Scholarlink

The study aims for low-energy consumption models that can detect obstacles on a head-mounted smart device for the vision impaired. Under this setting, long operational time with limited battery power is obviously important, so small but accurate models would be highly desirable. To achieve this goal, a comprehensive investigation is conducted to establish the most suitable model that can balance accuracy, real-time performance, and energy consumption. The results suggest that a highly efficient wearable obstacle detection solution for the blind is feasible. The established model can deal with the large variations introduced by the person’s natural head turns with a low energy footprint. In addition, our study shows that deep learning models, even tiny ones, are not ideal for this scenario, due to their high computational costs and delays.

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

Learning Low-Energy Consumption Obstacle Detection Models for the Blind

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Peijie Xu,
  • Andy Song,
  • Ke Wang

摘要

The study aims for low-energy consumption models that can detect obstacles on a head-mounted smart device for the vision impaired. Under this setting, long operational time with limited battery power is obviously important, so small but accurate models would be highly desirable. To achieve this goal, a comprehensive investigation is conducted to establish the most suitable model that can balance accuracy, real-time performance, and energy consumption. The results suggest that a highly efficient wearable obstacle detection solution for the blind is feasible. The established model can deal with the large variations introduced by the person’s natural head turns with a low energy footprint. In addition, our study shows that deep learning models, even tiny ones, are not ideal for this scenario, due to their high computational costs and delays.