The objective of first-person computer vision is to provide semantic understanding of videos taken from an egocentric perspective (i.e., from the subject's own point-of-view). These include videos captured from a wearable computer (e.g., 'glasses'), a vehicle 'black box' camera, and/or an autonomous robot. In contrast to previous recognition approaches for 3rd person viewpoint, first-person computer vision approaches aim to recognize ego-actions of the subject (e.g., a human wearing a camera) and its interactions with other appearing objects and persons. In this talk, we first discuss general computer vision approaches for images and videos, while focusing on semantic-level video understanding. Next, we introduce recent approaches designed to recognize activities from first-person videos. Particularly, we focus on (1) an approach to detect driving events from a moving vehicle, and (2) a new methodology to recognize physical interactions between a person and the subject, such as a person punching the camera or a person throwing objects to the camera.

Michael Sahngwon Ryoo (유상원) is a research staff of NASA's Jet Propulsion Laboratory (JPL), California Institute of Technology. His research interests include semantic recognition of human activities from videos, intelligent interactions between humans and machines, and smart context-aware environments. Dr. Ryoo received the B.S. degree in computer science from Korea Advanced Institute of Science and Technology (KAIST) in 2004, and the M.S. and Ph.D. degrees in computer engineering from the University of Texas at Austin, in 2006 and 2008, respectively. He was the lead organizer of the first ICPR contest on human activity recognition, SDHA 2010. He also has been providing tutorials on human activity recognition at major Computer Vision/AI conferences including CVPR 2011 and AVSS 2012, has authored a number of conference and journal papers, and is the corresponding author of the activity recognition survey paper published by ACM Computing Surveys in 2011.