Point Cloud And Image Data Fusion Approach For Three-dimensional Colored Surface Reconstruction

3D reconstruction is a key technology in computer graphics,virtual digital reality,automatic driving,autopilot and roboter application fields,which has been widely used in various research directions.At the same time,3D reconstruction and sensor fusion are the important solution to spatial environment perception.In this dissertation,the computer vision technology is used to reconstruct the interior 3D colored surface by using Visible Light Camera and Lidar Scanner.The research contents are as follows:(1)In the point cloud data acquisition process,considering the problem that the data collected by Lidar Scanner is large and contains discrete points,the statistical analysis method and the K-proximity algorithm are used to remove the discrete points and conduct downsampling processing on the data collected by Lidar Scanner.After preprocessing,the data volume and the number of discrete points decrease significantly.And the detailed textures of the point cloud are not affected,which is convenient for subsequent operation of the point cloud.(2)As there is no texture information on the surface of the scene in common 3D reconstruction,it is impossible to reproduce the real scenes.In this dissertation,point cloud and image information are fused,and a new MRF model is used.Through this algorithm,the reconstructed model can be attached with rich texture information,so as to obtain a more intuitive and real 3D scene model.In this dissertation,the method of one-to-one mapping is adopted to map 3D point cloud to 2D pixel space.Feature points are obtained from the mapped image and RGB image for registration.MRF model is adopted to supplement missing points on point cloud space,and finally a 3D colored space model with uniform spatial distribution is obtained.(3)In this dissertation,considering the problem of point cloud registration that the iteration takes a lot of time,Scale Invariant Feature extraction method is adopted.It can carry out point pair features matching based on the rotation invariant feature of image,so as to realize registration among point clouds and build a complete indoor three-dimensional colored model.Finally,a complete surface model is constructed by delaunay triangulation.The three-dimensional colored surface reconstructed method is proposed.It can be used for data collection and calculation with lower cost equipment,and efficiently complete the three-dimensional reconstruction of indoor scenes containing colored information.Compared with the reconstruction of the traditional methods,the experiment finally reconstruction of 3D scene model progress higher,spend less time,with good texture information,and more close to the real scene.In the actual production and daily life,it has a certain application value and development prospect.