A carregar...
Publicação
Avaliação da Condução baseada em Padrões utilizando Modelos baseados em LSTM
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 7.87 MB | Adobe PDF |
Autores
Resumo(s)
A crescente preocupação com a segurança rodoviária tem impulsionado o desenvolvimento de sistemas avançados para análise do comportamento do condutor. Este projeto apresenta um estudo focado na deteção de padrões de condução agressiva, recorrendo a sensores presentes em smartphones, como acelerómetros, giroscópios e GPS. Através da recolha e processamento destes dados, foi possível identificar comportamentos de risco no condutor que podem estar associados a diferentes fatores, como a distração ou fadiga.
A tecnologia moderna de sensores assume um papel fundamental na monitorização em tempo real, oferecendo um grande potencial para o fortalecimento dos sistemas de segurança automóvel. Os dados recolhidos pelos sensores representam sequências de eventos ao longo do tempo (acelerações, travagens, etc.), tornando-se essencial analisar estes padrões temporais para compreender o comportamento do condutor. Neste contexto, a análise dos padrões temporais foi realizada com recurso a redes Long Short-Term Memory (LSTM), explorando diferentes configurações e parâmetros de teste, alicerçados na utilização de um mecanismo de janelas deslizantes para estruturar os dados de forma sequencial e facilitar a aprendizagem dos modelos. Este mecanismo permitiu captar as dependências temporais presentes nos dados, reforçando a capacidade do modelo em distinguir entre condução agressiva e não agressiva.
O principal objetivo deste trabalho consistiu em encontrar a forma mais eficaz de detetar padrões de condução agressiva e não agressiva. Este processo implicou investigação contínua, envolvendo diversas abordagens e modificações ao longo do projeto, permitindo a comparação de diferentes resultados e, assim, identificar a melhor forma de alcançar o objetivo proposto.
Os resultados obtidos confirmam o potencial destas técnicas na monitorização em tempo real e na promoção de hábitos de condução mais seguros. Adicionalmente, a capacidade de recolher e analisar grandes volumes de dados em condições variadas de condução, abre perspetivas futuras para soluções mais personalizadas e adaptativas, reforçando o impacto da inteligência artificial na área da segurança rodoviária.
The growing concern for road safety has driven the development of advanced systems for driver behavior analysis. This project presents a study focused on detecting aggressive driving patterns by using sensors embedded in smartphones, such as accelerometers, gyroscopes, and GPS. Through the collection and processing of this data, it was possible to identify risky driving behaviors that may be associated with different factors, such as distraction or fatigue. Modern sensor technology plays a key role in real-time monitoring, offering great potential for strengthening automotive safety systems. The data collected by the sensors represent sequences of events over time (accelerations, braking, etc.), making it essential to analyze these temporal patterns to understand driver behavior. In this context, the analysis of temporal patterns was carried out using Long Short-Term Memory (LSTM) networks, exploring different configurations and test parameters, supported using a sliding window mechanism to structure the data sequentially and facilitate model learning. This mechanism enabled the capture of temporal dependencies within the data, enhancing the model’s ability to distinguish between aggressive and non-aggressive driving. The main objective of this work was to identify the most effective way to detect aggressive and non-aggressive driving patterns. This process required continuous research, involving various approaches and modifications throughout the project, which allowed for the comparison of different results and, consequently, the identification of the best approach to achieve the proposed objective. The results obtained confirm the potential of these techniques in real-time monitoring and in promoting safer driving habits. Additionally, the ability to collect and analyze large volumes of data under varied driving conditions opens future perspectives for more personalized and adaptive solutions, further strengthening the impact of artificial intelligence in the field of road safety.
The growing concern for road safety has driven the development of advanced systems for driver behavior analysis. This project presents a study focused on detecting aggressive driving patterns by using sensors embedded in smartphones, such as accelerometers, gyroscopes, and GPS. Through the collection and processing of this data, it was possible to identify risky driving behaviors that may be associated with different factors, such as distraction or fatigue. Modern sensor technology plays a key role in real-time monitoring, offering great potential for strengthening automotive safety systems. The data collected by the sensors represent sequences of events over time (accelerations, braking, etc.), making it essential to analyze these temporal patterns to understand driver behavior. In this context, the analysis of temporal patterns was carried out using Long Short-Term Memory (LSTM) networks, exploring different configurations and test parameters, supported using a sliding window mechanism to structure the data sequentially and facilitate model learning. This mechanism enabled the capture of temporal dependencies within the data, enhancing the model’s ability to distinguish between aggressive and non-aggressive driving. The main objective of this work was to identify the most effective way to detect aggressive and non-aggressive driving patterns. This process required continuous research, involving various approaches and modifications throughout the project, which allowed for the comparison of different results and, consequently, the identification of the best approach to achieve the proposed objective. The results obtained confirm the potential of these techniques in real-time monitoring and in promoting safer driving habits. Additionally, the ability to collect and analyze large volumes of data under varied driving conditions opens future perspectives for more personalized and adaptive solutions, further strengthening the impact of artificial intelligence in the field of road safety.
Descrição
Palavras-chave
Inteligência artificial LSTM Padrões de condução Sensores