Analysing Videos

(Source: Medium)

Moving from static images to videos, YOLO demonstrates its real power. With its high-speed analysis, YOLO can process each video frame individually and provide real-time insights. Whether detecting objects, segmenting images, or estimating poses, the procedure mirrors that used for static images but adds the dimension of handling sequential frames.

Project Setup

To start working with video analysis, we'll extend our previous project structure and create a new Jupyter notebook yolo_video.ipynb:

📁 computer_vision/
    ├── 📁 .venv/
    ├── 📁 pics/
    ├── 📄 yolo_detect.ipynb
    ├── 📄 yolo_segment.ipynb
    ├── 📄 yolo_keypoints.ipynb
    └── 📄 yolo_video.ipynb

Ensure your virtual environment (.venv) is active and that all necessary packages, including ultralytics and opencv-python, are installed.

OpenCV Installation

OpenCV is one of the few Python packages that have different names when installed and integrated into our notebook. The installation is carried out using:

pip install opencv-python

Capture Video Stream

To analyze video data, whether from a webcam or a saved file, we leverage the Python package OpenCV (cv2). Let's begin with a program to access your web camera and display its live feed.

Step 1: Import the Library

To use OpenCV, start by importing the library:

import cv2

This statement includes the OpenCV library in our program, giving us access to its methods and properties.

Step 2: Create a VideoCapture Object

In OpenCV, the VideoCapture() method allows us to capture the video stream from our webcam:

cap = cv2.VideoCapture(0)

The argument 0 refers to the first camera connected to the device. If additional cameras are connected, you can use 1, 2, etc.

Saved Video

You can also open videos from your hard drive or from a website like Youtube . Simply enter the path to your video instead of 0. Don't forget to read the docs.

Step 3: Read Frames

The read() method of the VideoCapture object retrieves each frame from the video stream:

ret, frame = cap.read()

• ret: Boolean indicating if the frame was captured successfully.
• frame: The captured frame as a NumPy array.

Step 4: Display Frames

To display the captured frames in a window, use the imshow() method:

cv2.imshow('Capturing Video', frame)

The first argument is the window title, and the second argument is the frame to display.

Step 5: Loop and Exit

To continuously capture frames, use a while loop and break it based on user input. Use cv2.waitKey() to listen for key presses:

if cv2.waitKey(1) & 0xFF == ord('q'):
    break

This stops the loop when the q key is pressed.

Step 6: Release Resources

Release the video stream and close any OpenCV windows:

cap.release() # release the resource
cv2.destroyAllWindows()     # closes all OpenCV windows

Release Resources

Releasing resources at the end is crucial to avoid issues. If the resource (e.g., webcam) is not released, it may remain locked, preventing further connections. This issue can occur if an error interrupts your code, skipping the release command. In such cases, manually execute the release method before attempting to use the webcam again. Alternatively, restarting the kernel can also resolve the issue:

Complete Program

Here’s the complete program to access your webcam and show the live feed.

import cv2

# Define the video source (0 for webcam or path to a video file)
video_source = 0  # Use "video.mp4" for a saved video
cap = cv2.VideoCapture(video_source)

# Process the video frame by frame
while cap.isOpened():
    ret, frame = cap.read()
    if not ret:
        break

    # Display the  frame
    cv2.imshow("Video Analysis", frame)

    # Exit when 'q' is pressed
    if cv2.waitKey(1) & 0xFF == ord('q'):
        break

# Release resources
cap.release()
cv2.destroyAllWindows()

Inference

YOLO processes each video frame independently, making it suitable for real-time applications like detection, segmentation, or keypoint extraction. To use any of these vision algorithms, simply analyze each frame in sequence and display the annotated frame.

results = model(frame)
annotated_frame = results[0].plot()

Task: Analyze a Video

1. Try to access the webcam and display the live video
2. Use the YOLOv8 Nano model to perform a:
   • Detection
   • Segmentation
   • Pose Estimation
   • Tracking
3. Visualize and save results for further analysis.