Traffic Rule Violation Detection — AID 728 Final Submission

This repository contains the final computer vision violation detection system for Group 28.

📂 Main Submission Folder

All submission files, report document, source code, and local model weights are organized under the Group28 directory.

Please refer to Group28/README.md for detailed installation, dependencies setup, and usage instructions.

🎯 Quick Overview

Inside the submission folder, we provide two distinct implementations of the violation detector:

1. PaddleOCR PP-OCRv5 Pipeline (Group28/solution.py): Uses custom YOLO models and offline PaddleOCR for maximum speed and accuracy.
2. 6-Stage EasyOCR Pipeline (Group28/solution_6stage.py): Incorporates a 6-stage image preprocessing pipeline with Test-Time Augmentation (TTA) and offline EasyOCR.
3. Project Report (Group28/Group28.pdf): Detailed report for the system architecture and evaluation results.

Deployment

This repo includes a simple Streamlit app in app.py.

Run locally

pip install -r requirements.txt
streamlit run app.py

Docker

docker build -t traffic-violation-demo .
docker run -p 8501:8501 traffic-violation-demo

Streamlit Cloud

1. Push this repo to GitHub.
2. Go to Streamlit Community Cloud.
3. Select this repository.
4. Set the main file path to app.py.
5. Deploy.

Note: the model files are already bundled under Group28/models/, so the app does not need to download weights at runtime.