Scholarlink

This paper presents a lightweight, real-time system for detecting environmental hazards faced by wheelchair users. Using a monocular camera and microphone, the system integrates visual object detection, speech recognition, and acoustic event analysis with large language model (LLM)-based reasoning to assess navigational risks. It delivers interpretable, context-aware feedback to support safe and autonomous mobility. Tested in outdoor environments, the framework reliably identifies hazards such as vehicles, curbs, and spoken warnings. By combining multimodal AI with low-cost hardware, this work contributes a scalable solution for enhancing assistive mobility in real-world settings.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Real-Time Multimodal Hazard Detection for Assistive Wheelchair Navigation

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Eisei Nakahara,
  • Xinxiao Li

摘要

This paper presents a lightweight, real-time system for detecting environmental hazards faced by wheelchair users. Using a monocular camera and microphone, the system integrates visual object detection, speech recognition, and acoustic event analysis with large language model (LLM)-based reasoning to assess navigational risks. It delivers interpretable, context-aware feedback to support safe and autonomous mobility. Tested in outdoor environments, the framework reliably identifies hazards such as vehicles, curbs, and spoken warnings. By combining multimodal AI with low-cost hardware, this work contributes a scalable solution for enhancing assistive mobility in real-world settings.