About the Event
Understanding human motions and activities in images and videos is an important problem in many application domains, including surveillance, robotics, video indexing, and sports analysis. Although much progress has been made in classifying single person's activities in simple videos, little efforts have been made toward the interpretation of behaviors of multiple people in natural videos. In this thesis, I will present my research endeavor toward understanding of behaviors of multiple people in natural images and videos. I identify four major challenges in this problem: i) identifying individual properties of people in videos, ii) modeling and recognizing the behavior of multiple people, iii) understanding human activities in multiple levels of resolutions and iv) learning characteristic patterns of interactions between people or people and surrounding environment. I discuss how we solve these challenging problems using various computer vision and machine learning technologies. I conclude with final remarks, observations, and possible future research directions.