DEVELOPMENT OF AN INTELLIGENT DISTRACTED DRIVER DETECTION SYSTEM USING DEEP LEARNING TECHNIQUE

Abstract

This study presents an intelligent approach to detect distracted driver by integrating a high-tech Deep Learning (DL) model. The DL model adopted in the system is based on the Efficient Channel Attention (ECA) mechanism integrated with the Swin Transformer. By strengthening the capacity of the model to focus on essential aspects of images, the goal is to enhance the classification of driver behaviour. The ECA module is integrated with the Swin Transformer, which is known for its window-based multi-head self-attention mechanism that can observe and document both local and global features. This allows the ECA module to dynamically modify channel weights, for the improvement of feature extraction process. A dataset from the 2016 State Farm Distracted Driving Detection Challenge, which is made up of 22,124 images depicting different distracted driving behaviours, is used to test the algorithm. To maximise the model's performance, the images go through preprocessing, which includes scaling and data augmentation. The implementation's results, which were achieved using MATLAB, demonstrate notable improvements in classification resilience and accuracy. The model applies ECA for improved feature focus and efficiently manages spatial connections among images using methods such as the Adam optimiser.