Unhealthy habits lead to a growing population of hemodialysis patients. The single treatment of hemodialysis is about four hours long. Therefore, patients usually go to the toilet during treatment and need to be checked for safety. However, existing fall detection techniques are often limited by factors such as privacy, signal interference, and the like. In this paper, we propose RF-IDH tackle the above issues, a dedicated system for detecting falls caused by complications in hemodialysis patients using RF signal. In RF-IDH, after collecting the signal, we process the collected data by three functional module clusters, namely signal preprocessing, residual feature extraction, hemodialysis patient’s fall detection, all of which are well-designed to achieve high performance in patient’s fall detection. In particular, we design a residual feature extraction (RFE) algorithm based on the hemodialysis patient safety process model, and the fall detection of hemodialysis patients is treated as a machine learning problem where four classification models are built via learning residual feature space. We implement our system on commercial off-the-shelf RFID devices and compared the evaluation metrics of four different methods in terms of system performance, efficiency, robustness, and latency. The evaluation results show that our proposed RF-IDH that optimizes the 2NN–RFE method achieves superior performance compared to other methods.

不健康的习惯导致越来越多的血液透析患者。血液透析的单次治疗大约需要四个小时。因此，患者通常在治疗期间上厕所，需要检查其安全性。但是，现有的跌倒检测技术通常受到诸如隐私，信号干扰等因素的限制。在本文中，我们提出RF-IDH解决上述问题，这是一种专用系统，用于使用RF信号检测血液透析患者并发症引起的跌倒。在RF-IDH中，我们在收集信号后通过三个功能模块集群处理收集的数据，即信号预处理，残差特征提取，血液透析患者的跌倒检测，所有这些都经过精心设计以实现高性能的患者跌倒检测。尤其是，我们基于血液透析患者安全过程模型设计残差特征提取（RFE）算法，并将血透患者的跌倒检测视为机器学习问题，其中通过学习残差特征空间构建了四个分类模型。我们在商用现成的RFID设备上实施我们的系统，并比较了四种不同方法在系统性能，效率，鲁棒性和延迟方面的评估指标。评估结果表明，与其他方法相比，我们提出的优化2NN–RFE方法的RF-IDH具有更高的性能。血液透析患者的跌倒检测被视为机器学习问题，其中通过学习残差特征空间建立了四个分类模型。我们在商用现成的RFID设备上实施我们的系统，并比较了四种不同方法在系统性能，效率，鲁棒性和延迟方面的评估指标。评估结果表明，与其他方法相比，我们提出的优化2NN–RFE方法的RF-IDH具有更高的性能。血液透析患者的跌倒检测被视为机器学习问题，其中通过学习残差特征空间建立了四个分类模型。我们在商用现成的RFID设备上实施我们的系统，并比较了四种不同方法在系统性能，效率，鲁棒性和延迟方面的评估指标。评估结果表明，与其他方法相比，我们提出的优化2NN–RFE方法的RF-IDH具有更高的性能。