Table 9.

Comparisons to state-of-the-art driver’s fatigue detection systems, in terms of multimodal (visual and non-visual); these comparisons are based on two-stage classification DFD systems, such as normal- and fatigue-based on 10 subjects under normal conditions.

Cited	Methodology	Detection Accuracy (ACC)	Time	Platform
(a) Classification driver fatigue without pre-training
[36] Simon_EEG (2012)	EEG with statistical analysis	FT: 83.5%, NM: 84.5%	6.7 s	No
[157] BJ Chang-smartphone (2012)	Different sensors, including video, electrocardiography, photoplethysmography, temperature, and a three-axis accelerometer	FT: 85.5%, NM: 86.5%	7.88 s	Yes
(b) Classification driver fatigue in cloud platform
[36] Simon_EEG (2012)	EEG with statistical analysis	FT: 83.5%, NM: 84.5%	4.33 s	No
[157] BJ Chang-smartphone (2012)	Different sensors, including video, electrocardiography, photoplethysmography, temperature, and a three-axis accelerometer	FT: 85.5%, NM: 86.5%	6.35 s	Yes
(c) M-DFD: Combine visual and non-visual features without smartphone
Visual and non-visual features	CNN + RNN without pre-training	FT: 89.65%, NM: 89.5%	3.45 s	NA
Visual and non-visual features	CNN + RNN with pre-training on scratch	FT: 90.40%, NM: 90.5%	3.75 s	NA
(d) M-DFD: Combine visual and non-visual features with smartphone
Visual and non-visual features	CNN+ RNN without pre-training	FT: 89.65%, NM: 88.5%	3.77 s	Yes
Visual and non-visual features	CNN + RNN with pre-training on scratch	FT: 94.50%, NM: 92.5%	3.85 s	Yes
(e) M-DFD: Combine visual and non-visual features with smartphone and Cloud
Visual and non-visual features	CNN+ RNN without pre-training	FT: 89.65%, NM: 88.5%	1.2 s	Yes
Visual and non-visual features	CNN + RNN with pre-training on scratch	FT: 94.50%, NM: 93.5%	1.3 s	Yes

CNN: Convolutional neural network, RNN: Recurrent neural network; EEG: electroencephalography, FT: Fatigue, NM: Normal state.