Scholarlink

<p>The applicability of gesture recognition technology in a variety of scenarios within the field of human–computer interaction has been demonstrated due to the flexibility and non-contact nature of the technology. In particular, thermal imaging technology is not limited by lighting conditions, which is effectively reduces risk by capturing only thermal radiation rather than detailed visual features. In this study, a low-resolution 32 × 24 pixels end-to-end embedded infrared thermal image camera gesture recognition system is developed. A thermal image gesture dataset of 4500 images is constructed to train and evaluate the system. This study investigates the effects of incorporating the spatial transformer network (STN) attention mechanism on improving gesture recognition accuracy. Thus, a new method combines lightweight convolutional neural networks and STN is proposed. Additionally, the proposed method achieves a recognition accuracy of 98.5% and inference time of only 59&#xa0;ms per frame on embedded devices when tested on a self-made infrared thermal image sign language gesture dataset, outperforming mainstream lightweight models.</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

Static gesture recognition based on thermal imaging sensors

  • 期刊论文

  • 发表日期：

    2025年3月13日

查看全文
  • Zhi-Yuan Zhang,
  • Hao Ren,
  • Hao Li,
  • Kang-Hui Yuan,
  • Chu-Feng Zhu

摘要

The applicability of gesture recognition technology in a variety of scenarios within the field of human–computer interaction has been demonstrated due to the flexibility and non-contact nature of the technology. In particular, thermal imaging technology is not limited by lighting conditions, which is effectively reduces risk by capturing only thermal radiation rather than detailed visual features. In this study, a low-resolution 32 × 24 pixels end-to-end embedded infrared thermal image camera gesture recognition system is developed. A thermal image gesture dataset of 4500 images is constructed to train and evaluate the system. This study investigates the effects of incorporating the spatial transformer network (STN) attention mechanism on improving gesture recognition accuracy. Thus, a new method combines lightweight convolutional neural networks and STN is proposed. Additionally, the proposed method achieves a recognition accuracy of 98.5% and inference time of only 59 ms per frame on embedded devices when tested on a self-made infrared thermal image sign language gesture dataset, outperforming mainstream lightweight models.