This is an inexpensive way to get a good quality realtime camera feed into your Pandaboard/BeagleBoard expansion port.
Purpose: To remove lens distortion from the incoming images for use on a Pandaboard with OpenCV.
Why XMOS: Timing and Concurrency. Interfacing with multiple camera modules requires accurate/consistent timing and must be done concurrently. If done serially, the images with be at slightly different times so if you are moving, they will be worthless for making a disparity/depth map.
Why XS1-G4: The algorithms are computationally intensive (especially without an FPU!) so it should require most if not all of the processing power of a XS1-G4. The two images are 2 bytes per pixel at a resolution of 320x240, so while they cant be 100% buffered, they can be mostly buffered.
Scalability: Due to the XMOS architectural design, this system can easily be scaled to use higher resolution images and/or a higher framerates. The bottle neck is the speed at which the images can be acquired from the cameras (port speed, port width and finally processor speed).
NOTE: This project is heavy on floating point computations, so if anyone would like to help by creating assembly code for software add/sub/mul/div/max/min operations for "float" and "double" types, that would be awesome.