A novel system architecture for real-time low-level vision

A novel system architecture that exploits the spatial locality in memory access that is found in most low-level vision algorithms is presented. A real-time feature selection system is used to exemplify the underlying ideas, and an implementation based on commercially available field programmable gate arrays (FPGAs) and synchronous SRAM memory devices is proposed. The peak memory access rate of a system based on this architecture is estimated at 2.88 G-Bytes/s, which represents a four to five times improvement with respect to existing reconfigurable computers.

[1]  John Woodfill,et al.  Real-time stereo vision on the PARTS reconfigurable computer , 1997, Proceedings. The 5th Annual IEEE Symposium on Field-Programmable Custom Computing Machines Cat. No.97TB100186).

[2]  Duncan A. Buell,et al.  Splash 2 - FPGAs in a custom computing machine , 1996 .

[3]  Viktor K. Prasanna,et al.  High-performance computing for vision , 1996, Proc. IEEE.

[4]  Pietro Perona,et al.  Real-time 2-D feature detection on a reconfigurable computer , 1998, Proceedings. 1998 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (Cat. No.98CB36231).

[5]  Zvonko G. Vranesic,et al.  Field-Programmable Gate Arrays , 1992 .