Research On Extraction,Matching And Bundle Adjustment Of Line Features For Structured Scenarios

In the application of urban objects in aerial photogrammetry,the acquisition of the three-dimensional(3D)structural models and true orthophotos of structured scenario,buildings for example,occupy an important position.At present,the theory and technology of photogrammetry based on point features are relatively mature.Almost all of the photogrammetry softwares and applications processing are based on point features,and the researches and applications of photogrammetric techniques based on line features are relatively few.Line features have some advantages that point features do not possess and have important application value in navigation & positioning,building structures acquisition,and 3D reconstruction.Similar to the accurate 3D reconstruction of point features,the extraction and matching of line features and the bundle adjustment optimization are also the basis and key technology guarantee for accurate 3D reconstruction.Especially in the city’s 3D modeling and mapping,the acquisition of building structure contours requires complete,sufficient,and accurate line feature information.The extraction and matching of lines is the first step;the adjustment solution based on the extraction and matching results is an important part of obtaining accurate 3D line segments.The current excellent line extraction methods have fast extraction speed and good results.However,almost all line extraction algorithms have a problem: the line segment is easily fragmented.That is,it should be a long line segment but is broken into many small short segments due to local minute changes or disturbances.Fragments can increase the amount of data,making processing more complicated.When line matching is performed using oblique photogrammetry technology,only very few matched lines with low accuracy are obtained due to the large viewing angle of the wide baseline.Although there are some studies based on line features to determine the orientation of the sensor,it is still not available to use line features to adjust the orientation of the sensor automatically and to obtain accurate coordinates of the object coordinates.In view of the above-mentioned several key issues,this paper focuses on how to achieve 3D reconstruction of high-quality line segments located on the building structure whose lengths are more complete,quantity is more and the positions are more accurate.These issues will be resolved and verified from the following four aspects.(1)Aiming at how to fundamentally provide a more complete two-dimensional(2D)line segment for 3D line segment reconstruction,an improved line extraction method based on affine invariance is proposed.According to the affine camera model,camera projections are simulated from different perspectives,additional observations are created,and the extracted line segments array is purified and optimized to obtain a more complete line segment.The experimental results show that the method proposed in this paper greatly increases the length of the obtained line segment,greatly reduces the segmentation effect,and can obtain a greater number of useful line segments while reducing redundancy.Applying the improved line extraction results to the line matching of the building,more complete,more quantitative,and higher accurate matched line segments can be obtained.(2)Aiming at how to provide more correct matched line segments for 3D line reconstruction,two kinds of line matching methods aiming at wide-baseline aerial images based on POS information assistance are proposed.The affine transformation model and the perspective transformation model are used to rectify the distortion of the oblique images,eliminate the difference in viewing angle,and match the lines from the conformal images to improve the success rate and accuracy of matching.The experimental results show that the two proposed models have greatly improved the number and correctness of the matched line segments without losing efficiency.At the same time,the perspective projection model is superior to the affine projection model.(3)In order to ensure a more accurate camera auto-orientation and 3D line segment acquisition in the absence of control points,a point-line hybrid bundle adjustment method using horizontal and vertical line constraints is proposed.Using the nonlinear objective function model established by the horizontal and vertical line constraints,the free network and absolute network adjustment optimization processing is performed.The results of simulation and actual test show that compared with the method of bundle adjustment without horizontal and vertical line constraints in the case of only three ground control points,the proposed method using the automatic extraction of line feature constrains to determine the positioning accuracy can be increased by 50%.(4)To verify the effectiveness of the method of extracting,matching and bundle adjustment in(1)(2)(3),3D line segments reconstruction is performed on the edges of the building and used for improving the quality of the ortho-rectified images.The experimental results show that the reconstructed 3D line segments can effectively assist the 3D points to solve the saw-teeth at the edge of the building.On the other hand,it shows that this article reconstructs high-quality 3D line segments with complete length,sufficient quantity,and accurate coordinates.