An Image Stitching Method Based on Best Stitching Plane and Local Features

Abstract

The invention relates to an image splicing method based on an optimal splicing plane and a local feature. The method comprises the following steps: S1: estimating the depth information of an image; S2: segmenting the image based on the depth information, dividing the original image into an area to be registered and a non-registration area, and determining the optimal splicing plane by using two segmentation lines as sides; S3: calculating rotation angles by using the optimal splicing plane as a reference and using the respective segmentation lines as axes, and transforming the image to be in a reference plane; S4: extracting FAST feature points and the feature description of the feature points in an area to be spliced; S5: calculating Hamming distance for feature point matching, and performing precise matching by using use depth information and an RANSAC algorithm to obtain matching point pairs; and S6: computing a transformation matrix by using the registered feature points to obtain a transformed registration area image, and obtaining a spliced image by using a weighted fusion method based on focus distance. The method can overcome tensile distortion caused by the global correction under a large viewing angle difference, and obtains the spliced image more conforming to human vision.

Description

technical field [0001] The invention belongs to the technical field of image stitching, and relates to an image stitching method based on an optimal stitching plane and local features. Background technique [0002] Image stitching technology is to stitch multiple small-view images with overlapping areas into a complete large-view image, which has better image quality than the large-view images obtained by hardware. Image stitching mainly includes the following three steps: preprocessing, image registration and image fusion. Among them, image registration is the core and key of image stitching, and image fusion is the last step in generating a wide-view image. The current image registration methods are mainly divided into three categories: methods based on transform domain, methods based on gray correlation and methods based on features. The feature-based image registration method is the best method, but the current algorithm generally uses an image to be stitched as a refe...

AI Technical Summary

Problems solved by technology

When there is a large viewing angle difference between the processed images to be stitched, it is easy to generate a large stretching distortion, which affects the stitching result

At the same time, in order to improve the real-time performance of the algorithm, most of the splicing algorithm fusion methods use the general weighted fusion method, which is prone to splicing traces and ghost images.

Images