In recent years there has been an increased interest in the modelling and recognition of human activities involving highly structured and semantically rich behaviour such as dance, aerobics, and sign language. A novel approach is presented for automatically acquiring stochastic models of the high-level structure of an activity without the assumption of any prior knowledge. The process involves temporal segmentation into plausible atomic behaviour components and the use of variable length Markov models for the efficient representation of behaviours. Experimental results are presented which demonstrate the generation of realistic sample behaviours and evaluate the performance of models for long-term temporal prediction.

For more details see:

A. Galata, N. Johnson, D. Hogg, Int. Journal of Computer Vision and Image Understanding (CVIU), Vol. 81, No. 3, pp. 398-413, March 2001 ( .pdf).
Also available as a SCS technical report [gzip postscript] [pdf]

Figure 1 illustrates stochastic synthesis of sample behaviour using the learnt behaviour model. An entirely hypothetical exercise routine sequence has been generated and is used to animate a virtual humanoid using the VRML modelling language.

Figure 2 illustrates prediction of future behaviour at a selected time instant. The learnt behaviour model is initially used in a recognition mode in order to locate the appropriate model state given the observed behaviour up to that time instant. Having located the current model state, prediction of future behaviour is achieved by using the model either as a stochastic or a maximum likelihood behaviour generator.