gem-hunter
Language: Python
The project focuses on Tracking missing people. We are using Image processing and Machine learning along with Postgres Database.
Endorse on LinkedIn if this project was helpful.
Disclaimer
All images of individuals appearing in the screenshots and used as sample data in this project were sourced from the internet solely for the purpose of demonstrating the facial recognition pipeline in a non-commercial, educational context. These images are the property of their respective owners. No claim of ownership is made. If you are the rights holder of any image and wish it to be removed, please open an issue and it will be taken down promptly.
This project does not store, distribute, or commercialise any personal images. The face data derived from sample images (landmark vectors) is used only locally for matching demonstration and is not shared with any third party.
Hundreds of people — mostly children — go missing every day in India. When a sighting is reported, officers have to manually compare photos, sift through paperwork, and coordinate across stations. By the time a match is confirmed, the trail has often gone cold.
Step 1 — Family files a report. Officer registers the case.
A family in Haridwar reports their child missing. An officer opens the portal, uploads a photo, and the AI immediately detects the face and extracts a 468-point mesh — no manual tagging needed.
Step 2 — The dashboard tracks every open case.
The officer's home screen shows live counts of found and not-found cases, and a map that plots where cases are concentrated across India.
Step 3 — A member of the public submits a sighting.
Someone recognises the person and submits a photo through the public portal (no login required). The same face mesh is extracted and stored.
When an admin clicks Match Cases, the KNN model compares all sightings against all open cases. If a face is close enough, the case is automatically flipped to Found and the complainant is notified by email.
Step 4 — The city map tells the bigger picture.
Admins can see which cities have the most unresolved cases and track resolution rates over time.
No manual photo comparison. No paperwork pile-up.
| Feature | Details |
|---|---|
| Face detection | MediaPipe Face Landmarker — highlights detected faces, handles multiple people in frame |
| AI matching | KNN on 1,404-dimensional face vectors; shows confidence % |
| Video sightings | Upload a video — unique faces extracted automatically per frame |
| Live map | Dashboard map showing case density by city across India |
| Email alerts | Auto-notifies complainant email when a match is confirmed |
| Role-based access | Admins can match, edit, delete; Officers can register and view |
| Public portal | Separate mobile-friendly submission page, no login needed |
git clone https://github.com/gaganmanku96/Finding-missing-person-using-AI.git
cd Finding-missing-person-using-AI
pip install -r requirements.txtRun the officer/admin portal:
streamlit run Home.py
Run the public submission portal:
streamlit run mobile_app.py
The SQLite database and face landmarker model (~30 MB, auto-downloaded on first use) are created automatically.
Set these environment variables to enable email alerts on match:
SMTP_HOST, SMTP_PORT, SMTP_USER, SMTP_PASSWORD
The complainant's email entered during case registration is used as the recipient.
Credentials are stored in login_config.yml. To add or change a user:
1. Generate a bcrypt password hash:
import bcrypt print(bcrypt.hashpw(b"your_password", bcrypt.gensalt()).decode())
2. Edit login_config.yml:
credentials: usernames: your_username: # used as the login username name: Your Display Name email: you@example.com city: Delhi area: Sector 1 role: Admin # Admin or Officer password: '$2b$12$...' # paste the hash from step 1
Roles:
| Role | Permissions |
|---|---|
| Admin | Register cases, view all cases, trigger matching, edit/delete cases |
| Officer | Register cases, view own cases |
login_config.ymlis git-ignored by default. Never commit real credentials.
The scripts/ folder contains two utilities for populating the database with demo data.
# Download ~2 images per celebrity into scripts/bulk_data/reported/ uv run scripts/download_celebrity_images.py --dest reported # Or split across both folders (reported + publicly_seen) uv run scripts/download_celebrity_images.py --dest both
This uses DuckDuckGo image search — no API key needed.
You can also drop your own images directly into:
scripts/bulk_data/reported/ ← missing person cases
scripts/bulk_data/publicly_seen/ ← public sighting submissions
python scripts/bulk_upload.py
This processes every image in both folders:
resources/ so the app can display themBy default submitted_by is set to gagan (the username in the default login_config.yml). If you've changed your username, pass it explicitly:
python scripts/bulk_upload.py --officer your_username
To reset and re-seed from scratch:
sqlite3 sqlite_database.db "DELETE FROM registeredcases; DELETE FROM publicsubmissions;"
python scripts/bulk_upload.pyQ: Can I run this without an internet connection? The face landmarker model (~30 MB) is downloaded once on first run and cached locally. After that, both portals work fully offline.
Q: How accurate is the face matching? Accuracy depends heavily on photo quality. Front-facing, well-lit photos work best. The confidence score shown on each match reflects KNN distance — higher is a stronger match.
Q: Can I add multiple officers?
Yes. Add as many usernames as needed to login_config.yml. Each officer sees only their own registered cases; Admins see all cases.
Q: The map doesn't show a city I entered.
The map uses a built-in city → coordinates lookup. If a city is missing, open an issue or add it to the CITY_COORDS dict in Home.py.
Q: Where is the data stored?
Everything is in sqlite_database.db (git-ignored) in the project root. Images are stored as JPGs in resources/ (also git-ignored). Nothing is sent to any external server.
Q: How do I reset the database?
sqlite3 sqlite_database.db "DELETE FROM registeredcases; DELETE FROM publicsubmissions;"Or simply delete sqlite_database.db — it will be recreated on next run.
Q: Can the public portal be hosted separately from the officer portal?
Yes. They are independent Streamlit apps (Home.py and mobile_app.py) and share only the SQLite database. Point both to the same database file path and they will work together.
Thanks to the MediaPipe team for the open-source face landmarker model.