Camera pose solving method applied to weak texture large rotation scene

Abstract

The invention discloses a camera pose solving method applied to a weak texture large rotation scene, and the method comprises the steps: recovering the normal information with fuzziness from a polarization image through a conventional SPF method, segmenting the obtained normal clustering into different normal plane blocks, extracting a plurality of random pixel points meeting the Gaussian distribution, tracking pixel points among the images to realize plane matching among the images, performing polarization-based RANSAC optimization processing on normal lines in the segmented plane block, removing outer points to obtain an average normal line value of the plane block, scoring the IMU data among the images at the same time, removing the normal fuzziness by combining the average normal values of the plane blocks corresponding to the two images with the relative attitude obtained by the IMU, and solving the camera attitude from the average normal pair by combining the linear optimization theory. According to the method, the defects of a visual method can be solved, the application range of a visual algorithm is expanded, and the camera attitude can still be estimated in extreme scenes such as weak texture and large rotation.

Description

technical field [0001] The present invention relates to the technical field of camera pose estimation, in particular to a method for solving the pose of a camera applied to a scene with weak texture and large rotation. Background technique [0002] Camera pose estimation plays an important role in the process of robot movement. It is currently widely used in robot autonomous navigation projects. It is a computer vision application field with great application prospects. [0003] Generally speaking, in the field of robotics, camera pose estimation is often classified as a VO (Visual Odometry) problem. In the current VO research issues, visual odometry based on feature points and optical flow method is recognized as a mature and robust VO method, but these methods require rich texture information, otherwise a reliable camera pose cannot be obtained. It cannot be applied to scenes with weak textures and large rotations. Contents of the invention [0004] The purpose of the ...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to overcome the deficiencies of the prior art, and to provide a method for solving the pose of a camera in a scene with weak texture and large rotation, which is different from the existing VO algorithm for camera pose estimation based on feature points. The present invention is based on The rotation solution of polarization is helpful to solve the problem that the traditional visual method is not applicable in the scene of weak texture and large rotation, so as to solve the shortcomings of the visual method, expand the scope of application of the visual algorithm, and still be able to robustly estimate in extreme scenes such as weak texture and large rotation camera pose

Images