Ditching expensive satellite technology may be a sign of the future in the race to create fail-safe camera-based navigation systems, says a leading QUT robotics researcher.
Dr Michael Milford is a chief investigator at the QUT-based headquarters of the Australian Research Council Centre of Excellence in Robotic Vision .
"Our group is one of the world's leading pioneers of developing camera-based navigation technology for robots and vehicles," Dr Milford said.
Dr Milford, PhD student Edward Pepperell and the Centre's leader Professor Peter Corke's latest research will be presented at the 2014 Australasian Conference on Robotics and Automation this week.
A short video explaining how QUT's camera-based GPS works is available on YouTube.
Professor Peter Corke said QUT was at the forefront of advances in technology to enable machines to see as well as people.
"This research is another step toward major advances in robotic innovation," Professor Corke said.
Dr Milford said the camera-based system was better than current technology on the market because it didn't rely on satellites and worked day and night in extreme conditions like storms.
"Current satellite-based systems don't work at all underground in tunnels and often drop out in city areas around tall buildings," he said.
"Camera-based GPS offers the potential to leverage extremely cheap mass produced cameras rather than more expensive lasers used on many existing platforms," he said.
Pepperell said visual place recognition technology had increasingly gained traction due to low costs, small size and low-power requirements.
"The ability to work in GPS-denied areas such as tunnels and adapt to any environmental or weather conditions is what we have achieved in several small-scale trials," he said.
He said the system was used to "fingerprint" locations in south-east Queensland.
"Using a new multi-scale image comparison technique, video images of a roadway are stored and then converted to contrast patterns," Pepperell said.
"The camera-based GPS then matches up the location based on the sequences stored in its memory."
Dr Milford said the trial demonstrated the system was successful across day and night cycles and weather and seasonal variations.
"We were surprised at how well camera-based systems can work in extreme conditions like night time and storms," he said.
"Because our system uses low resolution imagery, you could theoretically crowd source live navigation information from each car's camera GPS system to allow you to adapt to dynamic events like storms."
He said it was also capable of using street imagery databases such as Google Street View and other online images from road networks.
"A potential avenue for future work would be to leverage 3D reconstruction or scene understanding techniques," he said.
"The new research enabled the navigation system to recognise the same stretch of road, whether it's travelling along the same lane as the database images or four lanes over the on opposite side of a median strip."
Dr Milford said the low resolution used solved potential concerns about privacy ramifications because nothing was identifiable in the low resolution, grey images which used a low bandwidth to crowd source the data.