Multi-viewpoint image correction method for parallel panoramic camera array

Abstract

The invention discloses a multi-viewpoint image correction method for a parallel panoramic camera array. The method comprises the following steps: S1, extracting and matching feature points of viewpoint images: detecting sufficient SIFT (Scale Invariant Feature Transform) feature points to be matched from a plurality of viewpoint images, and determining positions and scales of the feature points; S2, accurately positioning key points: acquiring accurate positions, scales and curvature ratios by fitting surrounding data of determined feature candidate points; S3, correcting pairwise adjacent multi-viewpoint images according to selected refined matching point information; S4, projecting the multi-viewpoint images to a public correction plane; S5, generating a description vector; S6, matching features: determining corresponding feature point relationships in pairwise adjacent images; and S7, removing noise points from the extracted matched feature points in each three-dimensional pair by an improved RANSAC (Random Sample Consensus) refining algorithm. The method has a reasonable design and high feature matching performance, and contributes to increase of the parameter resolving accuracy and the multi-viewpoint image correction efficiency of the panoramic camera array.

Description

technical field [0001] The invention relates to the technical field of multi-viewpoint image correction methods, in particular to a multi-viewpoint image correction method for a parallel panoramic camera array. Background technique [0002] The multi-camera real-time acquisition stereo display system is an important application of 3D TV. It transmits the video captured by multiple cameras in real time to the server, processes the multi-viewpoint images in the server, synthesizes the stereo images and displays them on the stereo display. For the parallel camera array, due to the unavoidable operation error and equipment precision limitation, the method of manually adjusting the camera position cannot fully simulate the placement of the ideal parallel camera array. The multi-viewpoint images captured by this method of manually placing the camera, the corresponding feature points are not aligned in the vertical direction, and the parallax in the horizontal direction is uneven, ...

AI Technical Summary

Problems solved by technology

For the parallel camera array, due to the inevitable operation error and equipment precision limitation, the method of manually adjusting the camera position cannot completely simulate the ideal parallel camera array placement.

The multi-viewpoint images captured by this method of manually placing the camera, the corresponding feature points are not aligned in the vertical direction, and the parallax in the horizontal direction is uneven, resulting in obvious flickering of objects in the synthesized stereo image, which seriously affects the real-time Viewing effect of 3D stereoscopic video

The application of RANSAC algorithm in multi-viewpoint image correction is to calculate the mathematical model parameters of the data based on a set of sample data sets containing abnormal data, so as to obtain the algorithm of effective sample data. The traditional RANSAC algorithm has good performance in removing noise points , but less efficient