Optimized fixed point implementation of a local stereo matching algorithm onto C66x DSP

Stereo matching techniques aim at reconstructing disparity maps from a pair of images. The use of stereo matching techniques in embedded systems is very challenging due to the complexity of the state-of-the-art algorithms. An efficient local stereo matching algorithm has been chosen from the literature and implemented on a c6678 DSP. Arithmetic simplifications such as approximation by piecewise linear functions and fixed point conversions are proposed. Thanks to factorisation and precomputing, the memory footprint is reduced by a factor 13 to fit on the memory footprint available on embedded systems. A 14.5 fps speed (factor 60 speed-up) has been reached with a small quality loss on the final disparity map.