Scholarlink

<p>Human activity recognition in video is a popular research topic in present time with an aim to develop futuristic visual surveillance system. Common challenges in this field are sensor motion, cluttered background, illumination, variability in actions, etc. Several deep neural network models have been tested to classify human activities among which CNN continues to be the most popular one. The network complexity is critical for use in computationally limited environment. To address this issue, this paper proposes to consider various CNN models in the field of human activity recognition for use in real time. Various network parameters as well as the input image size are studied to achieve this goal. The UCF101 dataset is used to evaluate the models and it is observed that 3D-MobileNetV2 provides the best performance with a classification accuracy of 91.62%. A comparative study with the available models is also presented.</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

Revisiting human activity recognition using smaller DNN

  • 期刊论文

  • 发表日期：

    2025年6月30日

查看全文
  • Lakhyadeep Konwar,
  • Navajit Saikia,
  • Subhash Chandra Rajbongshi

摘要

Human activity recognition in video is a popular research topic in present time with an aim to develop futuristic visual surveillance system. Common challenges in this field are sensor motion, cluttered background, illumination, variability in actions, etc. Several deep neural network models have been tested to classify human activities among which CNN continues to be the most popular one. The network complexity is critical for use in computationally limited environment. To address this issue, this paper proposes to consider various CNN models in the field of human activity recognition for use in real time. Various network parameters as well as the input image size are studied to achieve this goal. The UCF101 dataset is used to evaluate the models and it is observed that 3D-MobileNetV2 provides the best performance with a classification accuracy of 91.62%. A comparative study with the available models is also presented.