基于可见光的室内定位与智能感知

发布时间：2018-03-20 03:14

本文选题：移动计算　切入点：室内定位　出处：《中国科学技术大学》2017年博士论文　论文类型：学位论文

【摘要】：移动计算的蓬勃发展和智能设备的更新换代,颠覆性地变革了现代人类的生活方式。用户随身携带的智能设备和日益普及的物联网设施不仅仅可以随时访问和处理信息,还可以实时感知用户行为和周围环境的变化。这些智能设备在知晓1)用户的位置信息以及2)行为信息之后,可以协助拓展单个人类个体有限的感知范围。为了实现这个目标,学术界和工业界提出了大量的1)定位系统与2)行为感知系统来满足日益增长的需求。但是,在定位系统方面,当前主流的基于WiFi信号的室内定位系统易受到来自环境变化的严重影响,而信源数目的不足也会严重影响其精度和覆盖范围。与此同时,在感知系统方面,主流的基于计算机视觉的行为感知与识别系统面临着较高的部署开销和高精度图像带来的隐私问题。可见光是人眼和智能设备均可感知的电磁波谱片段,是最为普遍的,无处不在的电磁波。可见光具有较强的方向性,不易受到环境变化的影响,且在室内场景中密集部署的照明用光源直接可用作信源,降低了系统部署的开销和复杂性。可见光的优良特性使其有望成为感知用户位置和行为的下一个主要媒介。然而,设计基于可见光的定位和感知技术存在若干难点。首先,有别于WiFi信号,智能设备采集到的接收光照强度(RLS,received light strength)仅仅是一个标量,缺乏用于标识信源身份的标志(光源的编号);这导致了一个相似的RLS可以在多个室内物理位置被捕捉到。其次,有别于WiFi信号,RLS对智能设备朝向,高度极度敏感;这使得影响RLS的变量空间具有较高维度,从而难以通过传统的非测距方法来建立RLS和室内物理位置的关联。再次,若干用户行为的高自由度(例如,手掌有23个自由度)使得借由粗粒度的接收光照强度信息恢复出细粒度的用户行为成为一个困难的任务。面临这些挑战和问题,本研究的主要研究和创新成果如下:·设计了一个基于可见光的室内定位原型系统,Lightitude。本系统的核心观察在于,不同室内物理位置的可见光信源的光照强度有差别,并且这个差别对智能设备上搭载的光传感器来说是非常明显的。鉴于此,智能设备采集到的信号强度可以对应特定的室内物理位置。本研究首先提出了一个适用于实际场景的可见光传播模型来描述RLS和设备当前状态的关联,并基于此设计了一个以粒子滤波器为核心的定位方案。真实环境中的实验证实了 Lightitude系统能够精确地确定用户在实验场景中的位置。与此同时,Lightitude系统在面对障碍物的遮挡,用户的意外行为以及日光的影响之下依然具有较强的鲁棒性。·设计了一个基于可见光的室内推荐原型系统,LiLoc。本系统的核心观察在于,特定的通路具有独特的光强特性。这样,服务提供商可以通过简单的一次性场景遍历,使用微小的人力代价建立场景信息数据库。借助用户在大型室内场景中的移动性,LiLoc系统首先确定用户在场景信息数据库中的相对位置,从而在线下向用户推送当前位置所对应的推荐服务信息。进一步地,LiLoc系统可以以高精度识别三种用户的典型行为(行走,驻足,伸手查看),从而可以进一步帮助用户离开室内场景后的辅助线上推荐。真实环境中的实验验证了 LiLoc系统具有较高的推荐精度。·设计了一个基于可见光的细粒度手掌轮廓恢复及移动监测技术,HandSense。本系统的核心观察在于,作为一个实心的障碍物,人的手掌既可以阻碍从光源(例如,日光,LED)到光传感器的可见光通路,也可以反射微弱的反向散射信号。这两种效应均可被预先部署的光传感器阵列所感知。这样,特定的手掌状态对应着一组特定的接受光照强度集合。实验者在服务器上实现了可见光遮挡监测和可见光反向散射监测,结合解剖学模型所描述的手掌固有限制,设计了一个启发式的优化算法,用于实现手掌的轮廓恢复和姿态监测,为进一步的高精度室内推荐提供客观依据。真实环境中的实验证实了 HandSense系统能够精确地感知用户手掌的轮廓,移动,以及姿态。
[Abstract]:The vigorous development of mobile computing and intelligent equipment upgrading, subversive to the revolution of modern human lifestyle. Intelligent equipment carried by the client and the increasing popularity of networking facilities not only can access and process information at any time, you can also change the real-time perception of user behavior and the environment. These smart devices in the known position 1) the user information and behavior information, 2) can assist the development of a single individual limited sensing range. In order to achieve this goal, academia and industry put forward a lot of 1) 2) positioning system and behavior perception system to meet the growing demand. However, in the positioning system, the current mainstream indoor positioning the system based on the signal of WiFi susceptible to serious impact from changes in the environment, rather than the number of source will seriously affect the precision and coverage. At the same time, in the The perception system, mainstream behavior perception and recognition system of computer vision facing privacy issues high deployment overhead and high precision image based on visible light is electromagnetic spectrum fragments and smart devices can sense the human eye, is the most common, the ubiquity of electromagnetic waves. The visible light has strong directivity and influence not susceptible to changes in the environment, and in the indoor scene in densely deployed light sources can be used as a direct source, reduce the cost and complexity of the deployment of the system. The excellent characteristics of visible light which is expected to become the next major media user position and behavior with perception. However, the design positioning of visible light and sensing technology there are a number of difficulties. Based on the first, is different from the WiFi signal, receiving intelligent equipment to collect light intensity (RLS, received light strength) is just a scalar, for lack of identification Mark the identity of the source (source code); this leads to a similar RLS can be captured in a plurality of indoor physical location. Secondly, different from the WiFi signal, RLS of smart devices to highly sensitive; this makes the variable space effect of RLS has high correlation dimension, and it is difficult to through the distance the traditional method to build RLS and indoor physical location. Again, a high degree of freedom some user behavior (for example, the palm has 23 degrees of freedom) made by receiving coarse-grained light intensity information to restore the fine-grained user behavior becomes a difficult task. Facing these challenges and problems, the main research and the innovation results of this study are as follows: the design of a prototype system based on indoor positioning of visible light, is to observe Lightitude. the core of this system, the physical location of the different indoor visible light source light intensity difference, And the difference is very obvious in the light sensor mounted on the intelligent equipment. In view of this, the signal intensity of intelligent equipment collected can correspond to the specific indoor physical location. The visible light propagation model, this paper puts forward a suitable for the actual scene to describe the association between RLS and the current state of equipment, and based on this the design of a particle filter core positioning scheme. The real environment experiment proved that Lightitude system can accurately determine the location of the user in the scene. At the same time, the Lightitude system in the face of occlusion avoidance obstacle, under the unexpected behavior of users as well as the effect of sunlight still has strong robustness and design. A recommendation system based on the indoor visible light is observed, LiLoc. the core of this system, the specific pathway has unique characteristics of light intensity. This service Service providers can one-time scene traversal simple, use the tiny human cost to build the scene information database. With the user in a large indoor scene in LiLoc mobility, the system first determines the relative position of the user in the scene information in the database, so online users push to the position corresponding to the information recommendation service. Further, typical the behavior of LiLoc system with high precision can identify three kinds of users (walk, stop, hand, view) so it can help online help users leave the indoor scene after further recommended. In real environment experiment verifies the accuracy of recommendation LiLoc system has high. Design a fine-grained palm contour recovery and visible light mobile monitoring technology based on the observation of HandSense. is the core of this system, as a solid obstacle, the palm can hinder from light source (for example, sun, LED) to the visible light channel light sensor, also can be reflected back scattering signal. The weak light perception sensor array of these two effects can be pre deployment. In this way, the specific state corresponds to a specific set of palm received light intensity collection. Experimenters in the server implementation the visible light shielding monitoring and visible light backscatter monitoring, combined with the anatomical model described by palm inherent limitations, a heuristic optimization algorithm design, for the realization of palm contour recovery and posture monitoring, provide objective basis for high precision indoor further recommended. The real environment experiment proved that HandSense system can accurately the perception of a user's palm contour, mobile, and attitude.

【学位授予单位】：中国科学技术大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP212.9

【相似文献】