Traffic-Surveillance-Application with Object Tracking and Detection using OpenCV & TensorFlow

An real time DVS-style traffic monitoring system for:

• Real-time vehicle detection and tracking
• Vehicle classification (Car, Bike, Truck) using a self trained CNN
• Event-frame generation inspired by Dynamic Vision Sensors (DVS)
• Speed estimation based on movement across predefined virtual lines

🎥 Watch Demo on YouTube

What's Special?

Unlike traditional frame-based detection, this project uses DVS-inspired motion detection — using pixel-level intensity changes between consecutive frames to generate event frames, making it very efficient and responsive to motion.

Features

• Motion detection via event-based frame differencing (DVS-inspired)
• Real-time object tracking with contour detection and centroid association
• Vehicle classification (Car, Bike, Truck) using a trained CNN model
• Relative speed estimation using timestamps between virtual lines across the road.
• Overlay visualizations with object IDs, motion arrows, and system metrics (CPU, latency, memory).

Tech Stack and Libraries

• Python 3.10+
• TensorFlow – Vehicle classification
• OpenCV – Image processing & tracking
• NumPy, SciPy – Centroid distance computation
• psutil – System performance monitoring
• os, time – Standard library utilities

📁 Project Structure

traffic-surveillance/
├── collect_data.py           # Save DVS-style ROIs from video
├── train_model.py            # Train CNN on vehicle images
├── main.py/                   # Real-time detection, classification & speed tracking
│   └── vehicle_cnn_model.h5  # Saved trained model
├── vehicle_dataset/
│   ├── Car/
│   ├── Bike/
│   └── Truck/
├── Resources/
│   ├── traffic_3.mp4         # For data collection
│   └── traffic_5.mp4         # For real-time inference
└── README.md  

---

Setup Instructions

1. Clone the repository

   git clone https://github.com/your-username/traffic-surveillance.git
   cd traffic-surveillance

2. Install dependencies

   pip install opencv-python tensorflow numpy scipy psutil

Step 1: Collect Training Data (DVS ROIs)

Extract grayscale event-based ROIs from a video for each class label.

   python collect_data.py

Controls:
Press s → Save current ROI (image)
Press n → Switch label (Car → Bike → Truck)
Press p → Pause and resume
Press q → Quit

🛠 Make sure Resources/traffic_3.mp4 exists before running.

Step 2: Train the CNN

Train a basic CNN to classify grayscale ROIs using the saved dataset.

   python train_model.py

Input: Grayscale images (120×100)
Output: Trained model saved as model/vehicle_cnn_model.h5

Step 3: Run the DVS Surveillance System

Run the real-time DVS-style vehicle detection, classification, and speed tracking system.

   python main.py

It performs -

• DVS-style motion detection via frame differencing
• Vehicle classification using the trained CNN
• Centroid-based object tracking with ID persistence
• Speed estimation from movement across lines (y = 240 → 200)
• Live overlay: object boxes, motion arrows, labels, and stats
  
  

Built with ❤️ by Shreya Verma