Research On Vision Based Outdoor Localization Method For Smart Device

The outdoor positioning service runs through the basic necessities of human life,and today's intelligent application services are inseparable from the positioning of this critical basis.Positioning technology has a wide range of applications in robot autonomous navigation,industrial manipulator automatic control,unmanned vehicle automatic driving and other projects.The outdoor positioning method of mobile intelligent terminal based on vision can be divided into two aspects to obtain estimates of point cloud and camera pose,including outdoor scene 3D reconstruction,point cloud registration,perspective point calculation position and coordinate conversion task.The visual outdoor positioning research based on the above two aspects,including: In view of the acquisition of point clouds in outdoor scenes,this paper studies the large-scale 3D reconstruction technology,including motion recovery structure,point cloud registration and fusion technology.For the large scale outdoor scene image data acquired by ourselves,the local point cloud data are constructed by local point cloud registration and fusion technology;In the aspect of camera pose estimation,on the one hand,on the basis of the traditional 2D to 3D matching,the camera pose is improved by using the perspective point method.On the other hand,presents a perspective point algorithm for objective assessment method.The main work and innovation of this paper are as follows:1)Deeply studied the motion recovery structure method,and carried out the outdoor scene 3D reconstruction for the experiment,and collected and established the point cloud database of some scenes in our school campus.Improved the 3D reconstruction method of UAV aerial images by fusing auxiliary information.2)Improved the traditional perspective method to calculate the camera pose,improved the matching and accuracy,and obtained more accurate camera pose results.Aiming at the position and pose of UAV aerial images,the idea of reference pose is proposed.The method is proved to be objective and feasible through experimental verification.3)Trying to find out whether large scale structured point cloud data is necessary for vision based localization methods.In this paper,the research results of mobile intelligent terminal localization based on vision can provide new methods and ideas for the construction of large scale scene point cloud database and the calculation of camera pose by perspective point.