Research And Implementation Of Indoor Pdr Positioning Based On The Fusion Of Landmarks And BMO-PF

PDR can be easily achieved in terminal by inertial measurement unit,so it was widely applied to indoor positioning system.Due to the cumulative drift,PDR was usually optimized by particle filter that fusion with other wireless positioning technology.The error of heading prediction in PDR is big,and wireless positioning is mostly achieved by fingerprint positioning which has heavy workload in database establishing.In order to solve these problems,a kind of Landmarks based on spatial partition was proposed in this paper and the indoor PDR positioning system based on particle filter fusion Landmarks was researched,the main contents and work includes:(1)A course estimation algorithm based on federate filter was proposed.On basis of the different characteristics of gyroscope,electronic compass and accelerometer,this paper uses the federate filter to optimize the heading predicting algorithm.Firstly,low-pass filter was used to purify electronic compass,then complementary filter was used to optimize the gyroscope and accelerometer,finally,the two filters' outputs was processed by Kalman Filter.The PDR positioning improved by the federated filter outperforms PDR based on Kalman Filter.(2)A kind of Landmarks based on spatial partition was proposed.The indoor space was divided into several regions according to the floor plan and Density-Peak-Cluster algorithm,the boundaries between different regions are regarded Landmarks.We then use Received WIFI Signal Strength to build fingerprint database with different regions,and adopt PN Learning algorithm to improve OS-ELM classifier which was used to identify the region where pedestrian stand.In online positioning phase,Landmarks are matched according to the change of pedestrian's regions and used to optimize the PDR positioning.Experiment show that the improved ELM algorithm outplays OS-ELM,SVM and Random Forest in regions classification.(3)A kind of particle filter improved by BMO algorithm was proposed.In the process of particle filtering,we divide particle set into five categories according to the principle of BMO algorithm,and different categories use different strategies to update particle set,what makes the particles distributed in the area with high posterior probability density as much as possible,and dynamic resampling algorithm was used to maintain the diversity of particles.The improved BMO-PF outplays Simple PF,PSO-PF and CS-PF.The improved particle filter algorithm and Landmarks were used to optimize the PDR positioning,which reduced the workload of fingerprint database establishing compared with the traditional method fused PDR and WLAN positioning.Finally,experiments were made to test the improved the improved positioning method,the results show that the average positioning error was 1.8m and the maximum error was 6.7m.