An Research On Key Technologies Of 3D Reconstruction Of Unordered Images

The rational use of land resources is directly related to the sustainable development of society.Rapid acquisition and accurate measurement of land data are an important prerequisites for land management,land planning and land consolidation and rehabitation.Aiming at the problem that traditional land surveying and mapping technology is time-consuming and laborious,this paper focuses on a 3D reconstruction technology based on unordered images.The main research contents are as follows:(1)An image mismatch elimination algorithm based on adaptive neighborhood testing is proposed.Aiming at the problem that the matching points obtained for the traditional ratio test cannot balance the matching correct rate and the matching number.The high threshold ratio test algorithm is designed to preliminarily filter matching points and ensure that the accuracy of filtered matching points reached more than 70%.An adaptive neighborhood testing algorithm for matching points is proposed.The constraints of the actual spatial distribution of matching points are established,and the accurate detection and elimination of mismatching points are realized.Compared with the traditional ratio test,the number of matching points obtained by this algorithm is increased by more than 30% without reducing the matching accuracy.(2)Two fast overlap detection algorithms for unordered images are proposed.Aiming at the problem that a large number of unnecessary matching computation seriously reduces the efficiency of the algorithm due to the lack of prior spatial distribution information in unordered images.For small data sets,a non-overlapping image discrimination algorithm based on color histogram is proposed.Fast recognition of non-overlapping image pairs is realized.The efficiency of unordered image matching is improved twice than before.For large data image sets,a region adaptive SURF algorithm is proposed.Establishment of similarity criterion based on uniform strong feature points matching number,fast ordering processing of unordered image sets is realized.The matching efficiency is increased by more than two times than before,and with the increase of data volume,the matching efficiency is improved more significantly.(3)An incremental 3D reconstruction algorithm for unordered images is studied.Fast matching of unordered images based on the above two algorithms.Reconstruction of initial point cloud model by Structure from motion Algorithms,Accuracy optimization of reconstruction results using bundle adjustment.Combined with the existing point cloud model,the other images are reconstructed incrementally and iteratively by DLT method,and finally the whole unordered image set is reconstructed rapidly.