Table 4.

Motion sensors.

Ref	Year	Experimental Setup	Number of Subjects	Driving Duration	Type of Sensors	Physical Measurand	Detection Algorithm	Problem Solved	Limitations
[80]	2024	Actual driving	9 Subjects	Total of 73 trips (0.5–10 h each trip)	wristbands, vehicle mounted equipment, and trip logs	Fatigue	Attention-BiLSTM	The authors developed a fatigue detection technique that incorporates non-visual features from customized wristbands, vehicle-mounted devices, and trip logs. This strategy solves the drawbacks of conventional methods that rely mainly on visual and physiological aspects, which can be obtrusive and less trustworthy.	The method for assessing feature relevance has inherent limitations. Certain qualities may only be beneficial when paired with others, resulting in an underestimate of their individual importance. This could have an impact on our general understanding of which traits are most important for detecting fatigue.
[81]	2022	Simulator	15 Subjects	1 h	Motion Capture Sensor the head pos	Fatigue and drowsiness	reLU-BiLSTMs	The study compared the performance of separate day and night models to a combined model and found that the models performed similarly. This finding implies that separate models for distinct driving conditions (daytime vs. midnight) can be useful for fatigue identification.	One important drawback is that the same dataset was used to construct the sleepiness detection algorithm and evaluate its performance. This dual use can produce biased results that do not accurately reflect the model's performance in real-world circumstances.
[82]	2021	Actual and simulated driving	5 subjects		microphones and speakers of smartphones	Drowsiness	LSTM networks	This work developed a drowsy driving detection system using only the embedded audio components (microphone and speaker) on cellphones. To identify tiredness in real time, the system recognized specific Doppler shift patterns induced by typical drowsy behaviors such as nodding, yawning, and aberrant steering wheel motion.	The reliance on smartphone audio devices to detect Doppler shifts may be susceptible to background noise, changes in driving situations, and smartphone location within the vehicle, potentially impacting detection accuracy.
[83]	2021	Simulator	40 subjects		impulsive radio ultrawideband (IR-UWB) radar	Drowsiness	Different Machine learning and Multilayer Perceptron	The study proposed a non-invasive device for detecting driver drowsiness based on Impulse Radio Ultra-Wideband (IR-UWB) radar technology. This technology recorded chest movements to determine breathing rates, allowing for precise sleepiness detection.	Using Impulse Radio Ultra-Wideband (IR-UWB) radar, required precise location and environmental conditions. Furthermore, major driving movements or body position changes that alter radar signal quality may jeopardize the system's performance.
[84]	2019	Actual driving	8 subjects		Steering wheel angle sensor (SWA)	Fatigue	Fuzzy recurrent neural network model (RNN)	The author developed a fuzzy recurrent neural network (FRNN) model that detects driver weariness using steering-wheel angle (SWA) data. This model captured the nonlinear properties and anomalies in SWA time series data under real-world driving settings.	The limitations of this work include the difficulties associated with relying simply on steering-wheel angle (SWA) data, which can be influenced by different external factors such as road conditions, driver habits, and vehicle dynamics, potentially reducing the accuracy of fatigue detection.