Design And Implementation Of Visual Positioning System Based On 3D Reconstruction

The location has been an indispensable technology in modern people's life.According to different usages of scenario,the location service can be differentiated from indoor positioning and outdoor positioning.Both of them use diverse technologies.Outdoor positioning mainly uses GPS signals.Compared with indoor positioning,it is more mature.Although the indoor location has various solutions,it always exists many problems like expensive cost,sophisticated installing,signal blocking,and so on.Computer Visual Positioning is a kind of solution in an indoor location,which utilizes image data from the sensor to locate.Compared with other indoor location technologies,it doesn't need extra signal sources.Therefore,it doesn't need much cost and can ensure high accuracy and speed at the same time.This article designs and implements a computer vision location system based on 3D-reconstruction.The system supports users in an image-based location both offine and online mode and aims at helping mobile applications in building computer vision visual positioning function conveniently.The system includes server and Android SDK(including map SDK and visual positioning SDK),3D-reconstruction tool,and map calibration tool.During the development process,this article builds the server and client simultaneously in order to improve code reusability and readability,reducing redundancy.On the aspect of the framework,it uses Mysql and Sqlite as server and application databases individually.3D-reconstruction mainly uses Colmap opensource tool.For better transmission e ciency,completing instant location function,the system uses Netty as a network framework and design a customized protocol.The Algorithm and image processing mainly use Open CV interfaces,and the article implements a 2D to 3D point matching algorithm on this basis.The system extracts key points from images first by feature extraction algorithm,then find the 4 sets of mapping between key points and point-cloud from 3D-reconstruction.finally,it uses P3P to calculate the real location where the image is captured.At last,this article uses and tests the system on a real large supermarket,summarizing the accuracy and speed of location,as well as stability and performance.As a result,the system satisfies the requirements of indoor positioning basically,which verifies the feasibility of the system and the rationality of the design.