Title: Optimisation enabled deep learning model for traffic sign recognition

Authors: Soja Salim; J.S. Jayasudha; B. Soniya

Addresses: Computer Science and Engineering, Sree Chitra Thirunal College of Engineering, APJ Abdul Kalam Technological University, Trivandrum, 695018, India ' Computer Science, Central University of Kerala, Kasargod, Kerala, 671316, India ' Computer Science and Engineering, Sree Chitra Thirunal College of Engineering, APJ Abdul Kalam Technological University, Trivandrum, 695018, India

Abstract: In this paper, a novel optimisation algorithm-trained deep learning network is developed to recognise all symbol-based traffic signs. The deep learning method employed here for the recognition is deep Q network (DQN) with an optimisation algorithm, named gradient descent-teamwork optimisation algorithm (GD-TOA), for training the DQN. Initially, the Gaussian filter is utilised for the pre-processing phase, in which the noise can be eliminated. Thereafter, the sign localisation is performed using SegU-Net with a modified loss function that includes balance cross-entropy, rescaled hinge loss, and insensitive loss to localise the sign region from the image. Finally, TSR is performed using the proposed GD-TOA-based DQN so as to improve the TSR performance. The experimentation reveals that the proposed GD-TOA-based DQN technique attains an overall improvement regarding the precision, accuracy, recall, and F-measure with values 0.977, 0.972, 0.971, and 0.973, respectively for TSR.

Keywords: traffic safety; TSR; traffic sign recognition; deep learning; sign localisation; optimisation algorithm.

DOI: 10.1504/IJHVS.2025.144165

International Journal of Heavy Vehicle Systems, 2025 Vol.32 No.1, pp.80 - 100

Received: 16 Jan 2024
Accepted: 07 Apr 2024
Published online: 30 Jan 2025 *

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