The evolution and proliferation of mobile sensing platforms such as mobile phones, enables services that analyze and adjust to the state of the real world. Billions of mobile phones around the globe seamlessly integrated into our life enable the vision of public sensing, i.e., monitoring and detecting a variety of physical phenomena by continuously collecting an abundance of sensor data. To exploit the enormous sensing capabilities, sensing may not interfere with normal operation of mobile phones. Furthermore, since mobile phones are battery powered, sensor data collection needs to be energy efficient and, thus, limited to the required data.

Therefore, this dissertation presents a public sensing approach that opportunistically collects sensor data. To specify and assess the quality of the data, spatial and temporal coverage metrics are devised. Virtual sensors are introduced as a data-centric abstraction to cope with the dynamic availability of mobile phones by decoupling applications from physical devices. More precisely, this dissertation addresses three major classes of virtual sensors that allow applications to request sensor data intuitively based on spatial, temporal, and quality requirements.

For each of the three classes of virtual sensors, this dissertation presents centralized and distributed algorithms for the selection and coordination of mobile phones according to the sensing requirements, while minimizing the energy consumption. In order to cope with the varying availability of physical sensors, the dissertation shows how to monitor the progress of sensing and how to adapt sensing to changes of movement. Moreover, this dissertation shows how to adapt the coordination mechanisms to the density of participating devices.

As a basis for the coordination algorithms, this dissertation presents basic group communication mechanism. These mechanisms allow to address specific devices based on their symbolic location. In essence, a routing structure that mimics the location model is created and proactively maintained. With a symbolic location model that matches the structure of the virtual sensors, this communication abstraction allows to easily identify and address nodes relevant for the coordinated data acquisition of the virtual sensors.