Real-Time Object Detection Warning System

Project conducted: May 2024 (was not uploaded at the time)

A real-time computer vision system using YOLO for object detection with danger zone alerts, sound notifications, and automated Telegram & Email warnings.

---

Introduction

This project aims to build an automated safety warning system that uses computer vision and object detection to identify intruding objects within a predefined danger zone. When a detected object enters the danger area, the system will:

• Play an alarm sound locally
• Send a warning message with image evidence via Telegram and Email

Key Applications:

• Intruder detection for security systems
• Safety monitoring in industrial zones
• Automated alerts in self-driving car assistance systems

---

Dataset

Source

• Images: 5,074 annotated images (4712 person, 362 non-person)
• Model: Pre-trained YOLOv4 model used for object detection

Performance

• Precision: 0.9985
• Recall: 0.9985

Tools & Technologies

• Languages: Python
• Libraries: OpenCV, YOLOv4, Shapely, smtplib, telegram API
• Hardware: Requires GPU for optimal real-time performance

---

Project Workflow

1. Setup & Running the System

• Run code/main.py
• The script will:
  • Open webcam feed
  • Allow the user to define a danger zone by clicking 4 points using the mouse (D key to finalize)
  • Start real-time object detection using YOLOv4
  • Continuously check if any detected object’s coordinates fall inside the defined danger zone

2. Object Detection

• Utilizes YOLOv4 to detect objects within the camera feed in real-time.
• Bounding boxes are drawn around detected objects with labels and confidence scores.

3. Danger Zone Checking

• Uses Shapely library to:
  • Define and process the user-drawn polygon as the danger zone
  • Check if object centroids fall within this zone

4. Warning & Notification

• If an object enters the danger zone:
  • Sound alarm: plays a local audio file as an immediate alert
  • Telegram alert: sends a photo with a warning message to a pre-configured chat
  • Email notification: sends an email with the captured frame and warning details

---

Result

Running the system displays:

• Real-time webcam window with bounding boxes and danger zone overlay
• Console log of detection events and notification status
• Sends alerts to configured Telegram chat and Email address when an intrusion is detected

---

How to Use

1. Clone the repository and install required libraries (code/setup.txt).
2. Configure your Telegram bot token, chat ID, and email SMTP settings in code/main.py.
3. Run:

   python main.py
   

Project Resources

To understand this project in detail:

• Presentation: Read the project presentation slides in slide_presentation.pdf
• Demo Video: Watch the demo video to see the system in action

Both files are included in this repository for better understanding of the workflow, functionality, and real-world application of the system.

Learning Outcomes

• Built a functional real-time object detection system with YOLOv4.
• Integrated computer vision with notification APIs for automated alerts.
• Applied spatial geometry checking (Shapely) for zone-based detection.
• Strengthened skills in Python, OpenCV, and deployment of AI-powered safety solutions.

Requirements

• opencv-python
• numpy
• imutils
• python-telegram-bot

Please install using pip and ensure compatible versions based on your Python environment.