System considered that links video camera with automatic target recognition

Imagine a series of video cameras surveying vehicular traffic around the St. Louis Arch. The video cameras are programmed to detect anomalies in traffic patterns, and one of them tracks a vehicle traveling erratically. Data from the video camera is relayed to a computer with specialized software that applies automatic target recognition (ATR) technology. This software, within two minutes of receipt, breaks the video data down and identifies the vehicle as one likely wired with explosives. An alarm is then issued to security personnel to intercept the vehicle or provide further barriers so that the vehicle cannot deliver its malevolent goods.

Two researchers at Washington University in St. Louis have the expertise and are in the early stages of collaboration to one day test such a system. Jody O’Sullivan, Ph.D., professor of electrical engineering, is an expert in ATR. Robert Pless, Ph.D., assistant professor of computer science, is an expert in computer vision and has developed several cameras that are “trained” to detect suspicious movement. The researchers are assistant directors of Washington University’s Center for Security Technologies, a research center devoted to developing technologies that safeguard the United States against terrorist attack.

The St. Louis arch has been thought to be a potential terrorist target. Researchers at Washington University’s Center for Security Technologies are considering an automated warning system that would spot suspicious traffic around the Arch or any site.

Ronald Indeck, Ph.D., Das Family Distinguished Professor of Electrical Engineering is director of the center, which features nearly 40 interdisciplinary collaborators. The Center was founded in early 2002 to address the fundamental scientific and engineering questions that arise in the design of advanced security systems. The primary motivation for the technical direction of projects is to provide the basis for new industries and to assist the government and existing corporations through collaboration. The goal is technology development to defend against an array of threats, including: cybersecurity, critical infrastructures, environmental catastrophes and natural disasters, among other possibilities.

“We are building on many years of knowledge, modeling and imagery expertise to make advances in specific areas important to homeland security,’ O’Sullivan said.

Video surveillance is being deployed in more and more places as a security tool, Pless said. With all the video data being collected, it is important to isolate important events and unusual behavior. Pless recently has developed new methods for capturing the “typical” motions that a surveillance camera might see. He and his research students have written programs to automatically mark objects that are not following the typical motions — that are behaving erratically. Whether these objects are threats depends on identifying what these objects are.

Automatic target recognition is a component of the nation’s military. Various aircraft from helicopters to fighter pilots to drones use the technology to spot tanks, troops, vehicles or missiles in uncertain visual environments. O’Sullivan has nearly two decades of experience in imaging science, radar and ATR, among other areas. His skills are in modeling the physics of scenes and sensors and then deriving algorithms — mathematical programs — based on these models. He and his research group implement these algorithms in software and then tests whether the software is an accurate model of how the sensor responds to the scene.

O’Sullivan’s ATR approach is different than prevailing ATR technology in that the systems he designs are capable of updating their knowledge by computing on the fly. Most ATR systems now take an input, process it and give an answer in a certain amount of time, say, 60 seconds. The time component always stays the same, and the algorithm cannot compute longer than that. O’Sullivan’s approach, on the other hand, would allow the system to make a decision taking longer, say two minutes, which makes the decision more reliable because of increased computations. It would also allow the system to take a shorter amount of time, say 25 seconds, to identify a target. His system also has a framework for predicting how the performance of the system varies as the amount of time allotted for recognition varies.

“No ATR technology can be 100 percent accurate,” O’Sullivan said. “We think our approach can produce more reliable estimates. We know the military is interested. They see our approach as a good way to think of adapting systems in the future.”

The fusion of O’Sullivan’s ATR theory and Pless’s video cameras also could be used in fingerprint identification, voice recognition, face recognition, and gait recognition applications, which could be used in a wide variety of homeland security devices.