Computing Techniques for Classifying Community Sensor Network Data

Abstract

Community sensing has emerged as an attractive research topic in the past decade. It involves public participation through an interactive sensor network via privately owned sensors embedded on devices, such as cellular phones. It enables the community to collect data, process and distribute resulted information that is useful in a variety of monitoring tasks. Important application areas of community sensing include infrastructural, health and environmental sensing. Tremendous increase in the number of smartphone users has been witnessed in the recent years. Smartphones owned by people are equipped with many low-cost sensors like tri-axial accelerometer, microphone, and GPS. Moreover, these phones have processing and communication capabilities and large storage. With such features, it is viable to use smartphones in different monitoring activities. In addition, to alleviate the necessity of costly traditional monitoring equipment, the novelty of such a paradigm is community sensing. It empowers the user community to deploy sensing applications at wider scale and collect data from heterogeneous sensors owned by a large number of people. This is a paradigm shift from the standard engineering practice where trained experts use reliable high precision equipment for monitoring. While the new paradigm promises to deliver a large amount of information with very little cost, reliability of the data obtained from heterogeneous sources must be critically examined. Sensor-enabled smartphones have been potentially used in different monitoring activities reliably and with high precision. Thereby, smartphones have the potential in precise motion sensing. Over the last few years, community sensing have been developed in domains ranging from information sharing to collaborative monitoring. Smartphone is identified as a powerful and widely used community sensing platforms in a number of such applications. This thesis mainly focuses on the development of a community sensing system through smartphones. It explains the proposed client-server framework involving development of client side applications to acquire data, server side algorithms to process data and sharing results to client devices. The proposed framework is evaluated as applications to infrastructure monitoring and environmental monitoring especially into continuous road surface roughness monitoring and noise level mapping respectively. The proposed framework has been used to sense and transmit the data over the web using inbuilt sensors and communication capability of smartphones. Firstly in this thesis, SenseMe- a monitoring xii application is presented. It captures the vibration data experienced by the driver while traveling over a road surface automatically by using the inbuilt sensors of smartphones. Signal processing using low pass filter is performed in order to remove high frequency components probably caused by the engine noise. The recorded data is processed further before communicating to the server. For this, an absolute average of vertical accelerations is calculated. The server assimilates all the data imported from the community of phones and fuzzy classes are produced to classify the road surface conditions. At the client devices (smartphone) the compressed codes representing severity of roughness are used to assign a particular colour to different road segments. The client device visualizes the road surface roughness on Google® maps. Secondly, this thesis presents DbSense - an environmental noise (from different locations and road traffic noise) monitoring application that can detect the noise patterns prevalent in the environment at different timings of the day. This thesis describes the developed client side application and server side algorithms. A noise level metric is calculated and used as a classifier based on fuzzy logic at the server side. The fuzzy logic based classifier has been used to model impreciseness. Compressed fuzzy classes are transferred to the client phone which in turn shows coloured Google® maps where each colour represents the level of noise impact on human beings. The popularity of distribution channels for the smartphone applications helps to distribute the developed framework to the users for prevailing conditions of the road network and environmental noise levels.