轨道交通低能见度场景图像增强系统研究

Research on Image Enhancement System of Low Visibility Scene in Rail Transit

摘要: 目标检测、语义分割等视觉任务应用于轨道交通的众多场景，大多数视觉系统被设计为在清晰的环境中执行，然而真实的轨道交通场景必然会包含退化的图像场景，列车一年四季在外遇到的恶劣的天气，以及面对隧道、夜间等环境中较差的照明，这些退化的图像会降低高级视觉任务的性能。本文关注于在恶劣天气(雾天，降雨)和弱光条件导致的能见度低的场景下进行图像增强，改善退化图像质量，为此，本文提出了一种轨道交通低能见度场景图像增强系统，可以自适应地对天气和照度进行分类并对分类后得到的低照度、雾天和雨天这三种场景图像进行增强。将该系统运用于轨道交通场景下的多目标检测以及语义分割应用中，实验结果表明，对于低照度图像以及雾天图像，本文系统可以提升多目标检测及语义分割约1%的准确率，对于雨天图像，本文系统可以提升多目标检测及语义分割4%以上的准确率。

Abstract: Visual tasks such as object detection and semantic segmentation are applied to many scenes in rail transit. Most visual systems are designed to execute in clear environments. However, real rail transit scenes inevitably contain degraded image scenes, such as the harsh weather encountered by trains outside all year round, as well as poor lighting in environments such as tunnels and nighttime. These degraded images can reduce the performance of advanced visual tasks. This article focuses on image enhancement in scenes with low visibility caused by severe weather (foggy, rainy) and weak light conditions to improve degraded image quality. For this reason, this paper proposes a low visibility scene image enhancement system for rail transit, which can adaptively classify weather and illumination and enhance the classified images of low illumination, foggy, and rainy scenes. Applying this system to multi-object detection and semantic segmentation applications in rail transit scenarios, experimental results show that the system can improve the accuracy of multi-object detection and semantic segmentation by about 1% for low illumination images and foggy images. For rainy images, the system can improve the accuracy of multi-object detection and semantic segmentation by more than 4%.