Scholarlink

<p>The recognition of thermal comfort based on occupants’ thermal adaptive behaviors has emerged as a significant area of research within the domain of smart buildings. However, existing vision-based thermal comfort recognition methods inevitably lead to privacy infringement and camera anxiety due to the capture of high-resolution occupant images. To address this, this paper proposes a thermal comfort recognition method based on a lightweight Time-of-Flight (ToF) sensor, termed Shift-RSTA. This method utilizes only a single 4 × 4, extremely low-resolution ToF sensor to achieve non-intrusive, complex thermal comfort recognition through deep learning optimization—a capability previously exclusive to high-resolution vision-based methods. Crucially, it attains recognition accuracy comparable to these visual methods while strictly preserving privacy. The main contributions are as follows. First, to enhance the representation capability of key spatio-temporal information and reconstruct high-resolution depth maps, a Residual Spatial-Temporal Attention module is designed, and an improved Pixels2Depth method, named RSTA-Pixels2Depth, is proposed. Second, an occupant 3D pose estimation method is proposed to precisely extract human body keypoints from the reconstructed depth map. Finally, to address the issues of non-adjacent joint synergistic movements in thermal adaptive behaviors and high computational complexity, a Shift-GCN-based thermal adaptive behavior recognition method is proposed. To support this research, a building indoor environment multimodal thermal adaptive behavior (BIEMTAB) dataset was released. The experimental results show that Shift-RSTA achieves a behavior recognition accuracy of 87.39%, while reducing the perceived privacy risk by over 98% compared to existing vision-based methods.</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

Thermal adaptation behavior-based personalized thermal comfort recognition using a lightweight Time-of-Flight sensor

  • 期刊论文

  • 发表日期：

    2026年4月1日

查看全文
  • Dongdong Shen,
  • Chenguang Ning,
  • Yu Wang,
  • Wenjun Duan,
  • Peiyong Duan

摘要

The recognition of thermal comfort based on occupants’ thermal adaptive behaviors has emerged as a significant area of research within the domain of smart buildings. However, existing vision-based thermal comfort recognition methods inevitably lead to privacy infringement and camera anxiety due to the capture of high-resolution occupant images. To address this, this paper proposes a thermal comfort recognition method based on a lightweight Time-of-Flight (ToF) sensor, termed Shift-RSTA. This method utilizes only a single 4 × 4, extremely low-resolution ToF sensor to achieve non-intrusive, complex thermal comfort recognition through deep learning optimization—a capability previously exclusive to high-resolution vision-based methods. Crucially, it attains recognition accuracy comparable to these visual methods while strictly preserving privacy. The main contributions are as follows. First, to enhance the representation capability of key spatio-temporal information and reconstruct high-resolution depth maps, a Residual Spatial-Temporal Attention module is designed, and an improved Pixels2Depth method, named RSTA-Pixels2Depth, is proposed. Second, an occupant 3D pose estimation method is proposed to precisely extract human body keypoints from the reconstructed depth map. Finally, to address the issues of non-adjacent joint synergistic movements in thermal adaptive behaviors and high computational complexity, a Shift-GCN-based thermal adaptive behavior recognition method is proposed. To support this research, a building indoor environment multimodal thermal adaptive behavior (BIEMTAB) dataset was released. The experimental results show that Shift-RSTA achieves a behavior recognition accuracy of 87.39%, while reducing the perceived privacy risk by over 98% compared to existing vision-based methods.