. In the subsequent fusion process with integrated navigation and lidar, the inaccuracy of monocular internal reference calibration will greatly affect the quality of fusion. [0003] If the calibration algorithm is not evaluated in a timely manner during the algorithm development process, resulting in poor robustness of the final algorithm accuracy, the calibration problem will only be reflected in the subsequent fusion process, whic...

AI Technical Summary

Problems solved by technology

Monocular camera internal reference calibration itself, if not combined with other sensors, can only be used for monocular image de-distortion, it is difficult to determine the quality of monocular calibration with the naked eye

In the subsequent fusion process with integrated navigation and lidar, the inaccuracy of monocular internal reference calibration will greatly affect the quality of fusion

[0003] If the calibration algorithm is not evaluated in a timely manner during the algorithm development process, resulting in poor robustness of the final algorithm accuracy, the calibration problem will only be reflected in the subsequent fusion process, which will greatly prolong the problem exposure and The time to solve it affects the progress of research and development

In the prior art, when evaluating a single-objective calibration algorithm, it is often evaluated by the consistency of the results of multiple calibrations. However, the stability of the results and the consistency of the values ​​can only guarantee whether the algorithm results are stable to data changes, and cannot explain the calibration results. The result is not biased, therefore, the precision is poor

Images