Otago Staff Bulletin, 9 Oct. 2011
Computer tracking and analysis of field sports using multi- camera angles is nothing new, but how about taking the brain’s own system for picking what it wants to watch and using it to improve live broadcast programming and post-match analysis?
Senior Lecturer in Information Science Dr Jeremiah Deng is working with PhD candidate Munir Shah, Dr Brendon Woodford and Professor Martin Purvis on sports-related computer vision research.
Dr Deng says computers are already able to do a lot of things, including tracking the ball and the players, and knowing which player belongs to which team.
“The idea is not new, but the approach we want to take is. Our model is inspired by the human biological neuron system. When you watch a movie or a rugby game you have numerous neurons working together processing the flow of information. We are adapting that model using smart algorithms implemented on multi-core computers and using multiple digital cameras. It will give us a better chance of understanding the game and detecting interesting events.
“In sports broadcasting you have a lot of cameras trained on a field and someone manually choosing which shot to use. Instead we can have the computer choose which angle gives the best view, instantaneously.”
Dr Deng envisages using the system with multiple overhead ‘spider cams’ that ‘fly’ over a field game, assisting referee decision- making.
“Fouls, faking, offsides, dives, and other infringements would all become much more apparent, reducing human error,” he says.
Mr Shah says the enhanced footage would help coaches and players analyse their performance.
PhD candidate Munir Shah, Dr Brendon Woodford and Dr Jeremiah Deng are part of a team looking to vastly improve the quality of sports video recording by mimicking our brain’s visual processing methods.
“Our model is inspired by the human biological neuron system.”
He says the research is primarily focused on soccer footage, but the technology could be fine-tuned to any field sport, rugby included.
Computer vision research associated this work has been supported by two University of Otago grants, with results published in the top tier journals including Pattern Recognition.
