Abstract: With
the launch of Microsoft Kinect sensors for Xbox 360, depth cameras are becoming
affordable for the vision and robotics communities because the mass game market
drives the cost down. This could present a revolution in our research. Previously
very difficult vision tasks with video cameras become easier, such as
foreground-background separation. However, there still exist many challenges.
In this talk, I will present some research projects conducted at Microsoft
Research, related to human activity understanding with
depth sensors. In particular, I will describe a person tracking algorithm based
on decision forests and a human action recognition algorithm based on action graph.