Research And Implementation Of Image Mosaic Technology Based On Video Stream

The image stitching technology based on video stream extracts several key frames in a video,and performs feature point matching and border stitching on the image of the overlapping area of the key frame to generate a complete stitching image.The technology is in computer graphics,video monitoring,UAV aerial image,remote sensing image and other fields show a broad application prospect and value.With the rapid development of computer technology and information technology,video surveillance has developed into an indispensable part of social life.Railway traffic surveillance video is a typical representative.Through camera monitoring,Internet transmission,and realtime display of surveillance images,effective image information can be used in the video.Complete image stitching,find hidden dangers in time,and effectively ensure the normal operation of the railway system.However,due to the large amount of video data,complicated calculation methods,and prone to distortion,there are still many problems.For example: automatic extraction of video keyframes,selection of image feature points and feature regions,smooth transition of image fusion,and improvement of stitching algorithms.The solution of these problems will have a certain impetus to the image stitching technology based on video stream.Therefore,based on the specific needs and improvement direction of video image splicing technology,this paper designs a set of video surveillance for railway based on a series of traditional image splicing algorithms and combined with the latest developments in target detection and deep learning.The image stitching algorithm,the main content can be summarized as follows:(1)Design a key frame extraction algorithm based on target detectionThrough the application of convolutional neural network(CNN)in the field of target detection,a key frame extractor is designed and applied.This includes selecting traditional SIFT feature points as rough recognition,and using convolutional neural networks to mine image depth features as fine recognition,and then obtain video key frame sequences.Finally,through the evaluation index of key frame rate,the experimental results show that the proposed algorithm extracts fewer frames and the average key frame rate is lower,reaching 24.60%,which improves the efficiency by nearly half.(2)An image mosaic algorithm based on optimized SIFT features is proposed.For the key frame image sequence,only the feature point pairs of the overlap region are considered,and the SIFT feature points are extracted using the maximum feature extraction method of the coincidence region.According to the sensitivity of the Harris corner point to the scale transform,the Harris corner points are detected in the extracted SIFT feature points.And removed more unnecessary points.The BBF feature search algorithm is then used to find unmatched feature points.The RANSAC feature extraction algorithm is used to remove invalid feature points.Finally,the improved fade-in and fade-out fusion weighting algorithm is used to eliminate seams.Feature matching,image transformation and image fusion are done.(3)Design and implementation of image stitching technology in railway traffic monitoring videoConstruct a complete technical framework to achieve key frame extraction and image stitching in railway surveillance video.Focusing on the rough matching rate,fine matching rate and image entropy,mutual information,average gradient and other index evaluation indexes in image matching,the SIFT algorithm and the algorithm are compared,and the efficiency is found.And real-time performance has been improved.