High-resolution Tilt Aerial Image Dense Matching Based On Improved SGM

In recent years,the development of tilted aerial photography technology and the widespread application of unmanned aerial vehicle platforms have greatly reduced the cost of acquiring multi-view aerial images,and it has high efficiency and flexible methods.However,its stereoscopic image has a large change in relative viewing angle,severe occlusion,and the presence of unstructured areas brings challenges for matching work.The quality of existing SGM methods of outputing disparity maps still needs to be improved.Therefore,this paper studies the improved algorithm of SGM dense matching in aerial imagery.First,the research background and significance of stereo matching in today's real society are expounded.Making a systematic elaboration of the current research status and existing problems of stereo matching at home and abroad in recent years.Secondly,it introduces SGM stereo matching and matching based on mutual information,and improves the traditional SGM algorithm.Using MVD one-dimensional dynamic planning for stereo matching,the initial disparity map and initial occlusion map are obtained.Taking this initial occlusion map as a priori condition,occlusion restrictions are introduced in the SGM algorithm.then,This article introduces Construction of Slanted Plane and calculation of penalty factor for pixel matching cost.The stereo images to be matched are primarily segmented in color space and for each segmented region,the Slanted Plane well presenting the local scene with gradually changed depth is constructed using initial disparity map.Furthermore,the penalty factor associated with certain Slanted Plane is outputted by predefined arctangent function and is imposed on pixel matching cost in SGM to ensure the proper propagation of depth information.Accurate disparity map is conducted using the scene structure knowledge from image segmentation and occlusion detection for interpolation operation.At last,the experiments implemented on standard test images from website and three typical urban aerial image stereos show that,compared with the traditional SGM,dynamic planning matching and their improved methods,proposed approach can output more accurate occlusion map and uniform disparity map in which the target boundary is more complete.Moreover,proposed approach is an additive algorithm that well follows the matching framework in original SGM and does not significantly increase computation.Therefore,it has good application value,especially for handling the urban aerial image with large relative angle of view and serious occlusions.