Road Surface Image Transformation And Stitching Method Under The Perspective Of Road Inspection Unmanned Aerial Vehicle

As the mileage of highways continues to increase,maintenance has become a top priority for road traffic,and the demand for pavement disease detection is increasing.Compared with commonly used manual and vehicle-based methods for detecting pavement diseases,inspection drones are widely used due to their strong flexibility,fast detection speed,and high safety.Due to limitations in camera perspective,the content covered by a single image is limited,so aerial images of roads need to be stitched together to obtain an overall view of the road.However,directly stitching oblique projection images can cause severe distortion of image resolution due to differences in shooting perspectives.Therefore,this study focuses on researching transformation and stitching methods for aerial pavement images.The main research contents are as follows:(1)A road image inverse perspective transformation method based on lane line detection is proposed to address the problem of perspective distortion caused by the camera axis not being perpendicular to the pavement during photography.By converting the collected image to the HSV color space,the information of the outermost left and right lane lines is detected and filtered through threshold segmentation,Canny edge detection,and accumulative probability Hough transform,and then an inverse perspective transformation is performed to obtain a rectified image of the road.(2)An aerial pavement image registration method based on pavement features is proposed to address the issue of non-pavement feature affecting homography accuracy.Firstly,SURF algorithm is used for feature extraction;then,a dataset is established and data augmentation is performed,and the DeepLabv3+ model of semantic segmentation is used to effectively extract the pavement area from the image,keeping only the feature points in the pavement area and removing those in the non-pavement area.Finally,the filtered feature points are subjected to FLANN coarse matching and RANSAC fine matching to calculate the homography matrix of the image.Experimental results show that this method effectively avoids the problem of misalignment caused by the existence of non-pavement areas,and its matching time is reduced by 26.4% compared to the traditional method.(3)To further improve the stitching effect of aerial pavement images,a road image stitching algorithm based on the best seam line is studied,which uses dynamic programming to find the best seam line to fuse the images.The effectiveness of this method compared to commonly used stitching methods is analyzed from subjective and objective perspectives.Experimental results show that the proposed improved SURF stitching method has advantages in stitching aerial pavement images.