Sensor Fusion for Health-tracking Wearable Devices and Internet of Things

近年来，随着智能手机和智能手表/表带及其消费应用的出现，可穿戴设备的传感器融合取得了长足的进步。对这些设备的各种用户行为检测和分类问题的自动解释的准确性和范围已显著提高。新的传感器技术、新颖的问题定义和表示、对大量多种传感器数据源类型的访问以及强大的信号处理和机器学习算法的复兴推动了这一进步。然而，目前的算法缺乏产生准确数据的能力，这些数据允许在现实生活环境中详细了解健康状况，并将这些数据转化为有意义的医疗结果。临床试验中的健康跟踪可穿戴设备面临着巨大的障碍和挑战，必须克服这些障碍和挑战，才能在医疗保健中采用和使用这些设备。 拟议研究计划的长期目标是开发基本和原则性的方法，通过用于个性化、预测性、预防性和精确化医疗保健物联网(IoT)的多模式健康跟踪可穿戴设备，细粒度地了解复杂的人类活动。我们的短期目标是开发基于手腕的惯性可穿戴设备的自动活动检测/分类/评估算法，目标是准确和持续地监测自由生活的老年人的健康状况(我们的目标领域)。为了实现这一目标，我们提出了五个具体的短期目标：(I)用于健康跟踪可穿戴设备的传感器融合：方位估计；(Ii)用于健康跟踪可穿戴设备的体力活动分类：深度学习；(Iii)用于健康跟踪可穿戴设备的基于活动的自适应/动态功率管理；(Iv)使用健康跟踪可穿戴设备的身体移动性评估：步行速度估计；以及(V)使用健康跟踪可穿戴设备的体力活动能量消耗估计。我们在这项研究中开发的这些算法将使数据丰富的多模式健康跟踪可穿戴设备能够产生必要的数据准确性和解释，以便它们被采用并用于老年护理(短期)和其他领域(长期)。 我们目前正处于医疗级、健康跟踪和高级可穿戴设备的最早阶段，拟议的开发可能会加快时间框架。拟议的工作将促进新一代可穿戴设备的实现，这些可穿戴设备足够可靠和准确，可以实现医疗保健物联网。它还将使这些设备早日进入临床试验并获得FDA的批准。

Great strides have been made within sensor fusion of wearable devices in recent years with the advent of smartphones and smartwatches/bands and their consumer applications. Accuracy and scope for automatic interpretation of a variety of user action detection and classification problems for these devices have improved dramatically. Progress has been driven by new sensor technologies, novel problem definitions and representations, access to volumes of multiple sensor data source types, and the resurgence of powerful signal processing and machine learning algorithms. However, current algorithms lack the capability to produce accurate data that permits detailed health understanding in the real-life setting and transforming such data into meaningful medical outcomes. Health-tracking wearable devices in clinical trials face significant hurdles and challenges that must be overcome in order for them to be adopted and used in healthcare. The long-term goal of the proposed research program is to develop fundamental and principled approaches for fine-grained understanding of complex human activity from multi-modal health-tracking wearable devices for personalized, predictive, preventive, and precision healthcare Internet of Things (IoT). Our short-term goal is to develop automatic activity detection/classification/assessment algorithms for wrist-based inertial wearable devices that are targeted for accurate and continuous health monitoring of free-living older adults (our targeted domain). To achieve this, we propose five specific short-term objectives: (i) sensor fusion for health-tracking wearables: orientation estimation, (ii) physical activity classification for health-tracking wearables: deep learning, (iii) activity-based adaptive/dynamic power management for health-tracking wearables, (iv) physical mobility assessment using health-tracking wearables: walking speed estimation, and (v) physical activity energy expenditure estimation using health-tracking wearables. These algorithms that we develop in this research will enable data-rich and multi-modal health-tracking wearables to produce the necessary data accuracy and interpretation in order for them to be adopted and used in elder care (short-term) and other domains (long-term). We are currently in the very earliest stages of medical grade, health-tracking, advanced wearables and the time frame could be accelerated by the proposed developments. The work proposed will facilitate a new generation of wearables that are reliable and accurate enough for realization of healthcare IoT. It will also enable early inroads for these devices being employed in clinical trials and meeting FDA approvals.