Predator: A Visual Tracker that Learns from its Errors

Google Tech Talk (more info below) May 4, 2011 Presented by Zdenek Kalal. ABSTRACT I will be talking about an algorithm that I have developed during my PhD thesis and which recently become popular on the internet: http://goo.gl/rC5Xj. The algorithm is called Predator and it is a visual tracker that has the property to improve its own performance during run-time. This is achieved by designing a feedback which is in computer vision not common. I will cover the journey that lead to the development of Predator, comment on the implementation and suggest potential directions for future research. About the Speaker Zdenek Kalal is a stubborn and hardworking person that enjoys working on problems, the solution of which may have high impact. His education started in Czech Republic, where he absolved business high school and later entered Czech Technical University to study control theory. His master thesis, dealing with face detection, was awarded the Prize of the Dean. In 2007, he started his PhD at the University of Surrey, UK. He focused on self-improving real-time vision algorithms and published 5 research papers at international computer vision conferences. His research effort was recognized by the UK ICT Pioneer 2011 prize which resulted in a significant attention from hackers, artists, universities up to top companies. Zdenek has released the source code of his PhD thesis and maintains a supporting discussion group with nearly 600 members to date. He is currently a postdoc at EPFL in Switzerland where he works on augmented reality applications.