Visual Models of Interaction
Neil J. Johnson, Aphrodite Galata, D. Hoggs · Research Explorer (The University of Manchester) · 1998
procedural variations between and within different types of human interaction. We describe recent work on interaction modelling in which the range of possible things that can happen is learnt automatically through passive observation of video sequences depicting typical interactions. The basis for the approach is the construction of probabilistic spatio-temporal models from training data extracted from the video sequences. Results will be presented for two kinds of application of this work. The first is concerned with interactions between a pair of individuals with application to human-computer interaction. The second is dealing with interactions between people and motor vehicles, intended for wide-area surveillance.