Apparatus and method for efficient point cloud feature extraction and segmentation framework

1. A non-transitory machine-readable storage media having instructions stored thereon, that when executed, cause one or more processors to perform an operation for a LIDAR system where point cloud feature extraction and segmentation is performed, the operation comprising:

receiving scan pattern information and point cloud data from the LIDAR system, wherein the scan pattern information and point cloud data are associated with a physical object;

generating a scan pattern grid from the scan pattern information and point cloud data;

detecting one or more silhouette edges using the scan pattern grid and the point cloud data;

detecting one or more intersection edges using the scan pattern grid and the point cloud data; and

growing a region based on the detected one or more silhouette edges and the one or more intersection edge,

wherein detecting the one or more intersection edges comprises:

generating a mesh around a point, which is not a silhouette edge point, of the scan pattern grid with neighbor points forming shared edges between regions of the mesh; and

computing normal gradient across an individual shared edge to determine whether the point is lying on an intersection edge of a smooth surface.

2. The non-transitory machine-readable storage media of claim 1 , wherein the scan pattern grid is a 2-D grid structure having cells organized in rows and columns, wherein an individual row corresponds to a vertical angle of a scan from a scanner of the LIDAR system, and wherein an individual column corresponds to a horizontal angle from the scanner of the LIDAR system.

3. The non-transitory machine-readable storage media of claim 1 , wherein the LIDAR system is a mobile LIDAR system, wherein the scan pattern grid is a 2-D grid structure having cells organized in rows and columns, wherein an individual row corresponds to a time of a scan from a scanner of the LIDAR system, and wherein an individual column corresponds to a scan angle from the scanner of the LIDAR system.

4. The non-transitory machine-readable storage media of claim 2 , wherein an individual cell of the scan pattern grid represents a laser pulse from the scanner, wherein the individual cell includes information of a return point, and wherein the information of the return point includes scan coordinates, color, and intensity.

5. The non-transitory machine-readable storage media of claim 1 , wherein the one or more silhouette edges comprises points lying on an edge between an object scanned by the LIDAR system and a shadow of the object.

6. The non-transitory machine-readable storage media of claim 1 , wherein detecting the one or more silhouette edges comprises:

determining, from the scan pattern grid, points with no return; and

identifying points in the scan pattern grid that neighbor the points with no return as silhouette edge points, wherein the silhouette edge points form the one or more silhouette edges.

7. The non-transitory machine-readable storage media of claim 1 , wherein detecting the one or more silhouette edges comprises:

determining mixed points, between two objects, from the scan pattern grid;

distinguishing silhouette edge points of the two objects from the mixed points by determining whether an incidence angle associated with a point in the scan pattern grid is on one of the two objects;

comparing the incidence angle, associated with a point, against a threshold;

identifying the point as a silhouette edge point if the incidence angle exceeds the threshold, wherein the silhouette edge point forms the one or more silhouette edges; and

identifying the point as a mixed point if the incidence angle is below the threshold.

8. The non-transitory machine-readable storage media of claim 1 , wherein the one or more intersection edges intersect two or more smooth surfaces.

9. The non-transitory machine-readable storage media of claim 1 , wherein detecting the one or more intersection edges is independent of a normal vector at each point of the scan pattern grid.

10. The non-transitory machine-readable storage media of claim 1 , wherein the mesh is a triangular mesh around the point, and wherein the regions of the mesh are triangles of the triangular mesh.

11. The non-transitory machine-readable storage media of claim 10 , wherein detecting one or more intersection edges comprises comparing a maximum normal gradient, from among the computed normal gradient across each shared edge, with a threshold of maximum normal gradient.

12. The non-transitory machine-readable storage media of claim 1 , wherein the scan pattern grid is a first scan pattern grid, wherein growing the region comprises:

analyzing points in the first scan pattern grid and a second scan pattern grid; and

generating clusters of homogeneous points continuous in space.

13. A system comprising:

a memory;

a processor coupled to the memory; and

a communication interface coupled to the processor, wherein the processor is to:

receive scan pattern information and point cloud data from a LIDAR system, wherein the scan pattern information and point cloud data are associated with a physical object;

generate a scan pattern grid from the scan pattern information and point cloud data;

detect one or more silhouette edges using the scan pattern grid and the point cloud data;

detect one or more intersection edges using the scan pattern grid and the point cloud data; and

grow a region based on the detected one or more silhouette edges and the one or more intersection edges,

wherein to detect the one or more intersection edges, the processor is to:

generate a mesh around a point, which is not a silhouette edge point, of the scan pattern grid with neighbor points forming shared edges between regions of the mesh; and

compute normal gradient across an individual shared edge to determine whether the point is lying on an intersection edge of a smooth surface.

14. The system of claim 13 , wherein the scan pattern grid is a 2-D grid structure having cells organized in rows and columns, wherein an individual row corresponds to a vertical angle of a scan from a scanner of the LIDAR system, and wherein an individual column corresponds to a horizontal angle from of the scan from the scanner of the LIDAR system.

15. The system of claim 13 , wherein the LIDAR system is a mobile LIDAR system, wherein the scan pattern grid is a 2-D grid structure having cells organized in rows and columns, wherein an individual row corresponds to a time of a scan from a scanner of the LIDAR system, and wherein an individual column corresponds to a scan angle from the scanner of the LIDAR system.

16. The system of claim 15 , wherein an individual cell of the scan pattern grid represents a laser pulse from the scanner, wherein the individual cell includes information of a return point, and wherein the information of the return point includes scan coordinates, color, and intensity.

17. The system of claim 13 , wherein the one or more silhouette edges comprises points lying on an edge between an object scanned by the LIDAR system and a shadow of the object, wherein to detect the one or more silhouette edges the processor is to:

determine, from the scan pattern grid, points with no return; and

identify points in the scan pattern grid points that neighbor the points with no return as a silhouette edge points, wherein the silhouette edge points form the one or more silhouette edges.

18. The system of claim 13 , wherein to detect the one or more silhouette edges the processor is to:

determine mixed points, between two objects, from the scan pattern grid;

distinguish silhouette edge points of the two objects from the mixed points by determining whether an incidence angle associated with a point in the scan pattern grid is on one of the two objects;

compare the incidence angle, associated with a point, against a threshold;

identify the point as a silhouette edge point if the incidence angle exceeds the threshold, wherein the silhouette edge point forms the one or more silhouette edges; and

identify the point as a mixed point if the incidence angle is below the threshold.

19. A method for point cloud feature extraction and segmentation from a LIDAR system, the method comprising:

receiving scan pattern information and point cloud data from the LIDAR system, wherein the scan pattern information and point cloud data are associated with a physical object;

generating a scan pattern grid from the scan pattern information and point cloud data;

detecting one or more silhouette edges using the scan pattern grid and the point cloud data;

detecting one or more intersection edges using the scan pattern grid and the point cloud data; and

growing a region based on the detected one or more silhouette edges and the one or more intersection edge,

wherein detecting the one or more intersection edges comprises:

generating a mesh around a point, which is not a silhouette edge point, of the scan pattern grid with neighbor points forming shared edges between regions of the mesh; and

computing normal gradient across an individual shared edge to determine whether the point is lying on an intersection edge of a smooth surface.

20. The method of claim 19 , wherein the LIDAR system is a mobile LIDAR system, wherein the scan pattern grid is a 2-D grid structure having cells organized in rows and columns, wherein an individual row corresponds to a time of a scan from a scanner of the LIDAR system, wherein an individual column corresponds to a scan angle from of the scan from the scanner of the LIDAR system, wherein an individual cell of the scan pattern grid represents a laser pulse from the scanner, wherein the individual cell includes information of a return point, and wherein the information of the return point includes scan coordinates, color, and intensity.