Bayesian classification of a human custom based on stochastic context-free grammar.

This paper describes a new approach for gesture recognition based on stochastic context-free grammar (SCFG). We focus on the actions in the Japanese tea ceremony, where structure of the action can be described by context-free grammar. Our aim is to recognize the actions in the tea ceremony. Existing SCFG approach consists of generating symbol stream, parsing, and recognition. The symbol stream often includes uncertainty. Therefore, the parsing process needs to recover from errors in the entry process. This paper proposes an error-free approach to segment an action into a set of events. This approach, based on acceleration of body motion, can produce events corresponding to a terminal symbol with little error. After translating the events into a set of symbol streams, parsing the set of the streams leaves a small number to be derived. The number of potential streams is not too large since each event is correctly segmented. Among the remaining streams, our SCFG can identify the action with a maximum posterior probability. Giving an SCFG rule the multiple probabilities, one SCFG can recognize multiple actions. © 2007 Wiley Periodicals, Inc. Syst Comp Jpn, 38(9): 52–62, 2007; Published online in Wiley InterScience (<URL>www.interscience.wiley.com</URL>). DOI 10.1002/scj.20531 [ABSTRACT FROM AUTHOR]