Bridging Conventional and Contemporary Approaches in Driver Distraction Detection: A Review with Proposed Graph-Based Model

Distracted driving is a major reason for collisions. Therefore, it is crucial to constantly monitor the driving condition of drivers and offer suitable solutions to those who are distracted. Cognitive distractions typically come from being tired, talking to a fellow passenger, listening to the radio, or engaging in other mentally taxing side activities that don't need a driver to take their eyes off the road. Because there are no outward signs of driver distraction, it is one of the most difficult diversions to identify. In this manuscript, Enhancing Road Safety through Attributed Multi-order Graph Convolutional Network based Cognitive Distraction Identification using Physiological Signals (ERD-AMGCN- CDI-PS) is proposed. The input data are collected from Emotions and Heart rate scale dataset. Then, the input data are fed into preprocessing stage. In preprocessing, Regularized bias-aware ensemble Kalman filter (RBAEKF) is used for removing noise, and artifacts. The pre-processed data are given into Synchro-Transient- Extracting Transform (STET) for extracting the features such as Heart Rate, and Emotion from the pre-processed data. Then, the extracted features are fed into Attributed Multi-order Graph Convolutional Network (AMGCN) for identifying the cognitive distraction. The proposed ERD-AMGCN- CDI-PS method is implemented on Python. Then effectiveness of the ERD-AMGCN- CDI-PS method is compared with other existing models. The ERD-AMGCN-CDI- PS method attains 16.28%, 30.78% and 25.29% higher accuracy when comparing with existing techniques, such as Driver cognitive distraction detection utilizing machine learning models (DCD-ML), Driver distraction detection utilizing bidirectional long short-term network under multiscale entropy of EEG (DDD-BLSN- EEG), and Multimodal driver distraction detection utilizing dual-channel network of CNN with Transformer (MDDD-DCN- CNN) respectively.