By Nick Gilbert
Christopher Amato, a postdoc at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL), has developed a mathematical system to accurately and quickly analyze numerous images that surveillance cameras capture at-risk locations 24 hours a day.
Taking into consideration the limitations of the existing system, Amato and Nisheeth Srivastava, Komal Kapoor, and Paul Schrater at the University of Minnesota use algorithms to browse hundreds of pages of images and match the face of a wanted criminal or terrorist within a fraction of time it would take a human camera operator.
An alarm has been incorporated into the system to realize the importance of immediate action even if the operator is not very sure as to what is happening. Accuracy is a key feature so as to prevent the alarm from going off unnecessarily. Amato’s system selects the appropriate algorithms based on the scene, airport, docks etc., which is being monitored, to determine the time taken to perform an analysis and the accuracy of the answer it delivers. This data is added to the system’s mathematical framework labeled as a partially observable Markov decision process (POMDP). Thus, the system can provide security personnel with most data in the shortest amount of time.
The framework can prompt the start and stop of the analysis once a clear picture emerges as to whether there is a known criminal around or not. It works like a human detective and can identify and track people, and can recognize and track movement of strange objects even amongst a group of objects. Other features include the ability to monitor video data obtained by unmanned aircraft, to analyze data supplied from weather monitoring sensors to ascertain locations of tornados and to analyze data from water samples obtained from autonomous underwater vehicles.
Matthijs Spaan, an artificial intelligence researcher at Delft University of Technology, believes that the POMDP will have high demand as it proves how decision-making using artificial intelligence techniques can benefit automated video surveillance.
The research team will present the paper on the system at the 24th IAAI Conference on Artificial Intelligence in Toronto this July.