Remote monitoring of older adults and detecting dangers in the state of human health have become essential elements of modern telemedicine. Falls are a frequent reason for deaths or post-traumatic complications in the elderly. Therefore, the early detection of falls can be crucial for the survival of a person or for providing necessary support. However, telemedicine data centers require scalable computing and storage resources for the growing number of monitored people. Dedicated approaches that allow for minimal data transmission of strictly interesting cases are also required.

In this paper, we show a scalable architecture of a system that can monitor thousands of older adults, detect falls, and notify caregivers. Scalability tests that disclose requirements to enable large scale system operations were also performed. Moreover, we validated several Machine Learning models to evaluate their suitability in the detection process. Among the tested models, Boosted Decisions Trees resulted in the best classification performance. We also experimentally tested the detection of falls inside a Cloud-based data center and on an Edge IoT device. Results of tests on the device-to-cloud data transmission confirmed that significant reduction in the size of stored and transmitted data can be achieved while performing fall detection on the Edge.

远程监控老年人并检测人体健康状况中的危险已成为现代远程医疗的基本要素。跌倒是老年人死亡或创伤后并发症的常见原因。因此，跌倒的早期检测对于一个人的生存或提供必要的支持至关重要。但是，远程医疗数据中心需要可扩展的计算和存储资源，以供越来越多的受监视人员使用。还需要专门的方法，以允许在非常有趣的情况下进行最少的数据传输。

在本文中，我们展示了一个系统的可扩展体系结构，该体系结构可以监视成千上万的老年人，检测跌倒并通知护理人员。还执行了可扩展性测试，该测试公开了实现大规模系统操作的要求。此外，我们验证了几种机器学习模型，以评估它们在检测过程中的适用性。在测试的模型中，Boosted Decisions Trees产生了最佳的分类性能。我们还通过实验测试了在基于云的数据中心内以及Edge IoT设备上的跌倒检测。设备到云数据传输的测试结果证实，在Edge上执行跌倒检测时，可以实现存储和传输数据大小的显着减少。