US 12,462,191 B2
Image processing methods and systems for training a machine learning model to predict illumination conditions for different positions relative to a scene
Matis Hudon, Charenton le Pont (FR); Guillaume Cortes, Villeurbanne (FR); and Toon Van Den Zegel, Goetsenhoven (BE)
Assigned to 7 SENSING SOFTWARE, Paris (FR)
Filed by 7 SENSING SOFTWARE, Paris (FR)
Filed on Jan. 4, 2022, as Appl. No. 17/568,386.
Prior Publication US 2023/0214708 A1, Jul. 6, 2023
Int. Cl. G06N 20/00 (2019.01); B60Q 1/24 (2006.01); G06T 15/50 (2011.01); G06T 19/00 (2011.01); G06V 10/60 (2022.01); G06V 10/75 (2022.01); G06V 10/77 (2022.01); H04N 23/74 (2023.01)
CPC G06N 20/00 (2019.01) [G06V 10/60 (2022.01); G06V 10/751 (2022.01); G06V 10/7715 (2022.01); B60Q 1/249 (2022.05); G06T 15/506 (2013.01); G06T 19/006 (2013.01); H04N 23/74 (2023.01); Y10S 209/938 (2013.01)] 22 Claims
OG exemplary drawing
 
1. An image processing method for generating a training dataset to train a machine learning model to predict different illumination conditions for different positions relative to a scene, said training dataset including training images and reference data, said image processing method comprising:
obtaining a training image of a training scene, to be included in the training dataset, acquired by a first camera having a first coordinate system associated thereto;
determining a plurality of local illumination maps, each of the local illumination maps being associated to a respective position relative to the training scene in a respective second coordinate system and representing in said respective second coordinate system the illumination received from different directions of arrival around said respective position;
transforming the position of each of the local illumination maps from the respective second coordinate system to the first coordinate system; and
responsive to determining that the transformed position of the respective local illumination map is visible in the training image: transforming said local illumination map from the respective second coordinate system to the first coordinate system and including the transformed local illumination map and the respective transformed position in the reference data associated to the training image to enable training the machine learning model to predict the different illumination conditions for the different positions relative to the scene.