Sensor Abstractions for Opportunistic Activity and Context Recognition Systems

Pervasive environments are inherently characterized to draw from sensor infrastructures in order to become situation aware. Very recent technological evolutions of sensor hardware (e.g. for geoposition, acceleration, orientation, noise, light, humidity, chemical properties, etc.) have fertilized an explosive growth of sensor infrastructures, introducing whole new challenges for sensor software architectures like heterogeneity, redundancy and replaceability, fault tolerance, mobility, massive deployment, but most of all frequency of change and spontaneous availability. In this paper we address the very fundamental issue of exploiting spontaneous sensor configurations by introducing mechanisms for sensor self-description, goal-oriented sensing missions and dynamic sensor ensemble management. The concept of "sensor abstractions" is introduced, making the use of physical as well as immaterial sensors transparent from any technical sensor properties. An opportunistic sensor software architecture has been implemented, reversing former architecture principles (e.g. fusion of all available sensors) into a spontaneous, selective, utility driven involvement of sensors based on their sensing mission contribution potentials. The framework is implemented in OSGi, and demonstrated for activity recognition missions.