Distracted Driver Classification for Low-computing Power Devices using Simple CNNs

Distracted driving is one of the major causes of traffic accidents. To help more drivers become aware of their driving behavior, this paper proposes a distracted driver classifier built specifically for low-computing power devices. The classifier is based on a simple Convolutional Neural Network (CNN) with Depthwise Separable convolution. The classifier was evaluated with the State Farm Distracted Driver Dataset and achieved a 99.15% accuracy while maintaining a speed of about 29.76 frames per second (FPS) on the Raspberry Pi 4 model B. Our proposed classifier will allow more cars to implement distracted driver detection systems, and thus, will help reduce the number of traffic accidents overall.