A Survey of Visual Attention Mechanismsin Degraded Environment Perception——Taking Foggy Vehicle Detectionas an Example

Wang Dejiao Xia Yuan Liu Yu

Chongqing Industry & Trade Polytechni

Abstract：

Environmental perception under complex meteorological conditions is a core bottleneck restricting the commercial deployment of high-level autonomous driving systems. The degradation of image contrast andtheloss of edge details caused by atmospheric scattering in foggy environments frequently lead to the failureof target feature extraction in visual detection algorithms. In recent years, visual attention mechanisms, with their inherent advantages in global receptive field modeling and anti-interference feature recalibration, have emergedas a critical breakthrough for addressing target detection challenges in degraded images. Focusing on the engineering application pain points of degraded environment perception, this paper uses foggy vehicle detectionas a concrete application scenario to systematically review the fundamental characteristics of channel attention, spatial attention, and self-attention mechanisms in feature restoration and object recognition. By comparing frontier studies since 2023, we analyze the performance differences among three mainstream technical routes: "dehaze-then-detect", "end-to-end feature enhancement", and "cross-domain adaptation". Utilizing quantitative metrics from public datasets such as Foggy Cityscapes, we discuss the feasibility of deploying lightweight attention modules on vehicle-mounted edge computing platforms. Finally, we project future development directions and emerging trends of attention mechanisms under multi-modal fusion, aiming to provide a valuable reference for optimizing intelligent connected vehicle vision systems under adverse weather conditions.

Key Words：

degraded environment perception; visual attention mechanism; foggy vehicle detection; deeplearning; autonomous driving; feature enhancement