System and method for rendering of texel imagery

• US 7,940,279 B2
• Filed: 02/14/2008
• Issued: 05/10/2011
• Est. Priority Date: 03/27/2007
• Status: Active Grant

• Current Assignee: Utah State University
• Original Assignee: Utah State University
• Inventors: Pack, Robert Taylor
• Primary Examiner(s): Yang; Ryan R
• Application Number:
  US12/031,088
• Publication Number:
  US 20080238919A1
• Time in Patent Office: 1,181 Days
• Field of Search: 345/581, 345/582
• US Class Current: 345/582
• CPC Class Codes: G06T 15/04 Texture mapping
• CPC Class Codes: G06T 17/005 Tree description, e.g. octree, quadtree
• CPC Class Codes: G06T 17/20 Finite element generation, e.g. wire-frame surface description, tesselation
• CPC Class Codes: G06T 2210/56 Particle system, point based geometry or rendering

Abstract

A point cloud data set may be pre-processed for fast and efficient rendering. The pre-processing may comprise creating an octree hierarchy from the data, generating a level of detail (LOD) representation for each octree node, simplifying the points in each node according to a simplification tolerance, and storing the data structure in a file. Textures associated with each node may be stored in a separate, compressed file, such as a texture atlas. At render time, the octree and LOD hierarchy may be traversed until a suitable LOD node is found. The associated texture data may be accessed, and the node may be rendered as a textured quadrilateral and/or a splat point primitive. In an alternative approach, multiple point cloud datasets may be merged using a global transform function. The merged dataset may be simplified using a hierarchical LOD tree. Textures may be ascribed to each LOD node. The resulting structure may be rendered using splat and billboard point primitives. The model may be streamed over a network to a client where the rendering may take place.

Claims

1. A method for pre-processing a point cloud data set, comprising:

constructing an octree hierarchy from the point cloud data set;

within the octree hierarchy, generating a level of detail (LOD) representative for each octree node;

replacing a set of points within the point cloud data set with a quadrilateral;

generating a texture for the quadrilateral;

defining a grid within each octree leaf node;

quantizing a position of a point remaining in the octree hierarchy to correspond to a grid cell within an octree leaf node; and

storing the octree hierarchy in a computer readable storage location for access by a renderer to thereby provide for fast and efficient rendering of the stored octree hierarchy on a display for viewing by a user.

2. The method of claim 1, further comprising rejecting the quadrilateral if the quadrilateral does not satisfy a compactness threshold.

3. The method of claim 1, wherein generating a level of detail representative for an octree node comprises determining a simplification tolerance of the octree node.

4. The method of claim 3, wherein replacing a set of points within the point cloud data set with a quadrilateral comprises detecting a plane within an octree node using a RANSAC method.

5. The method of claim 1, further comprising generating a splat data structure comprising a splat texture for the point.

6. The method of claim 1, wherein generating a texture for the quadrilateral comprises projecting a portion of the point cloud data set onto the quadrilateral, the portion comprising points within the octree node of the quadrilateral within the simplification tolerance, such that the distance between any two points is substantially the simplification threshold of the LOD representative of the octree node.

7. The method of claim 1, further comprising, within each octree node, generating a textured quadrilateral in an octree node if a texel-to-shot ratio within the octree node is greater than four to one, and generating a composite splat within the octree node otherwise.

8. The method of claim 7, further comprising:

for a first set of octree nodes, packing all of the textures into a texture atlas, and

storing the texture atlas in a storage location.

9. The method of claim 8, wherein the first set of octree nodes comprises octree nodes having a common grandparent octree node.

10. A method for rendering a point cloud data set, comprising:

obtaining an octree hierarchy comprising an octree node;

selecting a rendering point of view;

determining whether the octree node is visible and culling the node from the rendering process if the determining is negative;

determining a projection error of the octree node;

rendering the node if the projection error is below a projection error tolerance; and

displaying the rendered node on a display for viewing by a user.

11. The method of claim 10, further comprising, traversing the octree hierarchy to a second higher level of detail octree node if the projection error of the octree node is above the projection error tolerance.

12. The method of claim 11, wherein rendering an octree node comprises accessing a texture associated with the octree node from a texture atlas.

13. The method of claim 10, wherein determining whether an octree node is visible comprises checking the visibility of the octree node from the selected rendering point of view using frustrum culling.

14. A method for pre-processing a plurality of point cloud data sets, the method comprising:

calculating a point normal for each point in each of the plurality of point cloud data sets;

computing a global transformation using the point normals;

merging the plurality of point cloud data sets into a merged point cloud data set using the global transformation;

simplifying the merged point cloud data set using a hierarchical level of detail (LOD) tree structure;

for each tree node in the hierarchical LOD tree structure, generating a textured splat for the tree node if a point within the tree node comprises a point normal or generating a billboard point primitive for the tree node if a point therein does not comprise a point normal; and

serializing the hierarchical LOD tree structure to a storage location for access by a rendering engine to provide fast and efficient rendering of the merged point cloud data set on a display for viewing by a user.

15. The method of claim 14, further comprising determining a surface normal for each point in the merged point cloud data set before simplifying the merged point cloud data set.

16. The method of claim 14, wherein computing a global transformation comprises using an iterative closest point, point-to-point or point-to-plane variant.

17. The method of claim 14, wherein determining a normal for each point in a point cloud data set comprises:

selecting a point neighborhood associated with a point,

calculating a covariance matrix of the point neighborhood, and

selecting the normal as a unit eigenvector of the covariance matrix corresponding to a minimum eigenvalue of the covariance matrix.

18. The method of claim 14, wherein simplifying the merged point cloud data set using a hierarchical LOD tree structure comprises splitting the merged point cloud data set using a maximum eigenvector splitting approach, and wherein the maximum eigenvector splitting approach comprises:

calculating a covariance matrix of a point set, and

selecting a splitting plane based on an eigenvector of the covariance matrix corresponding to a maximum eigenvalue of the covariance matrix.

19. The method of claim 14, further comprising storing the serialized hierarchical LOD tree structure in a network accessible storage location.

20. The method of claim 14, wherein a node of the hierarchical LOD tree structure comprises a plurality of sixteen-bit color values.

21. A method for rendering a hierarchical LOD tree generated from a plurality of point cloud data sets, the method comprising:

obtaining a reference to a hierarchical LOD tree;

selecting a rendering point of view;

determining whether a node in the hierarchical LOD tree structure is visible and requesting the node from a network accessible storage location if the determining is positive;

determining a projected error ε

of the node and traversing the hierarchical LOD tree structure to a higher level of detail node if the error ε

is above an error threshold;

rendering the node as a textured splat if the node has a normal; and

rendering the node as a billboard if the node does not have a normal,

wherein the rendering is to a computer display for viewing by a user.

22. The method of claim 21, wherein obtaining a reference to a hierarchical LOD tree comprises requesting a portion of the hierarchical LOD tree from a network accessible storage location.

23. The method of claim 21, further comprising, determining whether a node is available and, if the node is not available, requesting the node from the network accessible storage location.

24. The method of claim 21, wherein rendering the node as a textured splat comprises limiting a radius or elliptical radii of the splat to a LIDAR footprint.

25. The method of claim 21, wherein rendering the node as a billboard comprises limiting a footprint of the billboard to a LIDAR footprint.