Apparatus and method for identifying an articulatable part of a physical object using multiple 3D point clouds
An apparatus comprises an input interface configured to receive a first 3D point cloud associated with a physical object prior to articulation of an articulatable part, and a second 3D point cloud after articulation of the articulatable part. A processor is operably coupled to the input interface, an output interface, and memory. Program code, when executed by the processor, causes the processor to align the first and second point clouds, find nearest neighbors of points in the first point cloud to points in the second point cloud, eliminate the nearest neighbors of points in the second point cloud such that remaining points in the second point cloud comprise points associated with the articulatable part and points associated with noise, generate an output comprising at least the remaining points of the second point cloud associated with the articulatable part without the noise points, and communicate the output to the output interface.
1. A computer-implemented method, comprising:
obtaining a first three-dimensional point cloud associated with a physical object having at least one articulatable part, the first point cloud associated with the physical object prior to articulation of the articulatable part;
obtaining a second three-dimensional point cloud associated with the physical object after articulation of the articulatable part;
coarsely aligning the first and second point clouds;
finely aligning the first and second point clouds after coarsely aligning the first and second point clouds;
operating on the finely aligned first and second point clouds by:
finding nearest neighbors of points in the first point cloud to points in the second point cloud; and
eliminating the nearest neighbors of points in the second point cloud such that remaining points in the second point cloud comprise points associated with the articulatable part and points associated with noise; and
generating an output comprising at least the remaining points of the second point cloud associated with the articulatable part without the noise points.
2. The method of claim 1 , wherein coarsely aligning the first and second point clouds comprises:
computing Fast Point Feature Histogram (FPFH) descriptors from the first and second point clouds; and
performing random sample consensus (RANSAC) registration between the first and second point clouds using the FPFH descriptors of the first and second point clouds.
3. The method of claim 2 , comprising downsampling the first and second point clouds prior to computing the FPFH descriptors.
4. The method of claim 1 , wherein finely aligning the first and second point clouds comprises performing Iterative Closest Point (ICP) registration on the coarsely aligned the first and second point clouds.
5. The method of claim 1 , further comprising:
applying clustering to the remaining points in the second point cloud to produce noise point clusters and a cluster of points associated with the articulatable part; and
distinguishing the cluster of points associated with the articulatable part from the noise point clusters;
wherein the generated output comprises at least the cluster of points associated with the articulatable part without the noise point clusters.
6. The method of claim 5 , wherein distinguishing the cluster of points associated with the articulatable part from the noise point clusters comprises:
receiving a user input identifying a location of the articulatable part in the first point cloud; and
selecting a cluster of points at the user identified location as the cluster of points associated with the articulatable part.
7. The method of claim 5 , wherein distinguishing the cluster of points associated with the articulatable part from the noise point clusters comprises:
algorithmically determining a size of each of the clusters of points; and
selecting a largest cluster of points as the cluster of points associated with the articulatable part.
8. The method of claim 5 , wherein applying clustering comprises applying density-based spatial clustering of applications with noise (DBSCAN) to the remaining points in the second point cloud.
9. The method of claim 1 , wherein finding the nearest neighbors of points comprises using a k-dimensional tree (k-d tree) and a predetermined distance threshold to find the nearest neighbors of points in the first point cloud to points in the second point cloud.
10. The method of claim 1 , wherein the generated output comprises points of the second point cloud associated with the articulatable part superimposed on the first point cloud.
11. The method of claim 1 , wherein the output comprises one or both of a graphical output and a data structure output.
12. An apparatus, comprising:
an input interface and an output interface, the input interface configured to receive:
a first three-dimensional point cloud associated with a physical object having at least one articulatable part, the first point cloud associated with the physical object prior to articulation of the articulatable part; and
a second three-dimensional point cloud associated with the physical object after articulation of the articulatable part;
memory operably coupled to the input interface and configured to store the first and second point clouds and program code; and
a processor operably coupled to the input interface, the output interface, and the memory, the program code when executed by the processor causing the processor to:
coarsely align the first and second point clouds;
finely align the first and second point clouds after coarsely aligning the first and second point clouds;
operate on the finely aligned first and second point clouds to:
find nearest neighbors of points in the first point cloud to points in the second point cloud; and
eliminate the nearest neighbors of points in the second point cloud such that remaining points in the second point cloud comprise points associated with the articulatable part and points associated with noise;
generate an output comprising at least the remaining points of the second point cloud associated with the articulatable part without the noise points; and
communicate the output to the output interface.
13. The apparatus of claim 12 , wherein the processor is configured to coarsely align the first and second point clouds by:
computing Fast Point Feature Histogram (FPFH) descriptors from the first and second point clouds; and
performing random sample consensus (RANSAC) registration between the first and second point clouds using the FPFH descriptors of the first and second point clouds.
14. The apparatus of claim 13 , wherein the processor is configured to downsample the first and second point clouds prior to computing the FPFH descriptors.
15. The apparatus of claim 12 , wherein the processor is configured to finely align the first and second point clouds by performing Iterative Closest Point (ICP) registration on the coarsely aligned the first and second point clouds.
16. The apparatus of claim 12 , wherein the processor is configured to:
apply clustering to the remaining points in the second point cloud to produce noise point clusters and a cluster of points associated with the articulatable part; and
distinguish the cluster of points associated with the articulatable part from the noise point clusters;
wherein the output generated by the processor comprises at least the cluster of points associated with the articulatable part without the noise point clusters.
17. The apparatus of claim 16 , wherein the processor is configured to distinguish the cluster of points associated with the articulatable part from the noise point clusters by:
receiving a user input identifying a location of the articulatable part in the first point cloud; and
selecting a cluster of points at the user identified location as the cluster of points associated with the articulatable part.
18. The apparatus of claim 16 , wherein the processor is configured to distinguish the cluster of points associated with the articulatable part from the noise point clusters by:
algorithmically determining a size of each of the clusters of points; and
selecting a largest cluster of points as the cluster of points associated with the articulatable part.
19. The apparatus of claim 16 , wherein the processor is configured to apply density-based spatial clustering of applications with noise (DBSCAN) to the remaining points in the second point cloud.
20. The apparatus of claim 12 , wherein the processor is configured to find the nearest neighbors of points using a k-dimensional tree (k-d tree) and a predetermined distance threshold to find the nearest neighbors of points in the first point cloud to points in the second point cloud.
21. The apparatus of claim 12 , wherein the output generated by the processor comprises points of the second point cloud associated with the articulatable part superimposed on the first point cloud.
22. The apparatus of claim 12 , wherein the output generated by the processor comprises one or both of a graphical output and a data structure output.