Detecção da direcionalidade do movimento humano utilizando perturbações do sinal eletromagnético de interfaces IEEE 802.11

• Ano de defesa: 2018
• Autor(a) principal: Silva, Bruno Soares da
• Orientador(a): Cardoso, Kleber Vieira
• Banca de defesa: Cardoso, Kleber Vieira, Laureano, Gustavo Teodoro, Abousheaisha, Abdallah S. Abdallah, Soares, Anderson da Silva, Viana, Aline Carneiro
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Goiás
• Programa de Pós-Graduação: Programa de Pós-graduação em Ciência da Computação (INF)
• Departamento: Instituto de Informática - INF (RG)
• País: Brasil
• Palavras-chave em Português: Detecção de fluxos de movimento; Direcionalidade de movimentos; Sensores não invasivos; Movimento humano; 802.11; Rede sem fio
• Palavras-chave em Inglês: Movement flow detection; Movement direction; Non-invasive sensors; Channel state information; Human movement; Wireless networks
• Área do conhecimento CNPq: CIENCIA DA COMPUTACAO::SISTEMAS DE COMPUTACAO
• Link de acesso: http://repositorio.bc.ufg.br/tede/handle/tede/9052

The movement flow detection in indoor environments requires the aquisition and implantation of specialized devices. The perturbations that can affect the electromagnetic signals used by 802.11 interfaces make this type of device a low-cost and widely available movement sensor. Most indoor environments have a 802.11 interface, which makes the use of this type of devices a good option as it doesn't requires any new device. In this work, we propose the WiDMove, a proposal to detect the movement flows in an indoor environment using the channel quality measurements (known as Channel State Information - CSI) offered by the IEEE 802.11n standard. Our proposal is based on signal processing and pattern recognition techniques, which allow us to extract and classify event signatures using the CSI. In lab tests with off-the-shelf 802.11 interfaces, we collected CSI samples that were affected by 8 different people. From this collected data we extracted the signature of the entry and exit events using some techniques such as Principal Component Analysis (PCA), Short-Time Fourier Transform (STFT) and Continuous Wavelet Transform (CWT). We trained two model types, the first based on a Support Vector Machine (SVM) classifier and the second based on a Multi Layer Perceptral (MLP) neural network. We validated this models with average accuracy experiments and with the cross-validation, including the K-Fold and Leave-One-Out techniques. WiDMove presented that can reach an average accuracy above 93% and that we can train neural networks that can reach an accuracy above 97%.