Gold Detection and Segmentation System

Real-time gold/jewellery detection system using YOLO models with person segmentation to suppress false positives, OCR weight reading, auto-recording, and a SQLite database for logging every detection event.

How It Works

1. Gold Detection — Custom YOLO11n model detects gold objects in a defined ROI
2. Person Segmentation — YOLO26-seg masks out people so gold worn by a person is ignored
3. Weight Reading — EasyOCR reads the weight from the scale display
4. Auto Recording — Video is recorded automatically when gold is detected, stops 10s after last detection
5. Database Logging — Every detection is logged to SQLite with video path, weight, timestamp, and extracted image

Quick Start

# 1. Install Python dependencies
pip install -r requirements.txt

# 2. Run the detection system
python3 GoldNormal.py

Press Q to quit.

GUI Viewer (browse detections)

pip install customtkinter pillow
python3 viewer/app.py

The viewer reads from the database and auto-refreshes every 5 seconds. It can run alongside the detection system.

Requirements

Python Dependencies

pip install ultralytics opencv-python easyocr numpy

Or use the requirements file:

pip install -r requirements.txt

Optional: Database Viewer

To visually browse the detection database:

sudo apt install sqlitebrowser
sqlitebrowser runs/jewellery_detections.db

Project Structure

Golddetection/
├── GoldNormal.py                  ← Main script (run this)
├── database/                      ← Database management package
│   ├── __init__.py
│   ├── db_manager.py              ← SQLite DB with deduplication
│   ├── image_extractor.py         ← Extracts best gold frame from videos
│   └── post_processor.py          ← Background thread for image extraction
├── viewer/                        ← GUI viewer application
│   ├── app.py                     ← Run this to launch viewer
│   ├── db_reader.py               ← Read-only DB queries
│   ├── file_opener.py             ← Cross-platform file opener
│   └── widgets/                   ← UI components
│       ├── sidebar.py
│       ├── topbar.py
│       ├── table_view.py
│       └── detail_panel.py
├── weights/                       ← All YOLO model files
│   ├── Yolo11n.engine             ← Gold detection (TensorRT)
│   ├── yolo26n-seg.onnx           ← Person segmentation (ONNX)
│   └── ...                        ← Other model formats (.pt, .onnx)
├── runs/
│   ├── recordings/                ← Auto-saved .mp4 clips
│   ├── images/                    ← Snapshots and extracted gold frames
│   ├── jewellery_detections.db    ← SQLite database (auto-created)
│   └── detection.log              ← Event log
├── test_db.py                     ← Database test suite
├── data.yaml                      ← Dataset config (Roboflow)
├── requirements.txt
└── ReadME.md

Models

Model	Format	Purpose
Yolo11n.engine	TensorRT	Gold/jewellery detection (primary)
yolo26n-seg.onnx	ONNX	Person segmentation

All models are stored in the weights/ directory.

Detection Classes (from data.yaml)

Bangles · Chain · Earrings · Gold Bar · Gold Coin · Ring

Database

Detection events are stored in runs/jewellery_detections.db (SQLite, auto-created on first run).

Checking the Database

Command line:

sqlite3 -header -column runs/jewellery_detections.db "SELECT * FROM gold_detections;"

GUI viewer:

sudo apt install sqlitebrowser
sqlitebrowser runs/jewellery_detections.db

Python:

from database.db_manager import JewelleryDBManager
db = JewelleryDBManager()
for row in db.get_all_detections():
    print(row)

Running Tests

python3 test_db.py
# Expected output: ALL TESTS PASSED

Dependencies

• PyTorch
• Ultralytics YOLO
• OpenCV
• EasyOCR
• NumPy
• CustomTkinter (viewer GUI)
• Pillow (image thumbnails in viewer)
• SQLite3 (built-in with Python)