Research On Location Fingerprint Localization Based On Matrix Completion And Bayesian Compressed Sensing

With the advent of the era of Internet of things, people's demand for location-based services is growing. As an effective complement to the outdoor GPS positioning technology, indoor positioning technology and its related research has attracted much attention in recent years. Location fingerprint localization method based on the received signal strength indication has become mainstream method of wireless indoor positioning technology and get more and more extensive application because of its high positioning accuracy, little influence of indoor factors, and flexibility to be realized. Location fingerprint location method includes two stages: off-line location fingerprint database construction and online location fingerprint matching and positioning. But the workload of the fingerprint information is usually large, which leads to the low efficiency of the construction of the fingerprint database. At the same time, due to the influence of the time variation of the wireless signal and the instability of the hardware, the on-line positioning accuracy is limited. The above characteristics of location fingerprint localization impact on its practicability to a certain extent, which needs to be further studied.In this paper, after the in-depth analysis of the problem in the off-line sampling phase and online positioning stage of the location fingerprint localization method, focusing on indoor location fingerprint location fingerprint database construction algorithm and the online phase of the positioning algorithm, the theoretical research and experimental study were carried out. The main research works are as follows:1.In the off-line stage of the fingerprint location method, the location fingerprint database which can accurately reflect the characteristics of the location need to be established according to the received wireless signal strength value of different reference points within the location region, which is the basis for the effective implementation of indoor positioning. The traditional location fingerprint database construction method mainly includes the entire collection and the interpolation method. The entire collection method is of great workload, which leads to the low efficiency of the construction of fingerprint database. Interpolation method usually can not make full use of global information, so the calculation of the location fingerprint database is often not accurate enough to affect the accuracy of positioning directly. In view of the above, in this paper, an efficient method to construct the offline location fingerprint database by Matrix Completion theory was proposed, using the location fingerprint information of a small amount of reference points collected in the location area, a Low-rank Matrix Completion model was established, and the model was solved by using the Singular Value Thresholding algorithm, then, the complete location fingerprint database can be reconstructed in the location area. At the same time, in this paper, the Backtracking Search Optimization Algorithm was introduced to solve the problem of the optimal solution and the poor smoothness of the traditional matrix filling theory, which takes the minimum norm of kernel norm as the fitness function. The optimization process of Matrix Completion algorithm is improved by using the global search characteristic and the convergence rate of the Backtracking Search Optimization Algorithm, which is a new novel location fingerprint database construction algorithm called BSA-SVT. On the basis of the reconstruction accuracy of the location fingerprint database, the workload of the off-line sampling phase can be reduced effectively.2. Online positioning stage is the core of location fingerprint location method, at present, there are many kinds of theories can be used in the design of positioning algorithm.The traditional K-Nearest Neighbor algorithm need to position the fingerprint data, one by one, resulting in a large amount of calculation, and the accuracy is limited. In recent years, some scholars have begun to try to introduce the new Compressed Sensing theory into the indoor localization algorithm, the Compressed Sensing theory transforms the localization problem into a sparse signal reconstruction problem, and it is proved to be one of the relatively high precision algorithms. On the basis of this development, the localization algorithm based on Bayesian Compressed Sensing is more accurate, efficient and better than the traditional Compressed Sensing algorithm. But the core of the traditional Bayesian Compressed Sensing is a kind of basis function selection mechanism based on fast RVM, in view of the problem that RVM is easy to fall into local optimum and can not get the optimal solution, in this paper, we introduce a Differential Evolution algorithm with relatively few adjusting parameters for high dimensional problems,a localization algorithm based on DE-BCS is proposed. DE-BCS algorithm based on the idea of Differential Evolution algorithm into the RVM basis function selection process, increase the diversity of the basis function selection, to ensure that the "update" operation in the RVM operation for the first time, to avoid the possibility of falling into the local optimal,on the basis of not increasing the operation time, the positioning accuracy is improved.According to the above algorithm, this paper carried out the research through theoretical analysis, computer simulation and experimental research combination, choosing a specific indoor scene in the campus, built a wireless indoor positioning experiment platform based on RFID, the algorithm proposed in this paper were sufficiently verified experimentally, further proof that the algorithm in this paper is of theoretical value and practical significance, provide a reference for the practical application of the algorithm.