Sign In

Communications of the ACM

ACM TechNews

Human Eye Inspires Advance in Computer Vision

View as: Print Mobile App Share:
Boston College Computer Science Assistant Professor Hao Jiang

Boston College Asst. Professor Hao Jiang said the method that he and Clare Boothe Luce Asst. Professor Stella X. Yu developed uses a linear approximation to explore a space globally and quickly.

Credit: Simon Fraser University

Boston College (BC) computer scientists have developed a program that enables computers to see quickly moving objects with nearly double the accuracy and 10 times the speed of previously developed methods. The researchers say their linear method has direct applications in action and object recognition, surveillance, wide-base stereo microscopy, and three-dimensional shape reconstruction.

Current computer visualization techniques use software to capture the live image, and then search through millions of possible object configurations to find a match, a process that becomes increasingly difficult when the objects move. Instead of searching through an image bank, which requires significant time and memory consumption, BC computer scientists Hao Jiang and Stella X. Yu based their approach on the way the human eye works. "When the human eye searches for an object it looks globally for the rough location, size, and orientation of the object. Then it zeros in on the details," Jiang says. "Our method behaves in a similar fashion, using a linear approximation to explore the search space globally and quickly; then it works to identify the moving object by frequently updating trust search regions."

Jiang and Yu's program focuses on the mathematically-generated template of an image, which appears similar to a constellation of stars with each star connected by a line. The new method enables computers to identify objects using the template of a trust search region, which the program can adjust as the object moves to find its mathematical matches. Jiang says using linear approximation in a sequence of trust regions enables the program to maintain spatial consistency as an object moves, reducing the number of variables that need optimization from several million to only a few hundred.

View a video of the technique inspired by the behavior of the human eye.

From Boston College
View Full Article


Abstracts Copyright © 2009 Information Inc., Bethesda, Maryland, USA



No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account