| With the maturity of 3D scanning equipment and depth sensors,the point cloud as the model of 3D object representation has also received more and more attention.It has been widely used in various fields such as 3D scene reconstruction,navigation,driverless and cultural relic reconstruction.However,due to the limitation of sensor accuracy,the impact of environmental factors and the loss of some data during transmission,the acquired point cloud data usually has holes.Incomplete point cloud data will reduce the accuracy of downstream work,so the repair of point cloud holes is an indispensable task in 3D vision.In addition,the point cloud is a disordered 3D model,and the formation of holes is random,and the shape of holes is irregular,or the size of incomplete parts is unpredictable.This also causes many methods to be unable to accurately restore the original shape of the missing area,and the repaired surface and the adjacent surface are too smooth.In view of these problems,the research work of this paper is as follows:For large point cloud holes,this paper proposes a specific sub-unit screening method based on Markov random.Generally,the smaller the hole area is,the less difficult the repair is,and the higher the accuracy will be.In view of this,this paper first discretizes the 3D space around the hole boundary point,and discretizes it into k sub-elements;Then,in the discretized sub-unit set,the sub-unit that needs to add the new filling point most is selected as the specific sub-unit.For the selection of specific sub-elements,Markov random field(MRF)is introduced in this paper to transform the selection of specific sub-elements into the maximum a posteriori probability problem(MAP)of random field label.Then,the variational inference is used to approximate the MAP,and the specific subunits that need to add new points are calculated to pave the way for the subsequent repair of new point clouds.To accurately restore the surface shape of the point cloud,this paper proposes a hole repair method based on convex sets.Considering the advantages of convex sets in dealing with uncertainty problems,the point cloud repair method proposed in this paper is to transform the point cloud repair problem into the construction problem of convex sets.Firstly,this paper weakens the convex set which is suitable for continuous space,so that it can be applied to the repair of scattered point clouds;At the same time,the property of the weakened convex set is used to determine the shrinkage direction of the hole region,so as to add new 3D points in the direction domain.After the specific subunit is determined,the area of the filling point is greatly reduced.Finally,according to the definition of convex set,new 3D points are added to the special sub-unit to make it a convex set.The method in this paper iteratively implements the discretization of the region and the addition of new filling points.With the increase of the number of iterations,the specific sub-elements gradually move towards the center of the hole region until the hole is completely repaired.Finally,the research method in this paper is tested in the Windows 10 system environment using C++ program language combined with open-source libraries such as PCL and Open CV.Experimental results show that the proposed method can repair 3D point clouds with high quality. |
