Unsupervised Learning of GDL Classifier

GDL (Gesture Description Language) is a pattern recognition method that enables syntactic description and real time recognition of static body poses and movement sequences. The syntax of context free GDL script (GDLs) language is intuitive and easy to learn for new user, however so far GDLs rules had to be implemented without feedback of machine learning methods. In this paper we present proposition and initial evaluation of unsupervised method of GDL classifier learning that enables automatic generation of GDLs descriptions using specified features and sample movements recordings. New automatically generated GDLs are well understandable the same as manually defined descriptions. This property enables easy interpretation of obtained training results in contrast to the results from others popular pattern recognition methods.