FROM IMAGES TO HUMAN RECOGNITION
Dr Shaogang Gong, Department of Computer Science, Queen Mary and Westfield College, University of London

Abstract: Advances in computer vision over the last decade have made considerable progress in understanding images of moving objects in dynamic scenes. One of the most active and fastest growing fields of this research has been in the recognition and interpretation of human subjects and their activities captured on live video cameras.

In this talk, I will describe some recent trends and progress in the effort to develop robust and tractable models for the recognition and interpretation of human faces, body gestures and behaviours. In particular, I will describe robust and real-time methods for the detection, segmentation, tracking and fusion of human subjects in a wide field of view, through a sequence of active views and multiple static views; learning multiview-based human face models without 3D reconstruction for real-time recognition and video-based communication; the representation and learning of human body and gesture models from examples, and the recognition of human behaviour and activities using density models of temporal structures in multivariant feature space. I will highlight in these models the exploitation of ideas from psychophysical studies on multiview-based representation, VR techniques for labelled data acquisition for learning, control theories for temporal prediction and tracking, stochastic processes such as hidden Markov models and conditional density propagation for modelling temporal structures, and Bayesian framework and support vector machines for learning decision boundaries in recognition. I will illustrate the use of such models through potential applications in visual surveillance, visually mediated human-machine interaction, access control in electronic banking and information retrieval, video-conferencing and virtual film production.

This seminar was held at the Department of Computer Science, Royal Holloway, University of London on 21 June 1999.

back