How to capture frames from recorded video with OpenCV

A fixed video clip gives computer-vision code the same encoded frames on every run, which makes decoding failures easier to reproduce than with a live source. OpenCV can open a local recording as a capture stream and stop immediately when the file or a requested frame cannot be read.

A filename passed to cv.VideoCapture selects a recorded-video source. The script checks isOpened() before reading, rejects an empty frame, and releases the capture object in a finally block after success or failure.

The bounded read keeps the decoded images in memory and reports their count and dimensions without writing extracted frames to disk. Use a clip with at least 60 decodable frames for the command shown here; shorter or damaged input exits before a success summary is printed.

1. Create capture_video.py with the input contract and insertion markers.

   capture_video.py

   #!/usr/bin/env python3
   import argparse
   import sys
   from pathlib import Path
    
   import cv2 as cv
    
    
   parser = argparse.ArgumentParser(description="Capture frames from a recorded video.")
   parser.add_argument("video", type=Path, help="path to a recorded video file")
   parser.add_argument("--frames", type=int, default=60, help="number of frames to capture")
   args = parser.parse_args()
    
   # VALIDATE_INPUT
   # OPEN_VIDEO
   # CAPTURE_FRAMES
   # REPORT_RESULT

2. Replace the VALIDATE_INPUT marker with the file and frame-count checks.

   if not args.video.is_file():
       sys.exit(f"Video file not found: {args.video}")
   if args.frames < 1:
       sys.exit("Frame count must be at least 1")

3. Replace the OPEN_VIDEO marker with recorded-video initialization.

   capture = cv.VideoCapture(str(args.video))
   if not capture.isOpened():
       sys.exit(f"Cannot open video: {args.video}")
    
   backend = capture.getBackendName()

   VideoCapture selects a decoder from the video backends included in the local OpenCV build. A file can exist but still fail here when its container or codec is unsupported.

4. Replace the CAPTURE_FRAMES marker with the bounded decoding loop.

   frames_captured = 0
   frame_size = None
    
   try:
       while frames_captured < args.frames:
           ok, frame = capture.read()
           if not ok or frame is None:
               sys.exit(
                   f"Cannot read frame {frames_captured + 1} from {args.video}"
               )
    
           frames_captured += 1
           frame_size = (frame.shape[1], frame.shape[0])
   finally:
       capture.release()

   capture.read() returns the decoded frame with a status value. End-of-file before the requested count and a decoder failure both stop the script instead of producing a partial success summary.

5. Replace the REPORT_RESULT marker with the recorded-video summary.

   width, height = frame_size
   print(f"Video: {args.video}")
   print(f"Backend: {backend}")
   print(f"Frames captured: {frames_captured}")
   print(f"Frame size: {width}x{height}")

6. Compare the assembled script with the complete capture_video.py file.

   capture_video.py

   #!/usr/bin/env python3
   import argparse
   import sys
   from pathlib import Path
    
   import cv2 as cv
    
    
   parser = argparse.ArgumentParser(description="Capture frames from a recorded video.")
   parser.add_argument("video", type=Path, help="path to a recorded video file")
   parser.add_argument("--frames", type=int, default=60, help="number of frames to capture")
   args = parser.parse_args()
    
   if not args.video.is_file():
       sys.exit(f"Video file not found: {args.video}")
   if args.frames < 1:
       sys.exit("Frame count must be at least 1")
    
   capture = cv.VideoCapture(str(args.video))
   if not capture.isOpened():
       sys.exit(f"Cannot open video: {args.video}")
    
   backend = capture.getBackendName()
   frames_captured = 0
   frame_size = None
    
   try:
       while frames_captured < args.frames:
           ok, frame = capture.read()
           if not ok or frame is None:
               sys.exit(
                   f"Cannot read frame {frames_captured + 1} from {args.video}"
               )
    
           frames_captured += 1
           frame_size = (frame.shape[1], frame.shape[0])
   finally:
       capture.release()
    
   width, height = frame_size
   print(f"Video: {args.video}")
   print(f"Backend: {backend}")
   print(f"Frames captured: {frames_captured}")
   print(f"Frame size: {width}x{height}")

7. Run capture_video.py against a recorded clip with at least 60 frames.

   $ python3 capture_video.py sample.avi --frames 60
   Video: sample.avi
   Backend: FFMPEG
   Frames captured: 60
   Frame size: 768x576

   The script prints all four lines only after VideoCapture opens the recording and decodes 60 nonempty frames. A missing file, unsupported video, or early read failure exits with an error instead.