3D Reconstruction With Dense Point Cloud Based On Image Feature

3D reconstruction has always been a hot research field in computer vision,and is an important means for acquiring object models and three-dimensional scenes.There are many ways for people to obtain three-dimensional information on objects.For example,the geometric modeling technology,this method requires a good professional level and large workload;3D laser scanning technology,this method is affected by the outdoor environment,and can not obtain the target texture information.In contrast,the 3D reconstruction method based on image feature points does not have many limitations in the above modeling methods.It requires only input images,has low cost,does not require other special prior information,and can recover three-dimensional information of objects and scenes in images through advanced algorithms.Not only is the equipment needed to be simpler,it has less restrictions on the scene,and it is also able to obtain accurate and realistic models.However,the existing image-based 3D reconstruction process is simple in equipment and low in cost.However,in recovering the 3D information of an object in the image,a large number of redundant points are obtained in the point cloud,and the degree of denseness of the point cloud is insufficient and the model is realistic.The degree is not high,and sometimes it even creates empty holes,making the final reconstruction unsatisfactory.In addition,the three-dimensional reconstruction time spent in the image feature matching process is very costly and inefficient.For these situations,this paper optimizes the image matching process,reduces the time cost of feature matching,and perfects the 3D reconstruction process to solve the shortcomings of the existing processes.Finally,the optimized and improved 3D reconstruction process is integrated into the designed 3D reconstruction system.In summary,this paper does the following work around the three-dimensional reconstruction based on image feature points:· Optimize the image matching process.In this paper,a selective image matching strategy is proposed.In the feature matching stage,all images are not matched in pairs,but by the initial calculation of the camera distance between the images,and for each image,only the similar parts are matched.The image,while not affecting the number of features,greatly reduces the time for feature matching,thereby reducing the time cost of 3D reconstruction.· Improve the existing 3D reconstruction method based on image feature points.On the basis of completing the original three-dimensional reconstruction process,aiming at the problems of point cloud redundancy and lack of visibility in the reconstruction process,the entire three-dimensional reconstruction process is improved and integrated into the following three-dimensional reconstruction system.Including: After the artificial interactive deletion processing of redundant point cloud,the point cloud is meshed to enhance the visibility of the model;the model void repair mode is added to repair the void generated by the point cloud meshing;The sub-processing mode solves the problem of the non-dense point cloud;adding a texture mapping process to the model makes it more realistic.In the end,this paper will show the reconstruction results through experiments,and compare and analyze them.· Development of three-dimensional reconstruction experiment system.Threedimensional reconstruction is a relatively complex process,and the experimental process involved often needs to coordinate the input and output of each process.Therefore,it is often impossible to complete it in a platform and experimental environment.The existing three-dimensional reconstruction system not only lacks pre-processing and post-processing,but also its internal three-dimensional reconstruction algorithm is generally a fixed algorithm and cannot be compatible with other algorithms.In addition,its internal modification is difficult,it is difficult to add the researchers' own algorithms,and also it has increased the time cost and difficulty of research to some extent.In view of this situation,this paper implements a three-dimensional reconstruction experiment platform “Realcon” system,which integrates the entire process of three-dimensional reconstruction ased on the addition of pre-processing and post-processing modes,while allowing researchers to design their own algorithms,added to the system framework Next,it greatly facilitated the progress of research on 3D reconstruction.