Indoor Mobility Map Construction And Localization Based On Wi-Fi-SLAM Pixel Template Matching

The fast development of Internet of Things(IoT)has driven an imperious demand for Location-based Service(LBS)such as shopping guidance,individual localization in emergency rescue and car tracing.In this circumstance,the indoor localization systems with high accuracy and low cost are particularly significant.Due to widely-deployed Wi-Fi infrastructure in various public places including school campus and shopping malls,the location fingerprinting based Wi-Fi localization has been one of the most priorities for the cost-efficient indoor localization.However,the process of fingerprint database construction is time-consuming and labor-intensive,and it will be invalid due to the time-varied environment.Meanwhile,crowdsourcing,which is an effective method to replace the large-scale off-line signal collection,has attracted much attention.The previous researches about the crowdsourcing paths based indoor localization usually need to plan the moving trajectories with fixed starting and ending positions beforehand.However,in practical applications,the crowdsourcing paths collected by users are often random,and the corresponding starting and ending positions are also unknown.In response to these compelling problems,this thesis constructs a crowdsourcing motion map with unmarked random crowdsourcing motion trajectories,and proposes a pixel template matching algorithm to achieve accurate mapping between the crowdsourcing RSS signals and the physical locations in target environment.The proposed approach can not only save the labor and time cost of fingerprint database construction in the offline stage,but also make the database comprising of crowdsourcing motion paths more practicable in real applications,which has important research significance.The main research work is as follows.First,this thesis builds a crowdsourcing based mobility map of the target environment.In concrete terms,based on unlabeled crowdsourcing sensor data,the relative location information of crowdsourced RSS signal sequences is obtained with the Pedestrian Dead Reckoning(PDR)algorithm.Then,the correlation sequencing is utilized to recompose the crowdsourcing motion path sequences with similar positions.After that,a density based clustering algorithm is adopted to divide the segmented path sequences into several motion areas,which consist of the mobility map of target environment.With the continuous enrichment of crowdsourcing paths,the constructed mobility map gradually approaches the global map.Meanwhile,with the collected multi-dimensional motion sensor data,some indoor important positions such as the area near staircases are firstly selected for possible performance enhancement.Second,this thesis constructs the relation between the crowdsourcing mobility map and the physical locations in target environment.Specifically,the pixel template matching algorithm is proposed to determine the physical locations of the constructed mobility map in the global map,and then the accurate mapping relations between the crowdsourcing RSS signal and the physical locations are achieved.Finally,the Kalman filter algorithm is performed to integrate the crowdsourcing sensor data with the collected RSS signal,which achieve the complementary advantages of the Wi-Fi based and motion sensors based localization systems.Moreover,the robust weight function is constructed to reduce the influence of coarse error,which further improves the positioning accuracy of the proposed uncalibrated data based localization system.