Logo Lanfrica

hatieku-boateng/road-defect-collector

Domaine:

mobility

Type de record:

software
Créateur:
hat
Hôte:
Mobile-friendly road image collection and review system for Ghana road defect monitoring. # Ghana Road Defect Monitoring A mobile-friendly road image collection and quality-review system designed to prepare reliable Ghanaian road-defect data for later AI modelling. ## Features - Collector ID and road-area capture - Manual camera capture or road image upload - Device GPS capture with accuracy and timestamp - Final record review before submission - Private image storage with Vercel Blob - Metadata storage with Neon Postgres - Password-protected administrator dashboard - Submission totals and filters - OpenStreetMap location overview - Approve or reject quality-review workflow - Local drone-video frame extraction - Open-source Grounding DINO pothole candidate detection - Human selection of AI candidates before upload ## Application routes - `/` — project landing page - `/collect` — collector submission form - `/drone` — browser-based drone AI analysis - `/admin/login` — administrator sign-in - `/admin` — protected review dashboard - `/api/health` — deployment configuration check ## Required environment variables Copy `.env.example` to `.env.local` and configure: ```text DATABASE_URL BLOB_READ_WRITE_TOKEN ADMIN_PASSWORD SESSION_SECRET ``` The database table and indexes are created automatically on the first database operation. ## Drone AI pilot The drone workflow samples up to 12 frames at two-second intervals. The Apache-2.0 Grounding DINO Tiny ONNX model runs in the reviewer's browser through Transformers.js. The original video is not uploaded. Only candidate frames chosen by a human reviewer are saved, and they enter the normal administrator review queue with a pending status. The model is downloaded from Hugging Face on first use and cached by the browser. This pilot searches only for potholes and must not be treated as a final trained road-defect model. ## Local development ```bash npm install npm run dev ``` Open `localhost` in a browser. ## Deploy a preview to Vercel ```bash vercel ``` Only promote to production after rev …