Student Imager FPGA Project

Image matching of Stout bottle top

Basil Esterhuysen’s imaging matching work in 2004

Basil used the Digilent Spartan 3 kit to test various imaging algorithms. On the PC the Open Computer Vision Library was used.

Image of Stout bottle top

One of the bottle tops used in the matching. No doubt someone had to drink all these beers to collect the bottle tops.

Basil Esterhuysen’s M.Tech Project

Professor Theo van Niekerk and I were promotors for Basil’s thesis, titled “Vision Based Property Extraction and Technology Benchmarking” — written up at the end of 2005 at the Nelson Mandela Metropolitan University, Port Elizabeth, South Africa.

We had several FPGA evaluation boards. Some with hardcore PowerPC, stand alone and card edge connector for the PCI bus. The PowerPC and Virtex-II Pro crashed regularly when trying to debug over the shared JTAG port and loading up memory was no simple matter. The upgrade from one release of the Xilinx tools (from version 6) caused earlier efforts to be abandoned as the designs could not be upgraded automatically—little guidance as to why the new tools would not compile prior VHDL designs did not help matters, so as another student who was slightly ahead on the FPGA journey jumped ship for Freescale 5200 PowerPC and standalone 500K gate Virtex FPGA, Basil settled on the Digilent 300K gate board. He pioneered methods of downloading image data and uploading to the PC, plus provided insight into the effort involved in a hardware- and a software approaches to solving a problem. As both the imaging library and FPGA workflow were fairly new to Basil, the two years set aside for the project would be useful to get an unbiased idea of the effort involved. The conclusion was that the software was a lot easier and for limited volumes for custom imaging projects, the standard PC was cost effective.

Basil brought years of motor vehicle assembly plant experience before being plunged into the deep end with pretty unfriendly FPGA tools. He travelled overseas and spent many years in the UK with Siemens on large industrial automation projects. We all learned valuable lessons and almost seven years later, the FPGA problem of simple generation and testing compares poorly to the more mature desktop or DSP devices.