Team earns top honors at Pascal Challenge a second time
Rather than basking in the success and recognition they earned in the Pascal Visual Object Classes Challenge, Tony Han, electrical and computer engineering assistant professor at the University of Missouri said he and his team are continuing to improve their performance and are far from reaching their ultimate goals. The Challenge measures the success of researchers’ algorithms in the areas of pattern analysis, statistical modeling and computational learning.
Han and his graduate assistants — Xuao Lv, Xiaoyu Wang and Xi Zhou — were joint winners with five other worldwide competitors in the action classification challenge in September 2010. In 2009, Han’s research team ranked third overall in object detection following the University of Chicago and Toyota Institutes joint team who placed first and the University of Oxford who placed second.
“The competition is a great way to receive feedback and work towards better detection for practical applications,” Han said.
The international competition invites researchers from universities and institutes around the world to submit their software and algorithms to compete for the best accuracy. The 2010 competition had three main challenges — image classification, object detection and object segmentation. There were also competitions in person layout, still image classification and large-scale recognition.
Action classification recognizes a human action, like running, from a simple action picture. Han and his team perfected their algorithms for over two years before submitting their work to the competition.
“We want a computer that performs as well or better than a human,” Han said. “Tasks that are easy for humans can be very difficult for a computer.”
He hopes to continue improving the accuracy of their algorithms to eventually reach that standard — similar to the ‘Watson’ computer software system designed for complex analytics that has beaten some top Jeopardy contestants.
“There are so many applications for this technology. Eventually it will be possible for a computer to recognize your facial expressions and understand your gestures and respond. Computers will be able to perform the tasks for you,” Han said.
For now, Han and his team are preparing for the 2011 Pascal Challenge, where they will compete in both categories of object detection and action recognition. They also are submitting papers to journals in hopes of publishing their work.
“The most important key to our progress is the hard work of my graduate students. They devote everything to their research,” Han said. “It may take years but we are always getting closer to our goal.”