Design And Implementation Of Image Acquisition And Stitching System Based On Improved SURF Algorith

With the increasing demand for image information,the limited view of traditional cameras limits their application in certain scenarios,while wide-angle cameras are expensive and difficult to popularize.So image stitching has become a solution of great interest,which stitches multiple images with small viewing angles and overlapping areas into one image with more image information and larger viewing angles.In this context,this paper proposes a new image stitching method and designs and implements an image acquisition and stitching system based on Raspberry Pi 4B to meet the needs of people in practical applications.The details of the research are as follows:First,in order to make the image acquisition and stitching system more pervasive,it is necessary to pre-process the image to be stitched to improve the stitching image quality.There are several options of image enhancement,image blur removal,defogging and pepper noise removal in the preprocessing process.The experimental results of using different pre-processing algorithms are compared,and the preprocessing algorithm with the best restoration is selected to be embedded in the system pre-processing module for users to choose according to the actual needs.Image registration is the basis of image stitching,and the single-response matrix obtained from the registration directly affects the quality of image stitching.The SFinROI image registration algorithm proposed in this paper addresses the problem of large computational effort of traditional SURF algorithm,which uses FAST to extract feature points,constructs the main direction of feature points by using the intensity centroid,and constructs SURF descriptors for feature points,the obtained feature points then have descriptors and rotation invariance to enhance the stability of feature points.A combination of two algorithms,the bidirectional FLANN and PROSAC,is used in the feature matching stage to improve the matching accuracy.According to the feature that the two images to be stitched must have the same region,the same region is set as ROI,and the registration is performed in the ROI area.After experimental verification,compared with the full-frame image,extracting ROI can effectively reduce the feature extraction time,improve the effective utilization of feature points,and effectively reduce the problem of error matching in complex image feature matching.The experimental results in Affine Covariant Features dataset show that the SFinROI algorithm in this paper has good performance in speed,accuracy and matching point pairs compared with traditional algorithms.It achieves the dual requirements of high accuracy and high real-time in the image registration stage,and lays a good foundation for the subsequent image stitching.Secondly,the SFinROI algorithm proposed in this paper is combined with bilinear interpolation to form a new stitching method.SFinROI can obtain a better single-strain matrix,which can align the two images to be stitched spatially more accurately,and perform pixel-by-pixel fusion of pixel points in the overlapping region using bilinear interpolation,which can effectively reduce the problems such as stitching seams and ghosting.The experiments compare the stitching method of this paper with two commonly used stitching methods,combining subjective evaluation,objective evaluation and pixel size of the stitched images,and prove that the SFinROI-based stitching method can effectively reduce the stitching seams and ghosting and other problems,and the large-view image stitching shape is more natural.Finally,the image acquisition and stitching system is designed and implemented based on Raspberry Pi 4B,and the Py Qt5 development tool is used to design the image acquisition stitching system software platform.The SFinROI stitching method and the corresponding algorithm of pre-processing proposed in this paper are embedded into the image acquisition and stitching system.The main interface of the system is designed to be more user-friendly and convenient for users to operate.In addition to the user interface,there is also a login interface and a registration interface to meet the user’s requirements for the image acquisition and stitching system to protect personal privacy and other more requirements.