Research On Registration Optimization Method Based On RGB-D Image Stitching

With the continuous development of computer vision technology,imaging devices have become more advanced and increasingly popular.In this context,how to transform a single small perspective image into a large perspective image has become a focus of attention.But to obtain a larger field of view image,multiple devices need to be used to obtain the image.In order to reduce costs,image stitching technology has become a focus of research and has been widely explored and applied.Image stitching refers to the stitching of two or more images containing overlapping parts to generate a large field of view image.RGB Depth(RGB D)images combine color and depth information,making them more advantageous than ordinary 2D images for stitching.In the process of stitching,image registration is a very critical step.The accuracy of registration results directly affects the stitching effect of the entire image.Therefore,this article focuses on the optimization of registration in image stitching.The research content of this article mainly includes the following three aspects:(1)Registration optimization method based on slope threshold constraint.The traditional image registration algorithm usually has the problem of feature point matching error.Firstly,use registration algorithms to extract feature points from the image,form lines on the initial matching points,and calculate the slope of all lines.Next,determine the overall reasonable range based on the average of all slopes,and ultimately eliminate those that are not within this reasonable range as mismatches.The experimental results verify that this method can effectively eliminate mismatches,optimize the registration effect,and improve the quality of stitching.(2)A registration optimization method based on neighborhood depth value constraints.The traditional image registration algorithm usually uses Euclidean distance,Hamming distance or other methods to measure the similarity between feature points.If depth images are introduced as an aid,depth values can be used as a new measurement method to improve the accuracy of registration.Therefore,this article studies a registration optimization method based on neighborhood depth value constraints for images under scaling.Specifically,it is to calculate the average of the depth values and other pixel points in the neighborhood of the feature points,and eliminate matching points with significant differences in the average values as mismatches.The experimental results verify that this method can also effectively eliminate mismatches,optimize the registration effect,and improve the quality of stitching.(3)Designed an image stitching system.This system can achieve image stitching under the method proposed in this article,as well as image stitching using traditional algorithms,and provide other relevant algorithm choices.In summary,this article conducted in-depth research on registration optimization problems,and studied a registration optimization method based on slope threshold and neighborhood depth value constraints,which was validated by multiple sets of experiments.The experimental results indicate that incorporating the method studied in this article can effectively optimize registration and improve the effect of image stitching.