Assistive Street Crossing: Integrating Object Detection and Multi-Frame Tracking for Visually Impaired Pedestrians

View at publisher Summarise with VIRAT
Authors: Sophie Hsu, Kayne Yao, Phuc Thanh-Thien Nguyen, Chung-Hsien Kuo
Cited by: 0
References: 21

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Navigating urban streets remains a major challenge for visually impaired pedestrians due to vehicle hazards and the absence of reliable acoustic signals from traffic signals. Existing assistive technologies mainly rely on object detection and depth estimation to improve environmental awareness; however, their reliance on single-frame detection can result in unsafe guidance when transient errors occur. To address this limitation, we propose a perception-to-decision framework that integrates temporal tracking with Simple Online Real-Time Tracking (SORT) enhanced by Kalman filtering. The system combines this tracking with a custom-trained YOLOv8 object detection model and 3D depth sensing to ensure robust vehicle threat analysis and consistent traffic light recognition. Experimental results show real-time performance, achieving 95.5% average crossing decision accuracy via temporal aggregation and motion-consistent threat estimation, with 52.1 ms latency. Overall, the proposed system significantly improves the safety and reliability of street crossing assistance for visually impaired users by unifying object detection with temporal tracking.

Fields of Study & Subjects

Autonomous Vehicle Technology and Safety Assistive Technology in Communication and Mobility Advanced Neural Network Applications

Publication details

Journal / source
International Conference on Consumer Electronics
Pages
1–4
Publication date
3 February 2026
Language
EN