| With the development of the Internet of Things,location information is closely related to people's lives.Outdoors,related applications and services such as ship navigation,route planning,shared bicycles,and Didi taxies have become an essential part of life travel.However,according to relevant research,more than half of human time is indoors.Because the outdoor satellite navigation system is weakened by the wall in the room,the signal is weak and cannot meet the requirements of the accurate indoor positioning.Therefore,indoor positioning technology has been studied intensively by researchers in various countries and regions in recent years.The widely deployed wireless networks and portable smart devices has made WLAN-based smartphone device positioning as a mainstream.However,most of the WLAN indoor positioning systems are not universal.Only the most popular WiFi fingerprinting method is attracting the attention because of its unique advantages.However,it also has challenges in practical applications: the offline acquisition phase requires manual survey and is costly.Therefore,how to ensure the accuracy of the acquisition efficiently becomes a difficult point.At the same time,a large number of users hold different types of smart phone devices in the actual application of indoor positioning,so that the heterogeneity of WiFi software and hardware between devices will cause deviations in the signal strength of the same AP signal source.Therefore,this paper focuses on reducing the heterogeneity of WiFi software and hardware between devices and building high-quality fingerprint databases.This paper first proposes a method for calibrating WiFi using BP neural network.Firstly,pre-processing the collected signal strength data pairs,that is,using outlier detection algorithm to denoise the outliers,and secondly,denoising the data is transmitted to the BP neural network for error back propagation training.In the whole process,the weights and offsets of each node of the network are updated repeatedly,so that the actual output value approaches the true value.The condition of the end of training is that the sum of the squares of the errors between the actual output value and the true value is less than the threshold.At this time,the corresponding weights and offsets of each node are determined.Finally,the ideal network calibration modelis obtained,and the model proposed can be used by the different kinds of the phones to correct the signal strength of the different smartphone devices.The experimental analysis shows that the positioning accuracy the network calibration method proposed in this paper is 39.72% higher than the uncalibrated,which indicates that the network model calibration method can greatly reduce the influence of the differences between the different types of the devices on the positioning accuracy.This paper also proposes a WiFi fingerprint indoor positioning algorithm that combines with the confidence to reduce the acquisition errors in real-time and build a high-quality fingerprint database.In the offline acquisition stage,the researcher collects the multiple AP signal strength feature values for positioning at each reference point,and processes the signal strength feature values,that is,calculates the feature means and mean square errors.After obtaining the mean square errors,the virtual mapping table constructed by using the mean square error range and the confidence degree of the paper is used to determine the confidence,and finally uploaded to the server location fingerprint database together with the serial number of the reference points,the signal strength feature average value,and the physical coordinates.In the online service phase,the real-time acquired signal strength feature value vector of the to-be-positioned point is added with its confidence when the Euclidean distance is matched with the feature vector in the database,and the aim of optimizing its distance can be achieved. |
