Research On Video Stitching Technology Based On Improved SURF Algorithm And Color Correction

Panoramic image stitching is one of the important research directions in the field of image processing,involving multiple steps such as registration,fusion,and stitching of multiple images.The development of panoramic image stitching technology can not only realize the presentation of a wider scene,but also play an important role in fields such as robot navigation,virtual reality,and 3D modeling.Video stitching is a further development of image stitching,which poses higher difficulties and challenges,requiring consideration of issues such as time synchronization,motion estimation,and lighting changes between multiple videos.Therefore,using traditional image stitching techniques alone may not achieve good results in some cases.This thesis focuses on improving the image registration and color correction aspects of image stitching technology.Aiming at the problems of high descriptor dimension,long feature extraction time and low matching accuracy caused by the complexity of data in the traditional Speeded Up Robust Feature(SURF)algorithm,an improved image matching algorithm based on SURF has been proposed.In the feature extraction stage of SURF,the number of feature points is reduced by limiting them to the overlapping area,and the 64-dimensional descriptor of SURF is reduced to 20 dimensions by replacing the rectangular area with a circular area to reduce data complexity.In the feature point matching stage,the adaptive KNN algorithm is first used for two-way initial matching of feature points,and then the two-way geometric constraints are combined with the RANSAC algorithm to remove the error matching pairs in the initial matching point pair set to further improve the matching accuracy.Experimental results show that this algorithm reduces the feature point detection time,improves the matching accuracy,and has good robustness.Aiming at the problem of color differences in the original video images,which can result in unnatural panoramic images and color transition marks,a histogram matching-based adaptive image color correction algorithm was proposed.The method uses the histograms of the overlapping areas between images to obtain a preliminary color correction mapping function,and adjusts the function based on pixel cumulative frequency thresholds.Finally,the color mapping correction function corresponding to the reference image is automatically selected using the minimum pixel mean difference method.Comparative experiments have validated the feasibility of the algorithm in achieving image color correction,effectively solving the problem of inconsistent color tones in video images due to objective reasons.Based on the above research results,a real-time panoramic video stitching platform was constructed.This platform mainly uses multi-threading to synchronously obtain multiple video frames,and uses an initialization module to obtain stitching parameters based on the first video frame.Then,based on the stitching parameters,the images are parallelized and accelerated on the GPU using CUDA streams to achieve real-time stitching.Finally,experimental verification was conducted,and the real-time performance of the platform was analyzed.