Bitte benutzen Sie diese Kennung, um auf die Ressource zu verweisen: http://dx.doi.org/10.18419/opus-9276
Autor(en): Chen, Na
Titel: Compressive sensing-based data uploading in time-driven public sensing applications
Erscheinungsdatum: 2017
Dokumentart: Abschlussarbeit (Diplom)
Seiten: vi, 66
URI: http://nbn-resolving.de/urn:nbn:de:bsz:93-opus-ds-92935
http://elib.uni-stuttgart.de/handle/11682/9293
http://dx.doi.org/10.18419/opus-9276
Zusammenfassung: Over the last few years, the technology of mobile phones greatly got increased. People gain and upload more and more information through their mobile phones in an easy way. Accordingly, a new sensing technology emerges, referred to as public sensing (PS). The core idea behind PS is to exploit the crowdedness of smart mobile devices to opportunistically provide real-time sensor data considering spatial and environmental dimensions. Recently, PS has been applied in many different application scenarios, such as environmental monitoring, traffic analysis, and indoor mapping. However, PS applications face several challenges. One of the most prominent challenges is the users acceptance to participate in the PS applications. In order to convince users to participate in the PS applications, several incentives mechanisms have been developed. However, the main two requirements - which should be met by any PS application - are the users privacy and the energy costs of running the PS application. In fact, there exist several energy consumers in PS applications. For example, many PS applications require the mobile devices to fix their position and frequently send this position data to the PS server. Similarly, the mobile devices waste energy when they receive sensing queries outside the sensing areas. However, the most energy-expensive task is to frequently acquire and send data to the PS server. In this thesis, we tackle the problem of energy consumption in a special category of PS applications in which the participating mobile devices are periodically queried for sensor data, such as acceleration and images. To reduce the energy overhead of uploading lots of information, we exploit the fact that processing approximately one thousand instructions consumes energy equal to that of transmitting one bit of information. Accordingly, we exploit data compression to reduce the number of bit that will be transmitted from the participating mobile devices to the PS server. Although, he technical literature has many compression methods, such as derivative-based prediction, Cosine transform, Wavelet transform; we designed a framework based on the compressive sensing (CS) theory. In the last decade, CS has been proven as a promising candidate for compressing N-dimensional data. Moreover, it shows satisfactory results when used for inferring missing data. Accordingly, we exploit CS to compress 1D data (e.g. acceleration, gravity) and 2D data (e.g. images). To efficiently utilize the CS method on resources-taxed devices such as the smart mobile devices, we start with identifying the most lightweight measurements matrices which will be implemented on the mobile devices. We examine several matrices, such as the random measurement matrix, the random Gaussian matrix, and the Toeplitz matrix. Our analysis mainly bases on the recovery accuracy and the dissipated energy from the mobile device's battery. Additionally, we perform a comparative study with other compressors, including the cosine transform and the lossless ZIP compressor. To further confirm that CS has a high recovery accuracy, we implemented an activity recognition algorithm at the server side. To this end, we exploit the dynamic time warping (DTW) algorithm as a pattern matching tool between a set of stored patterns and the recovered data. Several experiments have been performed which show the high accuracy of both CS and DTW to recover several activities such as walking, running, and jogging. In terms of energy, CS significantly reduce the battery consumption relative to the other baseline compressors. Finally, we prove the possibility of exploiting the CS-based compression method for manipulating 1D data as well as 2D data, i.e. images. The main challenge is to perform image encoding on the mobile devices, despite the complex matrix operations between the image pixels and the sensing matrices. To overcome this problem, we divide the image into a number of cells and subsequently, we perform the encoding process on each cell individually. Accordingly, the compression process is iteratively achieved. The evaluation results show promising results for 2D compression-based on the CS theory in terms of the saved energy consumption and the recovery accuracy.
Enthalten in den Sammlungen:05 Fakultät Informatik, Elektrotechnik und Informationstechnik

Dateien zu dieser Ressource:
Datei Beschreibung GrößeFormat 
chennaDA.pdf2,99 MBAdobe PDFÖffnen/Anzeigen


Alle Ressourcen in diesem Repositorium sind urheberrechtlich geschützt.