An image stitching method and device

Abstract

The image stitching method and device provided by the present invention can be used for adaptive multi-level layered stitching of aerial video images to reduce stitching cumulative errors, and can adaptively select appropriate processing levels and registration algorithms according to the characteristics of the target scene, effectively solving the problem of image stitching. The stitching error spreads and diffuses as the number of image frames increases. At the same time, the registered image is fused by using the gradual in and gradual out method, which can better eliminate the stitching gap problem caused by the cumulative error of the image transformation matrix and the difference in image brightness. A composite image with a wide field of view and high resolution is obtained, which has significantly improved the accuracy and robustness of image stitching; the "feature stack" is used to store the image features of each frame, and registration and stitching with subsequent images can effectively reduce the The small amount of calculation saves the time of repeatedly extracting image features, and further improves the speed of image stitching.

Description

technical field [0001] The invention relates to the fields of machine vision and image processing, in particular to an image splicing method and device. Background technique [0002] For aerial video image stitching, the traditional frame-to-frame registration stitching method causes the registration errors of adjacent images to accumulate during the stitching process, and these errors gradually become larger as the number of images increases, It directly affects the accuracy and effect of image stitching. [0003] The problem of error accumulation in aerial video sequence image stitching has always been a technical difficulty in image stitching. Researchers at home and abroad have tried to propose some solutions, such as beam adjustment, bundle adjustment, direct sparse Cholesky decomposition method, Kalman The global optimal registration method such as filtering method to obtain the optimal transformation matrix can reduce the global stitching error of multi-frame images ...

AI Technical Summary

Problems solved by technology

[0002] For aerial video image stitching, the traditional frame-to-frame registration stitching method causes the registration errors of adjacent images to accumulate during the stitching process, and these errors gradually become larger as the number of images increases, Directly affect the accuracy and effect of image stitching

[0003] The problem of error accumulation in aerial video sequence image stitching has always been a technical difficulty in image stitching. Researchers at home and abroad have tried to propose some solutions, such as beam adjustment, bundle adjustment, direct sparse Cholesky decomposition method, Kalman The global optimal registration method such as filtering method to obtain the optimal transformation matrix can reduce the global stitching error of multi-frame images to a certain extent, but in some practical applications (such as aerial reconnaissance and emergency disasters, etc.), using global registration Quasi-algorithms cannot meet real-time requirements

Images