You can use OpenCV, but by default this will only make use of the host (ARM) for processing.
It was done with pre-Parallella hardware and the SDK will have moved on quite a bit, but you may want to take a look at an early Epiphany-enabled OpenCV demo that Adapteva did as a collaboration with CVisionLab:
http://www.adapteva.com/white-papers/fa ... processor/This should give you an idea of the extra work required to enable Epiphany offload of compute intensive tasks within OpenCV.
Also, OmerK and his colleagues built a CV system in Erlang, using multiple Raspberry Pi + camera to provide the source video:
http://www.adapteva.com/white-papers/fa ... processor/Note that if you are looking for a plug-and-play computer vision library with Epiphany support, I do not believe that one exists at the present time. So, you will have to get to grips with the Parallella/Epiphany programming model first, and then extend whatever framework or library you decide to use.
Regards,
Andrew