Opencv VideoCapture always returns false on Heroku

Question

I'm using the following code to open a video stream:

import cv2
video = cv2.VideoCapture()
video.open("some_m3u8_link")
success, image = video.read()

However, even if the code works as intended locally, on Heroku success is always false.

I'm using cedar-14 stack with the following buildpacks:

heroku/python

https://github.com/jonathanong/heroku-buildpack-ffmpeg-latest.git

(I tried several buildpack options for ffmpeg)

Running ffmpeg --version on heroku instance will return ffmpeg version 4.0-static https://johnvansickle.com/ffmpeg/

Is there any setting/configuration I missed in order to make it work on deployment? Thank you!

Later edit: I tried several links for "some_m3u8_link" including from twitch and other streaming services (including traffic streaming li An example for reproducing:

python -c "import cv2; video=cv2.VideoCapture(); video.open('https://hddn01.skylinewebcams.com/live.m3u8?a=5tm6kfqrhqbpblan9j5d4bmua4'); success, image = video.read(); print(success)"

Returns True on local machine and False on Heroku.

(the link is taken from here)

Where does your **`some_m3u8_link`** point? To a local file or to a web resource? — Leon, May 03 '18 at 11:47
Can you then provide that link so that your code turns into a [Minimal, Complete, and Verifiable example](https://stackoverflow.com/help/mcve)? — Leon, May 03 '18 at 15:09
Please provide the link as well as a minimal deployable repo — Tarun Lalwani, May 03 '18 at 19:02
Can you perform [these tests](https://stackoverflow.com/a/47116576/2286337) on Heroku? See what the results are. — zindarod, May 06 '18 at 20:07
Inspecting the file using a hex editor seems like there isn't a link in there... can you confirm this file works as you say it does? — jmkmay, May 09 '18 at 01:17
@JonathanMay You are right, the links I provided were not fit for this use case (The read seem to work if the link points to a static resource I think). I updated with a new link (a traffic camera video streaming which behaves as expected) — Dacian Mujdar, May 09 '18 at 08:48

score 0 · Answer 1 · edited May 10 '18 at 10:29

0

You can try this:

import cv2
video = cv2.VideoCapture("some_m3u8_link")
success, image = video.read()

edited May 10 '18 at 10:29

mpromonet

11,326
43
62
91

answered May 09 '18 at 21:16

Uma Shankar Tewari

46
7

score 0 · Answer 2 · answered May 10 '18 at 06:40

0

You can use pafy module with cv2

-try opencv3 if its not working with cv2

    import cv2, pafy
    url = "Some url to stream"
    video = pafy.new(url)
    best = video.getbest(preftype="webm")
    video=cv2.VideoCapture(best.url)

pafyPYPI

answered May 10 '18 at 06:40

skysoft999

540
1
6
27

As I see from pafy project description it seems to be designed to work on youtube only ("Retrieve YouTube content and metadata") – Dacian Mujdar May 10 '18 at 06:59

score 0 · Answer 3 · answered Jun 27 '22 at 22:59

Specifying the mode to open it might work.

video.open("some_m3u8_link", "r")

If that doesn't work then specifying the file extension might help. You might also need to make a variable equal to the function

Ex:

""" replace .mp4 with the applicable file type, 
I don't know if you need to specify file mode"""
import cv2
video = cv2.VideoCapture()
video = video.open("some_m3u8_link.mp4") 
success, image = video.read()

If this doesn't work then I am out of ideas.

Opencv VideoCapture always returns false on Heroku

3 Answers3