Bourns College of Engineering

TechHorizons Attracts Top Experts to Research Community Meeting

Seventy-two of the top researchers in large-scale video sensor networks gathered at UC Riverside in May, evoking a future of intelligent machine surveillance, nanosensor networks, personal robots and a revolution in scientific observation.

The week of high-level technology talks at the Bourns College of Engineering's TechHorizons 2009 represented a "quantum leap" in the research agenda for that research community.

The military started funding research into the computer vision/machine understanding field more than 30 years ago. Now for the first time a critical mass of engineers and scientists representing universities, industry and government gathered to consider the future of their efforts. The workshop was sponsored by the National Science Foundation, the Office of Naval Research and the Army Research Office.

The gathering was held at the Bourns College of Engineering which has built "Vislab," the most extensive research network of wireless and mobile (robotic) cameras in the world. Using the 80 camera wireless network capable of 100 video streams in real time, UCR researchers are participating in the burgeoning research area which combines numerous fields including machine intelligence, pattern recognition, data mining, networking, and algorithms.

Electrical Engineering Professors Bir Bhanu, Amit Roy-Chowdhury, and Chinya Ravishankar, Professor of Computer Science and Engineering, are the principal investigators for a number of grants and contracts totaling more than $8 million supporting the work.

A MILLION CAMERAS

Indicating the breadth of the participants interested in the field, the workshop was organized to consider advancements in six areas: video processing and video understanding; simulation, graphics, cognition and video networks; wireless video sensor networks, communications and control; distributed embedded cameras and real time video analysis; and applications. The sixth group to be considered was the higher education implications and attracting young researchers to the field.

"Now that we can put a million cameras over a hundred kilometer area, what are the new problems that can be solved?" asked Rama Chellappa, Minta Martin Professor of Engineering and Director of the Center for Automation Research at the University of Maryland, College Park.

The ability to recognize objects and classes of objects in motion as either normal or abnormal will open the door to many important new applications, the researchers agreed. The developing technologies will permit monitoring of transportation systems, urban patterns and help in disaster recovery. It is also a powerful research tool for monitoring natural habitats to study like growth cycles in plants, migration patterns, animal habitats and colony collapse disorder in bees. It also opens the door to personal robots joining households.

The group recognized there are still many aspects of the basic technology to be worked out. The science will need to develop engineering models to identify group activities, crowd behavior and other dynamic properties. The technology will require a host of new algorithms to calibrate, integrate, control and interpret the massive data streams in a camera network.

TechHorizons 2009 was the third annual technology transfer conference at the Bourns College of Engineering. It focused on the long-term goal to achieve persistent awareness in large-scale video sensor networks.

Discussions and conclusions from the workshop are to be published and widely disseminated among the research community.

A STRONGER RESEARCH COMMUNITY

"We must build on each other's success," said Gerard Medioni, Professor of Computer Science and Electrical Engineering, and co-director of the Institute for Robotics and Intelligent Systems at the University of Southern California. "We have made great progress and our algorithms and data sets are available on line. Now we can build on each other's work where our findings can be tested against previous results. This is enabling," he added.

Among the highlights of TechHorizons 2009 were: a presentation by IBM Distinguished Scientist Arun Hampapur on current systems installed in Chicago and elsewhere; advances in nanosensors by Georgia Tech Distinguished Professor Ian Akyildiz; and a roundtable discussion on surveillance practices and needs by representatives of Caltrans, the FBI, Los Angeles World Airports, and a crisis response researcher from UC Irvine.

TechHorizons also highlighted the contributions to the field of 10 Bourns professors of Computer Science and Engineering, Electrical Engineering, and Mechanical Engineering.

Professor of Electrical Engineering Bir Bhanu, for instance, reviewed how game theory is being used to govern camera selection and handoff in the Bourns' experimental test bed. Game theory uses competition and cooperation algorithms which can be used to analyze interaction and conflict by the individual cameras in the network. Bhanu said his group is also investigating biological models for camera control.

Eamonn Keogh, Professor of Computer Science and Engineering, explained his ground-breaking data mining technique of translating video shapes, faces for instance, into "shapelets," providing a fast and accurate technique for classification of video subjects.

Ertem Tuncel, Associate Professor of Electrical Engineering, presented his work combining audio and video sensors to achieve cost reductions in effective multi-subject tracking. The technology also requires less bandwidth, video processing power and video storage, making it practical for a variety of applications. Using fewer cameras can also effectively monitor areas with fewer privacy concerns since cameras are only activated to resolve ambiguity in the sensor net.

Amit Roy-Chowdhury, Assistant Professor of Electrical Engineering, reviewed his work in active vision, an inexpensive scheme to get selective high resolution views in a large-scale video network. Processing power is embedded in each camera in the network. Special camera control algorithms allow the cameras to actively cooperate, selecting one camera to zoom in for a high resolution view of an object of interest, while the remaining cameras reconfigure themselves to cover the entire zone of view.

"We have developed algorithms for camera control and tracking, activity recognition and facial recognition," Roy-Chowdhury said. "The next step is to integrate all of them for robust video analysis."

The events fulfilled all of the goals the organizers had. "The workshop participants thoroughly enjoyed the overview talks given by the world class researchers in their respective fields of research," Prof. Bhanu said. "There was a lot of discussion of ideas and the workshop was productive."

"The field of distributed video sensor networks is evolving rapidly and we hope that the sponsoring organizations will create significant research and development opportunities in this important area," Bhanu added.