Scholarlink

<p>This paper proposes SenseDesk, a device designed to measure the center of gravity and weight on a desk to estimate the user and their desk activities. SenseDesk, equipped with four load cells on an acrylic plate, calculates the center of gravity and weight at 80 Hz and utilizes machine learning to estimate activities and users. We selected twelve desk activities, such as “Lying face down on the desk,” “Using a mouse,” and “Writing on paper with a pen,” and evaluated their estimation accuracy. Using the random forest method, we achieved an F value of 0.775 for Within-Individual activity estimation and 0.671 for Between-Individual activity estimation. Additionally, the F value was 0.716 for user estimation. We then aggregated these activities into four groups of high-level activity categories representing common work states: “PC work,” “Writing work,” “Smartphone operation,” and “Break.” This abstraction improved the classification performance, yielding an F value of 0.888 for Within-Individual and 0.796 for Between-Individual estimation.</p>

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

SenseDesk: a table-top device for activity and user recognition via center of gravity and weight analysis

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月7日

查看全文
  • Yukino Sato,
  • Kodai Ito,
  • Takumi Hasegawa,
  • Takashi Oshima,
  • Yuichi Itoh

摘要

This paper proposes SenseDesk, a device designed to measure the center of gravity and weight on a desk to estimate the user and their desk activities. SenseDesk, equipped with four load cells on an acrylic plate, calculates the center of gravity and weight at 80 Hz and utilizes machine learning to estimate activities and users. We selected twelve desk activities, such as “Lying face down on the desk,” “Using a mouse,” and “Writing on paper with a pen,” and evaluated their estimation accuracy. Using the random forest method, we achieved an F value of 0.775 for Within-Individual activity estimation and 0.671 for Between-Individual activity estimation. Additionally, the F value was 0.716 for user estimation. We then aggregated these activities into four groups of high-level activity categories representing common work states: “PC work,” “Writing work,” “Smartphone operation,” and “Break.” This abstraction improved the classification performance, yielding an F value of 0.888 for Within-Individual and 0.796 for Between-Individual estimation.