A Non-Intrusive Method for Detecting Visual Distraction Indicators of Transport Network Vehicle Service Drivers Using Computer Vision

Driving for a prolonged period of time exposes a driver to numerous components that influence his/her conduct on the road such as visual, cognitive, and manual distractions. Detecting signs of distraction is an essential component in lessening the possibility of road accidents. This paper presents a model that detects a Grab driver's lapse indicator that is mainly focused on visual distraction using a non-intrusive camera-based approach in hopes of contributing to the improvement of road safety technologies. The model primarily applies the concept of eye gaze to detect distraction cues. Manual annotation was conducted to compare it with the model's predictions to assess the model's effectivity. OpenFace was used to detect action units from the videos. The algorithm designed for the model reads the output file to fully detect distraction cues and apply time constraints. K-nearest neighbor was used to train the model and was validated by Kfold cross validation and has an 84% F-measure to indicate the detecting power.