Lunar Terrain Reconstruction Based On Sub Region Fast Stereo Matching Method

With the completion of the plan of China’s lunar exploration project,Chang’e-3and Chang’e-4 have carried out a series of inspection missions to the moon,and have achieved rich results.During this mission,high-precision 3D terrain reconstruction data played an important role in the navigation positioning and path planning of the rover.However,the lunar surface is an unstructured environment and the texture of the lunar surface environment is scarce and complex in features,making terrain reconstruction difficult.In addition,due to the timeliness requirements of the rover detection project,the running time of the lunar terrain reconstruction algorithm is relatively high.This paper studies the topographic reconstruction method of the Chang’e-3 lunar surface.The main research contents are as follows:(1)In the process of 3D reconstruction of the lunar surface based on the rover navigation camera,it is necessary to use the internal parameters of the two lenses of the navigation camera,the distortion parameters and the relative pose parameters between the lenses to perform epipolar correction and point cloud restoration on the stereo pair.Based on this,this paper uses the stereo image pair captured by the navigation camera in the rover ground calibration field,and based on the robust,high-accuracy and easy-to-operate Zhang Zhengyou camera calibration method to calibrate the camera of the Chang’e-3 rover navigation camera.The result of the navigation camera parameters.The calibrated results are used as known parameters in subsequent steps of lunar terrain reconstruction.(2)In order to obtain the disparity map of the lunar scene accurately and quickly,a fast stereo matching method by region is proposed in this paper.Firstly,a grid is constructed by a fixed spacing for initial matching and a Delaunay triangulation is constructed.Taking the weighted centroid of the triangle as the encryption point,the triangulation is encrypted,and the matching grid points and encrypted points are included in the set of support points,and the accurate disparity of the support points is solved by the bisection method based on the neighborhood interval.Then,the image is divided into two types of regions: edge triangles and non-edge triangles according to the disparity difference of the triangle vertices.In the edge triangle area,the optimal disparity in the area is obtained by generating disparity edges,constructing the initial disparity plane,and performing disparity space propagation and normal vector optimization on each pixel in the propagation area obtained by edge expansion.The disparity of the non-edge triangle area is selected as the disparity result by selecting the dot disparity with the smallest cost within the fixed range of the points to be matched.Finally,the disparity of edge triangles and non-edge triangle regions are fused to obtain the optimal disparity value of the entire image.Through a large number of experiments on images in the Middlebury dataset,the method of this paper and the methods of PMS(Patch Match),GEC(Global Edge Constraint),CSM(Consistent Stereo Matching),and SGF(Stereo Gait Feature)are used for comparative analysis.The stereo matching method in this paper is in Both nonocc and disc indicators have higher matching accuracy.In particular,the disc indicator is1.24% lower than the second place,and the matching time is 78.9% lower than that of PMS and 29.95% lower than that of SGF.Fast high precision matching is achieved.(3)After obtaining the disparity map of the lunar scene,this paper obtains the three-dimensional point cloud from the disparity results according to the internal and external parameters of the stereo camera and the principle of binocular stereo vision.Since there are mismatched points in stereo matching,there is noise in the 3D point cloud recovered from the disparity map,which will eventually affect the model results.Therefore,this paper uses the bilateral filtering method to denoise the 3D point cloud recovered from the disparity map.This method can denoise the surface of the 3D model while ensuring the geometric feature information in the point cloud data,ensuring the accuracy of the lunar surface model.(4)After the denoising step,the point cloud data can accurately reflect the topographic changes of the lunar surface.In order to obtain the 3D model of the lunar scene,the surface reconstruction of the point cloud data is required.The Poisson surface reconstruction algorithm is highly closed and watertight.In the process of converting point cloud data into a 3D model,it will automatically repair the loopholes in the point cloud and the uneven distribution of the point cloud.This paper uses the Poisson surface reconstruction algorithm to construct the lunar terrain grid using point cloud data,and further fills it into a solid patch,and finally completes the 3D terrain reconstruction,and obtains the high-precision lunar terrain reconstruction result.The paper have 40 figures,7 tables,and 63 references.