Driver Distraction Identification with an Ensemble of Convolutional Neural Networks

Table 1

Distracted driver posture classification results on randomly selected test data.

Model	Source	Loss (NLL)	Accuracy (%)
AlexNet	Original	0.3909	93.65
	Skin Segmented	0.3468	93.62
	Face	1.0516	84.28
	Hands	0.6186	89.52
	Face + Hands	0.8298	86.68
InceptionV3	Original	0.2654	95.17
	Skin Segmented	0.2903	94.66
	Face	0.6096	88.82
	Hands	0.4546	91.62
	Face + Hands	0.4495	90.88
AlexNet		0.2727	94.29
Majority Voting Ensemble		0.1661	95.77
GA-Weighted Ensemble		0.1575	95.98