High performance median FPGA implementation for machine vision applications

This paper presents a high performance median implementation targeting detection of microfluidic flows on Lab-on-Chips on a real-time machine vision implementation. We propose and implement a novel architecture which calculates the 2D median coordinates of a set of “active pixels” in a detection window of generic size. The proposed implementation takes full advantage of the FPGA´s characteristics to achieve full performance and is part of a machine vision system which achieves real-time microfluidic flow detection on 1Mpixel input videos at 60fps. The median module itself is flexible and generic and can be used for numerous machine vision applications achieving a median calculation in a very small number of clock cycles with an operational frequency of 204MHz.