US 11,670,038 B2
Processing point clouds using dynamic voxelization
Yin Zhou, San Jose, CA (US); Pei Sun, Palo Alto, CA (US); Yu Zhang, Santa Clara, CA (US); Dragomir Anguelov, San Francisco, CA (US); Jiyang Gao, San Jose, CA (US); Yu Ouyang, San Jose, CA (US); Zijian Guo, Sunnyvale, CA (US); Jiquan Ngiam, Mountain View, CA (US); and Vijay Vasudevan, Los Altos Hills, CA (US)
Assigned to Waymo LLC, Mountain View, CA (US)
Filed by Waymo LLC, Mountain View, CA (US)
Filed on Nov. 1, 2021, as Appl. No. 17/516,073.
Application 17/516,073 is a continuation of application No. 16/924,080, filed on Jul. 8, 2020, granted, now 11,164,363.
Claims priority of provisional application 62/871,676, filed on Jul. 8, 2019.
Prior Publication US 2022/0058858 A1, Feb. 24, 2022
This patent is subject to a terminal disclaimer.
Int. Cl. G06T 15/20 (2011.01); G06V 20/56 (2022.01); G06V 10/82 (2022.01); G06V 20/64 (2022.01); G06V 20/58 (2022.01)
CPC G06T 15/20 (2013.01) [G06V 10/82 (2022.01); G06V 20/56 (2022.01); G06V 20/64 (2022.01); G06V 20/58 (2022.01)] 20 Claims
OG exemplary drawing
 
1. A method comprising:
obtaining point cloud data representing a sensor measurement of a scene captured by a sensor, the point cloud data comprising a respective feature representation for each of a plurality of three-dimensional points in the scene;
generating, for each of one or more views of the scene, a corresponding dynamic voxel representation that assigns, to each voxel of a set of voxels for the view, a variable number of three-dimensional points, wherein each three-dimensional point in the point cloud data is assigned to a respective one of the voxels of the set of voxels in the corresponding dynamic voxel representation, and wherein the generating comprises:
assigning, based on positions of the three-dimensional points in the point cloud data according to the view, each of the three-dimensional points to a respective one of the voxels of the set of voxels; and
processing the dynamic voxel representations corresponding to each of the one or more views to generate an output that characterizes the scene.