Research

Research Interests

My research interests focus on pattern recognition, human computer interaction, video surveillance and analysis, multimodal interface, multimedia analysis, computer vision, and stochastic machine learning.

Currently my major efforts are put on assistive technology on healthcare and quality of life, particularly long-term behavior monitoring through multiple sensors, personalized activity and mood analysis, social interaction analysis in skilled facility, fundamental IT for care-media data indexing, retrieval and sharing and privacy protection in video.

Updates: 14,352 hours high-quality audio/video data are recorded from the dementia unit of the Longwood nursing home. (Caremedia Oct. 31 2005)

Updates: CMU has been granted the new NSF engineering research center: Quality of Life Technology whose mission is to transform lives in a large and growing segment of the population - people with reduced functional capabilities due to aging or disability. (July 2006)

Updates: ICME special session on Multimedia Technologies for Healthcare was hold in Toronto July 2006. The session attracted about 25 people. Some people believe that healthcare is a golden application for the multimedia community. Thanks for all the authors and audients.

Updates: ICARCV special session on Assistive Technologies for Healthcare and Quality of Life will be hold in Singapore on Dec. 6. I apologize for not able to chair it due to my wife’s delivery. Thanks for Prof. Eric Sung to chair this session.

	Elderly in a public nursing facility is separated from his/her family. Are they lonely? Are they healthy? How many social interaction an elderly has done during last week? What kinds of interactions? Through video and audio monitoring and multimodal based event detection techniques, we can answer those questions.
	Privacy protection is one of the key technologies in video monitoring in healthcare surveillance. This research identify and track a specified person in video and reserve his/her activities without exposing the identity.
	Tracking multiple people using online region confidence learning. We improved object tracking against occlusions and complex background (the background contains similar color as objects) by tracking only outstanding regions instead of the whole object.
	Unsupervised human activity analysis from video. The mission is to understand complex individual activities in public area of a nursing home. Activities are modeled automatically by using graphical models for recognition and categorization.
	Multiple cameras provide compensate information for tracking an object in 3-D world. A particle filtering based method has been developed to integrate tracking clues from multiple cameras and recover the 3-D trajectory. We use this method to measure the quality of walking such as speed and distance.
	Face detection and tracking in video. Both foreword and backward tracking are applied once a face is detected with a high confidence.
	Gender recognition from face by boosting MLPs in the Haar feature space. It is a hard problem. The recognition performance has a big variance on various data sets. So dress whatever you like, no one can tell your gender only from your face.
	Structuring surveillance video using event networks.
	Text detection and recognition from images and video sequences. This is the major contribute of my Ph.D thesis. I developed a system that can detect and recognize text in images and videos with complex backgrounds.
	I got involved in this project to build a mobile phone that can aware the environment. We put sensors on a phone to tell the lighting, noise and movement conditions and to automatically switch service modes.
	Recognition and Reasoning in an Awareness Support System for Generation of Storyboard-like Views of Recent Activity. This system automatically detects and summarizes remote working context and generate a storyboard-like views of recent activity for collaboration partners.