Table 3.

Hybrid measurements.

Ref	Year	Experimental Setup	Number of Subjects	Driving Duration	Type of Sensors	Physical Measurand	Detection Algorithm	Problem Solved	Limitations
[74]	2023	Simulator	13 Subjects	1 h	thermal camera, RGB camera, and environment thermometer	Drowsiness	YOLOv5	This study developed a vigilance detection system for drivers of autonomous rail rapid transit (ART) vehicles that combines face thermal imaging with ambient data. Thermal imaging was utilized to record physiological data, including skin temperature and respiratory patterns.	This work's limitations were its reliance on thermal imaging, which, while helpful in recording physiological signals, can be susceptible to external temperature changes and may require precise calibration and controlled circumstances to get reliable results.
[75]	2022		Used an online public data set and 2 subjects		Camera and AD8232 heart-rate sensor	Fatigue	A customized convolutional neural netWork	The system used the Nvidia Jetson Nano developer kit and Arduino Uno for embedded computing, combining eye and mouth localization techniques with heart rate monitoring to detect tiredness and the presence of a face mask accurately.	Although the system was designed to work in various situations, its effectiveness may be reduced in situations with poor lighting or different viewing angles. The heart rate monitoring module requires accurate electrode placement, which may not always be convenient or comfortable for users.
[76]	2021	Simulator	Public data set (SEEDVIG) and 23 Subjects		EEG and EOG electrodes And SMI eyetracking glasses	Driver's Vigilance and Fatigue detection	A capsule attention mechanism with (LSTM)	A capsule attention method is developed, allowing the model to focus on the most critical aspects of the learned multimodal representations. This multimodal system combines EEG and EOG data for real-time driver vigilance evaluation.	The inclusion of EEG and EOG data complicates the model. The research noted that multimodal analysis is complicated due to including complementary and conflicting information in the signals.
[77]	2020	Simulator	Public data set		Camera and ECG sensors	Fatigue	(CNN) and deep Belief network (DBN)	The authors developed a Hybrid Fatigue system that integrated visual information, such as the PERCLOS measure, with non-visual features, particularly heart-rate signals from ECG sensors.	The suggested approach is extremely dependent on the quality of the sensors employed. The system's performance may suffer if the sensors fail to record reliable data due to environmental conditions or hardware problems.
[78]	2019	Simulator	21 subjects	110 min	Physiological sensors, Camera	Drowsiness	Artificial neural network models	The study utilized a range of data sources, including physiological indicators (heart rate, breathing rate), sensorimotor indicators (blink duration, PERCLOS), and driving performance measurements (lane position, steering wheel angle).	Using a controlled, monotonous driving simulator setting was one of the drawbacks since it may need to fully capture the complexity of real-world driving situations and their impact on drowsiness detection and prediction.
[79]	2018	Simulator	29 Subjects		Infrared camera and PVT	Drowsiness	CNN	The system used convolutional neural networks (CNNs) to extract data-driven features associated with eye closure dynamics across four timescales (5 s, 15 s, 30 s, and 60 s), allowing it to balance accuracy and responsiveness.	The drawbacks included the reliance on video-based facial analysis, which can be affected by differences in lighting conditions, head positions, and potential occlusions, such as glasses or facial hair, influencing the accuracy of eye closure recognition.