High-Precision Indoor Positioning By Integrating WiFi Fingerprint And Vision

In recent years,with the rapid development of the mobile Internet and people's increasing demand for location based services,indoor localization technology has become a key topic for major institutions at both home and abroad to compete for research.Indoor positioning technology urgently needs to be used in shopping malls,underground parking lots and other consumer places to facilitate people's indoor activities.In addition,indoor navigation technology based on indoor positioning technology can also be applied to airports,hospitals and other places that require high public safety.However,as GPS signals cannot be received in indoor environments,there is no accurate and reliable indoor positioning system.This thesis first proposes a WiFi localization algorithm based on MAC address.In the offline stage,the Android phone is used to acquire the MAC addresses of the wireless AP(Access Points)at the WiFi sampling points,and WiFi fingerprint database containing the WiFi fingerprint and coordinates of WiFi sampling points.In the online stage,the Android mobile phone is utilized to collect the measured fingerprint.The fingerprint matching algorithm and the KNN algorithm are used to obtain the k fingerprints in the fingerprint database that have a high degree of match with the measured fingerprint,thereby obtaining the WiFi positioning range and the positioning coordinates.Secondly,this thesis proposes a fast visual localization algorithm based on doorplates,In the offline stage,collects all the doorplates in the positioning area and generates a training image set,and then forms a visual positioning database together with the coordinates of the sampling points of doorplates,then calculates the SURF global feature descriptors and the ORB global feature descriptors of the training images,by extracting the SURF and ORB global features of the training image can greatly improve the computational speed.In the online stage,the image feature matching algorithm and the KNN algorithm are used to obtain the training image which matching the tested image most in the training image set,that is,the matching image,and then the visual positioning database is queried to obtain the visual positioning coordinates.In the visual positioning stage,this article also used the QR codes as a visual marker to perform positioning experiments.The QR code pictures have low production cost,flexible deployment,and convenient user reading,and can achieve accurate positioning.The positioning accuracy is close to 100%.There are obvious technical advantages in indoor environments with less WiFi signals and where the QR codes are allowed to deploy.Finally,considering the advantages and disadvantages of the WiFi localization based on MAC address and the visual positioning based on the doorplates,the fusion positioning is needed after obtaining the WiFi positioning coordinates and the visual positioning coordinates,and the visual positioning is added on the basis of the WiFi coarse positioning to avoid the large errors caused by mis-matching in traditional vision positioning,by integrating WiFi fingerprint and vision,we can effectively improve the positioning accuracy and robustness.The proposed method has been implemented into an Android App and tested in three typical scenarios:a 7,200 m~2 shopping mall,a 24,000 m~2 office building,and a5,800 m~2 conventional hall.Experimental results show that the average position errors from the proposed method are 0.7m in these three test sites.The paper suggests a novel solution to accurate and effective indoor positioning.