A low power architecture of real time stereo vision algorithm using discrete cosine transform

This paper presents a low power architecture of a stereo vision algorithm, that runs in real-time. This algorithm is a modification to the conventional Sum of Absolute Difference (SAD) algorithm. Conventional SAD uses absolute difference between block of pixels in reference image and the other image to calculate the disparity cost function. The number of computations can be significantly reduced by taking the Discrete Cosine Transform(DCT) of the block of pixels and using only the lower frequency components of DCT for SAD calculation. An architecture which can perform these DCT and SAD could save the power up to 23% making it suitable for real time robotic applications.