TRACE, a graphical tool for the acquisition and detection of signal patterns

In this work we present TRACE, a tool for projecting the description of a signal pattern, obtained directly from an expert in the application domain, onto a computational model, and automatically identifying the pattern over a recording of the temporal evolution of a physical system. Knowledge acquisition and the visualization of results from detection are based on visual metaphors that enable experts to simply and intuitively describe patterns and identify them over a signal recording. TRACE has been used in the domains of mobile robotics and patient supervision in the intensive care units, with highly satisfactory results. The multivariable fuzzy temporal profile model (MFTP), which supports the tool, describes a signal pattern as a network of fuzzy constraints between a set of points from the evolution of the system which are especially relevant for experts. The constraint network formalism permits a single pattern to be described as a set of increments, temporal durations and slopes between the relevant points of the system’s evolution. Thanks to the use of fuzzy logic, the vagueness and uncertainty that are characteristic of human knowledge can be captured by means of the representation of imprecise values for the aforementioned parameters.

keywords: Structural pattern recognition, Knowledge acquisition, Knowledge representation, Fuzzy constraint satisfaction problems