A camera pose estimation method, device, equipment and storage medium

Abstract

The invention discloses a camera pose estimation method, device, equipment and storage medium. By acquiring two adjacent frames of images, feature points in the two frames of images are extracted, so that the two frames of images are divided into image blocks, and generated and Extract the image block sequence corresponding to each frame of image; input the image block sequence into the trained neural network model, so that the neural network model outputs the estimated fundamental matrix, and decompose the fundamental matrix to obtain the camera pose estimation result. Compared with the existing technology, by training a neural network model to solve the pose, the problem of solving the camera pose is unified into a joint optimization problem, which avoids the complicated calculation process in the traditional method, reduces the calculation amount of the model, and improves the performance of the model. The acquisition efficiency of the camera pose has strong practicability.

Description

technical field [0001] The present invention relates to the technical field of computer vision, in particular to a camera pose estimation method, device, equipment and storage medium. Background technique [0002] Camera pose estimation refers to the calculation of camera motion parameters, including visual methods that rely only on sequence image data, fusion methods that rely on lidar point clouds, and direct measurement methods that rely on high-precision pose measurement sensors. Traditional camera pose estimation It is generally divided into four steps, mainly including feature point detection, feature point matching, screening of credible matching point pairs and optimized estimation of camera pose. Traditional methods are difficult to define feature points, the repeatability of feature point detection cannot be guaranteed, and matching feature points requires strong matching features, which can easily lead to more mismatches. [0003] Existing methods based on deep l...

AI Technical Summary

Problems solved by technology

The traditional method is difficult to define feature points, the repeatability of feature point detection cannot be guaranteed, and matching feature points requires strong matching features, which can easily lead to problems such as more mismatches

[0003] Existing methods based on deep learning have made some progress in feature point detection, feature point feature learning, and feature point screening and pose optimization. However, these methods still follow the traditional camera pose estimation process and only solve certain problems. The problem of improving the accuracy of one step or some steps fails to solve the problem of solving the camera pose as a whole joint optimization problem

Images