A company formed using government and university funds has begun to commercialise a video technology that learns to recognise unusual behaviour from moving video footage, such as someone falling down the stairs at a train station.
The software, developed by a team of researchers at Curtin University, is designed to piggyback on existing video management programs such as Milestone Systems to provide more effective surveillance capabilities.
The software will be commercialised by a company called iCetana Pty Ltd, and is supported by $1.2 million in investment from venture capital firm Yuuwa Capital and $300,000 in Curtin University pre-seed funding.
The difference between this software and existing video analytics programs is that users don't have to tell the software what is unusual, according to Matthew Macfarlane, interim CEO of iCetana. Instead, exceptions are automatically defined through a learning phase, which could only last a few hours (depending on the area being monitored).
For instance, a camera trained on a train tunnel would disregard trains running regularly through it, but would notice if someone slips into the tunnel and walks down the tracks. An alert could be sent to security workers for the station, and they could notify police, or in the case of someone falling down the stairs as in the first example, an ambulance.
The software works using component analysis of motion vectors, looking at where pixels are moving. The constraint of the software is that it won't notice unusual occurrences in a motionless environment.
Macfarlane said that although such software might be more prone to false positives than other methods of watching footage, even a busy camera, such as one trained on people milling outside a bus stop, would only send five to six alerts a day. The next release of the software would also have an option for users to notify the system when it had returned a false positive.
Belmont City Council in Western Australia has already been trialling the technology for the last six months, according to the head of the team which developed the software, Professor Svetha Venkatesh.
"During this pilot program the software was able to identify behaviour such as loitering in a normal social and built environment, arson attempts, unusual sized groups, incorrect vehicle traffic direction, and anti-social and illegal behaviour," she said in a statement.
The company has also been in talks with the Western Australia public transport authority about using the system, according to Macfarlane. The thought was to offer it initially as an unpaid pilot and then work from there.
The venture capital funding available to Yuuwa has come from the Federal Government's Innovation Investment Fund, where $20 million at a time is doled out to form new venture capital funds that invest in start-ups. Yuuwa was formed last year. This has been its first investment.
According to Macfarlane, the company could release the software for sale in around three months. He didn't think that any other company had software which didn't require the definition of normal conditions in order to work, but did say that competitors working in the area would be Agent Vi, ObjectVideo and Australian company iOmniscient.