Roeselare Road Line Quality Classification (Case Study)

Location: Roeselare, Germany

Project Overview

Nazru’s AI platform was deployed in Roeselare, Belgium, to automatically assess the quality of road markings (lines) using satellite and panchromatic (pan) imagery. This project involved classifying each road line segment into three quality levels: Good, Medium, and Bad. All outputs are delivered as polygon features only.

The Challenge

The client needed a city‑wide, objective, and repeatable assessment of road line visibility and integrity. Traditional manual surveys were slow, inconsistent, and disruptive to traffic. There was no automated method to distinguish between high‑quality, partially faded, and severely deteriorated lines from remote sensing data.

The Nazru Solution

Nazru’s platform integrated high‑resolution satellite imagery and panchromatic (pan) imagery with deep learning models. Our algorithms automatically:

• Detected road line segments (both continuous and dashed)

• Assessed line quality based on contrast, continuity, and visibility against the road surface

• Classified each line segment into three categories:

  • Good – clearly visible, high contrast, intact

  • Medium – partially faded or worn but still discernible

  • Bad – heavily faded, broken, or barely visible

• Fused panchromatic (high‑resolution structural detail) with multispectral satellite data for improved accuracy

Key Results & Benefits

• City‑wide digital map of road line quality for Roeselare

• Objective, automated, and scalable quality grading (Good/Medium/Bad)

• Prioritised maintenance planning based on bad/medium line segments

• Reduced manual inspection cost and improved safety

Key Technologies Used

• AI‑based line detection and segmentation

• Quality grading using contrast, edge strength, and pattern consistency

• Fusion of panchromatic and multispectral satellite imagery

Output Format (Only Polygon)

Deliverables are provided exclusively as vector polygons in the following formats:

• Shapefile (.shp)

• GeoJSON

• KML/KMZ

Polygon attribute table includes:

• marking_type – one of: dashed_line, solid_line, arrow, other

• quality – Good, Medium, Bad

• subtype – (for arrows: straight, turn_left, turn_right, turn_left_right, etc.; for other: crosswalk, stop_line, no_parking_zone, etc.)