ENPI / e-JIKEI Network Promotion Institute

Journal of Mechanical and Electrical Intelligent System (JMEIS, J. Mech. Elect. Intel. Syst.)

An international open-access peer-reviewed journal

ISSN 2433-8273

Home

Home > Archives > Vol.6, No.3

Vol.6, No.3

TABLE OF CONTENTS

Articles

Convolutional Neural Network Based Zebra Crossing and Pedestrian Traffic Light Recognition

Shinji Eto, Yasuhiro Wada, and Chikamune Wada

Journal of Mechanical and Electrical Intelligent System, Vol.6, No.3, pp.1-11, 2023.

Abstract: The study proposes a convolutional neural network (CNN) method for providing guided assistance for the visually impaired and blind (VIB) at zebra crossings with pedestrian traffic lights (PTL). It seeks to improve existing methods by providing locational orientation as well as navigation guidance in travel direction for the VIB at the crossing. Images of zebra crossings with a PTL were taken with a camera under sunny conditions. However, classifying PTL labels is difficult due to image background assimilation. Nevertheless, this could be resolved by using high resolution images or applying preprocessing to improve image contrast for highlighting the PTL outline. Although the study CNN outperforms state-of-the-art methods in accuracy and calculation time, a limitation of the model performance is that it would decrease when the input image is taken in different environments from the dataset. This problem can be addressed by collecting images under various environmental circumstances.

Full Paper (PDF)