Key Technologies Research Of UAV-based Push-broom Aerial Image Stitching

In recent years,with the increased automation of UAV platforms and their convenient and fast deployment advantages,UAV-based infrared aerial imagery has been widely used in various fields such as geological analysis,natural disaster rescue and military reconnaissance.Due to the low spatial resolution and narrow field of view of infrared images,scanning imaging is often used to obtain a large number of scanned image sequences of the observed scene in order to achieve a larger range of scene observation.Afterwards,image stitching methods are used to obtain panoramic images.The main work for UAV-based push-broom aerial image sequences is as follows:The effect of the recursiveness of the SURF-based transformation matrix on the alignment accuracy of the image sequences is investigated.The SURF algorithm itself has high accuracy and can detect and match feature points in the images accurately.By using the recursion of the transform matrix,the transform matrix of the previous image is used as the initial value of the transform matrix of the next image,which preserves the feature information of the original image and improves the accuracy of image alignment.A global optimisation method is introduced and a global optimisation objective function is constructed,which effectively improves the image alignment effect.In the task of stitching large batches of infrared images,after the initial alignment of the image sequence,the accumulation of errors can lead to serious stitching misalignment and ghosting problems in local areas.Based on the initial global position coordinates of the infrared image sequence and the alignment results of adjacent images,a global optimisation objective function is constructed and the most rapid descent method is used to optimise the objective function.Then optimised global image position coordinates are obtained,which reducing the alignment errors caused by the accumulation of errors.The search strategy for the optimal stitching line is improved.The image texture complexity is introduced,and the image texture information,gray difference and gradient difference are integrated to jointly determine the energy function.The traditional search strategy is to find several stitching lines with different starting points in the overlapping region,and the one with the lowest energy value is taken as the optimal stitching line.The improved search strategy uses the boundary points of the overlapping regions as starting points,improving the search efficiency and eliminating as many visible stitching marks as possible.Based on the improved energy function,the energy values of the pixel points in the overlapping region are compared and the pixel point with the lowest energy value is used as the extension point of the stitching line to obtain an optimal stitching line that passes through a region of similar grey scale and low texture complexity,effectively avoiding the pixels at the stitching line locations from passing through the complex texture areas of each stitched image in the overlap region.The optimal stitching line fusion image is obtained that is more in line with the subjective visual perception of the human eye.The experimental results show that the proposed method is feasible and effective in terms of both subjective visual effects and objective indicators.