Adaptive Dense Matching Base On Texture Aware For Oblique Image

Oblique photography can obtain the information of building facade accurately and efficiently,which has a unique advantage in the 3D digital city reconstruction.Dense image matching technology is to establish the corresponding relationship between the stereo image pixels,obtain the pixel three-dimensional location according to the triangulation principle.Oblique Images contain the ground object with complicated depth change,and has obvious occlusion.These make a challenge for oblique image dense matching.Aim to overcome adaptive matching problem due to depth change and the lack of texture,we use semi global matching as framework through comparing the existing stereo matching method.Texture information is introduced as a constraints in the matching process,making adaptive adjustment for parameters,to improve the matching ability of different terrain.The main contents of this paper are as follows:(1)We use a more robust texture detection method through the comparison of different algorithms,which can give a quantitative description of the image,also can reflect the changes of the depth of pixels.Taking into account the texture information in the process of matching can significantly improve the matching accuracy.(2)In order to reduce the sensitivity of the algorithm to matching parameters,we combine different matching cost based on the textural attributes of the pixels,and in order to enhance the robustness of the initial matching cost,cost aggregation in the weak texture region happen.In the matching stage,the parameters of the semi global matching of each pixel are adaptively calculated,and different smoothness constraints are used in different texture regions.(3)We uses GPGPU to realize the matching process,which can improve the efficiency of the UAV image processing,so that the algorithm can be applied in the actual three-dimensional production.