Visual positioning method and device based on scene 3D point cloud

Abstract

The invention discloses a scene three-dimensional point cloud vision location method and device. the method comprises dividing scene point cloud into a plurality of disjoint subsets; performing priority ranking on all the subsets according to global characteristic descriptors of images to be searched and global characteristic descriptors of all subsets to obtain subset queues; selecting the first subset in the subset queues, searching for the matching relationship between scene points in the first subset and local characteristics of the images to be searched, determining whether the searched matching relationship meets ending conditions, and performing ending if the conditions are met; otherwise, selecting the second subset in the subset queues, searching for the matching relationship between scene points in the first subset and the second subset and local characteristics of the images to be searched, determining whether the searched matching relationship meets ending conditions, and performing ending if the conditions are met; estimating positions and directions of cameras for shooting images to the searched according to the searched matching relationship. By the aid of the method, the matching relationship searching efficiency can be improved.

Description

technical field [0001] The invention relates to computer technology, in particular to a visual positioning method and device based on a three-dimensional point cloud of a scene. Background technique [0002] At present, with the development of intelligent terminals, location-based services have received more and more attention, making positioning technology a research hotspot. Common positioning technologies are generally based on GPS or Wi-Fi wireless networks. However, there are many problems in these positioning technologies, such as susceptibility to interference and limited scope of application. [0003] Image-based Localization technology, that is, based on the three-dimensional point cloud of the scene to calculate the position and direction of the camera in the point cloud of the query image, can effectively solve the above problems. [0004] However, visual localization techniques determine the camera position and orientation of the query image based on the matchi...

AI Technical Summary

Problems solved by technology

At present, the scale of scene point clouds is often on the order of millions or even larger. Therefore, finding the matching relationship between the local features of the query image and the scene points requires a lot of calculations, resulting in the inefficiency of finding matching relationships in visual positioning technology.

Images