---

Smart Traffic AI System

Embedded Computer Vision | Real-Time Traffic Optimization | Wrong Parking & Wrong Way Detection | Accident Detection

---

Project Overview

This project contains a suite of AI systems for smart traffic management, designed for real-time vehicle monitoring, anomaly detection, and traffic optimization:

1. Traffic Light Optimization – Dynamically controls traffic signals based on real-time vehicle density.
2. Wrong Parking Detection – Detects stationary vehicles and identifies wrongly parked cars.
3. Wrong Way Detection – Monitors traffic lanes for vehicles moving in the wrong direction.
4. Accident Detection [prototype only - planned for final phase] – Identifies accidents on the road by detecting abnormal vehicle motion or stoppages.

The system combines YOLOv8 object detection, DeepSORT tracking, and optional Firebase integration for real-time reporting.

---

Hardware & Prototype

The project has two phases: prototype and full-scale embedded deployment.

Prototype Phase

Used for testing the algorithms on a small scale:

Component	Purpose
ESP32	Simulate traffic light signals on small-scale intersections
Toy vehicles	Represent real traffic for testing traffic optimization logic
Tapo C100 camera	Capture video of toy traffic environment for detection and analysis
Laptop	Run YOLO inference and simulate control logic
Lightweight YOLO model	Detect vehicles in real-time on the prototype setup

---

Full-Scale Embedded System (Funded Upgrade)

for on-road deployment:

Component	Purpose
Raspberry Pi 5 (8GB RAM)	On-device AI inference and traffic control
Hikvision DS-2CD1027G0-L IP camera	Outdoor video capture for real-time traffic monitoring
Traffic lights	Controlled by Raspberry Pi for dynamic signal changes
YOLOv8 (optimized)	Real-time vehicle detection and traffic density estimation
Firebase or MQTT	Optional cloud integration for status reporting, alerts, and dashboards
Mobile App	Allows citizens to monitor traffic and authorities to receive violations and anomalies

💡 Note: The accident detection module is currently at prototype stage and will be part of the next phase.

---

Architecture

             ┌─────────────────────────┐
             │        Camera/Video     │
             │  (Webcam, IP, MP4)     │
             └─────────┬──────────────┘
                       │
                       ▼
             ┌─────────────────────────┐
             │   Preprocessing Module  │
             │ - Resize frames         │
             │ - Split lanes (traffic)│
             └─────────┬──────────────┘
                       │
                       ▼
    ┌───────────────────────────┐
    │       YOLOv8 Detection     │
    │ - Vehicle detection        │
    │ - Wrong parking detection  │
    │ - Wrong way detection      │
    │ - Accident detection       │
    └─────────┬───────────────┘
              │
              ▼
    ┌───────────────────────────┐
    │   Tracking & Analysis      │
    │ - DeepSORT tracking        │
    │ - Stationary vehicle calc  │
    │ - Lane direction analysis  │
    │ - Abnormal motion analysis │
    └─────────┬───────────────┘
              │
              ▼
    ┌───────────────────────────┐
    │ Visualization & Control    │
    │ - Traffic light signals    │
    │ - Bounding boxes & labels  │
    │ - Accident alerts          │
    │ - Live video stream        │
    └─────────┬───────────────┘
              │
              ▼
    ┌───────────────────────────┐
    │   Optional Cloud Services  │
    │ - Firebase status updates  │
    │ - Alerts & dashboard       │
    └───────────────────────────┘

---

Folder Structure

traffic_ai_system/
│
├── inference.py         # Unified inference runner
├── traffic.py           # Traffic signal system
├── parked.py            # Wrong parking detection
├── wrong_way.py         # Wrong-way detection
├── accident.py          # Accident detection
│
├── models/              # Pretrained YOLO models
│   ├── yolov8n.pt
│   └── yolov8s.pt
│
├── outputs/             # Store processed videos & frames
│
├── media/               # Input images or videos
│   ├── traffic_videos/
│   ├── parking_videos/
│   ├── wrongway_videos/
│   └── accident_videos/
│
├── requirements.txt     # Python dependencies
└── README.md

---

Preparing Media Files

Place videos or images in the media/ folder:

• Traffic: media/traffic_videos/
• Wrong Parking: media/parking_videos/
• Wrong Way: media/wrongway_videos/
• Accidents: media/accident_videos/

Example:

media/
├── traffic_videos/traffic1.mp4
├── parking_videos/parking1.mp4
├── wrongway_videos/road1.mp4
└── accident_videos/accident1.mp4

---

Running the System

Traffic Light Optimization

python inference.py --mode traffic --source media/traffic_videos/traffic1.mp4

Wrong Parking Detection

python inference.py --mode parking --source media/parking_videos/parking1.mp4 --output outputs/parking_result.mp4

Wrong Way Detection

python inference.py --mode wrongway --source media/wrongway_videos/road1.mp4 --output outputs/wrongway_result.mp4

Accident Detection

python inference.py --mode accident --source media/accident_videos/accident1.mp4 --output outputs/accident_result.mp4

Firebase Integration (Optional)

• Configure Firebase in traffic.py or accident.py:

cred = credentials.Certificate("path/to/your/firebase.json")
firebase_admin.initialize_app(cred, {
    'databaseURL': 'https://your-database-url.firebaseio.com/'
})

• Status updates are automatically sent:

update_firebase_status('road1', 'Accident')

---

Demo phase model results

---

Project Book and Demo Video

Book[https://drive.google.com/file/d/11Q4jhJbcuXXnUtl_lyaJlae39yWCtJ-m/view?usp=sharing]

Video[https://drive.google.com/file/d/1aWYLqYVTHfPA0joM5Z6hLrywhT6NqOr_/view?usp=sharing]