| With the rapid development of intelligent terminal equipment and Internet of Things technology,people's demand for location-based services is increasing.Outdoors,with the development of Beidou navigation satellite positioning system in China,GNSS system can provide high-precision real-time positioning and navigation technology,while indoors,due to the influence of building occlusion,complex indoor environment and other factors,there is no unified dominant navigation technology for indoor positioning and navigation,so indoor positioning and navigation is the "last kilometres." in the field of location services.This paper focuses on the characteristics of low power Bluetooth Low Energy(BLE)positioning technology,such as high positioning accuracy,simple hardware deployment but obvious signal fluctuation,high positioning accuracy in short time with mobile phone inertial sensors,but large accumulation of long-time errors.While improving the positioning accuracy of the two positioning methods respectively,the advantages of the two positioning methods are complemented,and the integration location.of the two methods is achieved.The main contents of this paper are as follows:(1)In order to solve the problem of determining the ambiguity of reference points and their membership matrix in the initial class when clustering the reference points with the fuzzy C-means,this paper proposes that when the same AP number at the reference points is greater than 50%-70% of the total recorded maximum signal strength AP,these reference points are grouped into one group,and the corresponding membership matrix is set up.In this way,the initial class is determined,and then the fuzzy C-means is used.Value clustering reduces the clustering error caused by the artificial determination of reference points and their membership matrix in the initial class,reduces the clustering time and improves the positioning accuracy.Hidden Naive Bayesian method and clustering fingerprint database are used to improve the accuracy of location.(2)The total acceleration of the mobile phone in the forward,lateral,vertical and triaxial directions is analyzed,and the vertical or triaxial total acceleration is determined to be used as pedestrian step detection.In the step detection,the peak detection algorithm is improved.On the basis of the peak detection algorithm,the differential acceleration value is added,and the positive and negative changes of thedifferential acceleration symbol are used to assist the output of the mobile phone.The judgment of velocity wave crest and trough,combined with the relationship between wave crest and trough number,eliminates the incorrect estimation of wave crest and trough,and then collects the acceleration output of mobile phone when pedestrians stand still,run,walk uniformly and walk fast.The frequency analysis of these accelerations is carried out to determine the amplitude threshold and periodic threshold when acceleration judges the motion state.In the gait detection of smooth walking,uniform speed,fast walking and running gait,the correct rate is more than95%.(3)Firstly,the weighted fusion of pedestrian track estimation and Bluetooth location results is carried out.The weighted fusion weights are divided into constant weights and dynamic weights.When setting constant weights,Bluetooth positioning is the main method,while pedestrian trajectory estimation assists Bluetooth positioning.So the former is set to 0.6,while the latter is set to 0.4.The dynamic setting weights are to calculate the distance between the results of Bluetooth and PDR positioning at k-2,k-1 and K moments and the results of fusion at the previous moment,respectively.The variance of distance value is taken as dynamic coefficient by reciprocal variance.The weighted sum fuses the location results of Bluetooth and PDR.The experiment compares Bluetooth location,constant coefficient weighted fusion location and dynamic coefficient weighted fusion location,and the positioning accuracy is improved in turn.Finally,aiming at the abnormal situation of the noise covariance of the extended Kalman filter measurement equation,an error judgment method is introduced to correct the measurement noise covariance matrix.For the problem that the system noise covariance matrix is not suitable to be set as a constant value when Bluetooth and PDR are fused,an improved Sage-Husa algorithm is used to constantly revise the value of the matrix.The experiment compares Bluetooth positioning,classical EKF fusion Bluetooth and PDR positioning,and the robust adaptive EKF fusion Bluetooth and PDR positioning introduced in this paper.The positioning accuracy of the three experiments is improved by 24.96% and 24.4%respectively. |
