Detección de somnolencia en conductores: Análisis de señales fisiológicas durante la simulación de conducción.
Loading...
Date
2025
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad de Concepción
Abstract
La somnolencia y el cansancio son factores importantes en los accidentes de tráfico, sobre todo en las carreteras. Estadísticas recientes relacionan sustancialmente los accidentes mortales con el cansancio y la fatiga. En Chile, se reportaron 86,050 accidentes de tráfico en 2022, lo que provocó 1,745 muertes. De estos accidentes, 820 fueron causados por condiciones físicas deficientes como el cansancio y la fatiga. En este contexto, la investigación sobre seguridad vial se enfoca en el desarrollo de métodos automáticos para detectar la somnolencia del conductor y advertirle a tiempo mediante tecnologías de sensores y procesamiento de datos. El propósito de este proyecto es detectar signos iniciales de somnolencia mediante el examen de diversas señales fisiológicas como EEG, ECG, EMG y EDA para desarrollar un algoritmo que pueda identificar relaciones entre estas señales y la somnolencia durante la simulación de conducción.
Durante la etapa inicial de este estudio, se realizó una descripción detallada de los principios fisiológicos y de adquisición de señales relevantes, destacando sus características en relación con los estados de transición de vigilia y somnolencia. Se realizó una revisión del estado actual de conocimiento en el campo, la cual permitió identificar las características clave, modelos y algoritmos exitosos empleados en la detección de somnolencia.
Se utilizaron los datos obtenidos del estudio en simulador de conducción, realizado por Hermes Javier Mora, “Predicción de eventos tempranos de somnolencia mediante un conjunto de datos multifactoriales en conductores de vehículos”, usando los bloques para el estudio de somnolencia (bloque 2 y 3), de donde se obtuvieron señales fisiológicas de EEG, ECG, EMG y EDA. El estudio presente considera las características extraídas de las señales de EEG y ECG.
Se implementaron tres tipos diferentes de modelos de clasificación (redes neuronales, SVM y kernel), de los cuales se obtuvieron rendimientos positivos en la relación de las señales y los estados de somnolencia, destacando el modelo redes neuronales que logró una precisión del 91.23% en el bloque 3 y 86.76% en el bloque 2. Respecto a las características utilizadas, se calcularon la potencia de las bandas de frecuencia (delta, theta, alpha y beta), relaciones entre la potencia de las bandas (índice 1: Pθ/Pα ; índice 2: Pθ +Pα/Pα +Pβ; índice 3: Pθ + Pα/Pβ), frecuencia cardíaca y variabilidad de la frecuencia cardíaca. Los canales frontales y occipitales tienen una contribución significativa en la detección de somnolencia, especialmente en la potencia de las bandas delta y theta, lo cual está alineado con estudios previos. Además, se destaca que los índices 1 y 2, aunque muestran una relación general con la somnolencia en varios canales, no son tan específicos como el índice 3. Este último demuestra una relación más fuerte y precisa con las etiquetas de somnolencia en canales específicos, lo que sugiere que el índice 3 es más relevante para la identificación precisa del estado de somnolencia.
Drowsiness and fatigue are significant factors in traffic accidents, particularly on highways. Recent statistics show a substantial link between fatal accidents and fatigue. In Chile, 86,050 traffic accidents were reported in 2022, resulting in 1,745 deaths. Of these accidents, 820 were caused by poor physical conditions such as fatigue. In this context, road safety research focuses on developing automated methods to detect driver drowsiness and warn them in time using sensor technologies and data processing. The goal of this project is to detect early signs of drowsiness by examining various physiological signals such as EEG, ECG, EMG, and EDA, in order to develop an algorithm that can identify relationships between these signals and drowsiness during driving simulations. During the initial phase of this study, a detailed description of the physiological principles and ac quisition methods of relevant signals was carried out, emphasizing their characteristics in relation to the transition states between wakefulness and drowsiness. A review of the current state of knowledge in the field was conducted, which allowed identifying key features, models, and successful algorithms employed in drowsiness detection. Data obtained from a driving simulator study conducted by Hermes Javier Mora, titled *“Prediction of Early Drowsiness Events Using a Multifactorial Dataset in Vehicle Drivers,”* was used, focusing on the blocks designed for studying drowsiness (blocks 2 and 3), where physiological signals such as EEG, ECG, EMG, and EDA were collected. The present study considers the features extracted from EEG and ECGsignals. Three different types of classification models were implemented (neural networks, SVM, and kernel), achieving positive results in associating the signals with drowsiness states. Notably, the neural network model achieved an accuracy of 91.23% in block 3 and 86.76% in block 2. Regarding the features used, the power of frequency bands (delta, theta, alpha, and beta) was calculated, along with ratios between band powers (index 1: Pθ/Pα; index 2: (Pθ + Pα)/(Pα + Pβ); index 3: (Pθ + Pα)/Pβ), heart rate, and heart rate variability. Frontal and occipital channels contribute significantly to drowsiness detection, parti cularly in the power of delta and theta bands, aligning with previous studies. Furthermore, it is highlighted that indices 1 and 2, while generally showing a relationship with drowsiness across several channels, are not as specific as index 3. The latter demonstrates a stronger and more precise relationship with drow siness labels in specific channels, suggesting that index 3 is more relevant for accurately identifying drowsiness states.
Drowsiness and fatigue are significant factors in traffic accidents, particularly on highways. Recent statistics show a substantial link between fatal accidents and fatigue. In Chile, 86,050 traffic accidents were reported in 2022, resulting in 1,745 deaths. Of these accidents, 820 were caused by poor physical conditions such as fatigue. In this context, road safety research focuses on developing automated methods to detect driver drowsiness and warn them in time using sensor technologies and data processing. The goal of this project is to detect early signs of drowsiness by examining various physiological signals such as EEG, ECG, EMG, and EDA, in order to develop an algorithm that can identify relationships between these signals and drowsiness during driving simulations. During the initial phase of this study, a detailed description of the physiological principles and ac quisition methods of relevant signals was carried out, emphasizing their characteristics in relation to the transition states between wakefulness and drowsiness. A review of the current state of knowledge in the field was conducted, which allowed identifying key features, models, and successful algorithms employed in drowsiness detection. Data obtained from a driving simulator study conducted by Hermes Javier Mora, titled *“Prediction of Early Drowsiness Events Using a Multifactorial Dataset in Vehicle Drivers,”* was used, focusing on the blocks designed for studying drowsiness (blocks 2 and 3), where physiological signals such as EEG, ECG, EMG, and EDA were collected. The present study considers the features extracted from EEG and ECGsignals. Three different types of classification models were implemented (neural networks, SVM, and kernel), achieving positive results in associating the signals with drowsiness states. Notably, the neural network model achieved an accuracy of 91.23% in block 3 and 86.76% in block 2. Regarding the features used, the power of frequency bands (delta, theta, alpha, and beta) was calculated, along with ratios between band powers (index 1: Pθ/Pα; index 2: (Pθ + Pα)/(Pα + Pβ); index 3: (Pθ + Pα)/Pβ), heart rate, and heart rate variability. Frontal and occipital channels contribute significantly to drowsiness detection, parti cularly in the power of delta and theta bands, aligning with previous studies. Furthermore, it is highlighted that indices 1 and 2, while generally showing a relationship with drowsiness across several channels, are not as specific as index 3. The latter demonstrates a stronger and more precise relationship with drow siness labels in specific channels, suggesting that index 3 is more relevant for accurately identifying drowsiness states.
Description
Tesis presentada para optar al título de Ingeniero/a Biomédico.
Keywords
Conducción de automóviles, Fisiología Aparatos e instrumentos, Fisiología Procesamiento de datos