Scholarlink

<p>Unobtrusive detection of human activity and analysis of activity trend is the most elementary but challenging part of Ambient Assisted Living (AAL). Usually, it requires multi-parameter monitoring round the clock demanding huge computational and storage requirements. Efficient algorithms to handle such data burden are of foremost importance for practical implementation of AAL. In general, technologies for assisted living identify dynamic human activity recognition (HAR) as the first step. In this work a novel method for HAR from a wrist worn tri-axial accelerometer data is proposed. The HAR is performed by an artificial neural network (ANN) classifier which is modeled using phase space reconstruction (PSR) based features extracted from the accelerometer data. Observing the chaotic nature of accelerometer data for different activities of daily living (ADLs) and during transition of activities, PSR is employed to represent raw data into image form. Significant information is extracted from these images by calculating quadrant wise black pixels density. These feature set is tested by a Bayesian Regulation back-propagation ANN model for activity classification that results in 96.4% accuracy with HMP dataset and 93% accuracy with USC-HAD dataset. This proves the applicability of the proposed feature extraction method for a resource efficient HAR model which uses only one tri-axial accelerometer irrespective of mounting position either on wrist or waist.</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

PSR image-based novel human activity sequence monitoring towards ambient assisted living

  • 期刊论文

  • 发表日期：

    2026年3月9日

查看全文
  • Rohan Mandal,
  • Saurabh Pal,
  • Uday Maji

摘要

Unobtrusive detection of human activity and analysis of activity trend is the most elementary but challenging part of Ambient Assisted Living (AAL). Usually, it requires multi-parameter monitoring round the clock demanding huge computational and storage requirements. Efficient algorithms to handle such data burden are of foremost importance for practical implementation of AAL. In general, technologies for assisted living identify dynamic human activity recognition (HAR) as the first step. In this work a novel method for HAR from a wrist worn tri-axial accelerometer data is proposed. The HAR is performed by an artificial neural network (ANN) classifier which is modeled using phase space reconstruction (PSR) based features extracted from the accelerometer data. Observing the chaotic nature of accelerometer data for different activities of daily living (ADLs) and during transition of activities, PSR is employed to represent raw data into image form. Significant information is extracted from these images by calculating quadrant wise black pixels density. These feature set is tested by a Bayesian Regulation back-propagation ANN model for activity classification that results in 96.4% accuracy with HMP dataset and 93% accuracy with USC-HAD dataset. This proves the applicability of the proposed feature extraction method for a resource efficient HAR model which uses only one tri-axial accelerometer irrespective of mounting position either on wrist or waist.