MicoFaith/rwanda-water-meter-reading-system

# Rwanda Water Meter Reading System An AI-based computer vision project for automatic water meter detection and reading using YOLO object detection. This project includes: - dataset collection - image labeling - dataset preparation - YOLO training - retraining - prediction - reading reconstruction - a Flask REST API for serving predictions - a SQLite database recording every reading - a lightweight web UI for field agents (upload + history) The system is trained to detect: - water meter - reading window - digits `0–9` - unclear digits (`unknown`) --- # Features The project provides tools for: - collecting water meter datasets - labeling water meter images - rotating and cleaning images - validating YOLO datasets - preparing train/val/test datasets - training YOLO models - retraining from latest best model - predicting on test images - reconstructing final meter readings --- # Project Structure ```text rwanda-water-meter-reading-system/ ├── raw_dataset/ │ ├── images/ │ └── labels/ │ ├── dataset/ │ ├── images/ │ │ ├── train/ │ │ ├── val/ │ │ └── test/ │ │ │ ├── labels/ │ │ ├── train/ │ │ ├── val/ │ │ └── test/ │ │ │ ├── reports/ │ └── data.yaml │ ├── 01_label_images.py # labeling GUI (PySide6) ├── 02_prepare_dataset.py # build train/val/test splits + data.yaml ├── 03_train.py # train a YOLO model ├── 04_predict.py # inference + reading reconstruction (single source of truth) ├── 05_retrain.py # continue training from latest best.pt │ ├── app.py # Flask API + web UI ├── database.py # SQLite persistence for readings ├── templates/ # index.html, history.html ├── static/ # style.css, script.js, history.js │ ├── runs/detect/training_runs/ # YOLO training output (best.pt lives here) ├── prediction_outputs/ ├── logs/ # app.log (created at runtime) ├── wmrs.db # SQLite database (c …