| Recently,with the popularization of intelligent equipment as well as the construction of ubiquitous sensor networks,digital cities,smart homes,and the Io T,people have an increasingly urgent demand for indoor positioning.Due to the signal obstruction and multipath effects,mature satellite positioning technology is difficult to provide reliable positioning services indoors.It is extremely challenging to find an accurate and stable indoor positioning method.Abundant sensors deployed in the environment and powerful sensor systems on mobile platforms such as smartphones constitute a wireless sensor network providing conditions for indoor positioning implementation.Scholars have explored numerous indoor positioning technologies.The mainstream technologies have their advantages as well as have their limitations.The RFID technologies,ultrasonic,UWB,infrared,LED,and pseudo satellites are limited by expensive hardware foundation and implementation environment.Wi Fi,Bluetooth,and geomagnetic positioning are faced with uncertain environment interference and other problems.Inertial positioning is also limited by error accumulation.Thus,multi-source information fusion and location technology under a typical indoor scene has become a current research hotspot.In this paper,we study a fusion localization method based on low-cost MEMS/ signal localization technology with a smartphone as the carrier.Using ubiquitous Bluetooth /Wi Fi signals and inertial sensor information obtained from mobile phones,an indoor IMU/ Bluetooth fusion localization algorithm based on pattern awareness and map matching is proposed.First,to reduce the cost of fingerprint construction,the sparse fingerprint database can be expanded rapidly under the condition of limited sampling by using a spatial interpolation algorithm.Dynamic range measurement adjustment strategies have been explored to improve positioning accuracy and algorithm adaptability.Then,the data characteristics of indoor pedestrian behavior sensors are analyzed and build a machine learning model to percept and recognize the indoor multi-scene pedestrian behavior and movement modes.Moreover,sensor error correction,robust step counting algorithm,and pedestrian heading estimation method are further systematically studied.Last,combined with the positioning results of Bluetooth and PDR,the indoor positioning fusion algorithm assisted by behavioral information and map information is deeply explored.The following are the main contents of this study:(1)For the problems of time-consuming construction of the fingerprint database and the poor positioning accuracy due to the insufficient collection of fingerprint samples,a fingerprint automatic constructed and expanded method based on the Kriging algorithm and AP point selection is proposed.The proposed method obtains abundant fingerprint samples without increasing the collection cost and time;An improved LWKNN algorithm is proposed to solve the mismatching of adjacent points and weak adaptability of fixed K value in the WKNN algorithm.The experimental results show that the proposed fingerprint database expansion method and matching algorithm have significantly improved database building efficiency,positioning stability,and accuracy.For the problem of invalid position solution caused by inaccurate distance in the trilateral ranging method,a dynamic range adjustment strategy is proposed to improve the positioning accuracy and the adaptability of the algorithm.(2)Analyzed the characteristics of indoor pedestrian behavior sensor data and explored the mining and extraction of sensor data that effectively reflect behavior characteristics.Based on the above work,SVM and FSM-DT algorithms were used to constructed a lightweight pattern recognition model to implement the monitoring and recognition of typical indoor pedestrian movement states and equipment usage modes.(3)Introduced the basic principle of PDR and analyzed the characteristics of multi-posture sensor data and the problems and limitations in mainstream step detection and heading estimation methods.To solve the over-counting or under-counting problem in step counting,a robust and accurate step counting solution based on movement mode recognition was proposed,which effectively improves the performance of step counting in typical carrying mode and false walking state.Besides,the heading calculation method by the magnetometer and accelerometer data was compared and further combined with the gyroscope information to improve the accuracy of heading estimation.(4)Introduced the basic principle and process of the PF algorithm;a PF algorithm based on map matching was proposed to fusion BLE and PDR positioning.The algorithm utilized the map boundary information and behavioral landmark information in the resampling process to improve the trajectory’s rationality.Also,the timeliness and complexity of the standard PF algorithm and the proposed algorithm under different particle numbers were compared and analyzed;Finally,EKF,UKF,PF,and the improved PF algorithm were utilized to carry out the comparison test of fusion results.The experimental results show that the proposed algorithm can effectively improve the positioning’s performance and obtain more accurate and robust results.This study has 71 figures,17 tables,and 115 references. |
PDF Full Text Request