[Opalvoip-devel] Work in progress: get more stability in V4L2 video input plugin

[Denis L. <den...@go...>]

Hi Robert, hi mailing list,

we are currently on heavy duty to make the V4L2 plugin more stable in
some circumstances (especially when resizing the frame size or change
the frame rate and so on).

We were successful to keep the camera alive when it comes to sizing
changes (for instance from QCIF to CIF or vice versa). This will
probably fix the bug with "Segmentation fault with V4L2", however I
have to re-check this in some more test cases.

Additionally, we implemented the setNearestFrameSize function which
enables ptlib to create a colourconverter from the one dimension to
the other. This works quite well so far.

But, and now I require some help:

- I found out, that some camera devices do NOT support YUV420P. Most
of them seem to support only YUYV and/or MJPEG. I have some major
issues when using YUV420P as it results in using MJEPG for grabbing
frames and then converting them with libv4l-convert. Sometimes this
library seems to break and cannot decode frames properly.
Could someone tell me, how to tell opal or ptlib that YUV420P is not a
proper format and it should use YUY2 (which is YUYV)? In this case, it
could create a colour converter from one to the other.

- I wanted to replace the colour conversion algorithms with swscale
from ffmpeg, i.e. if libswscale can be used, it should do it so, and
if not to use the existing code.
Do I have to extend the code only in vconvert.cxx?
How do I add a check in configure if libswscale is available? And how
do I integrate a switch to enable/disable swscale? I haven't done so
much yet with configure scripts.

I was hopping to get some help.

I will apply a patch very soon, when I am done with the work.

Denis