Research On Indoor Location Algorithm Based On Fingerprint Library

As the increase of indoor activities,indoor positioning services,especially with the boosting usage of smartphones,researchers are increasingly focusing on reliable indoor positioning using wireless receivers and inertial sensors from Smartphones.This thesis mainly focuses on the practicability,accuracy,and robustness of indoor positioning system with the smartphone in WIFI environment,especially Pedestrian Dead Reckoning(PDR)based on inertial sensors and WIFI positioning techniques based on wireless receivers on smartphones.Experiments show that PDR positioning has good accuracy in distance detection and step detection,while the accuracy of positioning decreases as the positioning distance rises,due to the inadequate accuracy and accumulated bias of gyroscopes.Besides,we also experimented on 3 fingerprint positioning algorithms on WIFI positioning(K-Nearest Neighbor,Weighted K-Nearest Neighbor,and Confidence Weighted K-Nearest Neighbor),results showed that the 3 algorithms resulted in 3.82 meters,4.90 meters,and 5.29 meters as average bias distance,respectively,which means WIFI fingerprint positioning only take an average accuracy and low robustness.Based on the results we get,this thesis proposed an algorithm on fusion positioning combining Conditional Random Field(CRF)with WIFI and inertial sensors information based on Kalman filter algorithm,especially focusing on the scenario with the unknown initial position of pedestrians.The experiment showed that the Kalman filter fusion positioning algorithm can reduce the cumulative bias of PDR heading angle detection and improve the positioning accuracy under the condition that the initial position is known.However,when using WIFI positioning as the initial location in the context of the unknown initial position,initial bias will be accumulated to the final positioning result,which makes the result inaccurate.Besides,this algorithm also inherits the low robustness of WIFI fingerprint positioning.This thesis also constructs a location model based on CRF combining map information,inertial sensor information and limited WIFI data.The experimental verification shows that the CRF fusion positioning algorithm retains good robust performance,and achieved a good positioning effect when the initial position is known,with an average positioning bias of 1.79 meters,and a maximum bias of 4.41 meters.When the initial position is unknown,the CRF fusion positioning algorithm takes WIFI positioning as the initial position and uses the special position as markers to correct the bias of the WIFI positioning by the added walking mode determination module.Comparing to the scenario that the initial position is known,the performance of the algorithm is slightly depleted.The average bias of two positionings starting from different positions is 2.85 meters and 2.75 meters and the maximum bias is 5.20 meters and 5.24 meters,respectively.