Research On WLAN Semi-supervised Indoor Positioning And Location Fingerprint Updated Algorithms

With the widely usage of handheld intelligent terminal devices and coverage of wireless networks,the requirement for location based services(LBSs)indoor and outdoor manifests a tendency of increasing and rapid expansion.LBSs have been increasingly used in areas such as emergency rescue,medical service,travel navigation and surveillance,shown favorable market prospects.The demand to provide precise and real-time location information for LBSs ties positioning technique to its service provided closely.With the usage of existing network infrastructure,recently,the RSS indoor positioning technique based on WLAN is considered to be the researching hotspot in indoor positioning area.Meanwhile,it satisfies the positioning process as a variety of portable mobile terminals using pure software mode.However,in view of complex and changeable of indoor radio propagation environment will result RSS signal in serious time-varying characteristic.Considering the mapping relationship between signal space and physical locations is non-unique,that affects the localization accuracy greatly.In turn,it brings many problems to indoor positioning technique based on location fingerprinting,and presents more challenges for majority of researchers.The key technologies of location fingerprinting indoor positioning based on WLAN are studied in this paper.Start with main factors that influence on positioning performance,the paper focuses on deficiencies of fingerprint matching algorithm to decrease the influence of RSS time-varying property and reduce positioning computational complexity,while achieving balance between validity and reliability of the positioning system.Make full use of easy-collected unlabeled samples,to some extent,the fingerprint matching technology based on WLAN is improved through adopting the theories of semisupervised manifold learning,cluster analysis,and data mining.The major research contents and innovative points of this paper can be summarized as follows:Firstly,classical indoor positioning systems are compared and analyzed.Focusing on the positioning principle,construction cost,positioning accuracy and applicable scenarios are described in detail.Then,on this basis,the WLAN based location fingerprinting positioning technology is further studied.We discuss the key technologies in two phases respectively,including fingerprint collection and location computation.Base on summarizing the current research status and existing problems of fingerprinting positioning algorithm,clustering analysis,feature extraction and fingerprint data update.Part of the aforementioned algorithms are introduced and studied,in order to provide theoretical basis for the subsequent algorithm improvement.Secondly,dimensionality reduction and feature extraction of high-dimensional signal is studied.Focusing on noise and redundant information will be brought under a wireless network environment deployment with dense access point.Then the performance is affected by using the RSS signals as positioning algorithm input data directly.A novel positioning algorithm(named SDE)based on semi-supervised learning discriminate embedding feature extraction is proposed in this paper.It can project high dimensional data to lower one and excavate low dimensional manifold structure from high dimensional data space.In the case of improving reliability of RSS signals on the premise of maintain its discriminant ability,positioning accuracy is improved.The proposed algorithm takes advantage of random sample data collecting by mobile users as volunteers,and then extracts the most discriminating lower positioning feature.The proposed algorithm not only has low complexity but also reduces the online computation cost to save energy consumption on mobile terminals.Thirdly,reducing RSS signal position calculation searching space algorithm in indoor positioning system is investigated.According to the number of reference points and dimensions of signal are the main parameters affecting fingerprint matching algorithm searching area,which are components of location fingerprint database construction during the offline stage.Therefore,it is essential to constraint the feature extraction learning model into sub-regions through RPs partitioning clustering algorithm.In view of low classification accuracy of the existing clustering algorithm and invalid to overcome the lack of signal nonlinear and non-stationary,the semi-supervised affinity propagation clustering algorithm(named SAPC)combines with c means is proposed.The adjusting process of similarity matrix through making use of labeled data is firstly presented,which participate in clustering analysis.Then the clustering results adjustment is performed.Compared with the other algorithms,the results show that the proposed algorithm accuracy improves significantly,meanwhile,it keeps better balance relationship between the positioning accuracy and computational complexity.Finally,the reconstruction and updating algorithm of fingerprint database is studied.According to the static characteristic of fingerprint database brings large deviation to positioning results,the fingerprint updating algorithm based on mobile user's trajectory is proposed.The hidden Markov model(HMM)established based on subscriber locations is used to update the fingerprint database timely.Specifically,it treats the real-time RSS as the observation sequence of HMM,and then the hidden relevant location can be obtained through solving the HMM parameters.The proposed algorithm not only reduces labor consumption requirement on fingerprint database reconstruction and updating,but also benefits indoor positioning system to popularize and exploit in large-scale.Compared with the algorithms using static fingerprint database,the proposed algorithm improves positioning accuracy and robustness of the system,at the same time,it better overcomes signal fluctuations caused by environmental changes.