Speeding up machine vision
There is some way to go to reach the sight capabilities of the best science-fiction machines, but HYMOD - CSIRO's hybrid modular processor system - is making it easier, better and cheaper for machines to 'see', as well as providing solutions for many other high-speed, real-time, data-processing applications.
The HYMOD system has jumped ahead of its international competition by connecting different computer modules with very high-speed networks that replace the conventional 'computer bus' by developing software methods to program configurable logic processors (CLP), instead of programming these devices using factory-set hardware configurations.
The particular CLP used in the HYMOD system is a field programmable gate array (FPGA), which can contain millions of logic gates (ie, switches).
Each FPGA can be configured as an array of parallel processors that are 'hardware programmed' by setting switches in this massive array of logic gates. A conventional computer chip (a CPU - central processing unit) can now do this switch setting, such as a Pentium processor, linked to the FPGA.
This is a major advance on the procedure where a hardware engineer had to design the CLP configuration to suit each application - one CLP, one program and no program change after it leaves the factory.
Complex 'hardware programs' to suit particular image (or other) processing applications are set up in the FPGAs by software running on the CPU - a 'meta-level' program.
A HYMOD system can therefore be programmed in the field using conventional software programming and the configuration of the FPGA can even be varied in real-time in response to data received by the CPU. CSIRO has developed a special C-like language for programming the FPGAs in the HYMOD system.
The 'hybrid' tag is used because the HYMOD system blends both hardware and software programs to solve a task.
The basic HYMOD system has the CPU interfaced with one or more FPGAs. These are networked by very high-speed serial connections that enable information transfer at full on-board data rates between the various modules.
And the parallel operation enables data items to be processed simultaneously. For example, at 100 MHz, a 600,000 gate FPGA can run about 100 times faster than the fastest Pentium, which executes instructions one at a time.
HYMOD systems can use three different modules: a novel input/output module that interfaces with cameras, monitors, etc; FPGA parallel processors with multiple independent memories for very high speed processing; and high-speed computer chips for processing that cannot benefit from parallelism.
The input/output module has an FPGA coupled to a CPU and can use almost any type of interface to other devices, and the multiple memories can be accessed simultaneously, providing an unprecedented capacity for parallel data flow.
In each module, while the FPGA is running its hardware program, the CPU can be running a software program to control FPGA parameters and deal with the FPGA output.
The potential applications for this technology appear almost unlimited - from remote web devices for data acquisition, to machine and process control, to traffic applications such as advanced traffic sensing for intersection control, vehicle driver authentication, driver fatigue monitoring, navigation devices, data capture or even anti-collision devices.
The HYMOD platform is a generic design; daughter designs include an I/O board for interfacing to specialist equipment, a high-end processing board for substantial computing and a board for conventional signal processing.
HYMOD has enabled the development of cheap, flexible, real-time very high speed systems that can be readily configured by software programmers in the field for machine vision and other applications.
The CSIRO engineers who developed this system believe they have only scratched the surface of potential applications.
Robots that can work side by side with humans are changing the way manufacturing is done.
With more than 620 companies queuing up to participate in this year's annual Australasian Oil...
With the growing amounts of data being stored by industrial organisations today, understanding...