Study On Indoor Positioning Method Based On Fusion Of Bluetooth Technology And PDR

With the rapid development of Internet of Things(Io T)technology and the continuous improvement of human living standards,the role of location services is becoming increasingly important.People’s demand for geographic location information is increasing,and it is gradually extending from outdoors to indoors.Although global satellite navigation systems can provide real-time and accurate positioning results in outdoor environments,this positioning system is difficult to provide stable and accurate indoor positioning results because satellite signals cannot penetrate buildings.How to achieve low-cost and high-precision indoor positioning has become a research hotspot.Currently,there are many indoor positioning technologies available,and considering the comprehensive cost and technical difficulty,this article chooses to use Bluetooth positioning technology and Pedestrian Dead Reckoning(PDR)positioning technology based on inertial sensors for research.Each indoor positioning technology has its own advantages and limitations.For Bluetooth positioning technology,it has the advantages of easy deployment,high positioning accuracy,wide coverage,and error accumulation not increasing over time.However,its positioning frequency is low and the positioning results are not continuous.For Pedestrian Dead Reckoning(PDR)positioning technology based on inertial sensors,it has high positioning accuracy in a short period of time,continuous positioning results,and autonomy,but its positioning error will accumulate over time and diverge.Therefore,the main content of this paper revolves around the optimization of a single indoor positioning technology and the fusion of these two indoor positioning methods based on this optimization.The main research results are as follows:(1)The optimization of Pedestrian Dead Reckoning(PDR)positioning technology.Firstly,XGBoost classification model is used to identify the pedestrian activity pattern and determine whether the pedestrian is in a walking state.The accuracy of the classification recognition can reach 98.21%.Compared with SVM and RF classification models,the XGBoost model has higher classification accuracy and shorter time.For step frequency detection in PDR,a method of setting local threshold is proposed in this paper,which simplifies the parameters in the method and improves the accuracy of step frequency detection.At the same time,the peak-to-valley values of the original acceleration are retained,which can be further used for step length estimation.The method is convenient and effective,and has strong universality,with the accuracy of step frequency detection reaching more than 99%.Finally,the original heading of the mobile phone is calculated using the acceleration and geomagnetic data.The geometric information of the map is further used for heading correction,and the average heading accuracy is improved by69.45%.(2)The optimization of Bluetooth ranging positioning technology.Firstly,a DGMM weighted mixture filtering method is proposed to process RSSI values,which better eliminates signal anomalies and reduces signal errors caused by device shake and other factors.Secondly,a distance estimation model based on Rician channels is proposed to address the multipath effect in signal propagation,and experiments show that this model outperforms traditional logarithmic signal loss distance estimation models.Finally,considering that the distance formula is a nonlinear function,a ranging and positioning method based on nonlinear least squares is proposed.Comparison experiments show that the Gauss-Newton method in nonlinear least squares reduces positioning errors by 14.84% and 24.40% compared to the least squares and trilateration methods,respectively,and the Levenberg-Marquardt method reduces positioning errors by 27.47% and36.84% compared to the least squares and trilateration methods,respectively,improving the positioning accuracy.(3)A dynamic factor graph fusion model was proposed to fuse a single positioning result.An adaptive weight factor was added to the model to dynamically adjust the weight of Bluetooth positioning in the fusion model.When a large error was detected in this positioning method,its weight in the fusion model was adjusted in a timely manner to prevent the large error from affecting the positioning result.In addition,coordinate inversion was used to correct heading based on the fusion result,eliminating the error accumulation of PDR positioning.Experiments were conducted with three different phone postures: holding the phone upright,putting the phone in a pocket,and putting the phone in a backpack.The average error after fusion was 37.15%lower than that of Bluetooth positioning and 79.96% lower than that of PDR positioning,respectively.Compared with the fusion models using Extended Kalman Filter(EKF)and Unscented Kalman Filter(UKF),the positioning errors using the dynamic factor graph fusion model were reduced by 18.95% and 15.65%,14.72% and 8.55%,16.09% and 11.52%,respectively,demonstrating that the proposed dynamic factor graph fusion model has higher positioning accuracy.