A Self-Location Method for Moving Objects in Known Scenes

Abstract

The invention relates to a self-positioning method of a moving target in a known scene, which registers an image sample collected by a camera with a panoramic image of the scene, based on the registration relationship between the image sample and the panoramic image and the registration of the panoramic image and the three-dimensional model of the scene Quasi-relationship, estimate the camera pose of the image sample, use the camera pose as the mark of the corresponding image sample for regression training, obtain the regression model from the scene image to the camera pose, use the camera set on the moving target to collect the scene image in real time, and use it from the scene The image-to-camera pose regression model performs camera pose estimation, and then realizes the positioning of moving objects. The present invention is based on the regression model from the image to the camera attitude obtained by the supervised deep learning method training, and realizes the self-positioning of the moving target in the known scene, so as to partially overcome the poor real-time performance and low texture image stability of the traditional visual positioning method. Insufficiency and other issues.

Description

technical field [0001] The invention relates to a method for self-locating a moving target in a known scene. Background technique [0002] In recent years, in known scenarios such as large-scale square security, airport surface activity guidance and control, port production operation area operation status monitoring, and industrial park management and control (referring to scenarios where information such as images and 3D models can be obtained in advance), AR (Augmented Reality, AR) technology has been applied more and more. To realize AR, it is necessary to place virtual objects in the real environment or real-time video of the real environment, and perform information amplification on the real environment or real video. high demands. In addition, in the above-mentioned scenarios, autonomous driving shows a more promising application prospect than general open scenarios. In autonomous driving, vehicles need to perform high-precision positioning on themselves. [0003] T...

AI Technical Summary

Problems solved by technology

[0007] In order to overcome the above-mentioned defects of the prior art, the present invention provides a self-positioning method for moving objects in known scenes, which is based on a regression model from image to camera pose obtained through supervised deep learning method training, and realizes moving objects in known scenes In order to partially overcome the problems of poor real-time performance and insufficient stability in dealing with low-texture images in traditional visual positioning methods

Images