Research On Wlan Indoor Localization Approaches Based On Manifold Alignment

With the development of portable computational devices and wireless technology, position and navigation systems are attracted more and more attention of researchers and its applications are more and more comprehensive. Satellites are used to position and navigation in outdoor environment, however, satellite positioning navigation systems cannot adapt to indoor environment for lacking of available satellites. Recent years, several indoor positioning systems based on different standards were proposed, including infrared-based, ultrasonic-based and radio frequency(RF) –based. Among them, WLAN based indoor positioning system(WILS) is a hotspot for its realization is depended on the available WLAN without additional infrastructure.To tackle the problems of low efficient and high workload in constructing Radio Map and multi-approaches in processing received signal strengthen(RSS), Radio Map and localizing, algorithms which are cored with manifold alignment(MA) for fast construction of Radio Map and WLAN based indoor localization: direct mapping localization(DML) and KNN jointed with reconstructed Radio Map(re KNN) are proposed in this paper.A brief introduction to the three stratums: ultimates, wireless access points, and positioning servers and its procedures of WILS is presented first. And then Detailed information about fingerprint based localization algorithms(FLAs), both techniques based on matching and probability distribution are formulated. In this essay, methods to build Radio Map are also presented and four mobiles devices: OPPO T29, Google Nexus5, Samsung Note II and Lenovo V450 are selected as sampling and performance testing ultimate to collect all RSS information in HIT-WILS10/12, two WILSs deployed in Communication Research Center, Harbin Institute of Technology(CRC, HIT). And to facilitate the real-time performance of online positioning, AP-KNN algorithm, KNN jointed with affinity propagation(AP), is proposed. According to simulating results in passageway environment in HITWILS12, AP-KNN achieved comparable performance with KNN, and it gained stable performance with single-sample Radio Map as well as the multi-sample Radio Map. Taking localizing accuracy in 3 and 4 meters for example, it is up to 83% and 93% respectively.As for the theoretical analysis, basic conceptions about manifold learning, and typical algorithms, such as LLE, ISOMAP, LDE, and SDE are described at the first place. Then detailed introduction to the theory of coupled metric learning is presented. Starting from the analysis on correlation based and Fisher criterion based coupled metric learning, theoretical descriptions on universal coupled metric learing(UCML) are illustrated. On the basis of UCML, a novel manifold alignment algorithm(UCML-MA) and super resolution analysis based on UCML-MA(SRAMA) are proposed. By applying UCML-MA and SRA-MA approaches into WILS, schemes for WLAN indoor positioning and fast contructiong of Radio Map are showed in this paper. On the one hand, an approach based on SRA-MA for reconstructing high sampling resolution Radio Map from sparsed Radio Map is proposed; on the other hand, two methods for indoor positioning based on UCML are proposed: DML and re KNN.To certify the performance of proposed approaches, four difference mobile devices applied in two WILSs are used. According to simulating consequences, proposed SRA-MA, DML, and re KNN solved problems of current WILS, and realized the combination of core theory of constructing Radio Map and indoor localizaiton. As to the influences of different samples of Radio Map, it has little influenced on the overall performance of KNN, however, the localizing accurcy of DML and re KNN increased by 5%, from 55% and 75% to 60% and 80% when the error radius is within 3 and 4 meters respectively and almost achieved the same performance of original KNN.From the perspective of equal resolution analysis and SRA, the proposed approach for constructing Radio Map via SRA-MA achieved stable performance of DML and re KNN under the condition that the workload of building Radio Map reduced significantly. From the results, SRA-MA method obtained comparable and superb overall performance when sampling only half of reference points(RPs) in the indoor environment, which means the workload reduced at least 50%. As to the highest SRA, it gained comparable performance of DML and re KNN while only taking 1/16 RPs, reconstructing a high resolution Radio Map which is up to 4 RPs/m2 from 0.25 RPs/m2.