Question

我的代码处理一个框架，需要花费几秒钟来执行。如果我从相机流式传输，我自然会丢帧并每隔几秒钟获得一帧，对吗？我想在重放视频文件时模拟同样的事情。

通常，当您致电vidcap.read()时，您会获得视频中的下一帧。这实质上会减慢视频的播放速度，并且不会丢失任何帧。这不像处理实时摄像机流。有没有办法在处理过程中（例如处理摄像机流时）处理视频文件和丢帧？

我想到的解决方案是自己跟踪时间，并在每个vidcap.set(cv2.CAP_PROP_POS_MSEC, currentTime)之前致电vidcap.read()。这是我应该怎么做，还是有更好的方法？

Answer 1

一种方法是跟踪处理时间并跳过该数量的帧：

import cv2, time, math
# Capture saved video that is used as a stand-in for webcam
cap = cv2.VideoCapture('/home/stephen/Desktop/source_vids/ss(6,6)_id_146.MP4')
# Get the frames per second of the source video
fps = 120
# Iterate through video
while True:
    # Record your start time for the frame
    start = time.time()
    # Read the frame
    _, img = cap.read()
    # Show the image
    cv2.imshow('img', img)
    # What ever processing that is going to slow things down should go here
    k = cv2.waitKey(0)
    if k == 27: break
    # Calculate the time it took to process this frame
    total = time.time() - start
    # Print out how many frames to skip
    print(total*fps)
    # Skip the frames
    for skip_frame in range(int(total*fps)):  _, _ = cap.read()
cv2.destroyAllWindows()

这可能总比没有好，但是它不能正确模拟丢帧的方式。似乎在处理期间，网络摄像头数据已写入缓冲区（直到缓冲区填满）。更好的方法是使用虚拟过程捕获视频。此处理器密集的虚拟过程将导致丢帧：

import cv2, time, math
# Capture webcam
cap = cv2.VideoCapture(0)
# Create video writer
vid_writer = cv2.VideoWriter('/home/stephen/Desktop/drop_frames.avi',cv2.VideoWriter_fourcc('M','J','P','G'),30, (640,480))
# Iterate through video
while True:
    # Read the frame
    _, img = cap.read()
    # Show the image
    cv2.imshow('img', img)
    k = cv2.waitKey(1)
    if k == 27: break
    # Do some processing to simulate your program
    for x in range(400):
        for y in range(40):
            for i in range(2):
                dummy = math.sqrt(i+img[x,y][0])
    # Write the video frame
    vid_writer.write(img)
cap.release()
cv2.destroyAllWindows()

OpenCV重播保存的流并对其进行处理

1 个答案: