OpenCV C++ does not load ANY video on Ubuntu 16.04 under VideoCapture - c++

Question

I installed OpenCV 2.4.13 after installing the required dependencies and I'm running it with Qt Creator 5.7 for programming in C++. Following some tutorials I managed to load still images, process them and so on. But when I try to use the cv::VideoCapture, for example:
cv::VideoCapture cap;
cap.open("<File location/file name>");
or:
cv::VideoCapture("<File location/file name");
cap.isOpened() always return false. Even if I try to ignore this and go further with Mat, read, imshow and so on, the program crashes. I've already tried everything about loading the file: giving a std::string as argument, giving a *char, putting the video file on easier locations, in the project directory and nothing works. I tried with two different .mp4, with a .mov and with the cube4.avi file given in the "samples" folder (from openCV examples). All these files are perfectly played by my VLC. I've already tried in Qt with a QtWidgets project and with a plainCpp project. It never works. After searching about it, I've seen that this bug is very recurrent, but I've just found solutions for Windows, regarding on adding a .dll file to the project. But what should I do on Ubuntu, since Linux uses no dlls?

Answer 1

I figured out how to solve this and I'm here to share with everyone who encounter similar problems. First of all, my FFmpeg installation was for any unknown reason ignored on the OpenCV compilation. Since FFmpeg is needed to read and write videos, OpenCV wasn't able to do these jobs and displayed no error or warning messages about that. Anyway I decided to try everything again from the beginning. Keep in mind: if you just reinstall FFmpeg, it's not going to solve your problems; after that you have to do cmake, compile and install OpenCV again. I downloaded the most recent version of FFmpeg from the repository, compiled and installed it again. I tried to compile OpenCV again, but it was always stuck at about 15% when processing a FFmpeg (Libav) .h file with the message:
<file_name.h> can not be used when making a shared object; recompile with -fPIC
So I searched for it and with some effort I found out that it was necessary to compile the FFmpeg with the following configure line before doing the make:
.\configure --enable-pic --enable-shared
Some ffmpeg installation tutorials include even more commands after "configure". Then I compiled and reinstalled FFmpeg and compiled and reinstalled OpenCV. After that, I was able to load videos (the cap.isOpened() was no longer false), but nonetheless I was getting the following error for any video and they weren't being read:
Assertion desc failed at libswscale/swscale_internal.h:674
The way I found to solve this was downloading the newest stable FFmpeg version instead of the newest on, compiling it again, compiling OpenCV again and then it was successful! Now I can both load and write video files (I haven't tested it with multiple codecs yet). I wonder why they don't launch pre-compiled versions of OpenCV with everything so that we don't need to deal with all these stuffs...
Summarizing everything:
Download the last stable release of FFmpeg. If it doesn't work, try an older one. (I'm using the 2.8.6, the last stable right now).
Unpack it, open the terminal from the ffmpeg folder, compile and install it typing:
.\configure --enable-pic --enable-shared
make
sudo make install
Download OpenCV (if you haven't done it yet), unpack and install it following the official linux installation tutorial:
http://docs.opencv.org/2.4/doc/tutorials/introduction/linux_install/linux_install.html#linux-installation

Answer 2

I solve the problem by install "opencv3.2.0-dev" instead "opencv3.2.0";
pip install opencv-python is not enough to install OPENCV.