USRNet: Unified Scene Recovery Network for Enhancing Traffic Imaging under Multiple Adverse Weather Conditions

TL;DR

USRNet tackles the challenge of traffic-imaging under adverse weather by unifying restoration across haze, rain, snow, and mixed degradations. It introduces a modular architecture with a scene encoder, a Node Independent Learning Mechanism (NILM), an edge detector, and a scene restorer, all trained with a hybrid loss to preserve edges and capture degradation-specific features. NILM provides per-degradation specialization during training and sequential inference, enabling robust restoration of complex mixtures. Experimental results on RESIDE, Rain100L, CSD, and CDD-11 show state-of-the-art restoration and improved downstream object detection under adverse conditions, advancing reliable visual perception for intelligent transportation and surveillance in real-world weather.

Abstract

Advancements in computer vision technology have facilitated the extensive deployment of intelligent transportation systems and visual surveillance systems across various applications, including autonomous driving, public safety, and environmental monitoring. However, adverse weather conditions such as haze, rain, snow, and more complex mixed degradation can significantly degrade image quality. The degradation compromises the accuracy and reliability of these systems across various scenarios. To tackle the challenge of developing adaptable models for scene restoration, we introduce the unified scene recovery network (USRNet), capable of handling multiple types of image degradation. The USRNet features a sophisticated architecture consisting of a scene encoder, an attention-driven node independent learning mechanism (NILM), an edge decoder, and a scene restoration module. The scene encoder, powered by advanced residual blocks, extracts deep features from degraded images in a progressive manner, ensuring thorough encoding of degradation information. To enhance the USRNet's adaptability in diverse weather conditions, we introduce NILM, which enables the network to learn and respond to different scenarios with precision, thereby increasing its robustness. The edge decoder is designed to extract edge features with precision, which is essential for maintaining image sharpness. Experimental results demonstrate that USRNet surpasses existing methods in handling complex imaging degradations, thereby improving the accuracy and reliability of visual systems across diverse scenarios. The code resources for this work can be accessed in https://github.com/LouisYxLu/USRNet.

Figures (12)

• Figure 1: Severe weather conditions can significantly compromise the performance of VITS, resulting in decreased traffic efficiency and safety risks. Notwithstanding the propensity for weather-related disruptions to compromise transportation imaging sensors, the integration of AI and CV capabilities can effectively alleviate these disturbances, thereby ensuring an efficient and secure transportation system that can operate optimally even in the most adverse environmental conditions.
• Figure 2: Overview of the proposed USRNet for image restoration under complex imaging conditions, demonstrated through edge detection and image restoration tasks. The scene encoder extracts multi-scale generic visual representations from the degraded image. NILM incorporates a dedicated training node for each type of degradation, enabling each node to learn more specific and focused features, thereby enhancing the overall restoration performance. The edge decoder generates potential edge features, assisting the scene restorer in producing the final restored image.
• Figure 3: Illustration of the imaging degradation model under complex weather conditions, where uncertain combinations of factors yield a diverse range of degraded images.
• Figure 4: The pipeline of proposed dual residual (D-Res) block and standard residual (S-Res) block. D-Res will provide two output heads, each dedicated to learning and reasoning about edge features and global features of degraded images, respectively.
• Figure 5: The pipeline of proposed NILM. The standard convolutional layer (SCL) and global context attention (GCA) are used to extract long-range dependencies and global context information. During inference phase, NILM can generate latent features by adaptively calling parameters to specific training nodes for each type of degradation.