Publicação
Development of a BCI-Based application for remote TV control
| Resumo: | The goal of Brain Computer Interfaces (BCI) has always been to discover new pathways to interact with technological systems. Despite the numerous BCI applications developed throughout the years, there has not been a single project that used brain signal to interact with the television, functioning as a mental remote. To tackle the task at hand a complex system was developed where an electroencephalography (EEG) headband, designed to be in contact with the frontal and prefrontal cortices, sent acquired data via Bluetooth to a developed mobile application. Said app would then send the received data to a server running in a computer, where it was analyzed, processed and some relevant features were extracted. Those features were then fed into a previously trained Machine Learning model. According to the prediction output, a Hypertext Transfer Protocol (HTTP) POST request was sent to a cloud API, which would send a signal to an Infrared device connected to the Television (TV) box. The infrared (IR) device sent a signal to the box itself which caused it to execute a command, like changing the channel, volume, turn on/off etc. Several BCI paradigms were tested such as artifact detection, Steady State Visual Evoked Potential (SSVEP), focus/relaxation paradigm, motor imagery and mental task such as. Both motor imagery and mental task proved to grant the least sense of control, since the user could not interact with the box intentionally. Both artifact detection and focus/relaxation presented a rather satisfactory testing accuracy and granted the user an adequate sense of control over the TV, which was one of the goals of this project. The solution which presented the best outcome was the SSVEP based paradigm, which could execute numerous commands and had an accuracy above 90% in some users. Random forest was the machine learning algorithm used due to its robustness and prediction accuracy during the testing step. |
|---|---|
| Autores principais: | Amante, João Luís Neto dos Santos |
| Assunto: | BCI EEG Inteligência Artificial Televisão Aplicação mobile Teses de mestrado - 2023 |
| Ano: | 2023 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso restrito |
| Instituição associada: | Universidade de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório da Universidade de Lisboa |
| Resumo: | The goal of Brain Computer Interfaces (BCI) has always been to discover new pathways to interact with technological systems. Despite the numerous BCI applications developed throughout the years, there has not been a single project that used brain signal to interact with the television, functioning as a mental remote. To tackle the task at hand a complex system was developed where an electroencephalography (EEG) headband, designed to be in contact with the frontal and prefrontal cortices, sent acquired data via Bluetooth to a developed mobile application. Said app would then send the received data to a server running in a computer, where it was analyzed, processed and some relevant features were extracted. Those features were then fed into a previously trained Machine Learning model. According to the prediction output, a Hypertext Transfer Protocol (HTTP) POST request was sent to a cloud API, which would send a signal to an Infrared device connected to the Television (TV) box. The infrared (IR) device sent a signal to the box itself which caused it to execute a command, like changing the channel, volume, turn on/off etc. Several BCI paradigms were tested such as artifact detection, Steady State Visual Evoked Potential (SSVEP), focus/relaxation paradigm, motor imagery and mental task such as. Both motor imagery and mental task proved to grant the least sense of control, since the user could not interact with the box intentionally. Both artifact detection and focus/relaxation presented a rather satisfactory testing accuracy and granted the user an adequate sense of control over the TV, which was one of the goals of this project. The solution which presented the best outcome was the SSVEP based paradigm, which could execute numerous commands and had an accuracy above 90% in some users. Random forest was the machine learning algorithm used due to its robustness and prediction accuracy during the testing step. |
|---|