【作者】 路知远；

【导师】 周平； 陈香；

【作者基本信息】 中国科学技术大学 ， 生物医学工程， 2014， 博士

【摘要】 目前老龄化形式严峻,随时随地的健康监护和自然直观的人机交互成为了迫切的需求。穿戴式健康监护系统可以在日常生活中提供随时随地的健康监护而不会干扰被监护者的正常生活。基于穿戴式设备的多模态人机交互系统可以为老年人提供自然的人机交互,方便老年人使用电子设备。本文从基于穿戴式设备的健康监护和人机交互两方面开展研究,并将研究成果与实验室的研究基础整合,构建了老年人健康监护与交互平台。本文为移动健康监护设计了一个脉搏血氧仪。为了在不影响用户日常生活的情况下实现实时的心率和血氧饱和度监护,本文采用嵌入型的设计,将反射式传感器嵌入到掌上电脑(PDA)的后盖中。这一设计有别于传统的采用透射式传感器的指套型脉搏血氧仪,可以从身体多个部位进行测量,更适用于穿戴式和移动场景下。由于PDA后盖没有遮蔽环境光的作用,而反射式传感器容易受到环境光干扰,本文设计了新颖的光源调度策略和斩波网络以分离信号并滤除环境光。鉴于反射式传感器采集到的信号非常微弱且容易受到高频噪声干扰,本文设计了一个放大滤波网络,大幅提高了信噪比。针对反射式嵌入型脉搏血氧仪信号个体差异大且不稳定的特点,本文设计了一套基于波谷检测的心率和血氧饱和度计算算法,该算法可以根据输入信号自适应地调整参数,并针对移动设备或穿戴式设备做了简化。本文还针对反射式传感器设计了PPG信号(photoplethysmogram)判别算法,提高了测量结果的可靠性。临床标定和测试实验显示,此脉搏血氧仪和临床使用的设备对血氧饱和度的测量结果相差(0.3+0.9)%,心率相差(0.4±2.4)次/分。配对T检验的结果显示,此脉搏血氧仪可以达到与临床设备相同的精度(p<0.05),证明了该嵌入型脉搏血氧仪的可行性。本文还针对移动设备设计了一种可穿戴的实时手势交互系统,该系统包含一个自行研制的手势捕获设备,一套高效、可移植的手势识别算法,以及针对移动设备开发的手势识别和交互控制程序,支持使用19种预定义的动作或者用户自定义的动作来控制移动设备。为了允许用户自然、随意地执行手势,本文在已有的活动段提取算法的基础上,提出了对表面肌电信号(SEMG)和加速度信号(ACC)分别提取活动段的方法,并针对ACC信号活动段提取的难点,设计了以SEMG信号标记的方法。该活动段提取算法有效提高了ACC信号的一致性。本文提出了两种新的加速度特征,提高了手势识别的准确率；提出了基于改进型动态时间规整算法和基于状态转移模型的两种ACC信号识别算法,其计算量小,且识别准确度与主流的基于隐马尔科夫模型(HMM)的算法相当；设计了基于打分的决策级融合策略,充分考虑了不同特征的权重以及SEMG信号和ACC信号活动段在时间上通常不同步的问题。在用户有关和用户无关的测试中,19类手势识别率分别达到了95.0%和89.6%。交互测试和用户体验问卷调查的结果表明该系统可在移动设备和可穿戴设备上提供实时的手势交互,并提供良好的用户体验。鉴于穿戴式设备在健康监护和人机交互领域具有独特优势,本文在上述研究成果的基础上,构建了一个基于穿戴式设备的老年人健康监护与交互平台。该平台可以实时监测老年人的心电、心率、血氧饱和度、皮肽水分、体脂含量等生理参数,提供日常保健功能；该平台实现的跌倒检测功能可以及时发现意外情况并进行报警；基于全球卫星定位系统和步行者航位推算算法的室内外无缝定位功能可以在任何时间定位被监护者。同时,该平台针对手机和电视设计的手势交互功能,为老年人提供了简单方便的交互方式,极大方便了老年人的生活。本论文研究工作得到了国家863高科技研究发展计划“基于肌电传感器和加速计的手势交互设备研究”(2009AA012322)、国家自然科学基金项目“基于表面肌电的中国手语手势识别研究”(60703069)、中央高校基本科研基金“基于情境感知的多源信息分析与理解”(WK2100230002)的资助。更多还原

【Abstract】 Due to the growing population aging, the development of ubiquitous healthcare and natural human-computer interaction (HCI) is of great demand. Wearable healthcare systems are able to offer ubiquitous healthcare services in daily life without affecting users’daily activities. Multimodal HCI based on wearable devices is able to offer natural interaction experience and an easy way to operate smart devices for the elderly. This dissertation focuses on wearable device-based healthcare and interaction, and on these bases, develops a healthcare system with such functions, especially designed for the elderly.This dissertation introduces a prototype of pulse oximeter designed for mobile healthcare. It is designed to be embedded into the back cover of a personal digital assistant (PDA) to offer the convenient measurement of both heart rate (HR) and arterial oxygen saturation (SpO2) for home or mobile healthcare applications. As opposed to conventional transmission pulse oximeters with finger cots, a reflection pulse oximeter is implemented, which can work on various parts of the body, thus facilitating its applications in mobile and wearable use case. Considering that reflection sensor is easily interfered by ambient light due to the lack of shading effect on flat surface of the back cover of PDAs, a novel lightening modulation, along with a novel circuit module named chopper network, is designed for signal separation and ambient light removal. Aiming at the obtained weak signal amplification and denoising, a novel filtering amplifier is designed to overcome the influence of high-frequency interferences and to improve signal-noise-ratio. Furthermore, a method based on trough detection for improved HR and SpO2estimation is proposed with appropriate simplification for its implementation on wearable devices or mobile devices like PDA. In addition, with the purpose to process unstable PPG (photoplethysmogram) signals acquired by the embedded sensor, adaptive thresholds and parameters are applied to the HR and SpO2estimation algorithm. A PPG validation algorithm is also designed to reject invalid PPG or non-PPG signals towards the embedded oximeter to make measurement results more reliable. Clinical experiments are carried out to calibrate and test our oximeter. Our prototype oximeter can achieve comparable performance to a clinical oximeter according to the statistical analysis using paired T-test, revealing insignificant difference between the two oximeters at (0.3±0.9)%in SpO2measurement and (0.4±2.4) beats per minute in HR measurement (p<0.05). The experimental results demonstrate the feasibility of this proposed prototype.In this dissertation, a wearable gesture-based interaction prototype for mobile phone is developed. More specifically, a homemade wearable gesture capturing device is designed to acquire acceleration (ACC) and surface electromyography (SEMG) signals; an algorithm framework is proposed to process the signals for gesture recognition, and an application program is developed to realize gesture-based real-time interaction. Users are able to manipulate the mobile phone using19predefined gestures or even personalized ones. In the novel segmentation scheme based on the prior one designed for only SEMG signals, a gesture has two asynchronous signal segments, one from SEMG signals and the other from ACC signals. SEMG marked ACC signal segmentation algorithm is proposed to overcome the segmentation challenge for ACC signals. ACC signals shows improved internal consistency when processed by this novel segmentation scheme, facilitating the free and natural performance of gestures for users without cumbersome restraints like grasping hand simultaneously during waving arm. Two new features for ACC signals are proposed and achieve considerable improvement of recognition accuracy. Classifiers based on improved dynamic time wrapping (DTW) and state transition model (STM) respectively are designed to recognize ACC signals. Both can achieve the comparable accuracies to those of hidden Markov model (HMM), but cost much lower computational power. Considering the SEMG segment and ACC segment are usually asynchronous, a score-based fusion scheme is proposed to make final recognition decisions by combining the both using predefined weights. The proposed system can achieve an average accuracy of95.0%in user-dependent testing and89.6%in user-independent testing, offering practical solution to real-time gestural interaction on mobile or wearable devices. Such promising performance during the interaction testing, along with positive user experience from a questionnaire survey, demonstrates the feasibility of our prototype.Taking advantage of the wearable devices in mobile healthcare and intelligent HCI, a healthcare and interaction system is developed for the elderly. It can offer ubiquitous and real-time monitoring of electrocardiograph, HR, SpO2, skin moisture, and body fat in daily life for healthcare. Fall detection service activates the alarm after it detects any emergency. The position information is calculated by our novel location service based on both global position system (GPS) and our pedestrian dead reckoning (PDR) algorithms. Gestural interaction designed for mobile phone and smart TV offers simple and natural control using wearable devices, which facilitates the interaction between the devices and the elderly population.This work was supported in part by National High Technology of Research and Development Program of China (863Program) under Grant No.2009AA01Z322, Fundamental Research Funds for the Central Universities of China under Grand No. WK2100230002, and National Nature Science Foundation of China (NSFC) under Grant No.60703069.更多还原