Eyeing up the future of real-time image processing
Leading the way in real-time image processing are two spin-off companies whose state-of-the-art microprocessors are opening up a new range of applications in areas as broad as communications, manufacturing and the military.
Inspired by the workings of the human eye, the IST project DICTAM developed a series of mixed-signal visual microprocessors that are among the fastest and most complex ever created, capable of processing up to 50,000 images per second. Baptised by the project partners as Analogic Cellular Engines (ACE), these ACE chips represent the core of a new generation of artificial vision systems that promise to enhance fault-detection in manufacturing, increase transportation safety and provide new communications services among an almost endless range of uses.
"It is impossible to imagine what these chips will do in the future," explains project coordinator Angel Rodríguez-Vázquez of the Seville Microelectronics Institute. "They have the potential to change our vision of what computers are capable of."
Now, a year after the project ended, DICTAMs two spin-off companies AnaFocus in Spain and AnaLogic in Hungary are commercialising the results, working in research in both the public and private sectors, with companies such as Siemens, Hewlett-Packard and Volvo. Within two years, AnaFocus will have developed a commercially available ultra-fast image processing chip no larger than 5mm x 5mm that will cost no more than 18 euros.
"Artificial vision is not just capturing images - cameras do that," Rodríguez-Vázquez notes. "Vision is having stand-alone integrated circuits that process, interpret and, further down the line, make decisions based on the images they see."
A revolutionary approach
Chips with those capabilities have in the past proven prohibitively expensive, power-hungry and often too slow to provide real-time processing. Those chips, however, carried out all processing in the digital domain; with analogue used as a mere interface with the real world. But DICTAM took an unconventional approach, using analogue transistors to carry out pre-processing and by doing so achieve nothing less than a revolution in vision systems.
"With digital systems all the information from an image is processed and that causes bottlenecks at the transition between the analogue optical sensors and the digital processing domain, draining power and computing resources," the project coordinator explains. "With the analogue cells in the ACE chips carrying out pre-processing, information is cut down to only what is essential for interpretation."
In effect, the DICTAM project moved the border between analogue and digital to create mixed-signal chips that capture and process images in parallel, thereby approximating to the functions of the human eye.
"To risk oversimplifying, it could be said that the analogue domain represents the work of the retina, it does the pre-processing, and the digital domain the brain, which does the post-processing," Rodríguez-Vázquez says.
Three generations of chips
During the course of the DICTAM project two generations of chips were designed, as well as a third diversification that far surpassed the project partners original goals. This sub-generation chip, called the CACE1k, approximates most closely to what we know about human vision functions. Its memory cells save intra-frame, i.e. not only moving images broken down into static frames but also the dynamic information contained within each frame. The other two chips, the second-generation ACE4k and the third-generation ACE16k, save frame-by-frame. All, however, represent a significant advance in terms of speed, power consumption and cost compared to previous systems due to their unique analogue-digital, or mixed signal, architecture.
"The ACE16k, for example, has almost four million transistors of which 80 per cent are analogue," Rodríguez-Vázquez notes. "Each analogue transistor has 128 states, i.e. 128 things it can do, whereas digital transistors have two states - that is where we gain the acceleration in processing speed."
The three chips have their own specific properties and further development promises to convert them, or future variants of them, into everyday aspects of life.
"Chips such as these have an unimaginable range of uses," Rodríguez-Vázquez emphasises. "To name but a few, we have been working on their application in consumer electronics and toys; in surveillance cameras; in the automobile industry for increasing safety in airbag-release systems and as sensory systems to detect objects around vehicles; in manufacturing to detect faults in objects on production lines; in real-time video compression and communications; and obviously there are security and military uses."
In the manufacturing sector for example, around 60 per cent of European companies could probably make use of vision systems, which when employed with the ACE chips would increase productivity by up to 10 per cent and inspection accuracy on production lines by up to 30 per cent, the project estimates.
"Humans dont take into account the number of things we do with our ability to see, we take it for granted despite the fact that we obtain 60 per cent of our sensory information from sight," Rodríguez-Vázquez notes. "If we give computers just a fraction of that ability... well, the possibilities are endless."
Instituto de Microelectrónica de Sevilla
Avda. Reina Mercedes s/n (Edificio CICA)
E-41012 - Seville
Tel: +34-95-5056666/ +34-95-4081251
Source: Based on information from DICTAM
Tara Morris | IST Results