The demand of Indoor Location Based Services LBS has increased over the past years as smart phone market expands. As a result, there's a growing interest in developing efficient and reliable indoor positioning systems for mobile devices. Wi-Fi signal strength fingerprint-based approaches attract more and more attention due to the wide deployment of Wi-Fi access points. Indoor positioning problem using Wi-Fi signal fingerprints can be viewed as a machine learning task to be solved mathematically. This thesis proposes an efficient and reliable Wi-Fi real-time indoor positioning system using machine learning algorithms. The proposed positioning system, together with a location server equipped with the same algorithms, are tested and evaluated in several indoor scenarios. Simulation and testing results show that the proposed system is a feasible LBS solution.

The demand of Indoor Location Based Services LBS has increased over the past years as smart phone market expands. As a result, there's a growing interest in developing efficient and reliable indoor positioning systems for mobile devices. Wi-Fi signal strength fingerprint-based approaches attract more and more attention due to the wide deployment of Wi-Fi access points. Indoor positioning problem using Wi-Fi signal fingerprints can be viewed as a machine learning task to be solved mathematically. This thesis proposes an efficient and reliable Wi-Fi real-time indoor positioning system using machine learning algorithms. The proposed positioning system, together with a location server equipped with the same algorithms, are tested and evaluated in several indoor scenarios. Simulation and testing results show that the proposed system is a feasible LBS solution.

Submitted by Rafael Santana (rafael.silvasantana@ufpe.br) on 2017-08-30T17:07:46Z No. of bitstreams: 2 license_rdf: 811 bytes, checksum: e39d27027a6cc9cb039ad269a5db8e34 (MD5) Antonio_Thesis.pdf: 3530353 bytes, checksum: 0ba09bccbe8eb163cd5b4646977ba882 (MD5) / Made available in DSpace on 2017-08-30T17:07:46Z (GMT). No. of bitstreams: 2 license_rdf: 811 bytes, checksum: e39d27027a6cc9cb039ad269a5db8e34 (MD5) Antonio_Thesis.pdf: 3530353 bytes, checksum: 0ba09bccbe8eb163cd5b4646977ba882 (MD5) Previous issue date: 2016-02-12 / Atualmente as redes locais sem fio (WLANs) em ambientes internos estão presentes na maioria dos prédios públicos. Estas redes, além da sua função principal, podem ser utilizadas para localização de pessoas e objetos, salientando que nestes ambientes não é adequada a utilização do sinal de GPS para esta finalidade. Diversos estudos e pesquisas nesta área têm sido realizados. Serviços baseados na localização interna possuem inúmeras aplicações nas áreas de segurança, médica, monitoramento, navegação, auxílio a deficientes, gerenciamento de pessoas, entre outras e hoje já movimentam um mercado de US$1 Bilhão. Com a proliferação da Internet das Coisas (IoT), estes valores serão ainda maiores. Os sistemas de localização interna utilizam tecnologias como Ultrassom, Infravermelho, RFID, Bluetooth e WLAN, variando conforme a precisão, exatidão, custo, velocidade de resposta, infraestrutura e aplicação. O presente trabalho propõe uma técnica de localização interna que utiliza a intensidade do sinal recebido (RSS — Received Signal Strength) das redes WLAN presentes como medida para localização. Na técnica proposta, é feito inicialmente um mapeamento das intensidades dos sinais da WLANs existentes. Estes valores são classificados através de um Algoritmo de Agrupamento (clustering) e, posteriormente, são aplicados, a cada agrupamento, algoritmos de regressão para o cálculo da localização. Associada a estas técnicas são aplicados filtros visando minimizar as variações do sinal medido decorrentes de interferências do meio. Esta técnica não necessita de grandes esforços de calibração nem alterações na estrutura existente, apenas utilizando a rede WLAN já instalada, obtendo assim uma precisão compatível com aplicações de localização de pessoas e objetos e auxílio em navegação em ambientes internos. Na implementação e testes da técnica proposta, foi empregado o processador Edison da Intel para a coleta das intensidades dos sinais — RSS e como plataforma de servidor foi utilizada a estrutura de nuvem da Microsoft através do Azure-Studio Machine Learning, apropriada para a análise e predição de dados da técnica utilizada. As medições para composição dos conjuntos de testes e validação foram realizadas no prédio do Centro de Informática da UFPE, demonstrando que apesar do baixo esforço de calibração, sem alteração da estrutura existente, atendem aos requisitos necessários. Resultados preliminares mostram que 60% das amostras estavam com erro inferior a 5 metros. / Currently, wireless local networks (WLANs) in internal environments are present in most of the public buildings. These networks, in addition to their main function, can be used to locate people and objects, stressing that in these environments it is not adequate the use of the GPS signal to this goal. Several studies and researches in this area have been made. Services based in internal location have many applications in security, health, monitoring, navigation, disabled assistance, and people management, among other areas. Nowadays, they already move a US$ 1 billion market. With the proliferation of the Internet of Things (IoT), these values will increase even further. Internal location systems use technologies such as Ultrasound, Infra-red, RFID, Bluetooth, and WLAN, varying according to the required precision, accuracy, cost, response speed, infrastructure, and application. The following work proposes an internal location technique that uses the received signal strength (RSS) from existing WLAN networks as a location measurement. In the proposed technique, is initially made a mapping of the existing WLANs signals intensities, these values are classified through a Clustering Algorithm and, after that, regression algorithms are applied to each group towards a location classification. Associated to these techniques, filters are applied aiming to minimize the measured signal variations due to the environment interferences. This technique doesn’t require big calibration efforts, nor changes in the existing structure, just uses the already installed WLAN network, obtaining a precision compatible to the one required for people and objects location and assistence in internal environments navigation. In the proposed technique’s implementation and tests, it was used Intel’s Edison processor to collect RSS signal’s intensities. As a server platform, it was used Microsoft’s cloud structure through the Azure-Studio Machine Learning, appropriate for the used technique’s analysis and data prediction. The main set of tests and validation was accomplished in the UFPE Informatics Center building, showing that despite low calibration effort, without changing the existing structure, it complies with the necessary requirements. Preliminary results show that 60% of the samples had errors under 5 meters.