AI-based Road Defect Detection System

We are building a lightweight MVP to detect potholes & road defects from regular mobile-mounted videos (similar to RoadMetrics). The system will be used by delivery-partners (Zomato/Swiggy-style) who automatically capture road conditions during rides.

Scope of Work

We need an AI/ML + backend developer to build:

1. AI / Computer Vision Model

Detect potholes, cracks, rough patches from dashboard-style video (1080p/720p)

You can use YOLOv8/YOLOv11/Detectron2/RT-DETR or any optimized model

Output required:

Pothole bounding boxes

Severity score

GPS tag (from metadata or manual input)

JSON summary

2. Backend + API

Simple Python API (FastAPI preferred)

Accepts uploaded videos

Returns processed JSON + snapshot images

Store data in lightweight DB (SQLite or Firebase)

3. Simple Dashboard (optional if you can do frontend)

View processed data

Map with pothole markers

Download JSON/CSV

You Don’t Need to Build a Full App

Just MVP: upload → detect → output.

Skills Needed

Python + FastAPI

Computer Vision (YOLO/Segmentation/Video processing)

PyTorch / TensorFlow

Basic cloud deployment (AWS/Linode/VPS)

Budget

₹60,000 – ₹90,000 (fixed price)
Paid in milestones.
Small paid test-task will be required (detect 5 potholes in sample video).

What to Include in Your Proposal

Past work in computer vision (especially object detection)

Model you plan to use

Delivery timeline

Links to GitHub or portfolio