Surveillance cameras may be a tool for solving crimes, but what about using them to prevent or stop criminal and terrorist acts? That calls for someone or some "thing" to keep an eye on the video feeds 24-7.
While computerized monitoring would seem to be the obvious answer, creating software programs that can recognize suspicious activities or suspect individuals has proven highly difficult -- a solution that Rama Chellappa, a professor in the department of electrical and computer engineering of the University of Maryland's A. James Clark School of Engineering, is hoping to provide through the development of a real-time computer monitoring system using artificial intelligence.
A pioneer in the development of pattern recognition and computer vision software, Chellappa and his research assistants have developed a compact, digital signature for characterizing human gait and corresponding activities, such as humans carrying objects like backpacks, handbags, or briefcases, using video data from digital surveillance cameras and corresponding algorithms.
When a person's limbs are unencumbered, gait movements are symmetrical. Represented graphically, these movements form a twisted helical pattern resembling a "figure 8" called a double helical signature.
Chellappa and his team call this pattern, which is slightly different in each individual, "human gait DNA."
An individual's gait pattern is changed by any activity that changes the symmetry of the movements, such as carrying a package. By defining these signatures, the system can recognize unique patterns in human gait and automatically detect asymmetric movements like an individual walking with a hidden object tied to an ankle or wrist. Hidden objects secured to the body in ways that don't affect movement symmetry, for example, a fanny pack that is belted around the waist, aren't currently detected by this technology.
Chellappa and his team have integrated human gait DNA into a real-time video surveillance system and used it to study and locate pedestrians. The experimental results have demonstrated the effectiveness of the system under lighting changes, shadows, camera motion, various viewing angles, as well as significant obstacles in the cameras' viewing angles. The results also indicate that the approach is superior to many existing methods in terms of accuracy and reliability.
"These capabilities are extremely useful in creating a surveillance system intended to address security concerns," said Chellappa.
His research team is also "teaching" their gait recognition system to identify individuals by their unique gait. This is a much more difficult task, since subjects may deliberately attempt to walk in an uncharacteristic manner in order to try and cheat the system and avoid detection. If the suspect is unaware of the surveillance system, their normal walking style is more easily identified.
The Maryland team has also developed advanced face recognition software that can be combined with their gait recognition technology to watch for known terrorists, spies or criminals and help to identify unknown individuals who might turn up repeatedly in sensitive locations or who have been present during multiple criminal or terrorist acts.
Chellappa's team additionally developed two other recognition technologies that can add to the capabilities of automated video surveillance systems. Through work supported by the department of Homeland Security, Chellappa and one of his students have developed an algorithm to accurately estimate the heights of subjects in the field of view of a camera, providing what they call an important additional way to recognize and track subjects in crowded settings.
The team has also developed a program that detects unattended packages using a structured representation known as attribute grammars. Both of these technologies were demonstrated at the recently held 25th Army Science Conference.