Identification of human behavior patterns based on the gsp algorithm

Abstract

The analysis of the algorithms dedicated to the identification of sequential patterns described in the literature, shows that not all are suitable for the type of scenarios with which video surveillance often deals, in particular for the recognition of behavior patterns suspects to classify human behavior as normal or suspicious, it is necessary to analyze all the monitored actions. This is the reason why in this study the main proposal is a modification of the Generalized Sequential Patterns, which we call Generalized Sequential Patterns+memory, which mainly incorporates a module that manages the number of repetitions and combinations of actions (and not only of the sequence) that make up patterns. For the experimentation scenes of theft in supermarkets have been recorded with labels representing states that we assume can be recognized by an artificial vision system. The results obtained were analyzed and their performance was evaluated by comparing them with the results obtained from the GSP application.