This paper investigates the real time implementation requirements for several stereo matching algorithms. The area-based matching techniques sum of absolute differences (SAD), sum of squared differences (SSD), and normalised cross correlation (NCC) are considered as well as the non-parametric census and rank methods. A generic stereo matching framework is presented and algorithm specific implementation requirements compared. The suitability of field programmable gate array (FPGA) based custom computers to the stereo matching problem is discussed and the use of partial reconfigurability and adaptive stereo vision is suggested.
展开▼
