@article{ZHANG2025, 
author = {Tao ZHANG and Bo GAO and Qijun LUO},
title = {Adaptive Levenberg-Marquardt optimization for accurate calibration of multi-camera systems with low overlap fields of view},
year = {2025},
journal = {Acta Aeronautica et Astronautica Sinica},
volume = {46},
number = {4},
keywords = {multi-camera system, camera calibration, adaptive Levenberg-Marquardt optimization algorithm, global calibration, cylindrical projection},
url = {https://www.sciopen.com/article/10.7527/S1000-6893.2024.30860},
doi = {10.7527/S1000-6893.2024.30860},
abstract = {To acquire situational awareness of the surroundings, pilot assistance systems require high-precision camera calibrationand hazardous target position measurement in large-scale scenes with low overlap. To address the issues such as insufficient calibration accuracy in existing low-overlap multi-camera systemsand reduced visual measurement accuracy due to distortion, this paper proposes a precise multi-camera calibration method based on adaptive Levenberg-Marquardt optimization. First, a binocular camera calibration model is established using a rigid composite target. By employing a rigidly fixed composite targetand an auxiliary camera, the transformation relationship between the two cameras is derived. Then, an adaptive Levenberg-Marquardt optimization algorithm is introduced, which uses the reprojection error of the target's feature points as the objective function. An iterative optimization method with adaptive step sizes is utilized to optimize the transformation matrix between the binocular cameras, thereby enhancing calibration accuracy. Finally, a co-centric cylindrical projection of multi-camera images is performed to reduce target parallaxand improve the measurement accuracy of the stitched panoramic view. Calibrationand measurement experiments demonstrate that the proposed method effectively addresses the challenges of precise camera calibrationand hazardous target position measurement in low-overlapping fields of view, reducing the reprojection error by 10%and the measurement error by 13%.}
}