Research On The Crowdsensing And Low-power Bluetooth Based Indoor Localization Technologies

With the enhancement of mobile terminals in sensing and computing abil-ities, the inertial navigation based localization scheme has drawn much atten-tion. Due to the limited precision of inertial sensors, such localization scheme usually needs to be calibrated through indoor landmarks. However, the deploy-ment, excavation and maintenance of landmarks restrict the long-running and self-adaptive abilities of indoor localization systems.Nowadays, with the popularity of wireless mobile terminals, a new sens-ing paradigm has emerged in the research field of IoT, namely "crowdsensing".In crowdsensing networks, large amount of ubiquitous users, performing as the basic sensing units, can collaboratively complete the complex social tasks. In-spired by the new concept of crowdsensing, where every person in the monitor-ing area can be used as a sensing component, we are motivated to excavate and maintain the indoor landmarks via crowdsensing. This thesis focus on making the indoor localization system applied in the practice and adaptively run for a long time. We propose a serious of novel models and methods for solving the following three key issues: how to optimize the inertial navigation and land-mark based localization system via crowdsensing; how to excavate the potential landmarks via crowdsensing; how to monitor the status of indoor landmarks via crowdsensing. The main contributions of this thesis are as follows:(1) Crowdsensing based optimization scheme for the inertial navigation.In practice, user diversity, device heterogeneity and surrounding interference are main factors which significantly reduce the precision of inertial navigation and restrict it to be applied practically. Given this, we propose crowdsens-ing based optimization scheme for the inertial navigation. Firstly, for the step lengths' difference issue caused by participant diversity and device heterogene-ity, we propose a personalized step length estimation model. We obtain the walking distance between neighboring landmarks and build the correspond-ing dataset which contains step counts, acceleration information and walking distance via crowdsensing schema, consequently, the personalized walking pa-rameters could be learned from the acquired dataset. In addition, for the low accuracy issue of direction inference caused by the surrounding disturbance,we choose some appropriate areas for correcting users' direction, called "Di-rection Correcting Area (DCA)". Specifically, in such DCA, large amount of crowdsensing participants share their inferred direction information, including both correct directions and wrong directions, then we utilize the correct in-ferred users to adjust the wrong users, in order to improve the global accuracy of direction inference.(2) Recognition mechanism for indoor landmarks based on the combina-tion of multiple sensing signals. Whether the signatures of landmark will be affected by the users' operation modes and the surrounding environments, is the metric for measuring the landmark recognition degree. We define the im-proved landmark signature as the combination of motion signal and bluetooth signal. We benefit from the bluetooth signal to solve the problem of high false positive rate caused by users' operating mode diversity. Meanwhile, the combi-nation of motion signal can solve the interference of surrounding for the blue-tooth signal. The experimental results show that the signature defined by the combination of motion signal and bluetooth signal can improve the recognition accuracy of landmarks, which ensures the calibration precision of landmarks.(3) Robust estimation based radio map building scheme. Device hetero-geneity and outliers are two main bottlenecks for building high-quality radio map. Given this, we introduce linear regression model based robust estima-tion to acquire the linear mapping coefficients from the uploaded measure-ments which include outliers. In this way, we can project the signal measure-ments which acquired by different devices into a uniform space. After that, we leverage multivariate robust statistics to determine the signal strength for each position in the trajectory-dense area. Meanwhile, we acquire the weights of the participants via estimating the uploaded signal measurements of each user.Thus, for the trajectory-sparse areas which contain few signal measurements,we can acquire a relatively accurate signal strength with the help of partici-pants' weights information.(4) Context-aware based bluetooth monitoring scheme. Estimating the re-al status of bluetooth devices according to the uploaded signals information by users, becomes a big challenge for our monitoring system. Given this, we design a crowdsensing based monitoring framework which combines the mov-ing and static schemas of participants. Based the framework, we analyze the sensing context that affects the accuracy of the uploaded signal information and quantify the influence degree. Considering these factors, we build a prob-abilistic model which takes into account the error rate of participants and the sensing context effect to judge whether the status of device changes. We pro-pose a weighted fixed point calculation technique to estimate the real status of iBeacons and error rate of participants. The experimental results demonstrate that the proposed estimation approach outperforms two popular algorithms, i.e.,three-estimates algorithm and OtO EM algorithm.In summary, based on crowdsensing schema, this thesis proposes a series of models and methods for landmark and inertial navigation based localization system, excavation of indoor landmarks and monitoring landmark status. In order to further verify the effectiveness of the proposed methods and models,we conduct a case study for measuring indoor 4G signal via our localization system, and achieve ideal results.