Title: Advancing transport safety with faster pre-convoluted neural networks and lightweight multi-scale fusion for driver distraction detection

Authors: M. Joel John; K. Dinakaran; N. Bharathiraja

Addresses: Department of Computer Science and Engineering, Saveetha Engineering College, Chennai, 602105, Tamil Nadu, India ' Department of Artificial Intelligence and Data Science, S.A. Engineering College, Chennai, 600077, Tamil Nadu, India ' Chitkara University Institute of Engineering and Technology, Chitkara University, Punjab, 140401, India

Abstract: Automated vehicle technology aims to improve driving safety by removing human errors but driver distraction remains an issue, spurring interest in driver-assistance systems that recognise and support safe driver activities. Complex model topologies enable convolutional neural networks (CNN) to acquire more data characteristics but reduce generalisation and risk overfitting; despite regularisation methods, improving generalisation performance in robust CNN training remains a concern. For classifying and detecting driver distraction, an architecture based on Faster Pre-convoluted neural networks (FPCNN) is suggested in this research. An efficient P-FRNN with excellent accuracy is built, and a novel lightweight multi-scale fusion (LMF) architecture is presented to create intensive convolutional networks for multi-scale image increased generalisation performance. To improve the generalization effectiveness of deep learning models for driver distraction created using raw data from various activity monitors and the proposed design proved accurate to 97.8% and resulted in reduced computational complexity and shorter training times.

Keywords: automated vehicles; CNN; convolutional neural networks; transport safety; multi-scale fusion; driver distraction; generalisation; overfitting prevention.

DOI: 10.1504/IJHVS.2025.144167

International Journal of Heavy Vehicle Systems, 2025 Vol.32 No.1, pp.101 - 121

Received: 26 Jan 2024
Accepted: 18 Apr 2024
Published online: 30 Jan 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article