Custom-Object-Detection-from-Drone

A real-time object detection system using a Tello drone and OpenCV. This project enables autonomous object detection and tracking using a pre-trained SSD MobileNet V3 model.

Features

• Real-time object detection using a Tello drone
• Pre-trained SSD MobileNet V3 model for accurate detection
• Support for 91 different object classes (COCO dataset)
• Real-time video streaming and display
• Confidence threshold adjustment for detection accuracy
• Non-Maximum Suppression (NMS) for better detection results
• Battery status monitoring
• Automatic drone takeoff and altitude adjustment

Prerequisites

• Python 3.x
• Tello drone
• OpenCV (cv2)
• djitellopy library
• cvzone library

Installation

1. Clone the repository:

git clone https://github.com/yourusername/Custom-Object-Detection-from-Drone.git
cd Custom-Object-Detection-from-Drone

2. Install the required dependencies:

pip install opencv-python djitellopy cvzone

Project Structure

• main.py - Main script for object detection and drone control
• coco.names - List of object classes (91 classes from COCO dataset)
• frozen_inference_graph.pb - Pre-trained model weights
• ssd_mobilenet_v3_large_coco_2020_01_14.pbtxt - Model configuration file
• LICENSE - MIT License file

Usage

1. Ensure your Tello drone is powered on and connected to your computer's WiFi network
2. Run the main script:

python main.py

3. The drone will:
   • Connect to the Tello drone
   • Display battery status
   • Take off and move to a safe altitude
   • Start real-time object detection
   • Display detected objects with confidence scores

Configuration

You can adjust the following parameters in main.py:

• thres - Confidence threshold (default: 0.55)
• nmsThres - Non-Maximum Suppression threshold (default: 0.2)

Notes

• Make sure you have enough space for the drone to take off safely
• Keep the drone within WiFi range
• Monitor battery levels during operation
• The system uses the COCO dataset for object detection, supporting 91 different object classes

License

This project is licensed under the MIT License - see the LICENSE file for details.

Contributing

Feel free to submit issues and enhancement requests!