Scholarlink

This paper introduces Sensei, a real-time embedded perception system designed to enhance the mobility and situational awareness of visually impaired individuals. The system integrates an Intel RealSense D455 depth camera and an NVIDIA Jetson Nano to perform edge-based, depth-aware object detection and conditional face recognition. By leveraging RGB-D sensing and efficient deep learning models, Sensei offers real-time feedback on nearby people and obstacles. It operates in two dynamic modes—walk and idle—allowing the system to adapt processing strategies based on the user’s activity, thus optimizing performance and power consumption. The architecture emphasizes modularity, embedded efficiency, and affordability, enabling practical deployment in real-world scenarios. Unlike conventional navigation aids, Sensei focuses on context and proximity awareness rather than explicit path prediction, making it suitable for semi-structured environments like hallways or indoor spaces. Through intelligent cueing and audio feedback, the system provides essential spatial and contextual information, supporting safer and more independent navigation. This work contributes to the growing field of assistive technologies by presenting a compact, low-cost, and robust solution tailored for the everyday needs of visually impaired users.

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

Sensei: Assistive Real-Time Perception for Visually Impaired Users with Depth-Aware Object and Face Recognition

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Deeksha,
  • Toshanlal Meenpal,
  • Jyothi Swaroop Perisetty

摘要

This paper introduces Sensei, a real-time embedded perception system designed to enhance the mobility and situational awareness of visually impaired individuals. The system integrates an Intel RealSense D455 depth camera and an NVIDIA Jetson Nano to perform edge-based, depth-aware object detection and conditional face recognition. By leveraging RGB-D sensing and efficient deep learning models, Sensei offers real-time feedback on nearby people and obstacles. It operates in two dynamic modes—walk and idle—allowing the system to adapt processing strategies based on the user’s activity, thus optimizing performance and power consumption. The architecture emphasizes modularity, embedded efficiency, and affordability, enabling practical deployment in real-world scenarios. Unlike conventional navigation aids, Sensei focuses on context and proximity awareness rather than explicit path prediction, making it suitable for semi-structured environments like hallways or indoor spaces. Through intelligent cueing and audio feedback, the system provides essential spatial and contextual information, supporting safer and more independent navigation. This work contributes to the growing field of assistive technologies by presenting a compact, low-cost, and robust solution tailored for the everyday needs of visually impaired users.