Stereo vision does need two sensors and two lenses, where a micro-lens approach would only require one of each. However, the micro-lens camera would need a much larger and higher resolution sensor to produce a depth image that has the same resolution of the equivalent stereo camera.
Ignoring the size of the sensor, producing two standard camera lenses will always be cheaper than producing an array of multiple (i.e. more than two) micro-lenses. This is doubly true considering that the micro-lens technology is already encumbered by patents.
Finally, stereo is very well understood and has already been implemented on the GPU, on FPGAs and in ASICs (commonly known as STOC, Stereo On-Chip). I would personally love to see a demo of a micro-lens array used for creating a depth map, but I just don't see any practical advantages over stereo.
Ignoring the size of the sensor, producing two standard camera lenses will always be cheaper than producing an array of multiple (i.e. more than two) micro-lenses. This is doubly true considering that the micro-lens technology is already encumbered by patents.
Finally, stereo is very well understood and has already been implemented on the GPU, on FPGAs and in ASICs (commonly known as STOC, Stereo On-Chip). I would personally love to see a demo of a micro-lens array used for creating a depth map, but I just don't see any practical advantages over stereo.