Skip to content
Conference

A Low-Latency Passive Vision-Based Assistive Infrastructure for Accessible Public Transportation

Aug 2026 · International Workshop on Artificial Intelligence and Cognition · pp. 1462-1468 · 0 citations · 8 references

Abstract

Visually impaired and mobility-limited passengers face an accessibility barrier in urban bus transit because route and destination information is primarily communicated through visual displays. Existing assistive approaches often require active user interaction, while conventional infrastructure-side announcements lack contextual awareness. Although previous bus-recognition systems have processed individual images, image sequences, or video, they generally do not maintain persistent identity-specific recognition state for multiple approaching buses.We propose a passive, infrastructure-based assistive framework based on temporally stateful video processing. A YOLOv8-m detector trained on a custom 800-frame dataset, partitioned into 400/200/200 train/validation/test samples and balanced across daytime, night-time, and high-glare conditions, achieves an mAP50 of 97.6%. ByteTrack maintains persistent bus identities, enabling multi-bus handling and duplicate-announcement suppression. A lexicon-guided temporal stabilisation mechanism combining Levenshtein correction with a similarity threshold of 0.5 and seven-frame sliding-window voting reduces announcement decision latency by 40.7%. A context-aware gate further conditions audio output on detection confidence, route stability, and passenger presence.The framework additionally estimates bus distance using a monocular focal-length method. Resource profiling records peak CPU RAM and GPU VRAM usage of 1289.5 MB and 255.1 MB, respectively, supporting the feasibility of edge deployment. Evaluation is performed on the held-out test split across all three lighting conditions and includes comparison with a recent single-image baseline.

View source

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.