Engineers from the machine learning capability team at Microsoft Research Cambridge are in line to follow up their Guinness Book of World Records sales success with a major UK award for masterminding the Kinect human motion capture system, used with Xbox 360.
Microsoft is one of four elite UK teams tilting at Britain’s biggest engineering innovation prize – the Royal Academy of Engineering MacRobert Award.
Kinect for Xbox 360 has secured a position in the Guinness Book of World Records as the fastest selling consumer electronics device, with eight million sold in the first two months after launch.
The other finalists for the £50,000 prize for 2011 are:-
• Radio Design Ltd for the radio frequency filter that allows telephone companies like Orange and T-mobile to share their networks
• Jaguar for the new lightweight aluminium XJ body, the first production car to be made using aerospace cold joining technology
• Defence Science & Technology Laboratory for a new modular ceramic armour system for armoured personnel carriers.
The judging panel of eminent engineers, innovators and entrepreneurs will unveil their decision at the Academy Awards Dinner in London on June 6.
Microsoft’s Kinect represents a significant advance in technology for ‘Natural User Interface’ between man and machine. It was launched last November as ‘Kinect for Xbox 360,’ enabling controller-free games and entertainment in a new way.
Movies and music can be controlled with the wave of a hand or the sound of your voice. Effectively, the user’s body is the controller. A major breakthrough in Kinect is the use of machine learning to classify parts of the body.
The Kinect sensor provides a stream of 3D ‘depth images’. This is analysed by software to give a moving interpretation of the human skeleton, at 30 frames per second.
Before Kinect, equipment for motion-capture was already available commercially but required instrumentation of the moving human subject, in the form of retro-reflective markers, placed on all body joints.
For user interfaces, however, it is imperative that motion capture is markerless. The Microsoft Research Cambridge laboratory applied machine learning techniques to build a capability to analyse depth images independently, classifying pixels in each depth image as belonging to one of 31 body parts.
The classifier is trained and tested using a very large database of pre-classified images, covering varied poses and body types. It is engineered so efficiently that it uses only a fraction of the total available computing capacity – essential to the practical success of Kinect.
The Microsoft team members cited for the Award are: Research scientist Dr Jamie Shotton; Principal Research Scientist , Dr Andrew Fitzgibbon; Software development engineer Mat Cook; senior research software development engineer Toby Sharp; and team leader Professor Andrew Blake, MD of the Cambridge research laboratory.