Atmospheric Polarization Pattern Restoration Orientation Method in Complex Navigation Environment
Yuan Li, Y. Bian, Caiming Ren, Yi‐Gang Li, Yanhong Lu, Shaoyu Qu
Abstract
Bionic polarized light navigation is a new navigation method that uses atmospheric polarization patterns to obtain compass information. It has been extensively studied in the fields of autonomous navigation and integrated navigation in recent years. However, studies have found that polarized light navigation is susceptible to occlusion interference when navigating in complex environments, which will lead to the loss of atmospheric polarization information and severely reduce navigation accuracy. Therefore, this paper presents an orientation method for inpainting atmospheric polarization patterns, employing an enhanced U-Net convolutional neural network combined with sample block matching. To address the issue of error accumulation during the restoration process, an image integrity evaluation index based on visual coherence and visual saliency is utilized as a constraint to guide the restoration. The outdoor experimental results demonstrate that the proposed method effectively reduces the heading angle error from 7.3505° to 1.6619° when multiple occlusions, including branches, clouds, and buildings, are present simultaneously. This corresponds to an approximately 77.3% improvement compared to the error before occlusion removal. These findings underscore the ability of the method outlined in this paper to obtain relatively accurate navigation information, even in cases where the atmospheric polarization mode is disrupted.