Research On Key Technologies Of WLAN Indoor Localization Using Crowdsourcing Method

With the rapid proliferation of wireless networks and intelligent mobile terminals,the Location Based Service(LBS)has been applied to all aspects of social activities and shown a good development prospects and huge market space in e-commerce,health care,emergency rescue,logistics management and other aspects.As the Wireless Local Area Networks(WLAN)has been deployed in public places such as shopping malls,college schools,hospitals and airports,the Received Signal Strength(RSS)based WLAN localization is considered to be the preferred technology for indoor localization and navigation and LBS since it exploits the existing WLAN infrastructure.However,due to the complexity of indoor electromagnetic environment,RSS data shows high uncertainty,therefore,the traditional WLAN localization system needs to collect a large number of RSS values to establish the radio map in the offline phase to suppress the uncertainty of the RSS signal,which leads to huge costs of time and labor.By assigning the work to a lot of volunteers,the crowdsourcing based WLAN indoor localization system can greatly reduce the workload in the offline phase,but there are still some key techniques in this system need to be further researched.Based on the extensive research of the WLAN indoor localization system,the problems in crowdsourcing system are gradually emerging: first of all,whether in the offline phase and the online phase,the mobile devices used by different users are different,which caused serious device diversity problem;second,a great number of unlabeled RSS values,which cannot be effectively used by the traditional localization algorithms,are brought to the localization system by the crowdsourcing method,and the weighted graph in Graphbased Semi-Supervised Learning(G-SSL)is easily affected by noise,which resulting in a serious waste of resources and decline of localization accuracy;finally,although the workload are reduced by using the crowdsourcing method,the number of RSS measurements collected by each volunteers in each location is smaller,the RSS fluctuations could affect the accuracy of the radio map significantly,which will weaken the performance of localization system.Aiming to solve the existing problems mentioned above,the main work and innovation of this paper are as follows:Firstly,a linear regression algorithm against device diversity problem for crowdsourcing based WLAN indoor localization system is proposed in this thesis.Because different antennas,chips and signal processing algorithms are used in different mobile devices,the RSS values collected at the same location and time from the same AP are different.However,the relationship between RSS values collected by different devices is linear,using the linear regression algorithm,all the RSS values can be mapped into the same signal space.Since the outliers appear in the collected RSS values frequently and seriously affect the performance of the linear least squares(LLS)algorithm,the fast least trimmed squares(FAST-LTS)algorithm is used in this thesis.As a result,we can obtain a uniform radio map in the offline phase and more accurate positioning results.After applying LR algorithm in the crowdsourcing systems,the device diversity problem is solved automatically and we verify the LR algorithm using the theoretical study of probability of error detection.Simulation results verify the effectiveness of the proposed algorithm,and all the RSS data are mapped into the same signal space.Secondly,a Compressive Sensing and G-SSL based WLAN indoor localization algorithm is proposed in this thesis,which cannot only reduce the collection workload in the offline phase,but also effectively use the unlabeled data to improve the performance of localization system.When we build the radio map,the sensors equipped in the mobile device cannot work well in all the time,and even no sensors are equipped in the mobile devices.As a result,lots of the RSS values are not labeled in the radio map.A small number of labeled data and a large number of unlabeled data can be used by semi-supervised learning to estimate the coordinates in the online phase,which greatly reduce the workload in the offline phase and achieve accurate positioning result.Based on the Compressive Sensing method,the weighted matrix could be reconstructed more accurately,and result in a further improvement on the performance of the G-SSL and the localization system.Simulation results show that the localization accuracy can be improved by accurately reconstructing the weighted graph in semi-supervised learning algorithm and effectively using the unlabeled data.Finally,a signal propagation model based RSS data smoothing algorithm is proposed in this thesis,which greatly improve the accuracy of the radio map.In the radio map,the intrinsic relationship can be mined between fingerprints in both coordinate space and data space,and these relationships can be mapped to each other by using signal propagation model.The outlier values can be identified and adjusted using the Signal Propagationbased Outlier Reduction Technic(SPORT)and the RSS values in the adjacent locations.Combine the more accurate relationship mined by the Compressive Sensing(CS)and Semi-supervised Learning with the signal propagation model,we can get a smoother radio map and more accurate positioning results.Simulation results show that the proposed method can effectively improve the accuracy of radio map,and the more accurate localization results can be obtained by using the smoothed radio map.