UCF SIGNALS NEW RESEARCH THRUST IN COMPUTER VISION

Anyone who has passed the corner of Orange Avenue and Pine Street in downtown Orlando recently has witnessed a UCF research project.

The unobtrusive nature of the KNIGHT surveillance system is indicative of the quiet way in which computer vision is increasingly becoming a part of all our lives.

Mubarak Shah, the founder of UCF’s Computer Vision Lab, is principal investigator of the KNIGHT project, which was developed in 2001 as a field test program to help the Orlando Police Department with electronic patrol.

Cameras were installed at four intersections along Orange Avenue in downtown Orlando and the accompanying computer surveillance system was designed to detect important changes, events, and activities. Significant events are flagged and presented in a summary fashion to a monitoring officer for final analysis and response decision.

Shah and his graduate students are studying the results and fine-tuning the program to make it even more effective at screening out unnecessary information while leaving important data intact.

For example, while KNIGHT initially did very well in identifying bjects that were some distance away from each other, it had some trouble detecting items in a crowd – such as a bag that people were continually walking in front of.

Subsequent enhancements have made the program highly efficient in detecting and classifying objects and tracking them across multiple cameras, traits that are essential in the types of defense and homeland security applications the program is now being utilized for.

Shah defines computer vision as the processing and interpretation of video images by computer. He said the field has been studied for about forty years and has received increased emphasis since the terrorist attacks of Sept. 11, 2001.

UCF has one of the oldest and most comprehensive Computer Vision programs in the nation, Shah said, and has consistently ranked among the top in national and international competitions. Six papers from the Vision Lab have been accepted for presentation at the IEEE Computer Vision and Pattern Recognition conference this summer, an event that accepts only six percent of papers proposed.

Many people make the mistake of confusing computer vision work with manual video surveillance, he said. In fact, there is one key difference – with computer vision the data analysis is done by machine, saving human expertise for the times it is required.

“The goal is you want to be able to have a screener so that humans will only look when there is something interesting,” Shah said.

KNIGHT is built around three key components of computer vision: detection, tracking and classification, Shah said.

Detection is comprised simply of determining whether an object under surveillance is moving. This can be accomplished by viewing an object through multiple cameras positioned in different locations. Often detection is obscured by color similarities between the object and the background or multiple moving objects in close proximity to each other. Once an object is detected, the movement is categorized, i.e. is it a person, vehicle, bicycle, animal, male or female, etc.

The object is then tracked from frame to frame to determine speed and direction of movement. And finally, the movement is classified based on a database of thousands of routine gestures categorized by environment and function, such as a video clip of the act of typing on a computer keyboard in an office or walking across the street at an intersection.

“We are just at the beginning of categorizing behaviors,” Shah said. “These programs are still at the university research level.”

In addition to the Orlando Police Department, variations of KNIGHT have been funded by the Defense Advanced Research Projects Agency Small Business Technology Transfer program, the Lockheed Martin Corporation and the Florida Department of Transportation. The aim of the FDOT project is to use KNIGHT to monitor railroad grade crossings and to automatically inform authorities in case of potential hazards such as the presence of a person or a vehicle on the tracks when a train is approaching.

In the coming months Shah will be working in his new role as was assistant vice president for research in computer vision in order to develop ways to integrate the science into disciplines across campus.

Specifically he will work closely with UCF’s areas of research strength including life science, nanoscience, and photonics.

“Dr. Shah is extremely well respected both in his field and across campus for the breadth and depth of his research work as well as his commitment to students and to UCF. We are fortunate to have him in this exciting new capacity,” said M.J. Soileau, UCF’s vice president for research.

shah@cs.ucf.edu