Scholarlink

<p>This study builds upon previous research on lightweight architectures for human activity recognition by optimizing a lightweight hybrid MobileNetV2-CNN model. Enhancing earlier approaches, this work improves feature extraction with additional convolutional layers that capture intricate spatial and temporal details of human actions. To avoid overfitting and strengthen generalization, a strong data augmentation technique of rotation, shifting, shearing, and brightness has been used. The model is evaluated using different metrics such as accuracy, precision, recall, F1-score, area under the curve, root mean squared error and achieves an accuracy of 91.97% on the UCF101 dataset. This model is less complex compared to other deeper networks like ResNet and VGG-19 and achieves faster inference, better classification accuracy, and lower complexity; which makes it more suitable for real-time surveillance, interactive systems and applications, and embedded systems. The results show that lightweight architectures with customized convolutional layers can improve the recognition of human activities and have potential uses in automated monitoring environments, interactive systems, and real-time surveillance.</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

Efficient Human Activity Recognition Using a Hybrid MobileNetV2-CNN Model

  • 期刊论文

  • 发表日期：

    2025年6月27日

查看全文
  • Manjot Rani,
  • Munish Kumar

摘要

This study builds upon previous research on lightweight architectures for human activity recognition by optimizing a lightweight hybrid MobileNetV2-CNN model. Enhancing earlier approaches, this work improves feature extraction with additional convolutional layers that capture intricate spatial and temporal details of human actions. To avoid overfitting and strengthen generalization, a strong data augmentation technique of rotation, shifting, shearing, and brightness has been used. The model is evaluated using different metrics such as accuracy, precision, recall, F1-score, area under the curve, root mean squared error and achieves an accuracy of 91.97% on the UCF101 dataset. This model is less complex compared to other deeper networks like ResNet and VGG-19 and achieves faster inference, better classification accuracy, and lower complexity; which makes it more suitable for real-time surveillance, interactive systems and applications, and embedded systems. The results show that lightweight architectures with customized convolutional layers can improve the recognition of human activities and have potential uses in automated monitoring environments, interactive systems, and real-time surveillance.