A VGA 30-fps optical-flow processor core based on Pyramidal Lucas and Kanade algorithm

Abstract

This paper describes an optical-flow processor core for real-time video recognition. The processor is based on the Pyramidal Lucas and Kanade algorithm. It has small chip area, a high pixel rate, and high accuracy compared to conventional optical-flow processors. Introduction of search range limitation and the Carman filter to the original algorithm improves the optical-flow accuracy and reduces the processor hardware cost. Furthermore, window interleaving and window overlap methods can reduce the necessary clock frequency of the processor by 70%. The proposed processor can handle a VGA30-fps image sequence with 332 MHz clock frequency. The core size and power consumption in 90-nm process technology are estimated respectively as 3.50 times 3.00 mm2 and 600 mW.