On OSX I can record from my webcam and write a video file with the following simple script:
import cv2
camera = cv2.VideoCapture(0)
# Define the codec and create VideoWriter object to save the video
fourcc = cv2.VideoWriter_fourcc(*'XVID')
video_writer = cv2.VideoWriter('output.avi', fourcc, 25.0, (640, 480))
while True:
try:
(grabbed, frame) = camera.read() # grab the current frame
frame = cv2.resize(frame, (640, 480)) # resize the frame
video_writer.write(frame) # Write the video to the file system
except KeyboardInterrupt:
camera.release()
break
The resulting avi file is quite big though. I want a smaller file, preferably an mp4. So I changed the filename to output.mp4 and the fourcc codec to H264. That writes a video file which works, but gives me the following error:
$ python write_video_file.py
OpenCV: FFMPEG: tag 0x34363248/'H264' is not supported with codec id 28 and format 'mp4 / MP4 (MPEG-4 Part 14)'
OpenCV: FFMPEG: fallback to use tag 0x00000021/'!???'
Since I thought I'm missing the H264 codec in ffmpeg I decided to uninstall ffmpeg and opencv and reinstall them again with H264 support. For this I used the following commands:
# First ffmpeg
brew install ffmpeg --with-fdk-aac --with-libvidstab --with-openh264 \
--with-openjpeg --with-openssl --with-tools --with-webp --with-x265 --with-zeromq
# then opencv3
brew tap homebrew/science
brew install opencv3 --with-contrib --with-ffmpeg --with-tbb
After this I ran the script again, using the following combinations:
output.mp4 with H264
output.mp4 with X264
Unfortunately I still get the OpenCV warnings/errors. The file is readable, but it still annoys me that I get these errors. Does anybody have any idea how I can make OpenCV write mp4 video file with the H264 codec?
All tips are welcome!
I spent ages trying to find a list of video codecs on macOS, and finding which codecs work with which containers, and then if QuickTime can actually read the resulting files.
I can summarise my findings as follows:
.mov container and fourcc('m','p','4','v') work and QuickTime can read it
.mov container and fourcc('a','v','c','1') work and QuickTime can read it
.avi container and fourcc('F','M','P','4') works but QuickTime cannot read it without conversion
I did manage to write h264 video in an mp4 container, as you wanted, just not using OpenCV's VideoWriter module. Instead I changed the code (mine happens to be C++) to just output raw bgr24 format data - which is how OpenCV likes to store pixels anyway:
So the output of a frame of video stored in a Mat called Frame becomes:
cout.write(reinterpret_cast<const char *>(Frame.data),width*height*3);
and then you pipe the output of your program into ffmpeg like this:
./yourProgram | ffmpeg -y -f rawvideo -pixel_format bgr24 -video_size 1024x768 -i - -c:v h264 -pix_fmt yuv420p video.mp4
Yes, I know I have made some assumptions:
that the data are CV8_UC3,
that the sizes match in OpenCV and ffmpeg, and
that there is no padding or mad stride in the OpenCV data,
but the basic technique works and you can adapt it for other sizes and situations.
Related
I am trying to write a video file without any loss in OpenCV, but so far any codec that I have selected from fourcc codec lists somehow results in loss of data.
regarding the recording parameters I am using:
fps = int(cap.get(cv2.CAP_PROP_FPS))
width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
used these codecs so far but they either to compression or upsize video bit rate
fourcc = cv2.VideoWriter_fourcc(*'MP4V')
fourcc = cv2.VideoWriter_fourcc(*'DIVX')
fourcc = cv2.VideoWriter_fourcc(*'RGBA')
fourcc = cv2.VideoWriter_fourcc(*'x265')
fourcc = cv2.VideoWriter_fourcc('H','2','6','4')
my video writer function is:
writer= cv2.VideoWriter(out_dest, fourcc, fps, (width,height))
Just to be clear, I do not want any sort of compression for the output video.
I also use
vid_format = int(cap.get(cv2.CAP_PROP_FOURCC))
to get the output video bit rate and compare it to the original video.
I also found someone on GitHub using skvideo but wasn't able to perform the same code
https://gist.github.com/docPhil99/a612c355cd31e69a0d3a6d2f87bfde8b
as it kept showing an extension error and couldn't find proper documentation on how to use it!
Thank you in advance
An Update on the topic:
the final output writer codec will be used as the video writer for BGR to RGB conversion in OpenCV, if you have any other ideas or suggestions that can do the job, I'm all ears!
If all you need to do is to get RGB data out of raw AVI file and feed the RGB frame data with MediaPipe, there is no need for intermediate file to store RGB data because FFmpeg can convert the pixel format on the fly by specifying output -pix_fmt rgb24 option.
To do this, you can try my ffmpegio package to load the data. My package is designed for your use case to remove the need to set up FFmpeg call. To install:
pip install ffmpegio
You also need FFmpeg in the system.
Then video.read() loads the video data:
import ffmpegio
fs, I = ffmpegio.video.read('datafile.avi',pix_fmt='rgb24')
# fs = framerate
# I = 250x480x640x3 numpy array containing RGB video data
If you don't want to read all 250 frames at once, you can use the stream interface to work on X number of frames at a time:
with ffmpegio.open('datafile.avi','rv', blocksize=25, pix_fmt='rgb24') as f:
for I in f: # loops 10 times, 25 frames at a time
# I = 25x480x640x3 numpy array
I am using cv2.VideoWriter() to record from a single camera and writing the video to a .mp4 file with a HEVC codec. I had this script previously working on a (Windows) laptop with the following code:
video = cv2.VideoCapture(0)
frame_width = int(video.get(3))
frame_height = int(video.get(4))
# create output file
out1 = cv2.VideoWriter(videoFileName, cv2.VideoWriter_fourcc(str('H'), str('E'), str('V'), str('C')), 30.0,
(frame_width, frame_height))
I am setting up the script on a new laptop and the above code no longer records and throws the following error:
OpenCV: FFMPEG: tag 0x43564548/'HEVC' is not supported with codec id 173 and format 'mp4 / MP4 (MPEG-4 Part 14)'
OpenCV: FFMPEG: fallback to use tag 0x31766568/'hev1'
[hevc_mf # 0000028eed319d00] COM must not be in STA mode
[ERROR:0] global /build/opencv/modules/videoio/src/cap_ffmpeg_impl.hpp (2794) open Could not open codec hevc_mf, error: Unspecified error
[ERROR:0] global /build/opencv/modules/videoio/src/cap_ffmpeg_impl.hpp (2811) open VIDEOIO/FFMPEG: Failed to initialize VideoWriter
I have read through many forums who suggest that this is not possible with cv2 because it does not support the HEVC codec. I have also read the .mp4 and HEVC are possibly not compatible. However, it was previously functioning on the other laptop. Therefore, I know it works in some manner. Note: I did not set it up initially on the other laptop.
I have also tried
# fourcc = cv2.VideoWriter_fourcc(*'mp4v')
# fourcc = cv2.VideoWriter_fourcc(*'h264')
# out1 = cv2.VideoWriter(videoFileName, fourcc, 30.0, (frame_width, frame_height))
with mp4v working but not the codec I initially had and h264 throwing an error similar to above.
Has anyone encountered a solution to this problem?
Additional info:
Windows 10, Python 3.8.10 through Pycharm, cv2 installed via opencv-python and imported with ´from cv2 import cv2`
I have a large set of videos that I need to process but a this is slow going since the processing involves running through the entire video at about 0.6FPS and the vast majority of frames have little change between them.
Is there some way I could sample the video say every two seconds and save this as another video cutting the framerate and duration in two? I am not worried about losing information by doing this, I would gladly cut a 10 minute video down to a few hundred frames. I do need it to be a video file however and not a set of images.
You may solve it using cv2.VideoCapture and cv2.VideoWriter or using FFmpeg.
The solution bellow uses ffmpeg-python.
I posted both OpenCV solution, and FFmpeg solution.
Solving using OpenCV:
Open input video file for reading using cv2.VideoCapture.
Open output video file for writing using cv2.VideoWriter.
Read all frames, but write only every 10th frame to the output file.
Solving using FFmpeg:
Solution is based on: Selecting one every n frames from a video using ffmpeg, but uses ffmpeg-python.
The following example is "self contained" - generates synthetic input video file for testing, and executes the frame skipping on the generated input video.
Here is the code (example using ffmpeg-python is at the bottom):
import ffmpeg # Note: import ffmpeg is not a must, when using OpenCV solution.
import cv2
import sys
out_filename = 'out.mp4'
# Build synthetic video and read binary data into memory (for testing):
#########################################################################
in_filename = 'in.mp4'
width, height = 320, 240
fps = 1 # 1Hz (just for testing)
# Build synthetic video, for testing:
# ffmpeg -y -f lavfi -i testsrc=size=320x240:rate=1 -c:v libx264 -crf 18 -t 50 in.mp4
(
ffmpeg
.input('testsrc=size={}x{}:rate={}'.format(width, height, fps), f='lavfi')
.output(in_filename, vcodec='libx264', crf=18, t=50)
.overwrite_output()
.run()
)
#########################################################################
# Open video file for reading
in_vid = cv2.VideoCapture(in_filename)
#Exit if video not opened.
if not in_vid.isOpened():
print('Cannot open input video file')
sys.exit()
# Read first image (for getting resolution).
ok, frame = in_vid.read()
if not ok:
print('Cannot read video file')
sys.exit()
width, height = frame.shape[1], frame.shape[0]
# Get frame rate of input video.
fps = in_vid.get(cv2.CAP_PROP_FPS)
# Create video writer
# 264 doesn't come by default with the default installation of OpenCV, but I preferred using H.264 (supposed to be better than XVID).
# https://stackoverflow.com/questions/41972503/could-not-open-codec-libopenh264-unspecified-error
# (I had to download openh264-1.8.0-win64.dll)
out_vid = cv2.VideoWriter(out_filename, cv2.VideoWriter_fourcc(*'H264'), fps, (width, height))
frame_counter = 0
while True:
# Write every 10th frame to output video file.
if ((frame_counter % 10) == 0):
out_vid.write(frame)
# Read a new frame
ok, frame = in_vid.read()
if not ok:
break
frame_counter += 1
out_vid.release()
# Selecting one every n frames from a video using FFmpeg:
# https://superuser.com/questions/1274661/selecting-one-every-n-frames-from-a-video-using-ffmpeg
# ffmpeg -y -r 10 -i in.mp4 -vf "select=not(mod(n\,10))" -vsync vfr -vcodec libx264 -crf 18 1_every_10.mp4
out_filename = '1_every_10.mp4'
# Important: set input frame rate to fps*10
(
ffmpeg
.input(in_filename, r=str(fps*10))
.output(out_filename, vf='select=not(mod(n\,10))', vsync='vfr', vcodec='libx264', crf='18')
.overwrite_output()
.run()
)
Note: The solution looses some of the video quality, due to decoding and re-encoding.
You may also try FFmpeg decimate filter, if you are looking for solution that skips frames according to metrics calculation.
I am using mac os x 1.7.5 with python 2.7.5_1 and opencv 2.4.4_0 installed via macports. I seem to have all the latest dependent ports.
In my code, the cv2.Videowriter() is successfully created and opened which produces a 6kb .avi file but videoFile.write(img0) doesn't write anything into that file. I really am not able to figure out why the video stream isn't written to the file. Any insights?
My code is as follows:
import cv2
import cv
cv2.namedWindow("Original")
cap0 = cv2.VideoCapture(0)
codec = cv.CV_FOURCC('D','I','V','X')
print codec
videoFile = cv2.VideoWriter();
videoFile.open('video.avi', codec, 25, (640, 480),1)
key = -1
while(key < 0):
success0, img0 = cap0.read()
cv2.imshow("Original", img0)
videoFile.write(img0)
key = cv2.waitKey(1)
cv2.destroyAllWindows()
I've tried these codecs and none of them worked: I420, AVC1, YUV1, PIM1, MJPG, MP42, MP4V, DIV3, DIVX, XVID, IUYV,FFV1, FLV1, U263, H264, ZLIB
I've also gone through all the quick time codecs mentioned here
Using ZLIB codec I get the error:
[zlib # 0x7fb0d130a000] Specified pixel format yuv420p is invalid or not supported
Using H264 codec I get an error:
[libx264 # 0x7fe423869600] broken ffmpeg default settings detected
[libx264 # 0x7fe423869600] use an encoding preset (e.g. -vpre medium)
[libx264 # 0x7fe423869600] preset usage: -vpre <speed> -vpre <profile>
[libx264 # 0x7fe423869600] speed presets are listed in x264 --help
[libx264 # 0x7fe423869600] profile is optional; x264 defaults to high
I didn't understand what the above errors meant. I tried reinstalling ffmpeg to the latest version (1.2.2_0+gpl2) but my script still doesn't work. All the other codecs did not give any error.
I've even tried the file extensions of .mpg and .mkv with the above codecs. Sometimes I would get an error saying that the codec was not suitable for the file extension but when I didn't get error I would simply get the unreadable video file of a minuscule size.
Any help is much appreciated.
ps: I've already gone through the following SO questions which didn't solve my issue:
using opencv2 write streaming video in python
Writing video with OpenCV + Python + Mac
Create an avi video with opencv and python on a mac
Python OpenCV, can't write a video (.avi) to file
Creating AVI files in OpenCV
read/write avi video on MAC using openCV
Python OpenCV 2.4 writes half-complete PNG video frames
The problem seems to have been with the image size in the function videoFile.open('video.avi', codec, 25, (640, 480),1)
So I updated my script to include
size = (int(cap0.get(cv2.cv.CV_CAP_PROP_FRAME_WIDTH)),
int(cap0.get(cv2.cv.CV_CAP_PROP_FRAME_HEIGHT)))
And then accordingly changed videoFile.open to
videoFile.open('video.avi', codec, 25, size,1)
Then my script started working.
I've tried the codecs with fourcc: IYUV, I420, PIM1, MJPG, FFV1 and DIVX with file extension .avi
Each of the above codec worked with frame rates 16, 20, 25 and 30 except for PIM1 which seems to work only on 20fps and above.
Also,
Codec with fourcc THEO worked with file extension .ogv
XVID worked properly with file extension .mkv and though the .mkv container should work with any encoding I got a variety of weird results with other codecs.
FLV1 with file extension .flv didn't work. It gave the error:
[flv # 0x7f8414006000] Tag FLV1/0x31564c46 incompatible with output codec id '22' ([2][0][0][0])
FLV4 with file extension .flv didn't give an error but opencv's videowrite outputted an error "Could not update video file"
Of the codecs that worked with .avi file container, DIVX produced the smallest video file (~4Mb for 4 sec video) and IYUV produced the largest file (~160Mb for 4 sec video)
Note:
fps = videoCapture.get(cv2.cv.CV_CAP_PROP_FPS) always returns 0.0 when capturing from webcam. It is a bug in OpenCV2.4.3 and OpenCV2.4.4
I also found out that ffmpeg cannot grab images from Mac's iSight and FacetimeHD webcams the way it can from Windows and Ubuntu because designers at Apple have prohibited easy access to the mac's camera... what a shame!
References:
http://en.wikipedia.org/wiki/Comparison_of_container_formats
Initially I was getting a 0kb video file. I changed the Codec from MJPG to iYUV. It worked for me. Python 2.7 and openCV 2.4.5.
cap = cv2.VideoCapture(0)
fourCC = cv2.cv.CV_FOURCC('i','Y','U', 'V'); # Important to notice cv2.cv.CV_FOURCC
size = (int(cap.get(cv2.cv.CV_CAP_PROP_FRAME_WIDTH)), int(cap.get(cv2.cv.CV_CAP_PROP_FRAME_HEIGHT)))
out = cv2.VideoWriter("Test.avi", fourCC, 15.0, size, True)
To get h264 working, install ffmpeg from the following link. Its quite a cumbersome installation but it ll get the encoder up and running.
https://trac.ffmpeg.org/wiki/CompilationGuide/Ubuntu
I can't seem to capture frames from a file using OpenCV -- I've compiled from source on Ubuntu with all the necessary prereqs according to: http://opencv.willowgarage.com/wiki/InstallGuide%20%3A%20Debian
#!/usr/bin/env python
import cv
import sys
files = sys.argv[1:]
for f in files:
capture = cv.CaptureFromFile(f)
print capture
print cv.GetCaptureProperty(capture, cv.CV_CAP_PROP_FRAME_WIDTH)
print cv.GetCaptureProperty(capture, cv.CV_CAP_PROP_FRAME_HEIGHT)
for i in xrange(10000):
frame = cv.QueryFrame(capture)
if frame:
print frame
Output:
ubuntu#local:~/opencv$ ./test.py bbb.avi
<Capture 0xa37b130>
0.0
0.0
The frames are always None...
I've transcoded a video file to i420 format using:
mencoder $1 -nosound -ovc raw -vf format=i420 -o $2
Any ideas?
You don't have the gstreamer-ffmpeg or gsteamer-python or gsteamer-python-devel packages installed. I installed all three of them. and the exact same problem was resolved.
I'm using OpenCV 2.2.0, compiled on Ubuntu from source. I can confirm that the source code you provided works as expected. So the problem is somewhere else.
I couldn't reproduce your problem using mencoder (installing it is a bit of a problem on my machine) so I used ffmpeg to wrap a raw video in the AVI container:
ffmpeg -s cif -i ~/local/sample-video/foreman.yuv -vcodec copy foreman.avi
(foreman.yuv is a standard CIF image sequence you can find on the net if you look around).
Running the AVI from ffmpeg through your source gives this:
misha#misha-desktop:~/Desktop/stackoverflow$ python ocv_video.py foreman.avi
<Capture 0xa71120>
352.0
288.0
<iplimage(nChannels=3 width=352 height=288 widthStep=1056 )>
<iplimage(nChannels=3 width=352 height=288 widthStep=1056 )>
...
So things work as expected. What you should check:
Do you get any errors on standard output/standard error? OpenCV uses ffmpeg libraries to read video files, so be on the lookout for informative messages. Here's what happens if you try to play a RAW video file without a container (sounds similar to your problem):
error:
misha#misha-desktop:~/Desktop/stackoverflow$ python ocv_video.py foreman.yuv
[IMGUTILS # 0x7fff37c8d040] Picture size 0x0 is invalid
[IMGUTILS # 0x7fff37c8cf20] Picture size 0x0 is invalid
[rawvideo # 0x19e65c0] Could not find codec parameters (Video: rawvideo, yuv420p)
[rawvideo # 0x19e65c0] Estimating duration from bitrate, this may be inaccurate
GStreamer Plugin: Embedded video playback halted; module decodebin20 reported: Your GStreamer installation is missing a plug-in.
<Capture 0x19e3130>
0.0
0.0
Make sure your AVI file actually contains the required information to play back the video. At a minimum, this should be the frame dimensions. RAW video typically doesn't contain any information besides the actual pixel data, so knowing the frame dimensions and FPS is required. You can wrong-guess the FPS and still get a viewable video, but if you get the dimensions wrong, the video will be unviewable.
Make sure the AVI file you're trying to open is actually playable. Try ffplay file.avi -- if that fails, then the problem is likely to be with the file. Try using ffmpeg to transcode instead of mencoder.
Make sure you can play other videos, using the same method as above. If you can't, then it's likely that your ffmpeg install is broken.