Formal programming language for Modeling Activities using Visual Primitives

摘要

Surveillance systems through vision techniques represent a non-intrusive alternative to analyze the dynamics of moving objects. However, it seems that in order to define a criterion to identify activities in several scenarios, it is difficult because of the number of variables involved. A particular case is the identification and inference of the activities that are observed in video sequences. For this purpose, literature presents a variety of approaches. Each approach is based on theoretical concepts that define a formal model to decide over evidence. However, many of these approaches tend to be adapted through practical implementations to computable models. Consequently, because of the difficulty of modelling scenarios to infer its activity, this article proposes a formal programming language to model the dynamics of objects in a scenario. This language is based on segmenting vision field of the camera in a set of states and a set of operators that uses information of how is presented the states movement, and the relation between them by PL0 (Propositional Logic). As a result, it has a highly expressive grammar that allows to model as a deterministic system the transition of the movement by the states. The results, showing that to model and identify activities is to develop programs in this language through a deterministic formation by means of the proposed operators. The results show how it is possible to analyze activities in open and closed scenarios, proving that it is a simple language for modelling activities through the primitives.