Resumen de A framework for context-aware sensor fusion
Enrique David Martí Muñoz
Sensor fusion is a mature but very active research field, included in the more general discipline of information fusion. It studies how to combine data coming from different sensors, in such way that the resulting information is better in some sense –more complete, accurate or stable– than any of the original sources used individually. Context is defined as everything that constraints or affects the process of solving a problem, without being part of the problem or the solution itself. Over the last years, the scientific community has shown a remarkable interest in the potential of exploiting this context information for building smarter systems that can make a better use of the available information. Traditional sensor fusion systems are based in fixed processing schemes over a predefined set of sensors, where both the employed algorithms and domain are assumed to remain unchanged over time. Nowadays, affordable mobile and embedded systems have a high sensory, computational and communication capabilities, making them a perfect base for building sensor fusion applications. This fact represents an opportunity to explore fusion system that are bigger and more complex, but pose the challenge of offering optimal performance under changing and unexpected circumstances. This thesis proposes a framework supporting the creation of sensor fusion systems with self-adaptive capabilities, where context information plays a crucial role. These two aspects have never been integrated in a common approach for solving the sensor fusion problem before. The proposal includes a preliminary theoretical analysis of both problem aspects, the design of a generic architecture capable for hosting any type of centralized sensor fusion application, and a description of the process to be followed for applying the architecture in order to solve a sensor fusion problem. The experimental section shows how to apply this thesis’ proposal, step by step, for creating a context-aware sensor fusion system with self-adaptive capabilities. This process is illustrated for two different domains: a maritime/coastal surveillance application, and ground vehicle navigation in urban environment. Obtained results demonstrate the viability and validity of the implemented prototypes, as well as the benefit of including context information to enhance sensor fusion processes.
