Research On The Optimization Method Of BLE/PDR Fusion Indoor Positioning Based On Smart Phone

With the continuous development of technology,smartphones,watches,tablets and other smart mobile devices are becoming an indispensable part of people's daily life.Location-based services(LBS)based on these smart mobile devices are becoming more and more common,and their market share is also growing.These developments and changes are promoting the continuous progress and improvement of indoor and outdoor navigation and positioning technology.Previous Market Research on indoor location services showed that most of people's lives are in the indoor environment,so it is necessary to obtain the location information of people in the indoor environment.Among many indoor positioning technologies,this paper selects three popular indoor positioning methods to optimize,namely BLE(Bluetooth Low Energy),PDR(Pedestrian Dead Reckoning),BLE and PDR fusion positioning,in order to improve the accuracy,stability and practicability of positioning system,and to provide theoretical basis and technical support for the subsequent indoor positioning practicalization and product.In this study,smartphone is used as the device to be positioned,and machine learning,deep learning and fusion algorithms are used as the theoretical basis for optimal positioning.Through theoretical analysis,experimental validation of ideal and real complex indoor environment,and validation with different mobile phones,improvements have been made in the reduction of fingerprint mismatch,the implementation of BLE large-scale multi-floor positioning,the optimization of PDR and BLE/PDR fusion positioning.The main contributions of this article are as follows:1.Through the study of common BLE fingerprint locating methods,the reasons for large positioning errors are found and optimized,and the method of using Ada Boost algorithm to enhance machine learning classification is proposed to realize the combination of machine learning,so as to improve the accuracy of fingerprint database data classification model,thereby improving the classification accuracy of the points to be measured,reducing the probability of occurrence of no matches,and improving the positioning accuracy.In this paper,the SVM algorithm is taken as an example,when enhanced by Ada Boost algorithm,its average classification accuracy is improved more than 10%.For those points misclassified by SVM algorithm,their positioning accuracy is improved by more than 40% when they are accurately classified by Ada Boost-SVM.At the same time,it is verified that this method can effectively improve the classification accuracy and positioning accuracy when locating different types of mobile phones.2.After comparisons of common BLE fingerprint locating methods,it is found that most of them cannot use the feature of signal strength to classify floors.In this paper,the CNN method in deep learning is introduced in BLE fingerprint locations to classify the floors.A CNN network structure suitable for BLE floor classification and positioning is selected and constructed,and then the data is trained.Experiments show that the accuracy of floor classification using CNN is more than 96%,which meets the needs.Meanwhile,CNN can classify the points to be located into the control range of the corresponding reference points,and the classification accuracy reaches more than 91%.The static positioning error is less than 1.33 m,and the dynamic positioning error is less than 1.55 m.Compared with common BLE fingerprint positioning,the accuracy is higher,and it meets people's daily needs.At the same time,it also verified the results of classifying the data received by different mobile phones during positioning.The accuracy of floor classification exceeded 90%,and the positioning accuracy was higher than that of traditional fingerprint positioning.3.Error analysis of traditional PDR algorithm shows that the accuracy of heading angle calculation has a great influence on the whole pedestrian track.Therefore,in order to improve positioning accuracy,this paper uses EKF algorithm to fuse the data from accelerometer and gyroscope in smartphone to calculate the heading angle of PDR.Experiments show that the results of angle measurement using fused data are closer to the true heading angle than those using gyroscope data alone,both in experimental and real indoor environments.4.Due to the low update frequency of BLE fingerprint positioning and the cumulative error of PDR method over time,this paper presents a method of combining BLE fingerprint positioning with PDR positioning using EKF to improve positioning performance,after comparing and analyzing the features of different fusion algorithms.Since the two positioning methods of BLE and PDR have been optimized,the positioning result after fusion of EKF method is compared with the result of BLE/PDR fusion positioning using EKF before optimization,and the error is reduced by0.14 m in the laboratory environment.In the hospital environment,the error is reduced by 0.18 m.At the same time,the experiment was conducted with different models of mobile phones.The results show that the optimized BLE/PDR fusion positioning algorithm has improved positioning accuracy by more than 0.1m compared with the unoptimized algorithm,which is about 10%.Compared with BLE and PDR separate positioning methods,the positioning accuracy and stability of this method are also better.