5D texture grid data structure

Abstract

The invention discloses a 5D texture grid data structure used for four-corner grid reconstruction of point cloud data. The 5D texture grid data structure comprises a plurality of half-edge four-cornergrids which are connected in a topological mode; each half-edge four-corner grid comprises a four-corner surface patch and four vertexes S, S<m,n+1>, S<m+1,n> S<m+1,n+1> arranged on the four-corner surface patch, and a plurality of half edges; based on any point Pij in the point cloud data and three adjacent points of the Pij, four corresponding vertexes are obtained; each vertex stores 3D position coordinates (x, y, z), normal vector and color of the corresponding point Pij in the point cloud data and pointer information of a half edge adopting the point Pij as an initial vertex; each half edge stores the pointers or indexes of an initial vertex, a termination vertex, an adjacent surface, an upper half edge, a lower half edge and an opposite half edge; and after the 3D point cloud isconverted into the 5D texture grid data structure, indexing is carried out by using the connection relationship among the vertexes, the half edges and the four-corner patch, so that the omnibearing information of a three-dimensional object can be effectively represented.

Description

technical field [0001] The invention belongs to the technical field of computer graphics and computer vision, in particular to a 5D texture grid data structure used for data gridding of point clouds. Background technique [0002] In the fields of industrial manufacturing and logistics, unordered sorting tasks such as loading and unloading of machine tools, sorting of parts, and sorting of goods are labor-intensive, low in sorting efficiency, and difficult to recruit for traditional manual operations. Robot disorder sorting technology is increasingly replacing manual labor, and has become an important means for related industries to reduce costs, increase efficiency, and upgrade intelligently. The random sorting of robots relies on visual recognition methods to locate the objects to be sorted, and then guides the end of the robot to complete the grabbing and placing operations of the objects. Generally, 2D images, 3D point clouds or grids are used to represent object informa...

AI Technical Summary

Problems solved by technology

[0003] At present, traditional 2D vision or 3D visual recognition methods cannot solve the problem of universal and accurate recognition of massive disordered posture workpieces, and it is difficult to meet the disordered sorting needs of e-commerce logistics

When the workpieces to be scanned have a large attitude inclination and are stacked in disorder, such as identifying instant noodles in barrels placed on the side of the turnover box, the traditional 2D method needs to make multiple templates with different viewing angles for matching and positioning. In this way, when facing a large number of workpieces, establish The workload of the template is too large; when the workpieces to be inspected are closely placed, it is difficult for the 3D vision method to segment different objects in the scene, and it is difficult to adapt to the workpieces to be inspected with small changes in object shape, such as identifying closely arranged mobile phone packages in a turnover box box

Only by recognizing the complete shape of the workpiece (including the side and bottom surface), can the workpiece be placed accurately and avoid damage to the workpiece when grabbing the precious workpiece. The traditional 2D / 3D visual recognition methods can only be based on the local information on the upper surface of the workpiece. Recognition, it is difficult to judge the complete shape of the workpiece to be measured, and it is difficult to adapt to accurate stacking

[0004] The above problems are mainly due to the fact that the existing visual recognition methods are limited to the data structure representing three-dimensional objects. Thanks to the matrix form of 2D images, the image processing methods in 2D vision are very mature, but it is difficult for 2D images to represent the three-dimensional objects. All-round information; in 3D vision processing, most of the existing 3D image processing methods are based on disordered point clouds, and it is difficult to achieve ordered neighborhood access similar to 2D, so that 2D texture features cannot be used; 3D images based on ordered point clouds Processing method, since the ordered point cloud can only express the upper surface information of the workpiece, it cannot express the bottom surface and side surface, nor can it realize the expression of the complete shape

Images