Research On 3D Reconstruction Of Spruce Based On Unmanned Sequence Image

With the increasing improvement of science and technology and living standards,two-dimensional information can no meet people's growing perception and visual needs,and three-dimensional reconstruction technology has brought a glimmer of life to solving such problems.At present,with different reconstruction equipment,3D reconstruction technology is mainly divided into:3D reconstruction based on active vision and passive vision.Based on passive vision,the method of 3D reconstruction using sequence images is widely used in various occasions because its low cost,easy acquisition of images,and high degree of automation.The 3D reconstruction technology based on UAV image sequences is more popular because of its ability to reconstruct large-scale objects or large-scale scenes.This paper mainly studies camera algorithms,feature point detection,sparse reconstruction,and dense reconstruction in 3D reconstruction.The main research contents are as follows:(1)Firstly,the more commonly used camera models and camera calibration methods are introduced;secondly,the widely used SURF and Harris feature point detection algorithms are studied.In view of the lack of scale invariance of Harris corner detection algorithms,an improved Harris detection algorithms is proposed.The point detection algorithm uses the scale invariance of the Sift algorithm.Before Harris detects the corners,a Gaussian pyramid scale space is established for the image,and then corner detection is performed.When corner judgment is performed,the simpler and faster minimum feature is used.The corner response function was calculated by using the method.The spruce image was used to verify the proposed algorithm.The experimental results verified the robustness of the improved algorithm proposed in this paper.(2)Followed by the research of sparse reconstruction algorithm based on the structure from motion,aiming at fundamental matrix estimation in sparse reconstruction precision is low,is proposed based on multiple kernel learning the basis of the peak density matrix estimation algorithm,using multiple kernel learning methods to improve density peak value of the local density is defined in the algorithm,and clustering center by gamma distribution automatic selecting,using the improved matrix estimation method to estimate the fundamental matrix,and use the triangulation method to calculate the three-dimensional coordinate of the feature points finally to BA the camera parameters and 3d coordinate optimization of sparse reconstruction results.The 3D sparse point cloud of spruce was reconstructed using UVA image and the proposed algorithm was verified by experiments.(3)Finally,the stereo vision dense reconstruction algorithm based on planar slices is studied.Based on the sparsely reconstructed sparse point cloud,a planar slice with size and direction is defined.Using the improved Harris and DOG feature point detection algorithm for each correct.The matching feature points are used to reconstruct a corresponding planar slice.In order to ensure the correct generation of all the planar slices,the incorrectly generated planar slices are filtered using the visible consistency constraint,and finally a finer dense 3D point cloud is reconstructed.The improved PMVS algorithm is used to perform densification experiments on multiple sparse point clouds.The experimental results show the effectiveness and robustness of the improved plane-based densification reconstruction,which provides strong evidence for the research on the interception of spruce canopy.