Year: 2025
Type: article
Abstract: In South Korea, the incidence of pedestrian traffic accidents is higher than the Organisation for Economic Co-operation and Development (OECD) average. In response, recent legal regulations have been strengthened to prevent accidents in school zones, with a greater focus on pedestrian safety. Consequently, the necessity for real-time pedestrian detection systems is becoming increasingly apparent. This study proposes the implementation of a deep learning-based pedestrian detection system, which would enable drivers to accurately detect and make informed decisions regarding pedestrians, vehicles, and crosswalks. The study employs a monocular camera and an image segmentation algorithm to compare the architectures of R-CNN and YOLOv8. Subsequently, the YOLOv8-seg model, which incorporates a Segmentation Branch structure for instance segmentation, was trained and tested on a variety of models. Subsequently, the system's functionality was validated through real-time streaming within a vehicle.
Source: Journal of Neonatal Surgery
Authors: Choi, Won-hyuk, Woo-Jin Jung
Institutions: Hanseo University, Hansei University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
Award: G1532178269
OpenAlex ID: W7219253900