Fuzzy-based sensor search on the web of things
A steadily increasing number of sensors worn by people (e.g., contained in mobile phones), embedded into the environment (e.g., sensor networks), and into objects (e.g., smart objects and appliances) are being connected to the Internet. This trend is leading to the formation of an Internet of Things (IoT) that is expected to interconnect billions of devices by2020.
By publishing the resulting sensor data streams in theWeb, novel real-world applications can be created by mashing up sensors and actuators with services and data available on the Web, leading to a Web of Things.
As in the traditional web, search will be a key service also in the WoT to enable users to and sensors with certain properties. Existing directories of online sensors such as Pachube, GSN, or Microsoft SensorMap support search for sensors based on textual metadata that describes the sensors (e.g.,type and location of a sensor, measurement unit, object to which the sensor is attached) and which is manually entered by the person deploying the sensor. Other users can then search for sensors with certain metadata by entering appropriate keywords.
Unfortunately, this approach does not work well in practice, as humans make mistakes when entering metadata, different users use different terms to describe the same concept, or important metadata is not entered at all.
There are several approaches to address this problem.Firstly, some metadata of a sensor such as sensor type canbe stored on the sensor during production using so-called Transducer Electronic Data Sheets (TEDS) as defined by IEEE.
However, most of the relevant metadata of a sensor depends on the deployment and use of the sensor (e.g., logical location of the sensor, object to which the sensor is attached) and cannot be provided by the producer of a sensor. Secondly, there are efforts to provide a standardized vocabulary to describe sensors such as SensorML or the Semantic Sensor Network Ontology (SSN).
Unfortunately, these ontologies and their use are rather complex and end users likely won’t be able to provide correct descriptions of sensors and their deployment context without help from experts.
In this paper we adopt a search-by-example approach to sensors, i.e., a user provides a sensor, respectively a fraction of its past output as an example, and requests sensors that produced similar output in the past. We call this sensor similarity search service.In this approach searches are based on the past output of a sensor, sensors with similar output are found.
We designed an efficient approach to compute a similarity score for a pair of sensors. All sensors compute fuzzy sets that represent their past output using only few tens of bytes. These fuzzy sets are indexed in a data base. Given the output of another sensor, similarity scores are computed for each indexed sensor, sensors are ranked by this score and returned to the user. Using sensor data from three realworld deployments, we could show the high accuracy of our approach.
As proof of concept, we built a working prototype to demonstrate the functionality of our service and to support experimentation in realistic environments. Building upon those results, we will explore scalable search algorithms to support searching among large numbers of sensors in the Web of Things. Eventually, we also plan to perform a user study to assess the accuracy of our approach.
This service could be used for different purposes. First, it could be used to find places with similar physical properties. For example, if one wants to find places that have similar climatic conditions as a known place A, one could pick a temperature sensor that is known to be at place A, and then search for other temperature sensors with similar output.
Secondly, it could be used to assist users with the formulation of a metadata description of a newly deployed sensor. A user would deploy a new sensor and then search for sensors with similar output, fetch the metadata of the found sensors, and reuse appropriate fractions of the metadata for the new sensor.
To read this external content in full, download the complete paper from the author archives online at Luebeck Unversity