不同场景下基于手机加速度传感器的人体活动端点检测研究

发布时间：2019-01-09 06:45

【摘要】：随着可穿戴设备的普及和普适计算的不断发展,越来越多的研究将关注点聚焦到加速度传感器数据上来。用户与终端便携设备逐步呈现紧耦合的态势,与此同时,智能手机的运算能力在近年来有飞跃性的提高,如何利用好智能移动设备高度的携带时间占比以及计算能力是重要的突破口。而加速度传感器数据作为描述人体活动信息的重要组成部分,其携带的步态特征、行为模式等信息对于人体活动语义理解具有至关重要的意义。由智能手机收集到的用户活动数据不仅时间跨度长而且规模庞大,对数据的处理与存储提出了挑战。因此,本文以人体活动端点检测为出发点,将长时间复杂人体活动加速度数据当中活动起始点与终止点的提取做为目标,提出了两种不同场景下的端点检测算法。具体地,所做工作如下:一、针对作为信号采集设备的智能手机计算能力和内存资源的限制,提出了一种改进的双门限人体活动端点检测算法。针对三维空间矢量数据定义了三种不同的短时过零率。该算法可以进行粗粒度的行为活动检测,避免上传全部数据,节省大量的网络传输带宽以及服务器端的存储资源。二、以改进的双门限判别人体活动端点检测算法为核心技术,提出了一种适用于客户端资源受限条件下人体活动数据(加速度)传输策略。该策略包括相应的动态采样策略、上传窗口判定、数据存储队列以及上传队列的建立等内容。通过本文提出的传输策略,可以有效降低传输成本以及数据存储成本。三、针对服务器端具体人体行为识别过程中需要更精确的活动段提取的需求,提出了基于加速度数据信息熵的人体活动端点检测算法。并且为避免在计算三轴数据均方根时对加速度矢量方向信息的丢失构建了三维加速度信源联合信息熵模型。该算法相较于双门限算法虽然复杂度更高,但检测结果更加精确,适用于在人体行为识别前期做为数据预处理步骤用以提取实际活动段数据。通过验证实验,证明了本文提出的双门限方法可以有效降低数据的产生量,传输策略可以节省传输成本,信息熵检测算法可以有效提高复杂情况下行为识别的准确率。
[Abstract]:With the popularization of wearable devices and the development of pervasive computing, more and more researches focus on acceleration sensor data. At the same time, the computing ability of smart phone has been improved by leaps and bounds in recent years. It is an important breakthrough how to make good use of the high carrying time ratio and computing power of intelligent mobile devices. As an important part of describing human activity information, acceleration sensor data, such as gait characteristics, behavior patterns and so on, are of great significance to the understanding of human activity semantics. The user activity data collected by smart phone not only has a long time span but also has a large scale, which poses a challenge to data processing and storage. Therefore, this paper takes the detection of human moving endpoint as the starting point, taking the extraction of the starting point and the ending point of the moving point from the long-time complex acceleration data of human body as the target, and proposes two kinds of endpoint detection algorithms under different scenarios. Specifically, the work is as follows: firstly, an improved dual-threshold human mobile endpoint detection algorithm is proposed for the limitation of the computing power and memory resources of the smart phone as a signal acquisition device. Three different short time zero crossing rates are defined for three dimensional space vector data. The algorithm can detect coarse-grained behavior, avoid uploading all data and save a lot of network bandwidth and storage resources on the server side. Secondly, based on the improved dual threshold discriminant human activity endpoint detection algorithm, a new method is proposed to transmit human activity data (acceleration) under the condition of limited client resource. The strategy includes the corresponding dynamic sampling strategy, upload window decision, data storage queue and the establishment of upload queue. The transmission cost and data storage cost can be effectively reduced by the proposed transmission strategy. Thirdly, aiming at the need for more accurate extraction of human activity segment in the process of human behavior recognition on the server side, an algorithm based on acceleration information entropy is proposed to detect human activity endpoint. In order to avoid the loss of direction information of acceleration vector when calculating the root mean square of triaxial data, a three-dimensional information entropy model of acceleration source is constructed. Compared with the two-threshold algorithm, the proposed algorithm is more complex, but the detection result is more accurate. It is suitable for extracting the actual active segment data as a data preprocessing step in the early stage of human behavior recognition. Through the verification experiment, it is proved that the double threshold method proposed in this paper can effectively reduce the amount of data generated, the transmission strategy can save the transmission cost, and the information entropy detection algorithm can effectively improve the accuracy of behavior recognition in complex cases.
【学位授予单位】：辽宁大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP212

【参考文献】