Wenhui Liao

IJCAI Conference 2007 Conference Paper

• Wenhui Liao
• Yan Tong
• Zhiwei Zhu
• Qiang Ji

The paper describes a simple but effective framework for visual object tracking in video sequences. The main contribution of this work lies in the introduction of a case-based reasoning (CBR) method to maintain an accurate target model automatically and efficiently under significant appearance changes without drifting away. Specifically, an automatic case-base maintenance algorithm is proposed to dynamically update the case base, manage the case base to be competent and representative, and to maintain the case base in a reasonable size for real-time performance. Furthermore, the method can provide an accurate confidence measurement for each tracked object so that the tracking failures can be identified in time. Under the framework, a real-time face tracker is built to track human faces robustly under various face orientations, significant facial expressions, and illumination changes.

AAAI Conference 2006 Conference Paper

Efficient Active Fusion for Decision-Making via VOI Approximation

• Wenhui Liao

Active fusion is a process that purposively selects the most informative information from multiple sources as well as combines these information for achieving a reliable result efficiently. This paper presents a general mathematical framework based on Influence Diagrams (IDs) for active fusion and timely decision making. Within this framework, an approximation algorithm is proposed to efficiently compute nonmyopic value-of-information (VOI) for multiple sensory actions. Meanwhile a sensor selection algorithm is proposed to choose optimal sensory action sets efficiently. Both the experiments with synthetic data and real data from a real-world application demonstrate that the proposed framework together with the algorithms are well suited to applications where the decision must be made efficiently and timely from dynamically available information of diverse and disparate sources.

AAAI Conference 2005 Conference Paper

A Decision Theoretic Model for Stress Recognition and User Assistance

• Wenhui Liao
• Zhiwei Zhu

We present a general unified probabilistic decisiontheoretic model based on Influence Diagrams for simultaneously modeling both user stress recognition and user assistance. Stress recognition is achieved through dynamic probabilistic inference from the available sensory data from multiple-modality sources. User assistance is automatically achieved by balancing the benefits of improving user performance and the costs of performing user assistance. In addition, a non-invasive real-time system is built to validate the proposed framework. Utilizing the evidences from four modalities (physical appearance features, physiological measures, user performance and behavioral data), the system can successfully recognize human stress and provide timely and appropriate assistance in a task-specific environment.