Scholarlink

<p>Aiming at the problem that traditional human motion recognition technology cannot extract human features well, the study proposes a human motion recognition model by combining Discrete Cosine Transform and long short-term memory network error correction. The model also utilizes Graph Convolutional Networks to further process graph-structured data and coordinate the relationships between movements. Experimental results show that the human motion prediction algorithm based on Discrete Cosine Transform and Graph Convolutional Networks, achieves an average prediction error of 91.32&#xa0;mm. Furthermore, validation analysis of the proposed hybrid human motion recognition model reveals that the recognition accuracy for lying-down actions remains at 98.88%, which indicates a high level of accuracy. The above results show that the proposed hybrid model of human motion recognition has better predictive ability and a smaller error range. The study provides effective solutions for fields such as wearable sensor devices and contributes to better processing of human motion sensor data in the future.</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

Human motion recognition based on discrete cosine transform and error correction

  • 期刊论文

  • 发表日期：

    2025年9月5日

查看全文
  • Yang He,
  • Yanchun Wang

摘要

Aiming at the problem that traditional human motion recognition technology cannot extract human features well, the study proposes a human motion recognition model by combining Discrete Cosine Transform and long short-term memory network error correction. The model also utilizes Graph Convolutional Networks to further process graph-structured data and coordinate the relationships between movements. Experimental results show that the human motion prediction algorithm based on Discrete Cosine Transform and Graph Convolutional Networks, achieves an average prediction error of 91.32 mm. Furthermore, validation analysis of the proposed hybrid human motion recognition model reveals that the recognition accuracy for lying-down actions remains at 98.88%, which indicates a high level of accuracy. The above results show that the proposed hybrid model of human motion recognition has better predictive ability and a smaller error range. The study provides effective solutions for fields such as wearable sensor devices and contributes to better processing of human motion sensor data in the future.