RoadWatch MP

Potholes in Madhya Pradesh are reported one at a time, by hand, if at all. A citizen has to stop, photograph the hole, open the Lokpath app, type the location and describe the damage. Most people never do, so the Public Works Department learns about a stretch of broken road only after enough complaints pile up or a crash makes the news. India records thousands of pothole-related road deaths every year, and two-wheeler riders, auto drivers and bus commuters bear most of that risk.

The people who suffer are on both sides of the counter. Road users hit the same holes for months. PWD engineers and the MPSeDC team running Lokpath receive sparse, duplicated, poorly located complaints with no severity signal, so they cannot prioritise repairs or prove a road was fixed. The missing piece is a way to survey a whole road in one pass and hand the government a clean, geotagged list it can act on.

RoadWatch MP turns any drive into a pothole survey. A user uploads dashcam or phone footage, or points a phone camera at the road live. A YOLOv8 model fine-tuned on potholes detects them in every sampled frame, and a tracker follows each hole across frames so it is counted once and given a stable ID such as PH-001, not reported ten times as ten complaints. Each pothole is geotagged from the ride's GPS track, from a simulated route for a demo, or from the phone's location in live mode, and graded Low, Medium or High from its size and how long it stays in view. The best frame is saved as evidence.

When the footage ends the user has a reportable collection: a map with the route and every pothole, a table with coordinates, severity, confidence and a snapshot, and exports as CSV, JSON and GeoJSON that drop straight into GIS tools. One button files the whole collection through Lokpath, the MPSeDC-operated grievance gateway, and returns a complaint ID per pothole with the responsible PWD division, ward and repair SLA. The Lokpath call is a mock in this prototype that mirrors the real payload, so a production integration is a URL change, not a redesign.

The result is that a single fifteen-minute drive produces dozens of deduplicated, located, prioritised complaints with photo evidence, in the format the government already uses, with no typing by the citizen. The stack is a FastAPI backend with real-time WebSocket streaming and a React frontend, running on a laptop GPU or a free CPU host.