Here's looking at you! Describing Spatio-Temporal Human Events

点击翻译按钮获取中文摘要

Here's looking at you! Describing Spatio-Temporal Events What does a person concentrate on when looking at another person? The answer is the face and the body. My research addresses both of these activities as observed in videos and based on automating the interpretation of human (i) behavior and (ii) emotions as indicated by their facial expressions. This process is referred to as automated "visual surveillance". It is concentrated on creating real-time systems for monitoring spontaneous (not posed or acted) human activities. The long-term goal of this research is to be able to simultaneously interpret both (i) and (ii). Neither real-time nor spontaneity are presently being emphasized in the research literature on visual surveillance. Furthermore, the literature shows that the performance of methods trained on acted or posed databases deteriorates by about 50% when applied to unconstrained videos. The practical objective of this research is based on the fact that currently there are thousands and thousands of indoor and outdoor video cameras in our living spaces, but no way to automatically analyze the obtained unconstrained data. The challenge is to significantly improve the available analytical and algorithmic tools for automated visual surveillance.