Visual positioning method and system and computer readable storage medium

Abstract

The embodiment of the invention provides a visual positioning method and system and a computer readable storage medium. The method comprises the steps: acquiring a to-be-positioned image, retrieving and obtaining at least one image in a pre-constructed three-dimensional visual map database to serve as a candidate image, wherein the global feature vector similarity of the candidate image and the to-be-positioned image is larger than a threshold value; respectively calculating respective relative pose information of the to-be-positioned image and the at least one candidate image, and obtaining at least one pose result of the to-be-positioned image relative to a three-dimensional visual map by utilizing the calculated relative pose information; and performing a verification test on at least one pose result, and outputting the pose result with the highest score in the verification test. According to the method, the system and the computer readable storage medium provided by the embodimentof the invention, visual accurate positioning can be realized in a complex dynamic scene and a shielded place.

Description

technical field [0001] The present application relates to the field of computer vision, in particular to a visual positioning method, system and computer-readable storage medium. Background technique [0002] Precise positioning is a key technology that needs to be used in applications such as unmanned driving, drones, augmented reality, location-based services (Location Based Services, LBS) games, and pedestrian navigation. Traditional positioning methods (such as Beidou system, Global Positioning System (Global Positioning System, GPS), Bluetooth beacon, etc.) are easily affected by buildings, pedestrians, etc., and their applicable scenarios and positioning accuracy are limited. In recent years, the visual positioning system based on camera images has been widely used in autonomous navigation, augmented reality, assisted driving, unmanned aerial vehicles, etc. and other fields. General visual positioning solutions are easily affected by lighting, dynamic scene changes, ...

AI Technical Summary

Problems solved by technology

Some of the existing methods for visual positioning combined with deep learning use direct training of neural networks to calculate the position and attitude of the camera, lack of geometric constraints, resulting in a low positioning success rate; some of the existing visual positioning based on deep learning are based on regression iteration each The three-dimensional coordinates of the pixel points are used to further calculate the position and attitude of the camera, but the coordinate regression error of the pixel points is usually large, resulting in limited final positioning accuracy

Images