US 12,488,580 B2
Method and system for identifying objects
Fan Bo Meng, Beijing (CN); Xiang Li, Beijing (CN); and Xiao Feng Wang, Beijing (CN)
Assigned to SIEMENS AKTIENGESELLSCHAFT, Munich (DE)
Appl. No. 18/044,443
Filed by Siemens Aktiengesellschaft, Munich (DE)
PCT Filed Sep. 11, 2020, PCT No. PCT/CN2020/114844
§ 371(c)(1), (2) Date Mar. 8, 2023,
PCT Pub. No. WO2022/052052, PCT Pub. Date Mar. 17, 2022.
Prior Publication US 2023/0360380 A1, Nov. 9, 2023
Int. Cl. G06V 10/82 (2022.01); G06V 10/75 (2022.01); G06V 10/774 (2022.01)
CPC G06V 10/82 (2022.01) [G06V 10/76 (2022.01); G06V 10/774 (2022.01); G06V 2201/07 (2022.01); G06V 2201/12 (2022.01)] 14 Claims
OG exemplary drawing
 
1. A method for identifying an object, the method comprising:
generating a plurality of synthesized images according to a three-dimensional digital model, each synthesized image of the plurality of synthesized images having a respective view angles;
extracting a respective eigenvector of each of the plurality of synthesized images using a convolutional neural network (CNN);
generating a first fused vector by fusing the eigenvectors of the plurality of synthesized images using an AutoML technology-based fusion scheme or a neural architecture search technology;
transmitting the first fused vector into a deep learning classifier to train the classifier;
acquiring a plurality of pictures of the object, each picture of the plurality of pictures having a respective view angle matching at least a portion of the plurality of synthesized images;
extracting respective eigenvectors for each of the plurality of pictures using the CNN;
generating a second fused vector by fusing the eigenvectors of the plurality of pictures; and
transmitting the second fused vector into the trained classifier to obtain a classification result of the object.