Image Mosaic Research Based On The Video Sequence

Image mosaic based on video sequence is an important research direction in computer vision,which is widely used in aerial panoramic image acquisition,large field of video surveillance,and other aspects.Unlike the conventional research on image mosaic with several pictures given,the research on image mosaic based on video sequence still needs to consider how to select the image frame that is suitable for mosaic from the video,and how to solve the problem of composite ghosts caused by moving objects in the video.This dissertation focuses on the above problems and conduct research from the following aspects1.How to get appropriate frames from video sequence for mosaic,which is a fundamental problem in image mosaic research based on the video sequence.Taking aerial video as an example,this dissertation proposes an effective video frame acquisition method based on multi-module cooperation to avoid video redundancy and shooting defects.Specifically,the video stability module,image quality module,and shooting distance module are designed,and the three modules cooperate with each other to complete the constraints of video stability,image quality,and content scale.As a result,experiments are carried out on an aerial video data set with diverse scenes,comprehensive challenges,and more suitable for mosaic research,and the results verify the effectiveness of the proposed scheme.2.To address the problem of moving object interference in the video,this dissertation proposed a single composite ghost recognition and elimination method based on object perception.Firstly,a composite ghost recognition algorithm based on motion perception is put forward to detect the moving objects that may cause composite ghost in two stitched images;then,a replacement region selection strategy based on object perception is designed to prepare for eliminating composite ghosts;finally,based on the identified composite ghost and the selected replacement region,the composite ghost is eliminated.Therefore,experimental results show that the proposed method can effectively eliminate the composite ghost.3.To solve the problem that multiple moving objects in complex scenes affect the matching accuracy and cause complex composite ghosts,this dissertation proposes an image mosaic method based on the comprehensive selection of feature points and the elimination of complex composite ghosts.Specifically,on the one hand,feature points are selected from overlapping region,moving objects and adaptive projection verification to improve the accuracy of image matching;on the other hand,multiple composite ghosts are identified and eliminated by describing the trajectories of multiple moving objects.Therefore,the experimental results show that the proposed method can deal with complex scenes well.