BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Body Part Recognition: Making Kinect Robust - Jamie Shotton\, MSR 
 Cambridge
DTSTART:20110601T131500Z
DTEND:20110601T141500Z
UID:TALK31013@talks.cam.ac.uk
CONTACT:Stephen Clark
DESCRIPTION:Last November\, Microsoft launched Xbox Kinect (http://www.xbo
 x.com/kinect)\, a\nrevolution in gaming where your whole body becomes the 
 controller -- you need\nnot hold any device or wear anything special.  Hum
 an pose estimation has long\nbeen a "grand challenge" of computer vision\,
  and Kinect is the first product\nthat meets the speed\, cost\, accuracy\,
  and robustness requirements to take pose\nestimation out of the lab and i
 nto millions of living rooms.\n\nIn this talk we will discuss some of the 
 challenges of pose estimation and the\ntechnology behind Kinect\, detailin
 g our new approach which forms one of the\ncore algorithms inside Kinect: 
 body part recognition.  Deriving from our\nearlier work that uses machine 
 learning to recognize categories of objects in\nphotographs\, body part re
 cognition uses a classifier to produce an\ninterpretation of pixels coming
  from the Kinect depth-sensing camera into\ndifferent parts of the body: h
 ead\, left hand\, right knee\, etc.  Estimating this\npixel-wise classific
 ation is extremely efficient\, as each pixel can be\nprocessed independent
 ly on the GPU.  The classifications can then be pooled\nacross pixels to p
 roduce hypotheses of 3D body joint positions for use by any\nsuitable skel
 etal tracking algorithm.  Our approach has been designed to be\nrobust\, i
 n two ways in particular.  Firstly\, we train the system with a vast\nand 
 highly varied training set of synthetic images to ensure the system works\
 nfor all ages\, body shapes & sizes\, clothing and hair styles.  Secondly\
 , the\nrecognition does not rely on any temporal information\, and this en
 sures that\nthe system can initialize from arbitrary poses and prevents ca
 tastrophic loss\nof track\, enabling extended gameplay for the first time.
   We further discuss\nthe huge promise this technology holds for many othe
 r applications.\n
LOCATION:Lecture Theatre 1\, Computer Laboratory
END:VEVENT
END:VCALENDAR