如何调整OpenCV代码提取更多视频帧并修正帧90°逆时针倾斜问题

Hey there! Let's work through your two OpenCV video processing issues step by step: getting more frames out of your 24-second video and fixing that annoying 90° counterclockwise tilt in the extracted frames.

Your original code reads frames at the video's native frame rate, which is why you're getting ~742 frames for a 24-second clip (that's roughly 31 FPS). To extract more frames than the video's native count, we can sample frames at a fixed time interval instead of relying on vidcap.read() to step through the video's built-in frames. This lets you control exactly how many frames you want (e.g., 60 FPS for 1440 total frames in 24 seconds).

The tilted frames are easy to fix with OpenCV's built-in rotation function. A 90° counterclockwise tilt just needs a 90° clockwise rotation to revert to the correct orientation. OpenCV's cv2.rotate() is optimized for these 90/180/270° rotations, so it's fast and clean.

Modified Full Code #

Here's the combined solution that addresses both problems:

import cv2

# Initialize video capture
vidcap = cv2.VideoCapture('20180530_115209.mp4')

# Get basic video info to understand the source
native_fps = vidcap.get(cv2.CAP_PROP_FPS)
total_native_frames = vidcap.get(cv2.CAP_PROP_FRAME_COUNT)
video_duration = total_native_frames / native_fps
print(f"Original Video Details: FPS={native_fps:.2f}, Total Frames={int(total_native_frames)}, Duration={video_duration:.2f}s")

# Set your desired frame extraction rate (higher = more frames)
target_extraction_fps = 60
# Calculate time between each frame in milliseconds
frame_interval_ms = 1000 / target_extraction_fps

count = 0
current_time_ms = 0

# Loop through the video by time intervals
while current_time_ms <= video_duration * 1000:
    # Jump to the exact time point we want to extract
    vidcap.set(cv2.CAP_PROP_POS_MSEC, current_time_ms)
    success, frame = vidcap.read()
    
    if not success:
        break  # Exit loop if we can't read a frame (end of video)
    
    # Correct the 90° counterclockwise tilt with a clockwise rotation
    corrected_frame = cv2.rotate(frame, cv2.ROTATE_90_CLOCKWISE)
    
    # Save the corrected frame
    cv2.imwrite("sushant_2/image/frame%d.jpg" % count, corrected_frame)
    print(f"Extracted frame: {count}")
    
    count += 1
    current_time_ms += frame_interval_ms

# Clean up resources
vidcap.release()
print(f"Done! Total frames extracted: {count}")

Key Explanations #

Frame Extraction Logic #

• Instead of reading frames sequentially, we use cv2.CAP_PROP_POS_MSEC to jump to specific time points in the video. This lets us sample frames more densely than the video's native FPS.
• Adjust target_extraction_fps to control how many frames you get: higher values mean more frames (e.g., 120 FPS would give 2880 frames for your 24-second video).

Tilt Correction #

• cv2.rotate(frame, cv2.ROTATE_90_CLOCKWISE) directly fixes the 90° counterclockwise tilt. If you prefer, you could also use a 270° counterclockwise rotation (it's the same result), but the clockwise option is more intuitive here.
• For reference, if you ever need to use manual rotation matrices (for arbitrary angles), you could do this, but it's overkill for 90° steps:

  height, width = frame.shape[:2]
  rotation_matrix = cv2.getRotationMatrix2D((width/2, height/2), 270, 1)
  corrected_frame = cv2.warpAffine(frame, rotation_matrix, (height, width))
  

内容的提问来源于stack exchange，提问作者Harshit Mishra