Research on Safe Driving Evaluation Method Based on Machine Vision and Long Short-Term Memory Network

Table 3

Test results of different methods in self-built database.

Method	Network structure	Accuracy (%)					Parameter (MB)	Calculations (flops)
		Fatigue (%)	Calling (%)	Eating (%)	Turn (%)	Talk
CNN	VGG-16	65.56	95.22	94.65	98.88	93.55%	1.185	1.963
LRCN	VGG-16	70.78	94.53	96.55	99.12	94.17%	1.177	2.101
SSD	Google	73.66	89.82	96.35	98.25	94.28%	1.165	1.995
iDT + FV	ResNet-50	76.56	94.76	95.82	98.81	93.55%	1.135	1.926
iDT + HSV	ResNet-50	77.53	88.71	95.52	98.18	93.32%	1.123	1.928
Dual-stream	VGG-16	78.22	93.65	94.8	98.22	92.52%	1.193	1.875
Ours	VGG-16+	80.33	97.35	96.22	98.30	95.52	1.056	1.736
	Inception