CRII: CHS: WiFi-Based Human Behavior Sensing and Recognition System for Aging in Place

随着“婴儿潮”一代步入老年，美国的老年人口在未来几年将出现相当大的增长。研究一致证实，大多数老年人希望尽可能长时间地呆在自己家里。因此，迫切需要以家庭为基础的辅助生活技术，能够持续而不显眼地监测日常生活活动（ADLs）并检测异常事件，以降低老年人护理成本并提高生活质量。目前的人类行为监测系统通常基于摄像头、智能手机/可穿戴设备或环境传感器，这些系统存在成本高、侵犯隐私等基本限制，无法广泛部署。PI在这个项目中的目标是建立在他之前工作的基础上，建立一个研究项目，研究一种新的方法，利用现在无处不在的商用家庭WiFi信号来监测adl和检测异常事件。其核心思想是，不同的人类活动导致WiFi信号发生不同的变化；通过分析这些变化，可以识别引起变化的活动。这项工作将在美国和国外产生广泛的社会影响，通过在单主体和多主体场景中为基于wifi的人类行为感知和识别做出贡献的新技术和系统。如果新系统有效，它将为老年人提供一种非侵入性、无设备、低成本和保护隐私的辅助生活技术。PI将把该项目的研究成果整合到他的本科和研究生课程中，以及K-12教育计划中；此外，本研究开发的硬件和软件将是开源的，在此项目中收集的数据集将提供给其他人进行进一步的研究。PI计划利用从WiFi信号中提取的细粒度物理层信道状态信息（CSI）作为统一方案的基础，以监测老年人每天在家中进行的最常见的固定和移动活动。他将通过跟踪由呼吸引起的微小但周期性的胸部运动来检测静止活动，他将提取频域特征，即使在不同的运动方向或不同的位置也能稳健地识别相同的运动活动。PI将开发马尔可夫模型来识别复杂的adl，他将利用呼吸和身体运动信息来检测异常行为，包括意外跌倒和睡眠紊乱，这些都是与衰老有关的潜在问题。最终，PI将扩展他的技术，以识别多个人同时执行的adl。为了成功实现这些目标，PI将需要克服许多重大的技术挑战，例如检测由于固定活动（如在电脑前工作或坐在沙发上看电视）而导致的WiFi信号的微小变化。鲁棒性识别以不同方式或在不同地点进行的相同移动活动（例如，打扫房屋）也会很棘手，因为不同位置的不同移动方向或不同布局会对WiFi信号产生不同的干扰。

As "baby boomers" age, the United States will experience considerable growth in its elderly population over the coming years. Studies consistently confirm that the majority of older adults would prefer to remain in their own homes for as long as possible. Therefore, there is a critical need for home-based assisted living technologies capable of continuously yet unobtrusively monitoring activities of daily living (ADLs) and detecting abnormal events, both to reduce the cost of elder care and to enhance the quality of life. Current human behavior monitoring systems for aging in place, which are typically based on cameras, smartphone/wearable devices, or ambient sensors, have fundamental limitations such as high cost and invasion of privacy that prevent them from being widely deployed. The PI's objective in this project is to build on his prior work to establish a research program to investigate a new approach to aging in place that harnesses the now-ubiquitous commercial home WiFi signals to monitor ADLs and detect abnormal events. The central idea is that different human activities cause different changes in WiFi signals; by analyzing these changes, the activity that caused the change can be recognized. This work will have broad societal impact both within the United States and abroad, by contributing to new techniques and systems for WiFi-based human behavior sensing and recognition in both single-subject and multi-subject scenarios. If the new system is effective, it will provide a non-intrusive, device-free, low-cost and privacy-preserving assisted living technology for aging in place. The PI will integrate research results from this project into both his undergraduate and graduate courses, as well as the K-12 education program; furthermore, the hardware and software developed in this research will be open-source, and the dataset collected during this project will be made available to others for further research.The PI plans to exploit the fine-grained PHY layer Channel State Information (CSI) extracted from the WiFi signals as the basis for a unified scheme for monitoring both the most common stationary and moving activities performed daily by older adults in their homes. He will detect stationary activities by tracking the minute but periodic chest movements caused by breathing, and he will extract frequency domain features to robustly recognize the same moving activity even with different movement directions or at different locations. The PI will develop Markov models to recognize complex ADLs, and he will leverage the breathing and physical body movement information to detect abnormal behaviors including accidental falls and disturbed sleep that are potential issues relating to aging in place. Ultimately, the PI will extend his techniques to recognize ADLs of multiple persons performed at the same time. To successfully achieve these objectives, the PI will need to overcome a number of significant technical challenges, for example detecting minute changes in the WiFi signal due to stationary activities such as working at a computer or watching TV while seated on a sofa. Robustly recognizing the same moving activity (e.g., housecleaning) performed in different ways or at different locations will also be tricky, because different movement directions or different layouts at different locations cause different disturbances to WiFi signals.