Research And Implementation Of Multi-view Geometric 3D Reconstruction Algorithms Based On Linear Features

With the development of computer vision technology,three-dimensional reconstruction algorithm is more and more widely used in robotic navigation,virtual reality and other fields.It is urgent for an economical,practical,efficient and convenient way of three-dimensional reconstruction.Based on the line feature at the bottom of the image,this paper extends the application and improves the performance of the existing light field reconstruction and multi-view geometric reconstruction algorithms based on image sequence.The main research contents are as follows:Firstly,a line detection algorithm based on single-pixel chain edge polygon is proposed.Based on Edge Drawing's idea of edge detection,the algorithm extracts the single-pixel chain-width edge of the input image,compresses the vector data of the edge segment with Douglas-Peucker algorithm,realizes polygonalization of the edge curve,obtains the information of the line segment,and filters the line segment that does not meet the requirements through the resolution adaptive threshold to ensure the extraction of key geometric structures and redundancy straight.Line segment removal.In this paper,the influence of piecewise parameter factors on the performance of line detection algorithm is analyzed,and the algorithm is compared with other line detection algorithms in terms of detection quantity,average length and algorithm speed.The experimental results show that the proposed algorithm has better performance.Secondly,the application of three-dimensional reconstruction algorithm of light field depth based on EPI image sequence and its linear feature analysis in micro-nano scene is realized.Under the platform of nano-robot operation,EPI slicing was performed on the serial images of target samples obtained by scanning electron microscopy(SEM).Linear features in EPI images were analyzed and scene depth information was restored.The three-dimensional point cloud model of target samples was reconstructed in micro-nano scale.In addition,the depth image mosaic method is used to reconstruct the model of the wide-view SEM samples,and the reconstruction performance of the regular and irregular SEM samples is analyzed.Finally,the proposed linear detection algorithm is used to extract the edge linear features of rotating image sequences,and the features are input into the Line3 D framework to reconstruct the linear geometry structure.Finally,the model reconstruction of the macro-building scene and the micro-nano-scale SEM sample with linear structure is realized.Compared with Visual SFM output point cloud model,the results show that the proposed algorithm has wide applicability and strong model structure expression ability.