基于多分类器投票的Wi-Fi室内定位系统的设计与实现

发布时间：2018-03-18 22:48

本文选题：无线信号　切入点：模式识别　出处：《华中科技大学》2014年硕士论文　论文类型：学位论文

【摘要】：在许多实际场景中，人们对室内定位都有迫切的需要，如商场中的购物定位导航、大型机场火车站的进站口、厕所、商店等重要设施的导航等等。然而，室内定位技术却没有像常用的室外定位技术，如GPS定位、移动网络基站定位等技术一样得到广泛商业化，原因在于：GPS的卫星信号和移动网络基站信号到建筑物的传播距离太远，而且在室内环境下，人流不断发生变动，室内环境相对复杂，信号因为反射、折射等现象而导致衰减严重，使得定位精度很低。但是，因为受到室内视野局限的影响，室内定位精度通常要大大高于室外定位精度，才能初步满足实际应用的需要。而当前采用的室内定位技术，难以解决把室外定位技术直接应用到室内定位环境中出现的问题，因而大多仅存在在实验研究环境中。当前，Wi-Fi作为建筑物中最常见的多对多的信号网络，能够应用模式识别技术，特别是统计模式识别技术来进行不同地址的位置指纹的区分。本文应用模式分类技术，，将室内定位问题转化为一个分类问题，进而完成室内定位。本文在前人工作的基础上主要做了以下工作：在手机端和PC端实现了整套室内定位系统的解决方案，整个分为信号采集部分、模板训练部分、室内定位部分。由于单个分类器难免会出现分类错误的情况，我们利用概率论数学证明了两个问题：（1）在每个分类器的分类正确率均高于50%的情况下，多个分类器进行多数投票决策的分类正确率一定会高于单个分类器的分类正确率。（2）分类器的数量越多，多个分类器进行多数投票决策的分类正确的概率就越发高于单个分类器分类正确的概率。由此我们完成了系统设计和实现，并在系统测试中，实际证实了以上两个结论的正确性，最终使得我们的室内定位系统的定位精度能够接近加州大学伯克利分校的室内定位实验结果。
[Abstract]:In many practical situations, people have an urgent need for indoor positioning, such as shopping positioning navigation in shopping malls, entrance to large airport and railway stations, toilets, shops and other important facilities. However, However, indoor positioning technology is not as widely commercialized as the common outdoor positioning techniques, such as GPS positioning, mobile network base station positioning and so on. The reason is that the satellite signal from the GPS and the mobile network base station signal spread far from the building. Moreover, in the indoor environment, the flow of people is constantly changing, the indoor environment is relatively complex, the signal is attenuated seriously because of reflection and refraction, and the positioning accuracy is very low. However, because of the limitation of indoor visual field, Indoor positioning accuracy is usually much higher than outdoor positioning accuracy, which can meet the needs of practical application. However, the current indoor positioning technology is difficult to solve the problem of applying outdoor positioning technology directly to indoor positioning environment. Therefore, most of them only exist in the experimental research environment. Currently, Wi-Fi, as the most common many-to-many signal network in buildings, can use pattern recognition technology, especially statistical pattern recognition technology, to distinguish location fingerprints of different addresses. The problem of indoor positioning is transformed into a classification problem, and then indoor positioning is completed. On the basis of the previous work, this paper mainly does the following work: the solution of the whole indoor positioning system is implemented in the mobile phone and PC, the whole system is divided into signal acquisition part, template training part, Indoor positioning part. As a single classifier will inevitably have classification errors, we use probability mathematics to prove two problems:. 1) when the classification accuracy of each classifier is higher than 50%, the classification accuracy of multiple classifiers for majority voting decision will be higher than that of single classifier. 2) the more the number of classifiers, the higher the probability of classifying the correct classification of multiple classifiers for majority voting decision is higher than that of single classifier. Therefore, we have completed the design and implementation of the system, and in the system test, the correctness of the above two conclusions has been verified. Finally, the positioning accuracy of our indoor positioning system is close to the results of the indoor positioning experiment at the University of California, Berkeley.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN92

【相似文献】