Research On Parallax Image Stitching Technology For UAV Aerial Photography

With its advantages of simple operation and strong maneuverability,unmanned aerial vehicle(UAV)has been widely used in military and civil fields such as regional reconnaissance and remote sensing mapping.Due to the limitation of matching accuracy and deformability,the stitching results of main current methods in large parallax scenes usually contain obvious stitching errors.It has always been one of the research hotspots in the field of image processing to realize seamless stitching of images with large parallax obtained by UAV aerial photography.In this paper,facing the stitching and evaluation requirements of parallax wide scenes in UAV aerial photography,a parallax image stitching technology based on optimized alignment and seam cutting is researched and designed.Aiming at the problems of mismatching and undermatching in traditional parallax image registration,this paper used the improved Oriented FAST and Rotated BRIEF(ORB)based on the quadtree model and Grid-based Motion Statistics(GMS)algorithm based on two-way matching and distance weighted threshold to perform feature extraction and feature matching respectively,to obtain evenly distributed,sufficient,and accurate feature matching pairs.Aiming at the limited deformation ability and the perspective distortion problem of local homography warping,the Thin Plate Spline(TPS)function was used to fit the projection bias through non-rigid deformation to further align the images,and a global similarity constraint strategy was integrated to correct the perspective distortion.Aiming at the problem of existing ghost after deformation under large parallax scenes,this paper used a seam cutting method based on human visual perception to bypass misaligned areas and achieved natural color transition through gradient fusion.At the same time,a perception-based stitched image quality assessment(SIQA)method was designed in this paper.Taking the input images as the reference,the chromaticity distortion,geometric distortion,and structural distortion in the stitched image were respectively quantified,and the final assessment model was obtained through the support vector regression algorithm fitting.The experimental results on the public stitched image dataset and the UAV aerial photography stitched image dataset show that compared with other methods,the parallax image stitching method based on optimized alignment and seam cutting proposed in this paper has obvious advantages in eliminating ghost and misalignment,and can effectively and seamlessly stitch UAV aerial images with parallax.Compared with other assessment methods,the SIQA method proposed in this paper increases the Spearman Rank Order Correlation Coefficient and Pearson Linear Correlation Coefficient of the experimental results from 0.3146 and 0.3358 to 0.7921 and 0.8044,respectively,which confirms that this method can effectively measure the quality of stitched images.