An optimized classification method for human behavioral patterns recognition

The paper proposes an innovative supervised learning method for human behavioral recognition in which the behavioral patterns are classified according to the classes importance. A detector classifier is trained to recognize the human behavioral patterns belonging to the most important class. The optimization is performed by fixing the classifier operating point to provide the appropriate performance trade-off between a target and non-target behavioral class. The applications include public and personal safety, for instance elderly people home tele-assistance and other systems in which the early detection of abnormal behaviors is required.