US 12,437,551 B2
Instance segmentation in a pseudo-image
Sourabh Vora, Marina Del Rey, CA (US); and Qi Chen, Baltimore, MD (US)
Assigned to Motional AD LLC, Boston, MA (US)
Filed by Motional AD LLC, Boston, MA (US)
Filed on Mar. 31, 2022, as Appl. No. 17/657,590.
Claims priority of provisional application 63/269,601, filed on Mar. 18, 2022.
Prior Publication US 2023/0298356 A1, Sep. 21, 2023
Int. Cl. G06V 20/58 (2022.01); G06V 10/26 (2022.01); G06V 10/764 (2022.01); G06V 10/82 (2022.01)
CPC G06V 20/58 (2022.01) [G06V 10/267 (2022.01); G06V 10/764 (2022.01); G06V 10/82 (2022.01); G06V 2201/08 (2022.01)] 18 Claims
OG exemplary drawing
 
1. A method, comprising:
receiving, with at least one processor, a pseudo-image of a 3D image of an environment, wherein the 3D image is received from a 3D image sensor coupled to a vehicle, and wherein the pseudo-image comprises a grid of sections;
concurrently determining, with the at least one processor;
an object classification for a a plurality of sections of the pseudo-image using a first neural network, and
an instance portion classification for the plurality of sections of the pseudo-image using a second neural network;
combining, for each section of the plurality of sections, the object classification for the section and the instance portion classification for the section to form combined features for the section;
grouping a first section of the plurality of sections with a second section of the plurality of sections using the combined features of the first section and the combined features of the second section;
identifying a particular instance of an object in the environment using the grouped first section and second section; and
causing the vehicle to navigate based at least in part on the identified instances.