Publication: Fusión de datos distribuida en redes de sensores visuales utilizando sistemas multi-agente
Loading...
Identifiers
Publication date
2010-05
Defense date
2010-07-16
Authors
Tutors
Journal Title
Journal ISSN
Volume Title
Publisher
Impact
Export
Abstract
Las técnicas de fusión de datos han surgido como una solución al problema de combinar de forma óptima múltiples fuentes de información redundantes; de esta manera, representan una alternativa que mejora (cuando se utiliza correctamente) los sistemas basados en una sola fuente de información. Esta mejora viene proporcionada por la redundancia inherente en los sistemas que utilizan múltiples fuentes de información. En esta tesis, se estudian las técnicas de fusión datos existentes en la actualidad y su aplicación al dominio de las redes de sensores visuales. Una red de sensores visuales consiste en un conjunto de cámaras, distribuidas de forma lógica, espacial o geográfica en un entorno y conectadas por una red de transmisión. La información del entorno, obtenida por las redes de sensores visuales, se encuentra distribuida y debe ser combinada o integrada utilizando técnicas de fusión de datos. Para que la fusión de los datos del entorno, obtenidos por la red de sensores visuales, pueda ser llevada a cabo de una forma óptima, es necesario disponer de una arquitectura software que proporcione la comunicación entre los nodos implicados y la suficiente flexibilidad al proceso de fusión. El paradigma de los sistemas multi-agente, se ha establecido dentro de la comunidad de inteligencia artificial, como el conjunto de tecnologías que pretenden proporcionar los principios para construir sistemas complejos que involucran múltiples agentes y los mecanismos para coordinar los comportamientos individuales de los agentes y las comunicaciones entre ellos. En este contexto, se propone el uso de los sistemas multi-agentes para construir redes de sensores visuales que utilicen técnicas de fusión de datos para mejorar el proceso visual. La primera parte de este documento, se dedica a revisar el estado actual de las técnicas, métodos y algoritmos de fusión de datos clásicos (en su modalidad centralizada y distribuida) y de los sistemas multi-agente, haciendo hincapié en aquellos que han sido aplicados en el dominio de las redes de sensores visuales. Debido a la gran extensión de los trabajos existentes en estas tres áreas (fusión de datos, sistemas multi-agente y visión artificial), esta sección no pretende ser una revisión exhaustiva, sino un marco para establecer el contexto de este trabajo y una forma de resaltar los trabajos más destacados en cada área. La segunda parte, proporciona una descripción de la propuesta realizada dentro de las redes de sensores visuales, que consiste en el diseño e implementación de una arquitectura multi-agente que realiza fusión de datos dentro de una red de sensores distribuidos. Entre todos los tipos de arquitecturas multi-agente se ha optado por el uso de una arquitectura híbrida (de tipo Belief-Desire-Intention), ya que representa de forma explícita el conocimiento frente a otras arquitecturas puramente reactivas. La principal aportación de la arquitectura propuesta es el uso de la tecnología de sistemas multi-agente para modelar las redes de sensores visuales y realizar la fusión de los datos. En concreto se propone utilizar una fusión activa, la cual aporta la ventaja de que los agentes involucrados en el proceso de fusión se auto-corrijan; además, se establecen las bases de protocolos para la formación de coaliciones dinámicas entre los agentes involucrados en el proceso. El sistema desarrollado se ha probado de forma experimental, por un lado utilizando información obtenida por varias cámaras desplegadas dentro del laboratorio de investigación y por otro lado utilizando conjuntos de datos disponibles dentro de la comunidad de visión artificial para la investigación.------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Data fusion techniques have been proposed as a solution to the problem of combining, in an optimal way, multiple redundant sources. Therefore, they are an alternative which, when are correctly employed, improve traditional systems that are based on only one source. This improvement is provided by the inherent redundancy of the systems that use multiple sources. In this work, data fusion state-of-art techniques are explored and their applications to the visual sensor domain are analyzed. A visual sensor network consists of a network of cameras, logically, spatially or geographically distributed in an environment and connected by a transmission network. Environmental information, obtained by the visual sensor network is distributed and should be combined or integrated using data fusion techniques. With the aim of fuse in an optimal way, the monitoring data, obtained by the visual sensor network it is required to have a software architecture which provides the communication between the nodes and the necessary exibility to the data fusion process. The multi-agent paradigm has been established by the artificial intelligence community, as the technology which provides the principles for building complex systems that involve multiple agents and which provide the mechanism to coordinate individual agent's behaviors and their communications. In this scenario, the application of multi-agent systems with the aim of building visual sensor networks that use data fusion techniques to improve the visual process, is proposed. The first part of this document, reviews the classic data fusion state-of-art techniques, methods and algorithms (in their centralized and distributed versions) and the multi-agent systems, providing special emphasis in those systems applied to visual sensor networks. Due to the enormous amount of works in these three areas (data fusion, multi-agent systems and computer vision), this section does not aim to be an exhaustive reviewing, instead aims to be a framework for this work and a way to highlight the most relevant works. The second part of the document, provides a detailed description of the developed proposed in the area of visual sensor networks. The work consists in the design and implementation of a hybrid multi-agent architecture (Belief-Desire-Intention model) which performs data fusion in a visual sensor network. The main contribution of this proposed architecture is the use of multi-agent systems for modeling visual sensor networks and fuse data. In particular, is proposed an active data fusion, which provides the advantage of auto-correct the state of the agents involved in the data fusion process. Moreover, the basis of protocols for dynamic coalition formation between the agents involved in the data fusion process, are proposed. The developed system has been experimental tested, on one hand using the information obtained from several cameras deployed in our indoor laboratory and on other hand using available datasets for the computer vision community.
Data fusion techniques have been proposed as a solution to the problem of combining, in an optimal way, multiple redundant sources. Therefore, they are an alternative which, when are correctly employed, improve traditional systems that are based on only one source. This improvement is provided by the inherent redundancy of the systems that use multiple sources. In this work, data fusion state-of-art techniques are explored and their applications to the visual sensor domain are analyzed. A visual sensor network consists of a network of cameras, logically, spatially or geographically distributed in an environment and connected by a transmission network. Environmental information, obtained by the visual sensor network is distributed and should be combined or integrated using data fusion techniques. With the aim of fuse in an optimal way, the monitoring data, obtained by the visual sensor network it is required to have a software architecture which provides the communication between the nodes and the necessary exibility to the data fusion process. The multi-agent paradigm has been established by the artificial intelligence community, as the technology which provides the principles for building complex systems that involve multiple agents and which provide the mechanism to coordinate individual agent's behaviors and their communications. In this scenario, the application of multi-agent systems with the aim of building visual sensor networks that use data fusion techniques to improve the visual process, is proposed. The first part of this document, reviews the classic data fusion state-of-art techniques, methods and algorithms (in their centralized and distributed versions) and the multi-agent systems, providing special emphasis in those systems applied to visual sensor networks. Due to the enormous amount of works in these three areas (data fusion, multi-agent systems and computer vision), this section does not aim to be an exhaustive reviewing, instead aims to be a framework for this work and a way to highlight the most relevant works. The second part of the document, provides a detailed description of the developed proposed in the area of visual sensor networks. The work consists in the design and implementation of a hybrid multi-agent architecture (Belief-Desire-Intention model) which performs data fusion in a visual sensor network. The main contribution of this proposed architecture is the use of multi-agent systems for modeling visual sensor networks and fuse data. In particular, is proposed an active data fusion, which provides the advantage of auto-correct the state of the agents involved in the data fusion process. Moreover, the basis of protocols for dynamic coalition formation between the agents involved in the data fusion process, are proposed. The developed system has been experimental tested, on one hand using the information obtained from several cameras deployed in our indoor laboratory and on other hand using available datasets for the computer vision community.
Description
Keywords
Visión artificial, Redes de sensores visuales, Inteligencia artificial, Fusión de datos, Sistemas multiagente