Vision Enhancement in Homogeneous and Heterogeneous Fog

Tarel, J-P; Hautiere, N.; Caraffa, L.; Cord, A.; Halmaoui, H.; Gruyer, D. · 2012 · OpenAlex-citations

DOI: 10.1109/mits.2012.2189969

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of enhancing visibility in foggy road images for camera-based Advanced Driver Assistance Systems (ADAS). Fog reduces contrast and fades colors, impairing the performance of detection algorithms and driver visibility. The authors identify that existing visibility enhancement methods are either too slow for real-time applications, rely on rigid geometric models that lack flexibility, or fail to distinguish between fog and large uniform gray regions like roads, leading to over-enhancement artifacts. The study aims to develop a fast, robust algorithm capable of handling both homogeneous and heterogeneous fog while specifically accounting for the planar nature of road surfaces. The proposed method, named NBPC+PA, formulates visibility restoration as the inference of the local atmospheric veil from three specific constraints. First, it utilizes photometric constraints based on Koschmieder’s law. Second, it applies a "no-black-pixel" constraint (NBPC), which ensures that the local standard deviation of enhanced pixels remains lower than their local average, preventing negative intensity values. Third, it introduces a planar assumption constraint (PA), leveraging the knowledge that a significant portion of the image corresponds to a planar road. This third constraint prevents the over-estimation of the atmospheric veil in the lower part of the image by using known camera calibration and a minimum visibility distance assumption. The algorithm combines these constraints to estimate the atmospheric veil, which is then removed to restore the scene luminance. To evaluate the algorithm, the authors conducted a comparative study against state-of-the-art methods, including Multiscale Retinex (MSR), Contrast-Limited Adaptive Histogram Equalization (CLAHE), Dark Channel Prior (DCP), and Free-Space Segmentation (FSS). The evaluation utilized two datasets: 66 synthetic images generated with physically-based fog models (uniform and heterogeneous) and 40 real camera images. Quantitative metrics and visual inspections were used to assess performance. The results demonstrated that the NBPC+PA algorithm outperforms existing methods, particularly in handling heterogeneous fog and avoiding the over-enhancement of road textures that plagues local methods like NBPC and DCP. Additionally, the proposed algorithm operates close to real-time, making it suitable for practical ADAS implementation. The significance of this work lies in providing a computationally efficient solution for visibility enhancement that is specifically tailored to road scenes. By integrating the planar road assumption into a local regularization framework, the method achieves superior contrast restoration without introducing artifacts. The paper also proposes a model to estimate the potential safety benefits of such systems, linking visibility enhancement settings to the probability of fatal injury in accident scenarios. This contributes to the development of more reliable Fog Vision Enhancement Systems (FVES) that can improve driver awareness and object detection performance in adverse weather conditions.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• weather rain fog snow