Research On Disparity Refinement Based On Line Segment Detection For Remote Sensing Image Depth Discontinuous Region

Remote sensing images contain rich spatial geographic information,and have important application value in military operations,urban planning,geological survey and other research fields.Remote sensing images can obtain important three-dimensional spatial information.The research on stereo matching technology of remote sensing image is complicated,the remote sensing image is more difficult than the traditional image matching because of the diversity of objects,large deformation and obvious illumination difference.Stereo matching technology often use a fixed window to find matching points,matching speed,wide disparity,but building edge area,belongs to the depth of discontinuous area,there are often jump of parallax,using a fixed window matching precision is low,easy to cause the matching ambiguity,additional buffer,on the edge of the roof edge expansion and the edge is not clear.In general,the effect of disparity refinement algorithm is not obvious in the depth discontinuous region,and will change the disparity of ground objects near the edge of buildings.The matching problem of deep discontinuous regions will be solved.In order to obtain dense and clear edge disparity map results,this paper proposes a disparity refinement algorithm based on line segment detection for remote sensing image depth discontinuous region.Aiming at the above problems,this paper selects semi-global Matching(SGM)algorithm as the Matching technology to study.On the basis of detecting the building edge line segment,line segment constraint is introduced to adjust the energy equation and optimize the final disparity result.The specific research content of this paper is as follows:First,extract the edge line segment of the building.Line Segment Detection(LSD)is used to detect the initial Line Segment of remote sensing image.Then,in order to extract the line segment that conforms to the edge of buildings,the direction of the line segment is adjusted and the significance calculation is made to fit the elliptic region of the line segment.Then a feature descriptor based on the support window of edge line segment is constructed to calculate the feature region and match the corresponding line segment.The stability of the stero matching proposed algorithm is revealed in different regions,and the matching accuracy of the algorithm is calculated on statellite images.The results indicate the validity of the algorithm.Secondly,an effective disparity refinement algorithm based on edge segment constraint is proposed to solve the problem of building edge spreading and unclear edge.In the cost calculation stage of stereo matching,three transformation methods are combined,and a new energy function is proposed by introducing line segment condition in the cost aggregation stage.The line segment detection algorithm extracts the line segment at the edge of the building,which is usually the disparity depth discontinuous area.Global energy function of smoothing according to punish disparity condition constraints,this article on smooth give similar buildings adjacent segment level edge pixels in a larger,the value of penalty and building a small edge line adjacent grayscale differences pixel value,the key into the edge of the disparity refinement problem energy function of the minimum value problem,The matching problem of depth discontinuous region is solved.In the post-matching stage,the problem area of disparity map is optimized.The proposed algorithm is compared with other algorithms,and the error matching rate of depth discontinuous region is calculated on Middlebury standard dataset and remote sensing image.The results show the improvement of the proposed algorithm.Finally,different line segment detection methods are compared,and the matching accuracy is calculated on remote sensing images,which reflects the robustness of the proposed method and the accuracy of line segment detection for different areas.In order to verify the effect of the improved cost calculation,experiments are carried out on the three cost transformations and the improved fusion transformation respectively,and the results show the effectiveness of the improved cost.After that,the effect of penalty coefficient of smoothing term is tested by parameter selection experiment and the effect of different post-matching processing methods is verified.In this paper,we compare different disparity refinement algorithms,calculate the mismatching rate of depth discontinuous region on Middlebury standard data set and remote sensing image,and make the disparity of building edge more fit the actual edge shape.