Research On Indoor Localization Optimization Method Based On Wi-Fi And Multisensor Data Fusion

Recently,with the rapid development of mobile internet technology and communications industry,Localization Based Services(LBS)has gradually become an important requirement for people,and the demand for indoor localization services is more intense.Increasing researchers focus their attention on the Wi-Fi fingerprint localization and Pedestrian Dead Reckoning(PDR)localization system which is based on inertial sensors for reasons of low cost,good compatibility,extendibility and so on.Wi-Fi localization builds a mapping between Received Signal Strength Indication(RSSI)and location.Therefore,its accuracy is limited by the stability of RSSI.In real applications,RSSI is easily affected by environment changes.Therefore,it has high volatility and makes the localization accuracy volatile.PDR shows high localization precision since it uses sensor information,such as accelerometer and gyroscope which are not liable to be affected by the external conditions.However,it is easy to generate drifting over time.The single-mode localization technology is unable to meet the demand of people in the complex indoor environment because of its own limitations.Therefore,integration of multiple localization techniques has been one of the hotspots in localization field.In this thesis,the following research work is based on the reference to relevant literature and the field experiment and analysis:(1)We collect RSSI samples in Wi-Fi network environment.After data analysis,it is found that there is a relationship between the Wi-Fi signal and the location,and the localization model can be built by the matching relationship of location and fingerprint.However,it is difficult to realize the fingerprint localization with high precision due to the redundancy and high volatility of Wi-Fi signal.For this reason,this thesis proposes Information Gain Pre-Processing based Principal Component Analysis(IG-PCA)to do data pre-processing and reduce the dimension of RSSI.In the localization phase,we use machine learning algorithm to build models.It has proved that this algorithm reduces the algorithm complexity and improves the localization efficiency.(2)The PDR localization includes three aspects:step length estimation,step detection and orientation calculation.Since the motion state of human is polytropic,simply setting the threshold or using wave detection cannot deal with the changes of motion pattern.Therefore,this thesis proposes an adaptive PDR localization method,which can be adjusted timely according to the changes of motion pattern,improving the accuracy of PDR algorithm.Wi-Fi fingerprint localization and PDR localization both have some limitations.Therefore,this thesis uses the Extended Kalman Filter(EKF)and Unscented Kalman Filter(UKF)to construct the fusion localization model.After testing,fusion model can combine the advantages of Wi-Fi fingerprint localization and PDR localization efficiently and improve the localization precision.(3)Because of the complexity of localization environment,fusion localization method still lacks stability under extreme conditions.In order to improve the localization stability,based on the characteristics of localization area,this thesis uses the K-means clustering algorithm to mine the "landmark" which has unique data identification.By detecting and matching landmarks,localization model can realize self-correcting.It is showed in experiments that this method can mine the landmarks in the localization area and correct the localization trajectory and thus improve the accuracy and stability of localization.