Research On Bridge Image Stitching Technology Based On Improved SURF Algorithm

Image stitching is an important research field of digital image processing,and it is widely used in many scenarios such as the shooting of panoramic images,the collection and postprocessing of high-resolution remote sensing images,and the processing of drone aerial images.In this article,it is mainly aimed at the application scenarios of using drones and other tools to carry cameras for bridge bottom quality inspection.Since the information of a single bridge image is limited.Image stitching algorithms are used to extend the image information.So that some disease information scattered in different images can be completely displayed in one image.In this application scenario,on the one hand,due to the use of high-definition cameras for shooting,the number of pixels in the image is so large,and excellent feature extraction algorithms such as SURF are gradually difficult to meet the real-time efficiency requirements of image stitching.On the other hand,due to the instability of the direction,height,jitter and other factors during the flight of the UAV or other tools,the acquired images are deformed and distorted,which affects the stitching quality of image stitching.In order to solve the above problems,this paper deeply studied the feature extraction mechanism of the SURF algorithm and the image fusion mechanism of the weight fusion algorithm,and proposed corresponding solutions.The specific innovations and main tasks of this paper are as follows:(1)An improved SURF algorithm is proposed,which improves the overall efficiency of the algorithm by dynamically programming the feature point detection area.In the traditional SURF algorithm,the detection and description range of feature points is the entire image.In the application of a single image,this feature point extraction method can effectively reflect the feature distribution of the entire image.But in the application of image stitching,the feature point extraction of two or more images will be involved.There is usually only partial overlap between multiple images.If you continue to perform feature extraction on the entire image,there will be a lot of redundancy that does not participate in feature matching.Based on the actual overlap ratio of the images,this paper dynamically plans the feature point extraction area to avoid redundant feature points from participating in feature extraction and matching.The experimental results show that compared with the classic SURF algorithm,the method in this paper can effectively improve the execution efficiency of the image stitching algorithm.(2)A non-linear weighted image fusion model is proposed,which can effectively reduce the blur,ghost,brightness transition and other distortions caused by image fusion for complex images with large pixels and fast color changes.In the classic progressive image fusion algorithm,the left and right(or up and down)images participate in the construction of the overlapping area of the stitched image with linear weights.In the image with simple information and less distortion,a higher image fusion effect can be achieved.But for the image with complex information and large distortion have poor fusion effects and often have serious ghosting phenomena.Aiming at this defect,this paper proposes a new nonlinear weighted image fusion model.Experimental results show that using this model for image fusion experiments can effectively reduce the possibility of ghosting and other phenomena.Compared with classic algorithms and recent excellent improved algorithms,it has achieved significantly better subjective effects and higher objective quality.(3)Designed and implemented an image browsing system with image stitching function.In this system,there are mainly four modules: image set preprocessing and loading module,image set viewing module,image tagging module,image stitching and mosaic processing module.The system can realize the overall preprocessing and import of the image set to be spliced,and display it in the adjacent position according to the actual position relationship of the image to be spliced.In the system,the displayed image can be zoomed,shifted,marked,spliced.It is convenient to view and process the image.