Skip to content
Badisa Srikanth
All projects
Research proposalAI

RoadSense AI

AI-based Indian road damage detection and research project.

Research proposal. No dataset, experiments or results are published yet, so every feature is listed as proposed.

Problem

Damaged roads in India (potholes, cracks, waterlogging and damaged manholes) are reported slowly and repaired without a clear priority order.

Solution

Detect road damage from images, estimate severity and risk, and prioritise repairs, with a corrective RAG layer for grounded explanations.

Architecture (planned design)

  1. Image input from road photos
  2. Damage detection: potholes, cracks, waterlogging, damaged manholes
  3. Severity and risk assessment for each detection
  4. Corrective RAG: retrieve guidance, check its relevance, re-retrieve when it is weak
  5. Prioritised damage report
Design only. Components are not verified as built.

Implemented

Nothing verified yet. Features move here once a public repository, demo or screenshots show them working.

Proposed (not built yet)

  • Pothole detection
  • Road crack detection
  • Waterlogging detection
  • Damaged manhole detection
  • Severity and risk prioritisation
  • Corrective RAG architecture
  • Documented research methodology, dataset and evaluation
ResearchPublication status: Not published

RoadSense AI: Indian road damage detection with severity assessment and corrective RAG

Abstract

A proposed system that detects potholes, cracks, waterlogging and damaged manholes on Indian roads from images, scores each finding for severity and risk, and uses a corrective retrieval-augmented generation (CRAG) layer to produce grounded maintenance explanations. This is a research proposal; no results are reported yet.

IntroductionIn progress
RoadSense AI studies whether road photos can be used to find and rank road damage on Indian roads, and whether a corrective RAG layer can explain each finding with grounded maintenance guidance.
MotivationIn progress
Potholes, waterlogged stretches and open or broken manholes put road users at risk. Faster detection and a clear repair priority could help authorities act on the most dangerous damage first.
Problem statementWritten
Road damage in India is reported slowly and repaired without a clear priority order, which leaves dangerous potholes, waterlogged stretches and open manholes unattended.
Literature surveyNot written yet
The survey is not written yet. Starting references to review:
Existing system limitationsNot written yet
Will be added once this stage is written.
Research gapNot written yet
Will be added once this stage is written.
Research objectivesProposed
  • Detect potholes, cracks, waterlogging and damaged manholes in road images
  • Assign a severity and risk level to each detection
  • Ground maintenance explanations with corrective RAG
  • Produce a prioritised damage report
Proposed Corrective RAG architectureProposed
  1. Image input from road photos
  2. Damage detection: potholes, cracks, waterlogging, damaged manholes
  3. Severity and risk assessment for each detection
  4. Corrective RAG: retrieve guidance, check its relevance, re-retrieve when it is weak
  5. Prioritised damage report
Dataset collection planProposed
Planned classes: potholes, cracks, waterlogging and damaged manholes. The data source, dataset size and per-class counts will be published once the dataset is collected.
Research methodologyNot written yet
Will be added once this stage is written.
Model training methodologyNot written yet
Will be added once this stage is written.
Evaluation methodologyNot written yet
Will be added once this stage is written.
ResultsNot written yet
No experiments have been run, so there are no results yet.
LimitationsNot written yet
Will be added once this stage is written.
Future workNot written yet
Will be added once this stage is written.
Current research statusIn progress
Research proposal stage. The problem statement, objectives and proposed architecture are written and starting references are collected. No dataset has been collected and no experiments have been run, so results are pending.