Python Real-Time Object Detection

A Python-based real-time object detection application using the YOLOv8 model, OpenCV, and PyTorch. The program captures video from a webcam, processes each frame to detect objects, and displays bounding boxes and confidence scores for detected objects in fullscreen mode.

Features

• Real-time object detection using YOLOv8.
• Supports GPU acceleration with PyTorch and CUDA.
• Displays detected objects with labels and confidence scores.
• Resizeable Window.

Requirements

• Python 3.8+
• OpenCV
• PyTorch
• Ultralytics YOLOv8
• pygame

Note

Anything above version 3.12.4 for python will not work given that not every library is updated for said verison) Install the required libraries.

pip install opencv-python torch torchvision torchaudio ultralytics pygame

Configuration

This section explains how to tweak specific settings in the script to modify its behavior and performance based on your preferences or hardware capabilities.

Change Precision (16-bit or 32-bit)

By default, the model uses 32-bit precision. To improve performance on GPUs, you can switch to 16-bit precision. Modify the following lines in the code:

Note

The file is currently on 16 bit it can be changed by removing "float16" and leaving it blank.

• 32-bit Precision (Default):

  if device == "cuda":
      model.model.half()```

• 16-bit Frame Processing:

  if device == "cuda":
    frame = frame.astype("float16")```

Tip: Use 16-bit precision if your GPU has limited VRAM or if you want faster inference times. However, 32-bit precision provides slightly better accuracy.

Webcam Resolution

Set the resolution of the webcam input using these lines:

cap.set(cv2.CAP_PROP_FRAME_WIDTH, 1920)
cap.set(cv2.CAP_PROP_FRAME_HEIGHT, 1080)