A flexible calibration method for large-range binocular vision system based on state transformation

Vision measurement is becoming an indispensable technology in construction scenarios. The calibration result of the binocular camera directly affects the accuracy of vision measurement. Aiming at the problem that it is challenging to calibrate the large-range binocular vision system conveniently in harsh construction environ-ments, a flexible binocular camera calibration method based on state transformation is proposed. First, the mapping between the pixel coordinates in the image of the incomplete target and the corresponding world coordinates is realized based on the global locator. Second, the calibration process is divided into two stages. The lenses are focused on the calibration position, and the non-overlapping fields of view calibration in the initial state is achieved using virtual projection point pairs. Then the lenses are focused on the measurement position. Some extrinsic parameters of the monocular camera are solved independently by using the properties of the rotation matrix, and the remaining parameters are calculated by the depth constraint. On this basis, the state transformation matrices are solved, and the relative extrinsic parameters of the binocular camera are obtained. Finally, a series of simulations and experiments are carried out. The results demonstrate that the mean repro-jection error of the monocular camera is only 0.0675 pixels. The relative measurement error of binocular vision is less than 1.75%. In addition, the proposed method shows good robustness under different target angles and calibration distances. The method can be applied to complex engineering sites.

Automated summary

Vision measurement is crucial in construction scenarios, and the accuracy of binocular camera calibration directly affects its precision. This study presents a flexible binocular camera calibration method using state transformation to address the challenge of conveniently calibrating a large-range binocular vision system in harsh construction environments. The proposed method achieves mapping between pixel coordinates and world coordinates and divides the calibration process into two stages. Simulation and experiments show that the method has low reprojection error and relative measurement error, demonstrating good robustness for complex engineering sites.