Original title (Spanish): Modelo para el Análisis de la Normalidad de Eventos y Conductas en Entornos Monitorizados: Aplicación a la Vídeo Vigilancia
Defense: 28 January 2010
Grade: Excellent Cum Laude
Download PDF (11.7 MB, in Spanish)In recent years, surveillance systems have been used as tools to monitor and maintain security in supervised environments. The main problem with traditional systems is their inability to interpret, on their own, the events that occur and the behaviors of the individuals acting in the environment.
Intelligent security systems have emerged as a solution to this problem. They use techniques from Artificial Intelligence and Computer Vision to interpret what happens in a monitored environment from the information provided by sensors. Most of these systems consist of different analysis modules designed independently, without following a common pattern. They can be seen as a set of pieces that act separately to detect specific types of events in very particular environments, without relating these analyses to one another to obtain global assessments. There is therefore a need for mechanisms to build flexible and scalable monitoring systems that make it easy to add new analysis modules without modifying the existing ones and that, in turn, can be easily combined with the rest to increase the overall analysis capability of the system.
To address this need, this thesis proposes a formal model for building flexible and scalable monitoring systems. The model is based on “normality components”, which are independent and reusable across environments with different characteristics. Each normality component specifies how an object should behave with respect to a given aspect to be monitored, such as the trajectories or the speeds of the objects detected by the system. The model also includes fusion mechanisms that combine the analyses performed by each of the components designed with the proposed model, thus obtaining a global analysis of the current state of the monitored environment.