Research On Real-time Video Mosaic Technology In Moving Foreground

At present,video stitching technology has been widely used in virtual reality,video surveillance,intelligent driving,digital entertainment and other fields.Due to the large amount of data to be processed by video images,improving the splicing speed while ensuring the splicing quality is an urgent problem to be solved in the video splicing technology.In addition,video stitching is likely to produce parallax artifacts and blurring in the case of moving foreground objects,improve the accuracy of stitching,and ensure that the video image is clear and natural,without ghosting and stitching seams.Another need to solve the video stitching technology The problem.In view of this,this subject has conducted in-depth research on real-time video stitching technology under the mobile prospects.The main research contents are as follows:First,understand the research background of video stitching technology and the current research situation at home and abroad,and elaborate the difficulties faced by video stitching technology in the mobile prospect.In-depth study of the relevant theoretical knowledge of image stitching,combined with the fixed feature of the camera,this article speeds up the projection transformation of video images by creating an index table,which lays the foundation for real-time video stitching.Secondly,a fast video registration algorithm based on improved grid motion statistics is proposed.The algorithm uses the SURF feature detector to detect feature points in the feature point detection stage,then uses the binary descriptor BRISK to describe the features,and finally uses the grid motion statistical algorithm to quickly remove the mismatched points,and quickly estimates the transformation model by using the PROSAC algorithm.Experimental results prove that,compared with traditional SIFT + RANSAC,SURF + RANSAC and other registration algorithms,the algorithm in this paper greatly improves the registration accuracy and matching accuracy,and also improves the real-time performance.Then,in view of the artifacts and blurring of video stitching caused by the mobile foreground,this paper proposes the best stitch search algorithm based on visual background extraction to fuse the video.First,the energy function is improved according to the foreground area,grayscale difference,texture and color difference of the overlapping area image.At the same time,the dynamic planning technology with energy minimization is used to find the best stitching line in the non-moving foreground area,and then the moving object is detected using the VIBE algorithm And update the best stitching in the subsequent video images,and finally combine the color difference adaptive fusion algorithm to fuse the two sides of the best stitching.The experimental results show that this method can effectively eliminate ghosting,and the stitching video has no obvious seam.Finally,a real-time video splicing system was built,and the design of video splicing software was completed in the Visual Studio2019 environment.The splicing software uses CUDA stream parallel computing to accelerate real-time video stitching.The video stitching method studied in this paper is applied to the system.The experimental results prove that the stitching method in this paper achieves a good stitching effect and can complete the video stitching in real time.