Seeing more with computer vision

Cameras have been used for surveillance for long, but can they ‘see’ beyond the plain image? A combination of cameras and computer vision programs now promises to do just that. The system can create a revolution in security and in non-security areas such as health, transport and business.
Graduating from plain surveillance, cameras can now track a variety of things — behaviour, movement and presence or absence of objects. These capabilities assume importance in security operations. In a busy railway station, an object such as a package or suitcase left unattended can be picked up by a computer vision system because it suddenly stops moving. But a security person monitoring display screens may miss that.
Computers and cameras engage Arun Hampapur, manager of the Exploratory Computer Vision Group at IBM’s industry solutions laboratory in Hawthorne, New York. He has been ‘seeing more’ through cameras and computer vision, and the technology is being commercialised as the Smart Surveillance System.
Dr. Hampapur, who was the design consultant for the CNN Video archive system created by IBM and Sony, says biometric recognition is an exciting area today. “In the area of recognition, if the situation is controlled, you can do very well,” he said. Iris recognition for immigration is an example. This is done with the subject standing close to the camera, under controlled lighting. But in a train station with a lot of people, the infrastructure is pre-existing and the challenge is to get the right picture. “The bottomline is that our recognition is only as good as the picture. Getting to high levels of recognition is the next challenge.”
Another challenge is presented by the need for a database. Governments invariably have no idea who they are looking for ahead of time. “If it is common criminals, they have been arrested three times, and you can get a database,” he says.
There are non-security applications too. Business optimisation is one. “Data come from somebody or some transaction system,” he explains. “You take that data and do a lot of things with it.” A camera could create a record. It can measure how long the luggage pick-up or fuel truck took at an airport. Through its ability to do a range of tasks, it removes the bottleneck .
“We have not even begun to understand the implications of all this. I can have five cameras and suddenly I have much more insight into what is progressing,” says Dr. Hampapur, who holds eight patents and has expertise in media indexing, video analysis and video surveillance.
Another promising field is healthcare. In the U.S. there is a great deal of concern about the aging demographic and reducing the cost of caring for the elderly. If a senior citizen could be kept at home for just a month longer it would result in a very significant cost benefit.
“You can think of not just a camera, but other sensors at home for potential alerting purposes and for long term tracking,” he says.
For instance, an aging person may be walking at speeds significantly slower than in previous years. With these systems, precise measurements can be made of the time taken to go from the bedroom to the kitchen. The data could be kept for years and combined with other tests, such as blood pressure. “If we had a system and someone had a stroke, you can go back and say, was there anything in my data which could have been an indicator? Today, the only way we can get data is when you go to the doctor.”
In commerce, computer vision can help organised retailing. It would relate marketing and money spent. But does not a pervasive surveillance camera culture affect civil rights? Dr. Hampapur says it is a broad concern not just with cameras, but with computers and data in general. It covers a broad range of information and not merely images. “People are more sensitive about being watched, but not about someone getting access to their social security number [in the U.S.] or even telephone numbers.”