Image Registration And Stitching Of Binocular Vision And Its Application

At present,computer vision technology is widely used in various fields,for example,Google,Baidu and other technology companies are deeply involved in the field of autonomous driving research.In the autonomous driving visual navigation system,the collection of pre-vehicle image information is considered a top priority.A single lens is used for image information acquisition in the traditional visual navigation system.However,since the single lens angle is small,the information of the captured image is limited.In contrast,binocular vision has more advantages.In the application of binocular vision,in order to obtain more image information of the field of view and break through the limitation of the field of view of a single image sensor,it is often necessary to use image stitching technology.The binocular vision lens has a large parallax when acquiring complex scene images.However,there are still many difficulties in the problem of stitching of parallax images.This thesis analyzes the current common image stitching technology and studies the image stitching technology of binocular vision.A method based on optimal registration and error compensation optimization alignment for divergence distribution of inner point set is proposed.Then,the algorithm of optimizing seam search based on inner point set is studied to realize seamless stitching of parallax images to improve image stitching quality.The main research work of this thesis:1.By analyzing the relationship between the inner point set and the projection plane,a global optimization method based on the divergence distribution of inner point set is proposed.The problem of the images stitching for binocular vision acquisition.The registration error of the quasi-densified feature points and the global error compensation optimization alignment strategy are designed,and most of the ghosting phenomenon in the image stitching is solved.2.Based on the two characteristics of color difference and geometric difference of the seam search algorithm,this thesis proposes a method for optimizing the seam search based on the inner point set for the weight constraint of the seam to solve the phenomenon of ghosting caused by large parallax.Finally,the registration error of the sparse feature points is used to narrow the search range of the seam,and to improve the efficiency of seam search.3.In order to verify the performance of the proposed algorithm in processing parallax images,multiple sets of comparative experiments were designed.Comparison of stitching effects with AutoStitch,APAP(As-Projective-As-Possible Image Stitching)and optimal seam search algorithms.The result shows that the proposed algorithm can effectively solve the problem of obvious ghosting and target mis-cutting in parallax image stitching,and the proposed algorithm has good robustness and applicability.4.A system of parallax image stitching system is designed and implemented.Based on the laboratory software and hardware environment,a parallax image stitching system was designed and implemented.The system mainly includes four major functional modules,namely image preprocessing,image registration,seam search and stitching algorithm comparison.