Surveillance images can be notoriously difficult for human eyes to keep track of -- is there a better way?
Security cameras. As commonplace as chewing gum on the sidewalk, wherever we are, it's almost expected now to be watched. Have a drink in a bar, pick out a new outfit, walk down the street --security technology can be used for safety, profiling, or even to analyze customer preferences.
It's normal. Whether we like it or not, whether it makes us feel safer or we think its an invasion of privacy, cameras are part of daily life in urban cities.
If you're in a high-security area, such as a country border, airport or governmental facility, these devices rear their heads in every corner. However, no matter how advanced or accurate the technology is, human eyes are required to verify and detect safety risks and dangerous situations.
The continual flow of information and images can be difficult for us mere mortals to keep on top of. Networks that cover areas 24/7 can be too much for human scanning, and so mistakes can be made -- sometimes with serious consequences.
In order to try and combat this problem, a new system is being developed by Christopher Amato, a postdoctorate at MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL). Instead of having one person straining their eyes by staring at a single screen for hours on end, the technology is designed to perform a real-time analysis of images automatically in a fraction of the time.
"A person is not going to be very good at searching through pages and pages of faces to try to match [an intruder] with a known criminal or terrorist.
Sometimes it's important to come up with an alarm immediately, even if you are not yet positive exactly what it is happening. If something bad is going on, you want to know about it as soon as possible."
With colleagues Komal Kapoor, Nisheeth Srivastava and Paul Schrater at the University of Minnesota , they developed a system based on the mathematical principles of accuracy and probability. Rather than alarms being triggered if a fly zooms across a camera, it is hoped that by using different algorithms, video feeds can be analyzed quickly so security staff can given as much time as possible to respond to real threats.
The system works through the input of different algorithms including skin detection, face recognition, background analysis (for unusual objects) and movement.
Once these parameters are in place, Amato’s system begins a "learning phase" where it automatically assesses how its software can be applied to different situations -- such as airport security or warehouses. This information is added to the mathematical framework it uses -- the partially observable Markov decision process (POMDP) -- in order to calculate high-risk or unusual situations.
If the system is used at an airport, it could be programmed to detect 'people of interest', abandoned objects or odd behavior, for example. An alarm can then be sounded to alert security staff. Amato said:
"We plug all of the things we have learned into the POMDP framework, and it comes up with a policy that might tell you to start out with a skin analysis, for example, and then depending what you find out you might run an analysis to try to figure out who the person is, or use a tracking system to figure out where they are [in each frame].
"You continue doing this until the framework tells you to stop, essentially, when it is confident enough in its analysis to say there is a known terrorist here, for example, or that nothing is going on at all."
In addition to general security concerns, it is possible the technology could be used to improve weather systems, natural disaster detection or underwater studies.
The system will be presented at the 24th IAAI Conference on Artificial Intelligence in Toronto in July.
Image credit: Howard Dickins