Driver Distraction Identification with an Ensemble of Convolutional Neural Networks

Figure 2

An overview of our proposed solution. A face detector, a hand detector, and a skin segmenter are run against each frame. For each output image (i.e., Skin, Face, Hands), an AlexNet and an InceptionV3 network are trained (i.e., resulting in 10 neural networks: 5 AlexNet and 5 InceptionV3). The overall class distribution is determined by the weighted sum of all softmax layers. The weights are learned using a genetic algorithm.