Abstract

In this talk, I am going to give a brief overview of our research projects at the University of Tokyo, focusing on two research topics. The first part of my talk will cover vision-based sensing and understanding of human activities. Sensing various kinds of human activities in real-time is a key component of human-machine interactions. We have been working on vision-based techniques for sensing human body motions at different levels of details: 3D tracking of people in complex scenes, estimation of facial pose and deformation, and gaze estimation from low resolution images. I will explain each of the techniques as well as some results from our on-going attempts to apply our techniques for human activity analysis. The second part of this talk will cover our recent project on photometric compensation of projector-camera systems to compensate irregularities in the output of digital projectors that occur when they are used in non-ideal situations, such as those with varying surface reflectance and ambient light.