Sick of having your GPS tell you to turn the wrong way up a one-way street or lead you to a dead end? Fear not: Linux-based technology developed at NICTA is on its way to help make personal navigation systems more accurate.

AutoMap, developed by National ICT Australia (NICTA), uses machine vision techniques that can detect and classify geometric shapes from video footage.

These shapes include things like signs and company logos: the type of fixtures that change frequently in a neighbourhood and make it difficult for digital map makers to keep their products up to date.

AutoMap project leader Lars Petersson says that on average 10 to 15 per cent of street signs in an area will change every year.

“Currently, to keep on top of this, the mapping companies need to get someone to physically drive up and down each street in a van with five or six cameras fixed in all directions. There will be a driver and a co-driver who will sit and make annotations. They then take this footage back to the office where they have an army of slaves who will look at this footage frame by frame and record where all the signs are,” he said.

“That is of course an extremely boring task and because it is so boring they make a lot of mistakes. What we provide is an intelligent solution that can automatically detect signs from video footage without having to employ an army of slaves to do it.”

The AutoMap system uses some of the technology developed as part of an earlier smart cars project.

Although the product is now ready for commercial deployment and discussions are underway with the significant mapping companies, research on the project will continue.

“Looking a year from now, we will have a more advanced way of gathering this data. [We are looking at placing] this technology inside a little camera and putting it in taxis, fleet vehicles, and garbage trucks that are going about their business,” Petersson said.

“These vehicles will traverse the whole road network on a regular basis. They will be able to automatically detect points of interest and automatically send this information back to base where a complete and constantly updating map emerges over time.”

The research team will also be developing methods to recognise three-dimensional images like park benches and speed cameras.

Petersson said that this research and technology is almost entirely Linux-based.

“Our use of Linux in the research and development goes without question. We don't run any of it on Windows. We have ported some GUI stuff to Windows but that is just because some of our helpers could not use Linux. But all the software is written using Linux. We have a big cluster of 100 to 200 computers running on Ubuntu that process the video. The in-car unit that we plan to place in taxis and vehicles will run on Linux.”

The research team also use an Intel-based UMPC (an ASUS R50A).

“Even though this unit has received some criticism from reviewers as far as normal usage goes, it is actually a good platform for us,” Petersson said.

NICTA predicts that the digital mapping market will expand significantly as companies like Google, Microsoft and Yahoo continue to develop and release location based services.

Whilst these companies currently purchase some mapping information from digital map producers it is expected they will quickly shift to developing and maintaining their own databases.

The entry of these companies is expected to grow the annual spend on map development by more than $US1 billion by 2010, according to the research organisation.