Even as we’re about to discuss more about gestures and motion, CEVA has announced a vision platform explicitly designed for mobile devices and with a specific eye to gesture recognition and other advanced applications.
The idea is to take the raw camera images directly onto its L2 memory and then do, at the very least, your basic image processing before sending completed jpegs or video files directly out to DDR memory, integrating all of the various functions in one engine and with only one trip out to memory.
But beyond this, further processing can do things like face recognition, gesture recognition, object recognition, 2D à 3D conversion – even face “beautification.” Boy, there’s a loaded algorithm if ever there was one… Gotta wonder whose standard they’re using… They have a software package with a lot of these functions in it. As usual, their approach is “do it in software” – especially since so many of these algorithms need to be adapted to the specific device that will house them.
They’re also working with a company called eyeSight for the gesture recognition part. They take the output of standard cameras and let users control the system (computer, TV, whatever) by interpreting hand gestures and tracking a finger as if it were a mouse.
More on gestures next week; more on CEVA’s MM3101 platform in their release…