Scholarlink

<p>Human Activity Recognition (HAR) is aimed at deducing characteristics and patterns in the activities of people, which is basic research issue in digital healthcare. Several approaches have been proposed for HAR, although the techniques fail to extracts the temporal and spatial feature suitable for attaining good results. Hence, this paper devises a novel Hybrid Efficient Xception network (HyEx-Net) for HAR. The implementation of the approach is elaborated as follows: Initially, the input color image is obtained from dataset and is subjected to feature extraction. On the other hand, a depth image is also acquired from database, which is transformed into the extraction of skeleton joints. The extracted skeleton joints are fed to the extraction of local features, where features like displacement vectors, and relative positions are extracted. Then, the extracted features from the input depth and color images are concatenated and allowed for HAR. Subsequently, HAR is performed utilizing HyEx-Net, which is designed by incorporation of the Xception model and Efficient Network (EfficientNet). The HyEx-Net is compared with other traditional schemes, and the HyEx-Net recorded superior accuracy, True Positive Rate (TPR), and True Negative Rate (TNR) with values of 92.79%, 94.79%, and 90.88% respectively.</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

Hyex-net: hybrid efficient Xception network for human activity recognition using multimodalities

  • 期刊论文

  • 发表日期：

    2025年2月25日

查看全文
  • Saurabh Gupta,
  • Rajendra Prasad Mahapatra

摘要

Human Activity Recognition (HAR) is aimed at deducing characteristics and patterns in the activities of people, which is basic research issue in digital healthcare. Several approaches have been proposed for HAR, although the techniques fail to extracts the temporal and spatial feature suitable for attaining good results. Hence, this paper devises a novel Hybrid Efficient Xception network (HyEx-Net) for HAR. The implementation of the approach is elaborated as follows: Initially, the input color image is obtained from dataset and is subjected to feature extraction. On the other hand, a depth image is also acquired from database, which is transformed into the extraction of skeleton joints. The extracted skeleton joints are fed to the extraction of local features, where features like displacement vectors, and relative positions are extracted. Then, the extracted features from the input depth and color images are concatenated and allowed for HAR. Subsequently, HAR is performed utilizing HyEx-Net, which is designed by incorporation of the Xception model and Efficient Network (EfficientNet). The HyEx-Net is compared with other traditional schemes, and the HyEx-Net recorded superior accuracy, True Positive Rate (TPR), and True Negative Rate (TNR) with values of 92.79%, 94.79%, and 90.88% respectively.