Skip to content

Civic Lens

Civic sense is a PoC app that analyzes dashcam videos for traffic violations, reckless driving incidents on roads.

Core trick

A local detector proposes violation windows, and only uncertain ones go from a cheap vision model to a stronger model for verification.

Tags

Description

Civic sense is a PoC app that analyzes dashcam videos for traffic violations, reckless driving incidents on roads. Indian roads as everyone knows are not for beginners, you require a special skill to navigate through them. Law enforcement is also not possible at every road and corners, that's why I built civic lens, an app that takes matter in it's own hands and identifies these bad characters and takes a note of them
priyanshu-09/civic-lens★ 00 forksPythonCreated Feb 14, 2026 · last push Feb 14, 2026

README excerpt

# Civic Lens (Hackathon PoC)

Civic Lens is a local-first PoC for detecting traffic violations in dashcam videos and generating a reviewable case pack.

## What it does
- Upload a video from UI
- Run local proposal engine to find candidate violation windows
- Verify and enrich candidates using Gemini (Flash -> Pro) when API key is available
- Route dynamically: local packets -> Flash, only uncertain Flash packets -> Pro
- Apply max caps and concurrent Gemini routing via `backend/config/perf_config.json`
- Review events manually (accept/reject, notes, include plate)
- Export a case pack ZIP with logs, JSON artifacts, report, and incident thumbnails
- Inspect full packet lineage (`local -> flash -> pro -> final`) via run trace
- See live packet/event updates in Review screen while pipeline is still running

## Violation types in this build
- `NO_HELMET`
- `RED_LIGHT_JUMP`
- `WRONG_SIDE_DRIVING`
- `RECKLESS_DRIVING`

## Demo Video

A professional Remotion demo video showcasing Civic Lens is available in `demo-video/`:

```bash
cd demo-video
npm install
npm run dev          # Preview the video
npm run build        # Render as MP4
npm run build-gif    # Render as GIF
```

The 90-second video walks through:
1. The problem (reckless driving, rule violations, zero accountability)
2. The solution (AI-powered detection + human review)
3. The user flow (upload → process → review → export)
4. Key features and impact

See `demo-video/README.md` for full details.

## Repository docs
- Arch

Team