US 12,449,766 B2
Holographic display calibration
Manoj Sharma, Troy, MI (US)
Assigned to GM GLOBAL TECHNOLOGY OPERATIONS LLC, Detroit, MI (US)
Filed by GM Global Technology Operations LLC, Detroit, MI (US)
Filed on Oct. 24, 2022, as Appl. No. 18/048,968.
Prior Publication US 2024/0134316 A1, Apr. 25, 2024
Prior Publication US 2024/0231277 A9, Jul. 11, 2024
Int. Cl. G03H 1/22 (2006.01); B60K 35/00 (2006.01); G02B 27/01 (2006.01)
CPC G03H 1/2294 (2013.01) [B60K 35/00 (2013.01); G02B 27/0103 (2013.01)] 16 Claims
OG exemplary drawing
 
1. A system for training a machine learning algorithm to generate a plurality of ideal hologram phase correction maps, the system comprising:
a holographic head-up display (HUD) configured to display a plurality of duplicates of a graphic based on a hologram phase map, wherein each of the plurality of duplicates of the graphic is displayed at one of a plurality of eyebox locations;
a camera system configured to view each of the plurality of duplicates of the graphic displayed by the holographic HUD; and
a controller in electrical communication with the holographic HUD and the camera system, wherein the controller is programmed to:
determine a plurality of ground-truth hologram phase correction maps using a genetic algorithm, the holographic HUD, and the camera system, wherein each of the plurality of ground-truth hologram phase correction maps corresponds to one of the plurality of eyebox locations;
generate a training dataset including a plurality of images of the graphic displayed at each of the plurality of eyebox locations, wherein to generate the training dataset, the controller is further programmed to:
display the graphic using the holographic HUD;
capture a plurality of images using the camera system, wherein each of the plurality of images includes one of the plurality of duplicates of the graphic displayed at one of the plurality of eyebox locations;
save the training dataset to a non-transitory memory of the controller, wherein the training dataset includes the plurality of images;
repeat the display, capture, and save steps using a plurality of different holographic HUD units such that influences due to manufacturing differences between individual holographic HUD units are included in the training dataset; and
repeat the display, capture, and save steps using a plurality of different graphics; and
train the machine learning algorithm to generate the plurality of ideal hologram phase correction maps, wherein the machine learning algorithm is trained using at least the plurality of ground-truth hologram phase correction maps and the training dataset.