[Paper Review] Measuring Visibility using Atmospheric Transmission and Digital Surface Model

Jean‐Philippe Andreu, Stefan Mayer+2 more|arXiv (Cornell University)|May 20, 2015

Image Enhancement Techniques3 references3 citations

TL;DR

This paper proposes an automated method to estimate airport visibility by combining image-based atmospheric transmission estimation with a high-resolution digital surface model (DSM). By detecting haze-free pixels via transmission maps and using DSM-derived depth maps, the approach computes visibility as the 99th percentile depth of visible pixels, achieving results comparable to METAR reports across visibility ranges from 250m to 10km.

ABSTRACT

Reliable and exact assessment of visibility is essential for safe air traffic. In order to overcome the drawbacks of the currently subjective reports from human observers, we present an approach to automatically derive visibility measures by means of image processing. It first exploits image based estimation of the atmospheric transmission describing the portion of the light that is not scattered by atmospheric phenomena (e.g., haze, fog, smoke) and reaches the camera. Once the atmospheric transmission is estimated, a 3D representation of the vicinity (digital surface model: DMS) is used to compute depth measurements for the haze-free pixels and then derive a global visibility estimation for the airport. Results on foggy images demonstrate the validity of the proposed method.

Motivation & Objective

• To overcome the subjectivity and limitations of human-based visibility reporting in air traffic control.
• To enable dense, pixel-level visibility estimation across an airport’s field of view, rather than relying on sparse landmark observations.
• To develop a physically grounded, image-processing-based method that leverages atmospheric transmission and 3D scene geometry for visibility estimation.
• To validate the method against official METAR reports using real foggy images from an airport environment.

Proposed method

• Estimates atmospheric transmission per pixel using a single-image haze removal technique based on the dark channel prior.
• Generates a depth map using a high-resolution (1m) digital surface model (DSM) to compute the distance to scene points in the camera's line of sight.
• Applies a global threshold (0.75) to the transmission map to identify haze-free pixels, assuming transmission > 0.75 indicates visibility.
• Computes visibility as the 99th percentile depth of all pixels with transmission above the threshold, reducing outlier influence.
• Corrects depth map artifacts caused by occlusions (e.g., buildings) by reconstructing facade projections from eaves lines.
• Uses a combined LiDAR and SRTM DSM to improve depth accuracy, especially in complex urban-like airport environments.

Experimental results

Research questions

• RQ1 Can atmospheric transmission estimation from a single image enable reliable visibility assessment in hazy conditions?
• RQ2 To what extent can a high-resolution digital surface model improve depth estimation for visibility computation?
• RQ3 How does the visibility distance derived from transmission and depth maps compare to official METAR reports?
• RQ4 What threshold value on the transmission map yields the most accurate visibility estimation?
• RQ5 Can local threshold adaptation improve visibility estimation accuracy over a global threshold?

Key findings

• The method achieved visibility estimates ranging from 102m to 12,178m, closely matching official METAR reports of 250m to 10km.
• For visibility distances of 250m, 500m, and 6000m, the measured values were 102m, 731m, and 6768m, respectively, showing strong agreement with reported values.
• The 99th percentile depth of haze-free pixels provided a robust visibility estimate, effectively reducing the impact of depth outliers.
• The use of a corrected depth map improved accuracy by resolving depth ambiguities caused by occlusions such as rooftops.
• The heuristic threshold of 0.75 on transmission yielded consistent results across diverse weather conditions, including fog banks and low-level clouds.
• Discrepancies between measured and reported visibility were attributed to human observer subjectivity and sparse landmark availability in certain distance ranges.