US 12,277,781 B2
Device and method for training a machine learning model for recognizing an object topology of an object from an image of the object
Andras Gabor Kupcsik, Boeblingen (DE); and Markus Spies, Karlsruhe (DE)
Assigned to ROBERT BOSCH GMBH, Stuttgart (DE)
Appl. No. 17/911,318
Filed by Robert Bosch GmbH, Stuttgart (DE)
PCT Filed Jun. 22, 2021, PCT No. PCT/EP2021/067011
§ 371(c)(1), (2) Date Sep. 13, 2022,
PCT Pub. No. WO2022/002698, PCT Pub. Date Jan. 6, 2022.
Claims priority of application No. 10 2020 208 211.1 (DE), filed on Jul. 1, 2020.
Prior Publication US 2023/0115521 A1, Apr. 13, 2023
Int. Cl. G06V 20/64 (2022.01); B25J 9/16 (2006.01); B25J 13/08 (2006.01); G05B 19/4155 (2006.01); G06N 3/09 (2023.01); G06T 7/73 (2017.01); G06V 10/46 (2022.01); G06V 10/774 (2022.01)
CPC G06V 20/653 (2022.01) [B25J 9/1697 (2013.01); B25J 13/08 (2013.01); G05B 19/4155 (2013.01); G06N 3/09 (2023.01); G06T 7/75 (2017.01); G06V 10/46 (2022.01); G06V 10/774 (2022.01); G05B 2219/40269 (2013.01); G06T 2207/20081 (2013.01)] 11 Claims
OG exemplary drawing
 
1. A method for training a machine learning model for recognizing an object topology of an object from an image of the object, the method comprising the following steps:
obtaining a 3D model of the object, the 3D model including a mesh of vertices connected by edges, wherein each edge of the edges has a weight which specifies a proximity of two vertices connected by the edge in the object;
determining a descriptor for each vertex of the mesh by searching descriptors for the vertices which minimize a sum, over pairs of connected vertices, of distances between the descriptors of the pair of vertices weighted by the weight of the edge between the pair of vertices, wherein the searching of the descriptors includes determining eigenvectors of a Laplacian matrix of a graph formed by the vertices and edges of the 3D model and taking components of the eigenvectors as components of the descriptors;
generating training data image pairs, wherein each training data image pair includes a training input image showing the object and a target image and wherein generating the target image includes:
determining vertex positions of the vertices of the object's object model that the vertices have in the training input image, and
assigning, for each determined vertex position in the training input image, the descriptor determined for the vertex at the vertex position to the position in the target image; and
training the machine learning model by supervised learning using the training data image pairs as training data.