Research On Indoor Mapping And Localization Algorithm For User Multi-pattern Behavior Sensing

The wide availability of mobile devices makes the user based applications promising.Specifically,the Location Based Services(LBS)have attracted significant attention due to its personalized,convenient,and smart user experience.Meanwhile,accurate mapping and localization algorithm plays a crucial role in a satisfying LBS.As the traditional mapping and localization approach,the location fingerprinting based method relies on the radio map constructed in offline phase to achieve online matching localization,but the calibrated radio map construction is time-consuming and labor-intensive especially in large scenarios.In recent years,there are many researches focucing on cost-efficient location estimation.Especially,the semi-supervised localization constructs a mapping relation between signal and physical spaces based on a few calibrated location fingerprints and completes satisfying positioning.Meanwhile,the semi-supervised localization is unsuitable for cases with increasing calibration-free data,thus the location fingerprinting independent estimation defines the correlation among crowdsourcing signal sequences to achieve demonded localization performance without any site survey.Relying upon the location fingerprinting based positioning,this thesis focuses on the quantification between calibrated fingerprints and localization performance as well as on the location estimation approaches with calibration-free data.In particular,the augmented manifold alignment based positioning with calibrated location fingerprints is proposed and the expansion of closed-form solution to multi-dataset cases is deduced for the sake of higher performance with more location related information.Meanwhile,as for the disposal of calibration-free data,an edge detection based signal correlation model in Wireless Local Area Network(WLAN)is proposed.On the other hand,the time-stamp relation in signal sequences is utilized for better representation of calibration-free data,and the hotspot mapping is determined with behavior transfer probability.In all,three contributions of this thesis are listed as follows.1.In order to quantify the relation between calibtated fingerprints and localization performance,the augmented manifold alignment based location estimation is proposed.The localization objective function between Received Signal Strength(RSS)and physical coordinates is constructed whose corresponding closed-form solution of transform matrices converts the online collected RSS into a uniform manifold space.Furthermore,the expansion of closed-form solution to muti-dataset cases is deduced for higher positioning performance with more location-aware information.2.As for the sporadically collected RSS sequences,an image where each pixel stands for the correlation between corresponding two RSSs is generated which is utilized to identify the correlated RSS blocks via edge detection.And the finer-grained relation in detected RSS blockcs is determined with proposed sequencing model.Finally,the mapping from constructed signal logic graph to physical logic graph is established with the definition of adjacent degree.3.The time-stamp relation in RSS sequences is considered for better representation of calibration-free data.Specifically,the line segments are constructed which are then clustered with density based clustering to establish the motion pattern based signal logic graph.And reling on the knowledge of behavior transfer probability,the hotspot mapping between signal and physical spaces is established for final localization.