Image based fog detection in vehicles

Pavlic, Mario; Belzner, Heidrun; Rigoll, Gerhard; Ilic, Slobodan · 2012 · OpenAlex-citations

DOI: 10.1109/ivs.2012.6232256

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 detecting fog in vehicle-mounted camera images to support intelligent driving assistance systems. While existing methods rely on analyzing local features such as lane markings, road regions, or the horizon, these approaches are often unreliable due to occlusions or varying feature quality. The authors propose a novel method that utilizes global image descriptors to distinguish between foggy and fog-free conditions. This approach is motivated by the observation that fog consistently causes a decrease in contrast and blurring across the entire image, regardless of specific scene content. Potential applications include automatic fog lamp control, high-beam adjustment, speed limit advice based on visibility, and local hazard warning systems. The methodology involves extracting global features from the power spectrum of the image, defined as the squared magnitude of the Fourier transform. Foggy scenes exhibit frequency components concentrated at zero frequency due to blurring, whereas clear scenes show a broadly spread spectrum. The process begins with preprocessing to normalize illumination and cancel mean intensity. Feature extraction is performed by sampling the power spectrum using a Gabor filter bank with varying frequencies and orientations, followed by Principal Component Analysis (PCA) to reduce dimensionality. The resulting feature vectors are classified using a Support Vector Machine (SVM) with a Radial Basis Function kernel. The authors also define fog categories based on visibility ranges derived from stopping distance calculations: No Fog (>1000 m), Low Fog (300–1000 m), Fog (100–300 m), and Dense Fog (<100 m). Experiments were conducted using data from the front camera of a BMW 5 series, capturing daytime highway scenes at 320x240 resolution. A dataset of 44,000 images was manually labeled into fog categories. The evaluation employed an 8-fold cross-validation scheme, ensuring diverse highway types were represented in both training and testing sets to assess generalization. The system achieved an overall accuracy of 94%, with a true positive rate of 93% for fog detection and a true negative rate of 96% for fog-free classification. The results demonstrate high recognition rates on approximately one hour of test video data. The study concludes that global spectral features offer a robust alternative to local feature analysis for fog detection, particularly for daytime driving scenarios. The high accuracy rates suggest the method is viable for real-time vehicle applications. However, the authors note limitations, including the exclusion of nighttime scenes and the need for further testing across varied road profiles and weather conditions. Future work aims to extend the method to nighttime detection and refine classifier parametrization for broader environmental applicability.

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.

• distraction detection algorithms