Roger Needham lecture 2006

Computer vision and the geometry of nature

Speaker: Dr Andrew Fitzgibbon, Microsoft Research, Cambridge.

Computer vision is the search for mathematical models and algorithms which can explain and emulate the tremendous visual abilities that most of us rarely notice we possess. For instance, we can easily recognise thousands of objects, follow complicated movements and almost subconsciously build a three-dimensional view of the world through stereo vision.

When a camera captures a movie of some scene in the world, the rich visual complexity of the scene is not lost - we can still enjoy the images and recognise the film's contents - but the visual patterns are translated to complex numerical arrangements which current mathematics and statistics strive to represent and understand. Although science is far from having a complete understanding of the processes of vision, the last decade has seen applications of artificial vision move out of the lab into the real world.

The lecture looks mainly at the use of computer vision in obtaining a 3D representation of the world and at the application of these techniques to cinematic special effects, in film series such as "Harry Potter" and "The Lord of the Rings". Commencing with some aspects of the human visual system, Dr Fitzgibbon will try to answer the question of whether it is even reasonable to expect to emulate human vision without first building an artificial intelligence. Is vision “AI-hard”?

Dr Fitzgibbon will describe several classic experiments which suggest that not all tasks require AI. He will then consider applications where these tasks arise, e.g. robot navigation and special effects, showing how a combination of engineering and geometry gives us reliable solutions in real scenes after looking at how a manmade world simplifies the solutions to these problems.