Towards a well-planned, activity-based work environment: Automated recognition of office activities using accelerometers
Towards a well-planned, activity-based work environment: Automated recognition of office activities using accelerometers
复制标题
DOI：
10.1016/j.buildenv.2018.07.051
复制
发表时间：
2018-10-15
影响因子：
7.4
通讯作者：
Koo, Choongwan
中科院分区：
文献类型：
--
作者：
Cha, Seung Hyun;Seo, Joonoh;Koo, Choongwan
随着工作需要更多的动态和协作环境，基于活动的工作环境（ABW）越来越受到关注。然而，如果不清楚办公室工作人员的活动模式，贸然采用ABW可能会带来各种不利影响，如工作站短缺和不适当的工作站安排。在这方面，使用加速计自动识别办公室活动可以帮助建筑师了解活动模式，从而实现ABW环境的有效空间规划。然而，据我们所知，主要需要人工活动的静态办公室任务尚未得到承认。因此，这项研究旨在确定使用加速计同时识别七种静态和非静态办公室活动的可行性。进行了一项实验研究，以收集来自七个活动的加速度数据。在不同的窗口大小下，分析了五种分类器（即k-最近邻、判别分析、支持向量机、决策树和包围式分类器）的准确性。最高的分类准确率，在96.1%，实现了包围分类器，窗口大小为4.0秒。此外，所有办公活动的召回率和准确率均大于0.9，显示出较高的预测可靠性。这些发现有助于建筑师更系统、更全面地理解静态和非静态的办公活动模式。
As work has come to require more dynamic and collaborative settings, activity-based work (ABW) environments have claimed increasing attention. However, without a clear understanding of office-workers' activity patterns the rash adoption of ABW may entail a variety of adverse effects, such as work-station shortages and inappropriate work-station arrangements. In this regard, the automated recognition of office activities with an accelerometer can help architects to understand activity patterns, thereby enabling effective space planning for the ABW environment. To the best of our knowledge, however, static office tasks requiring mainly manual activities have not yet been recognized. The study thus aims to determine the feasibility of recognizing seven static and non-static office activities simultaneously using an accelerometer. An experimental investigation was carried out to collect acceleration data from the seven activities. The accuracy of five classifiers (i.e. k-Nearest Neighbor, Discriminant Analysis, Support Vector Machine, Decision Tree and Ensemble Classifier), was analyzed with different window sizes. The highest classification accuracy, at 96.1%, was achieved by Ensemble Classifier, with a window size of 4.0 s. In addition, all office activities showed recall and precision greater than 0.9, demonstrating high prediction reliability. These findings help architects to understand static and non-static office activity patterns more systematically and comprehensively.