Driver Mirror-Checking Action Detection Using Multi-Modal Signals

Studies on driver distraction aim to identify features extracted from various sensory signals that can be used to distinguish between normal and distracted driving behaviors. A major challenge in these studies is to determine whether the observed behaviors are associated with the primary driving tasks (checking mirrors, monitoring speed, changing lines) or secondary tasks that deviate the attention of the drivers. This study focuses on detecting the important driving task of checking the mirrors. The study compares the duration and frequency of mirror checking actions observed when the drivers are engaged in secondary tasks with the ones observed in normal driving conditions. The analysis reveals that the drivers reduce the frequency of checking the mirrors when they are engaged in secondary tasks, which affects their situational awareness. Based on real world driving data, we extracted multimodal features associated with the driver and vehicle behaviors. We train binary classifiers to detect mirror checking actions. Even though the classes are highly unbalanced (most of the samples do not have mirror checking actions), the proposed system achieves an F-score of 0.65 (recall 73%, precision 68%) using the extracted features. This promising result suggests that it is possible to detect mirror-checking actions, which can be used to signal alarms, preventing collisions and improving the overall driving experience.