Driver Handheld Cell Phone Usage Detection

Abstract

The usage of cell phones by car drivers leads to a lack of attention to the road and an increased chance of accidents. The Dutch police is tasked with fining these drivers. Current fining methods require drivers to be caught red-handed. In this work, it is demonstrated that application of computer vision techniques can lead to a massive decrease in man-hours necessary by automating the phone usage detection process. 2038 images of drivers were collected and classified into risky (phone usage) and non-risky (no phone usage) behavior. A straightforward Convolutional Neural Network approach and a more intricate combination of phone, hand and face detection and hand classification were compared on this task. The combined approach performed best, with an accuracy of 86.4% and an F-Score of 0.70 (precision: 0.70, recall: 0.70). The study revealed that it is achievable to detect driver phone usage using computer vision.