“I consider myself an inventor. I like to create stuff,” says Jezekiel Ben-Arie, an engineer-turned-scholar whose avocation is painting and sculpture.

His vocation is learning ways to translate human activity images into information that computers can quickly recognize and analyze.

Ben-Arie’s latest invention, now protected through a provisional patent, presents a fast, efficient way of recognizing human activities through video analysis.

“It’s a new method for recognition of sequences, or vectors,” says Ben-Arie, professor of electrical and computer engineering. “A vector sequence could be a lot of things. It could be human activity, it could be speech or other things.”

“I developed a method based on video sampling. You only have to sample a few static images from the sequence to recognize what the person is doing. It’s a nice method which gives 100 percent recognition, judging from about 100 videos we tested so far.”

Ben-Arie’s method improves on current recognition methods by accurately analyzing activities faster, using less information. In computer parlance, there’s less bandwidth required to send the information needed. That should help keep costs down.

His invention has applicability to a wide range of scientific research areas, but has commercial potential by enhancing security systems that use video surveillance.

“It could be used with security surveillance systems, where you want to see if someone is committing an illegal activity, breaking into some place or placing a bomb,” says Ben-Arie.

“Security guards get bored and can’t keep watch on hundreds of security monitors 24 hours a day. If you use this to help identify suspicious activities and alert security officials, it could be very useful.”

“Analysis of human action is also important in many areas such as kinesiology, ergonomic research, anthropology, biomechanics, rehabilitation, sports medicine and athletic analysis,” he says.

“In addition, research in artistic areas such as dance choreography and figure skating require accurate analysis of human motion.”

Ben-Arie’s expertise in the fields of computer vision, image processing, neural networks, human computer interaction and related fields is underscored by his more than 120 technical publications on the subjects and numerous research projects supported by organizations such as the National Science Foundation, the Whitaker Foundation, the Defense Advanced Research Projects Agency and others.

He is associate editor of IEEE Transactions on Image Processing, a professional journal of the Institute of Electrical and Electronics Engineers, an international professional association.

Ben-Arie discovered a general visual-geometric phenomenon called the probabilistic peaking effect of viewed angles and distances, which serves as the basis of the area of quasi-invariants in image understanding.

He developed the non-orthogonal expansion matching method used in recognizing highly occluded objects, tracking and in motion video compression. Recent developments include the Affine Invariant Spectral Signatures and the Volumetric Frequency Representation, which provide novel, efficient means for object and face recognition and representation.

Ben-Arie’s research has also helped advance the work in such topics as shape from recognition, model-based segmentation, neural modeling of auditory localization, evolutional design of the human ear, conveying visual information with spatial auditory patterns, shape description by magnetic fields modeling, and pose invariant face recognition using 3-D frequency domain modeling.

While most of his time is devoted to research, he teaches both graduate and undergraduate courses.

“I like teaching. It is truly rewarding. I think I’m a good teacher,” says Ben-Arie.

“My wife says I’m a good teacher. She’s a teacher too. She must know.”

Photo: Grant Therkildsen