Research And Implementation On Image-Based 3D Line Segments Reconstruction For Building

3D reconstruction of buildings is one of the research focuses in computer vision and computer graphics.It has more and more applications in the fields of urban planning,game animation,virtual reality,battlefield simulation and so on.In the image-based building 3D reconstruction,the 3D point cloud model with complex background image is exist holes,and the 3D line segment model can better reflect the building topology,The3 D reconstruction method of building line segments have been proposed.First of all,for the large-scale buildings,the target building's photo collected by the UAV and other equipment are difficult to extract,and the background is cluttered.The network model based on YOLOv3(You Only Look Once)target detection algorithm in deep learning is improved.Perform line segment extraction matching in the detected target building,and eliminate the messy parts of the image sequence.And then a line segment detection method based on LSD(Line Segment Detector)is proposed,which uses the purely polar geometry relationship to detect and match.Finding adjacent pictures in the image sequence and matching only between adjacent images can increase the efficiency of line segment detection matching and provide effective two-dimensional line segments for subsequent 3D line segment reconstruction.Then a method for clustering two-dimensional line segments into a three-dimensional model is studied.Use the spatial and angular re-projection error to assign a correct score to each 3D line segment.When the score is the highest,the position of the specified 3D line segment is used.The 2D line segment is clustered by the Replicator Graph clustering algorithm in the 3D line cluster and optimized by bundle adjustment.Three important parameters,which include fixed threshold ?,nearest neighbors M and matched specific segment in neighbors K,are used in the model.Parameter evaluation has been done in different picture sets and get the suitable parameters by judging the model and the time and number of segments reconstructed.Based on the research of object detection and the three-dimensional reconstruction of line segments are combined to design a three-dimensional reconstruction method forthe image sequence of large buildings,and the feasibility of the above algorithm is designed by experiments.Experiments show that the method is more accurate than the traditional point cloud model,and faster than the method without object detection.What's more it has great application value.