AI System for Preliminary Visual Classification of Road Damage

The problem

When field photographs, locations, road classes and maintenance histories are held in separate sources, their fields must be matched by hand before a draft damage type and inspection queue can be prepared.

The road maintenance engineer must compare every suggestion in the draft damage type and inspection queue with the source record; missing or conflicting evidence must not replace the final decision.

If a shadow, patch or wet surface that resembles damage is not reviewed separately, a normal surface could be treated as critical damage or a pothole could be missed; the record must not progress without a human decision.

What we set out to improve

• Time taken for the draft damage type and inspection queue to reach the road maintenance engineer with its supporting evidence
• Outcomes of records referred for human review because a shadow, patch or wet surface resembled damage
• Number of suggestions corrected after identifying the risk of treating a normal surface as critical damage or missing a pothole

How the system works

1. 01 Retrieve the field photograph, location, road class and maintenance history from authorised sources
2. 02 Verify field and document versions
3. 03 Prepare the draft damage type and inspection queue with its supporting evidence
4. 04 Exception check: a shadow, patch or wet surface resembles damage
5. 05 Have the road maintenance engineer accept, correct or reject the suggestion
6. 06 Decision gate: road closure, repair and priority decisions
7. 07 Record the decision and source separately

Methods we used

• A source-identified data schema for the field photograph, location, road class and maintenance history
• A structure that separates deterministic rules from AI suggestions
• A screen showing the draft damage type and inspection queue beside its supporting evidence
• A human review queue when a shadow, patch or wet surface resembles damage
• A record identifier that prevents duplicate processing

Where people stay involved

Data and security

Only the field photograph, location, road class and maintenance history fields required for the task should be processed. Role-based access should be applied, and unnecessary identity, health, financial and internal organisational data must not be sent to the model. Source retrieval, AI suggestions and the road maintenance engineer's decisions should be recorded separately.

Who this suits

Frequently asked questions