M021V4L2: Capture images from Leopard Imaging LI-USB30-M021 camera on Linux in C++ and Python3

The LI-USB30-M021 camera from Leopard Imaging is a fast (up to 90 frames per second) global-shutter CMOS camera that captures images over USB 3.0. Because the camera serves up raw image bytes, getting it to work on Linux using V4L2 (Video For Linux 2) requires a bit of extra format-conversion work. With some help from the folks at Leopard Imaging, I was able to write a few simple APIs for the camera for people who want to use it on Linux without doing the conversion themselves.

The C++ and Python3 APIs are intended for OpenCV users who want to be able to capture images as a Mat object (C++) or NumPy array (Python3). Because the M021 camera supports three frame sizes (1280 x 720 at 60 FPS; 800 x 460 at 90 FPS; 640x480 at 30 FPS), I've provided a class for each frame size. The capture runs on its own automatically-launched thread. As the code fragment below shows, the classes are extremely simple to use:

    cap = Capture800x460()

    while True:

        ret, frame = cap.read()

        cv2.imshow('LI-USB30-M021',frame)

        if cv2.waitKey(1) & 0xFF == 27:
            break
 

To run the Python3 demo program, cd to opencv/python and do

  % sudo python3 setup.py install
  % make cap

If you get errors about OpenCV or NumPy not being installed, you should make sure that they're installed now:

  % sudo pip3 install numpy
  % sudo pip3 install opencv-python

Because Python2 will not be supported much longer, I've switched to Python3 for this project, and will not be supporting Python2.

To run the C++ OpenCV capture demo, cd to opencv/cpp and type make cap.

Both of these programs do some post-capture color balancing to compensate for the slightly dimmed / green appearance of the image from the camera.

I've also provided a C API (which is used by the C++ and Python3 code) for capturing images in YUYV format, along with a demo program (a cut-down version of Guvcview) that displays the images using GTK and SDL. To run the GTK/SDL demo, cd to gtksdl and type make run.

Known Issues

• Programs will occasionally seg-fault on exit.

• On ODROID XU4, Python3 version is much slower than C++ version.

• On desktop Ubuntu 16.04, an Unable to dequeue buffer error occurs in the OpenCV C++ examples, and no image is displayed.