US 11,818,328 B2
Systems and methods for automatically calibrating multiscopic image capture systems
Guy Satat, Sunnyvale, CA (US)
Assigned to Google LLC, Mountain View, CA (US)
Filed by Google LLC, Mountain View, CA (US)
Filed on Sep. 23, 2022, as Appl. No. 17/934,620.
Application 17/934,620 is a continuation of application No. 17/132,097, filed on Dec. 23, 2020, granted, now 11,496,722.
Prior Publication US 2023/0015589 A1, Jan. 19, 2023
This patent is subject to a terminal disclaimer.
Int. Cl. H04N 13/246 (2018.01); G06T 7/70 (2017.01); G06V 20/10 (2022.01); H04N 13/271 (2018.01); G06T 7/80 (2017.01); G06N 3/084 (2023.01); G06F 18/22 (2023.01)
CPC H04N 13/246 (2018.05) [G06F 18/22 (2023.01); G06N 3/084 (2013.01); G06T 7/70 (2017.01); G06T 7/85 (2017.01); G06V 20/10 (2022.01); H04N 13/271 (2018.05); G06T 2207/10012 (2013.01); G06T 2207/20224 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method comprising:
receiving, from a multiscopic image capture system, two or more images depicting a scene;
determining, based on the two or more images and using a neural network comprising a plurality of layers, a disparity map of the scene;
identifying a recalibration trigger, wherein identifying the recalibration trigger comprises determining that the multiscopic image capture system is facing a recalibration target; and
based on identifying the recalibration trigger, (i) determining an error of the disparity map based on values of pixels of the disparity map and (ii) back-propagating the error to one or more layers of the plurality of layers of the neural network, wherein back-propagating the error comprises updating one or more weights applied to the one or more layers.