CSR: Small: Realism in Activity Recognition for Long Term Sensor Network Deployments

无线传感器网络的研究已经非常成功地为许多应用领域（如家庭医疗保健、建筑节能、基础设施监测、农业和各种环境科学应用）创建了学术测试平台和短期实际部署。然而，在不受控制的环境中长期部署会出现重大的新问题。同样重要的是要注意，大多数这些应用程序执行活动识别。然而，对于长期部署来说，这些活动识别解决方案并不总是足够健壮。因此，这项工作的目标是为家庭部署开发健壮和可靠的活动识别，以解决长期部署的现实问题。为了实现这一目标，需要在以下方面取得新的研究成果：为训练活动识别系统获得标记的地面真相，识别重叠活动，检测发生在家中各个房间的活动，处理丢失的传感器事件和传感器故障，解决多人家庭和访客的问题，以及处理人类行为的演变。这些解决办法必须以整体的方式结合起来。此外，活动识别的效用往往取决于从典型的人类行为中识别异常。异常检测也可能受到长期部署的现实影响，因此也需要加以解决。基础研究方法包括采用数据挖掘、机器学习和其他健壮的技术，这些技术考虑了长期部署的现实情况。解决方案的实用性演示涵盖了从实验室实验到实际的9个月或更长时间的长期部署。这项工作具有广泛的意义，因为为长时间运行的无线传感器网络开发强大的活动识别方案意味着家庭医疗保健和家庭和建筑物节能等应用的改进。家庭保健可以挽救生命，改善老年人和慢性病患者的生活方式，提高他们的独立性，降低医疗费用，并通过纵向研究增加对疾病原因的了解。能源是一种稀缺资源，改进的活动识别可以用来执行节约能源的控制动作。这种节能可以节省资金，降低全球变暖的影响。

Research in wireless sensor networks has been very successful in creating academic testbeds and short term real deployments for many application areas such as home health care, saving energy in buildings, infrastructure monitoring, agriculture, and various environmental science applications. However, significant new problems arise for long term deployments in uncontrolled environments. It is also important to note that most of these applications perform activity recognition. Yet, these activity recognition solutions are not always robust enough for long term deployments. Consequently, the goal of this work is to develop robust and reliable activity recognition for in-home deployments that address the realism of long term deployments. To accomplish this goal requires new research results in: obtaining labeled ground truth for training activity recognition systems, recognizing overlapping activities, detection of activities that occur across rooms of a home, handling missing sensor events and sensor failures, addressing the issues of multiple person homes and visitors, and handling the evolution of human behaviors. These solutions must be combined in a holistic manner. In addition, the utility of activity recognition often depends on recognizing anomalies from typical human behaviors. Anomaly detection can also suffer from the realisms of long term deployments and, therefore, is also addressed. The basic research approach includes employing data mining, machine learning, and other techniques in robust ways that account for realisms in long term deployments. Demonstration of the utility of the solutions spans from lab experiments to realistic long term deployments for 9 months or longer.The broad significance of this work occurs because developing robust activity recognition schemes for wireless sensor networks that operate for long time periods implies improvement in applications such as home health care and saving energy in homes and buildings. Home health care can save lives, provide improved life style and greater independence for the elderly and chronically ill, lower medical costs, and via longitudinal studies, increase understanding of the causes of diseases. Energy is a scarce resource and improved activity recognition can be used to perform control actions that save energy. This energy savings can save money and lower the impact of global warming.