Sky Polarization Pattern Reconstruction and Neutral Line Detection Based on Adversarial Learning

The neutral line, a crucial feature found in sky light polarization pattern images, plays a crucial role in bio-inspired polarization navigation. However, current methods for detecting the neutral line typically assume clear and unobstructed conditions. Unfortunately, the sky polarization pattern (SPP) is susceptible to various weather conditions and geographic environments, leading to distorted patterns caused by random occlusions. Consequently, accurately determining the orientation information of the neutral line becomes challenging. To address this issue, we conduct an in-depth analysis of the distribution characteristics of the neutral line in the SPP using both simulation and real measurement experiments. In addition, we curate a comprehensive dataset known as the SPP dataset. Leveraging an adversarial learning-based approach, we propose a novel method for reconstructing SPP images. Subsequently, we perform neutral line detection on the reconstructed SPPs. Notably, our method showcases exceptional performance in detecting neutral lines, even in the presence of up to 80% occlusions within the area of interest. Experimental evaluations, conducted on both simulated and measured SPPs, validate the effectiveness and robustness of our proposed method. Furthermore, quantitative analyses of the experimental results confirm the stability and reliability of our approach. Our work makes a valuable contribution to the field of SPP reconstruction and neutral line detection under complex conditions. Moreover, it holds significant potential for advancing the measurement and application of polarization techniques. Data and models are available at:
https://github.com/WNCI
.