Measuring Fog Density With Pictures

Jim Crocker
3rd June, 2025

Measuring Fog Density With Pictures

Sample images from the Cityscapes dataset exhibiting increasing fog density levels of 0.005, 0.01, and 0.02 served as a synthetic benchmark to evaluate the accuracy of the proposed fog density estimation algorithm compared to ground truth data.

Image adapted from: Dai et al. / CC BY (Source)

Key Findings

  • In China, researchers developed an image-based method that combines dark channel, color saturation, and gray noise features to estimate fog density accurately in various traffic scenes
  • Tests on both synthetic images and real highway and airport videos show that the new method provides consistent and efficient fog assessments, outperforming previous techniques
Recent advancements have increased interest in using images to estimate fog density, an important factor for traffic management, surveillance, and safety systems. The China Academy of Railway Sciences and Guangdong University of Petrochemical Technology[1] have developed a method that improves both the accuracy and computational efficiency of fog density estimation based solely on images. This method is significant because it addresses limitations found in previous image-based techniques. In this study, researchers developed three distinct image features to capture the characteristics of fog. These features are based on the dark channel information, image saturation, and the proportion of gray noise points. The dark channel of an image refers to the minimal intensity value among color channels in a local region, often used to identify haze or fog effects in photographs. Image saturation indicates how vivid or pure the colors are; in foggy scenes, saturation usually decreases. Gray noise points are pixels that appear similar to a uniform gray, often the result of light scattering caused by atmospheric particles. By considering these features simultaneously, the method can better capture the varying effects of dense fog. The three features are then combined using a feature fusion process—merging different types of data into one powerful representation—that enhances the overall estimation accuracy. This integration of features ensures that the method encapsulates a more complete picture of fog properties than approaches which rely on a single indicator. The researchers also designed two performance indicators to evaluate fog density estimation methods. The sequential error indicator measures how well the method preserves the correct order of fog density values over time, while the proportional error indicator assesses the consistency of the estimated values relative to the actual fog levels. These new indicators provide a more robust framework for comparing various methods and ensuring that the estimates are both accurate and consistent. Previous research in related fields supports the use of image-based approaches for estimating atmospheric obscurants. For instance, one study[2] focused on haze density estimation by optimizing an objective function that considered brightness, saturation, and image sharpness, while also minimizing the dark channel. Although haze and fog are similar phenomena in that both reduce image clarity by scattering light, fog often represents a denser and more locally variable condition. The new study builds on these earlier findings by adapting the focus to fog density, which can benefit from a tailored combination of features. In related work, research in mobile systems emphasized the dangers of poor visibility due to weather phenomena such as fog, rain, and sun glare[3]. These studies underscored the importance of accurate visual estimation for decision-making in autonomous vehicles and driver assistance systems. The current study extends this line of research by providing a method that could ultimately support quicker and more reliable assessments of road conditions, especially in challenging weather. Another relevant study[4] explored fog density estimation by aligning video data with visibility sensor information to correct estimation errors. Unlike that approach, which integrates multiple sensor inputs, the current method relies solely on image data. This simplifies the implementation process and reduces dependency on additional hardware while still achieving superior performance compared to eight contemporary image-based methods. The fact that all processing is confined to image analysis reduces cost and complexity, and it allows easier research into dynamic, real-time fog monitoring in both indoor and outdoor surveillance scenarios. Experimental results from the new study show that the integrated method provides the best performance among its peers, clearly indicating that each of the three features—dark channel, saturation, and gray noise—significantly contributes to more accurate fog density estimation. This result is not only promising for researchers but also for practical applications, as it demonstrates that one can achieve high accuracy in a computationally efficient manner. The source code is publicly available, encouraging further research and development in this field. This latest development offers a practical solution to the long-standing challenge of accurately detecting and quantifying fog from images, marking progress over earlier attempts and consolidating the strengths of previous research efforts[2][3][4].

Environment

References

Main Study

1) Image based fog density estimation

Published 2nd June, 2025

https://doi.org/10.1371/journal.pone.0323536

Related Studies

2) Haziness Degree Evaluator: A Knowledge-Driven Approach for Haze Density Estimation.

https://doi.org/10.3390/s21113896

3) Visibility Enhancement and Fog Detection: Solutions Presented in Recent Scientific Papers with Potential for Application to Mobile Systems.

https://doi.org/10.3390/s21103370

4) Fog Density Analysis Based on the Alignment of an Airport Video and Visibility Data.

https://doi.org/10.3390/s24185930


Related Articles

An unhandled error has occurred. Reload 🗙