Smart Traffic Monitoring System

An AI/ML-based vehicle detection, tracking, counting, and license-plate/helmet identification script using YOLOv8, DeepSORT, and EasyOCR.

Features Included

1. Vehicle Detection: Ignores pedestrians, strictly detects car, motorcycle, bus, truck seamlessly using YOLOv8.
2. Object Tracking: Uses deep-sort-realtime to stably assign a unique ID across frames.
3. Line Crossing & Counting: Contains an internal counting logic. When a tracked centroid crosses the virtual line on the screen, the vehicle is counted exactly once.
4. License Plate Reading (ANPR): Automatically crops the detected vehicle bounding box, runs thresholding and passes it through easyocr to detect number plates. (Fallback logic implemented if custom weights aren't supplied).
5. Helmet Detection: Specifically targets motorcycles when they cross the capture line.
6. Data Storage: Exports events iteratively to output/traffic_log.csv.
7. Recording: Records annotated footage matching the input video specs automatically to output/annotated_traffic.mp4.

Project Structure

smart_traffic_system/
├── requirements.txt
├── README.md
├── main.py                - System pipeline architecture
├── detector.py            - Inference logic for YOLOv8 and EasyOCR
├── tracker.py             - DeepSORT wrapper module 
├── utils.py               - Vector math and OpenCV rendering tools
├── models/                - Required. Pretrained model files (.pt)
└── output/                - Stores output annotated video and CSV

Setup Instructions

1. Environment Setup Ensure you have Python 3.9+ installed. It is recommended to use a virtual environment.

   python3 -m venv venv
   source venv/bin/activate

2. Install requirements:

   pip install -r requirements.txt

3. Get weights & a video: You must provide your own target .mp4 file explicitly in main.py, or simply place a sample target traffic footage and name it sample_traffic.mp4. To use the sample video:

   curl -L -o sample_traffic.mp4 "https://github.com/intel-iot-devkit/sample-videos/raw/master/car-detection.mp4"

4. Run the tool:

   python main.py

Custom Domain Tracking

By default, the script dynamically pulls yolov8n.pt and utilizes mocked logic combined with OCR fallback methods for plates and helmets logic respectively if they are missing true training models. If you have trained models for Plate Detection and Helmet detection (like from Roboflow Universe), place the .pt files in models/ and link them directly during initialization: detector = VehicleDetector(plate_model_path="models/plate.pt", helmet_model_path="models/helmet.pt")