Research On 3D Point Cloud Growth Algorithm Based On Local Consistency

In this paper,we mainly study how to obtain the high-quality object point cloud by the 3D reconstruction algorithm based on image sequence.The purpose of this paper is to obtain the dense point cloud of object model with high-integrity and correct texture detail by analyzing the image sequence.Although the current 3D reconstruction algorithm based on image sequence is mature,its quality is often unsatisfied.Both sparse and dense point cloud obtained by traditional algorithm cannot meet the reconstruction requirements.Usually,the space structure of the point cloud is not complete,which means it cannot represent the space info of object completely.This problem is usually caused by the difficulty of extracting more feature points from the region with sparse texture on the object.Meanwhile,due to the error in the calculating processing,the point cloud itself has many disordered points.Finally,the point cloud with low quality leads to the final reconstruction of the object model cannot meet the requirements.Therefore,the main work of our algorithm is to fill the vacancy of the point cloud,increase the density of the point cloud and filter wrong points.Meanwhile,the color and texture of the point cloud is mapped correctly.Firstly,this paper proposed a point cloud growth algorithm based on local consistency matching.This algorithm is based on the traditional 3D reconstruction algorithm to obtain the camera transformation matrix and image sequence for the point cloud and image processing.We can obtain large number of matched feature points with our proposed algorithm,which based on the hypothesis we proposed that the adjacent frames change little.These matching points can be mapped to the real position of the 3d space by the proposed point cloud mapping algorithm based on the image model,and some wrong matching points are filtered in this algorithm.Finally,with the contour based point cloud filtering algorithm,we can obtain the point cloud with enough density,which fills the gap in the origin point cloud and can completely represent the structure info of the object model in 3d space.Then,to get the right color of the point cloud,this paper proposed a point cloud color remapping algorithm.The algorithm is based on the point cloud back projection to correct the point cloud's color and texture.In the process of back projection,there will be a problem that point in one side projected through another side of the object to the image plane.So,for filtering the noise caused by this situation,we proposed a noise filtering algorithm based on mean value.And for handling the blur caused by this noise filtering,we proposed a two-way verification algorithm.Finally,With this algorithm,we obtain the point cloud with high-quality where the noise is filtered and the detail is kept.The algorithm proposed in this paper can solve the problem that that the point cloud is missing due to local surface feature is sparse.Meanwhile,after the processing of our proposed algorithm,the color and texture of the cloud is corrected.Finally,through the test on a series of images,our proposed algorithm's effectiveness is verified,and point cloud with high-density and correct color is obtained.