Real-Time Object Detection with OpenCV and YOLO

This project demonstrates real-time object detection using Python, OpenCV, and the YOLO (You Only Look Once) object detection model. The application can identify and locate various objects in video streams from a webcam or a provided video file in real-time.

Features

• Real-time object detection using webcam footage or any video file provided by the user
• Integration with the pre-trained YOLO model for accurate object detection
• Bounding box visualization around detected objects with labels showing the classification
• Display of performance metrics such as frame rate (FPS) in real-time
• Non-maximum suppression to reduce overlapping detections

Prerequisites

Before running the script, ensure that you have the following dependencies installed:

• Python 3.x
• OpenCV: Install using pip install opencv-python
• NumPy: Install using pip install numpy

You will also need the YOLO model files:

• yolov4.weights: Download the pre-trained YOLO weights file from the official YOLO website or a reliable source.
• yolov4.cfg: Download the YOLO configuration file that defines the model architecture.
• coco.names: Download the COCO dataset class names file, which contains the names of the objects that YOLO can detect.

Place these files in the same directory as the Python script.

Usage

1. Clone the repository or download the Python script and the required YOLO model files.
2. Open a terminal or command prompt and navigate to the project directory.
3. Run the script using the following command:

   python object_detection.py
   

4. The script will open a window displaying the real-time video stream with object detection overlay and FPS counter.
5. To use a webcam, ensure that a webcam is connected to your computer. The script will automatically use the default webcam (index 0).
6. To use a video file instead of a webcam, provide the path to the video file in the cv2.VideoCapture() function call in the script.
7. Press 'q' to quit the application.

Customization

• To use a different YOLO model version (e.g., YOLOv3), download the corresponding model files and update the file paths in the script accordingly.
• Adjust the confidence threshold (currently set to 0.5) and non-maximum suppression threshold (currently set to 0.4) in the script to fine-tune the object detection performance.
• Modify the FPS display position or color by changing the cv2.putText() parameters in the script.

Performance

The script now includes real-time FPS (Frames Per Second) calculation and display, allowing you to monitor the performance of the object detection system. The FPS is shown in the top-left corner of the video frame.

Future Improvements

Potential future enhancements for this project include:

1. Error handling for missing files (weights, config, names).
2. Command-line arguments for input source and confidence threshold.
3. Saving the output video with detections.
4. Multi-threading to improve performance.

Acknowledgements

• The YOLO object detection model is developed by Joseph Redmon et al. For more information, refer to the YOLO website.
• The OpenCV library is used for video capture and image processing. Visit the OpenCV website for more details.

License

This project is licensed under the MIT License.