roboforum.ru

However, the odds are quite good that your camera will have some way
to retrieve an MJPEG stream over HTTP via a well defined URL. So two
fairly straight forward ways to handle it are:

1. Supply the appropriate URL to cvCreateFileCapture, and then use
a loop with cvQueryFrame just like you would with other cameras.

This can also work with an MPEG4 feed (such as an rtsp:// URL) if
that's what your camera supports, though in general I would prefer
an MJPEG feed if given the option for image processing if you can
support the bandwidth.

Note that this approach is dependent on your OpenCV library having
been built with appropriate support for handling an MJPEG and/or
MPEG4 stream. I've had good luck under Linux but not so much under
Windows.

2. Just make your own connection to the camera at the supplied URL, and
decode the frames yourself, independent of OpenCV.

Most MJPEG feeds are just a series of JPEG images in a multipart
HTTP response, which are fairly easy to parse and decode using
whatever popular library you prefer for web retrieval and JPEG
decoding in your language and platform of choice. Once you have
any individual JPEG frame decoded, you can then convert it over to
an OpenCV IplImage (there was a recent thread on this list about
that) and proceed from there.

Trying to decode an MPEG4 stream would be much harder, and isn't
something I've attempted.