US 11,886,995 B2
Recognition of objects in images with equivariance or invariance in relation to the object size
Artem Moskalev, Amsterdam (NL); Ivan Sosnovik, Amsterdam (NL); Arnold Smeulders, Amsterdam (NL); and Konrad Groh, Stuttgart (DE)
Assigned to ROBERT BOSCH GMBH, Stuttgart (DE)
Filed by Robert Bosch GmbH, Stuttgart (DE)
Filed on Jun. 28, 2021, as Appl. No. 17/360,709.
Claims priority of application No. 102020208080.1 (DE), filed on Jun. 30, 2020.
Prior Publication US 2021/0406610 A1, Dec. 30, 2021
Int. Cl. G06N 3/08 (2023.01); G06T 7/246 (2017.01); G05D 1/02 (2020.01); G06V 20/56 (2022.01); G06F 18/21 (2023.01); G06F 18/213 (2023.01); G06V 10/75 (2022.01); G06V 10/80 (2022.01); G06V 10/44 (2022.01)
CPC G06N 3/08 (2013.01) [G05D 1/0214 (2013.01); G06F 18/213 (2023.01); G06F 18/217 (2023.01); G06T 7/248 (2017.01); G06V 10/454 (2022.01); G06V 10/757 (2022.01); G06V 10/806 (2022.01); G06V 20/56 (2022.01); G05D 2201/0213 (2013.01); G06T 2207/10016 (2013.01); G06T 2207/20084 (2013.01); G06T 2207/30252 (2013.01)] 17 Claims
OG exemplary drawing
 
1. A method for recognizing at least one object in at least one input image, comprising the following steps:
processing a template image of the object by a first convolutional neural network to form at least one template feature map;
processing the input image by a second convolutional neural network to form at least one input feature map;
comparing the at least one template feature map to the at least one input feature map; and
evaluating, from a result of the comparison whether the object is contained in the input image;
wherein the first and second convolutional neural networks each contain multiple convolutional layers, and at least one of the convolutional layers is at least partially formed from at least two filters, which are convertible into one another by a scaling operation, and
the first and second convolutional neural networks each output multiple feature maps for the template image and the input image, respectively, and in the creation of each of the multiple feature maps a different one of the filters convertible into one another predominantly participates.