A method based on deep learning to obtain the position and attitude of the image city

Abstract

A method based on deep learning to obtain the position and attitude of an image within a city, involving the field of image location recognition and augmented reality. The method includes the following steps: 1) creating a city picture set; 2) training a mixed Gaussian model on the city picture set, and using the trained mixed Gaussian model to divide urban geographical regions; 3) training a joint learning picture pose estimation and scene recognition neural network; 4) Initialize, upload the user's GPS or network rough location information; 5) use the learned division function to divide the rough location information, download the corresponding network model and the rendering data to be displayed to the client; 6) collect the user input camera video stream, Use the downloaded network model of the current area to predict the positioning results of the three levels at the current moment. If the confidence of the prediction result output by the network is higher than the threshold, the rendered data will be rendered using the predicted position and attitude parameters.

Description

technical field [0001] The invention relates to the field of image geographic location recognition and augmented reality, in particular to a method for obtaining the location and posture of an image within a city range based on deep learning. Background technique [0002] With the explosive development of mobile Internet and smart devices, taking and sharing photos has become a part of people's daily life. How to deduce from the photo where the photo was taken and the perspective of the photo has become a very meaningful problem. The problem of inferring the shooting position and perspective from the photo is also called the pose estimation problem of the camera in Multi-View Stereo. It is a basic problem in the field of computer vision and robotics, and has a wide range of applications, such as enhancing Reality (Augmented Reality, referred to as AR), simultaneous positioning and map construction (Simultaneous Localization and Mapping, referred to as SLAM), and image-based...

AI Technical Summary

Problems solved by technology

Although the accuracy of coordinate regression forest is very high, its disadvantage is mainly that RGB-D images are required as input. In actual use, RGB-D images are only suitable for indoor scenes, and the RANSAC calculation process is very time-consuming.

Images