Imaging through the fog is valuable for many areas, such as autonomous driving and cosmic exploration. However, due to the influence of strong backscattering and diffuse reflection generated by the dense fog on the temporal-spatial correlations of photons returning from the target object, the reconstruction quality of most existing methods is significantly reduced under dense fog conditions. In this study, we describe the optical scatter imaging process and propose a physics-driven Swin Transformer method utilizing Time-of-Flight (ToF) and Deep Learning principles to mitigate scattering effects and reconstruct targets in conditions of heterogeneous dense fog. The results suggest that, despite the exponential decrease in the number of ballistic photons as the optical thickness of fog increases, the Physics-Driven Swin Transformer method demonstrates satisfactory performance in imaging targets obscured by dense fog. Importantly, this article highlights that even in dense fog imaging experiments with optical thickness reaching up to 3.0, which exceeds previous studies, commonly utilized quantitative evaluation metrics like PSNR and SSIM indicate that our method is cutting-edge in imaging through dense fog.
Read full abstract