Publicação
Human Posture Recognition through Passive RF Sensing
| Resumo: | Human activity recognition using passive Radio Frequency (RF)-sensing systems has been present on various studies in the recent past. This is due to its application potential in several domains. Besides this, the majority of passive RF-sensing systems propose low-cost, high portability and reliability, and non-intrusive solutions. In this thesis, a passive RF-sensing system based on two Software Defined Radio (SDR) devices is proposed. The system is operating in the Very High Frequency (VHF) range band at a frequency of 104.3 MHz. Several experimental datasets were generated to study and analyse the human postures to automatically recognise (standing, sitting, and the none class, which represents the absence of a person). Various features were computed from the collected data and its importance is evalu- ated through the Analysis of Variance (ANOVA) feature selection algorithm. Using the selected features, the data of the acquired datasets is classified using three different Ma- chine Learning (ML) techniques: Support Vector Machine (SVM), K-Nearest Neighbours (KNN), and one classification method proposed by the author. The performance of each classifier is assessed using different distance metrics or ker- nels to classify the data and an overall classification accuracy of 87.896% was achieved. A prototype for a real-time human posture classification service was also developed and its components are described. The results that were obtained validate the selected features and highlight the value of the classification techniques used in this area. The results confirm the great potential of passive RF-sensing systems for human posture recognition, which motivates future works on the field. |
|---|---|
| Autores principais: | Pereira, João Tiago Barata |
| Assunto: | Machine Learning Human Activity Sensing Passive RF Sensing Automatic Recognition |
| Ano: | 2022 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade Nova de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório Institucional da UNL |
| Resumo: | Human activity recognition using passive Radio Frequency (RF)-sensing systems has been present on various studies in the recent past. This is due to its application potential in several domains. Besides this, the majority of passive RF-sensing systems propose low-cost, high portability and reliability, and non-intrusive solutions. In this thesis, a passive RF-sensing system based on two Software Defined Radio (SDR) devices is proposed. The system is operating in the Very High Frequency (VHF) range band at a frequency of 104.3 MHz. Several experimental datasets were generated to study and analyse the human postures to automatically recognise (standing, sitting, and the none class, which represents the absence of a person). Various features were computed from the collected data and its importance is evalu- ated through the Analysis of Variance (ANOVA) feature selection algorithm. Using the selected features, the data of the acquired datasets is classified using three different Ma- chine Learning (ML) techniques: Support Vector Machine (SVM), K-Nearest Neighbours (KNN), and one classification method proposed by the author. The performance of each classifier is assessed using different distance metrics or ker- nels to classify the data and an overall classification accuracy of 87.896% was achieved. A prototype for a real-time human posture classification service was also developed and its components are described. The results that were obtained validate the selected features and highlight the value of the classification techniques used in this area. The results confirm the great potential of passive RF-sensing systems for human posture recognition, which motivates future works on the field. |
|---|
Atividades financiadas
Não foram encontrados projetos financiados.