Body posture recognition by means of a genetic fuzzy finite state machine

Body posture recognition is a very important issue as a basis for the detection of user´s behavior. In this paper, we propose the use of a genetic fuzzy finite state machine for this real-world application. Fuzzy finite state machines (FFSMs) are an extension of classical finite state machines where the states and inputs are defined and calculated by means of a fuzzy inference system, allowing them to handle imprecise and uncertain data. Since the definition of the knowledge base of the fuzzy inference system is a complex task for experts, we use an automatic method for learning this component based on the hybridization of FFSMs and genetic algorithms (GAs). This genetic fuzzy system learns automatically the fuzzy rules and membership functions of the FFSM devoted to body posture recognition while an expert defines the possible states and allowed transitions. We aim to obtain a specific model (FFSM) with the capability of generalizing well under different subject´s situations. The obtained model must become an accurate and human friendly linguistic description of this phenomenon, with the capability of identifying the posture of the user. A complete experimentation is developed to test the performance of the new proposal, comprising a detailed analysis of results which shows the advantages of our proposal in comparison with another classical technique.