Tweet
Surveillance: the next generation
How can you get CCTV cameras to not only record, but to recognize patterns of behavior.
Nobody would like to spend their time staring at a CCTV image of a bicycle rack and trying to spot a thief. Prompted by her PhD supervisor, Professor David Hogg, Damen realised that a computer vision application could help detect suspicious activity. "I thought that whatever vision could do would certainly be better than the current situation," she says.
The need was clear - half a million bicycles are stolen each year - but how could it be done? "In computer vision, there are techniques to track people when they're walking in front of the camera," says Damen.
She then realised that if a smart CCTV system tracked people dropping off and picking up their bicycles, perhaps any differences could be noted. In computer vision terms, this meant comparing moving blocks of pixels from different times. The initial tests - where 11 out of 13 simulated thefts were successfully detected - involved using clothes' colouring to compare individuals and raising a theft warning when they looked different.
"The hard bit, actually, is to link the drop at the beginning of the day to the pickup at the end of the day," says Damen.
"If you wanted to try and cover everywhere, you'd need a whole army of people watching CCTV. That's essentially the driver for this technology - it's automation, it's doing things more efficiently."
Dr James Orwell, Digital Imaging Research Center, Kingston University, London
How can you get CCTV cameras to not only record, but to recognize patterns of behavior.
Nobody would like to spend their time staring at a CCTV image of a bicycle rack and trying to spot a thief. Prompted by her PhD supervisor, Professor David Hogg, Damen realised that a computer vision application could help detect suspicious activity. "I thought that whatever vision could do would certainly be better than the current situation," she says.
The need was clear - half a million bicycles are stolen each year - but how could it be done? "In computer vision, there are techniques to track people when they're walking in front of the camera," says Damen.
She then realised that if a smart CCTV system tracked people dropping off and picking up their bicycles, perhaps any differences could be noted. In computer vision terms, this meant comparing moving blocks of pixels from different times. The initial tests - where 11 out of 13 simulated thefts were successfully detected - involved using clothes' colouring to compare individuals and raising a theft warning when they looked different.
"The hard bit, actually, is to link the drop at the beginning of the day to the pickup at the end of the day," says Damen.
"If you wanted to try and cover everywhere, you'd need a whole army of people watching CCTV. That's essentially the driver for this technology - it's automation, it's doing things more efficiently."
Dr James Orwell, Digital Imaging Research Center, Kingston University, London