US 11,861,480 B2
Orientation detection in overhead line insulators
Arun Innanje, Dayton, OH (US); Kuan-Chuan Peng, Plainsboro, NJ (US); Ziyan Wu, Lexington, MA (US); and Jan Ernst, Princeton, NJ (US)
Assigned to Siemens Mobility GmbH, Munich (DE)
Appl. No. 17/267,163
Filed by SIEMENS AKTIENGESELLSCHAFT, Munich (DE)
PCT Filed Aug. 21, 2018, PCT No. PCT/US2018/047138
§ 371(c)(1), (2) Date Feb. 9, 2021,
PCT Pub. No. WO2020/040734, PCT Pub. Date Feb. 27, 2020.
Prior Publication US 2021/0304437 A1, Sep. 30, 2021
Int. Cl. G06T 7/73 (2017.01); G06T 7/11 (2017.01); G06V 10/24 (2022.01); G06V 10/764 (2022.01); G06V 10/82 (2022.01); G06V 20/13 (2022.01); G06V 20/17 (2022.01); G06N 3/04 (2023.01)
CPC G06N 3/04 (2013.01) [G06T 7/11 (2017.01); G06T 7/74 (2017.01); G06V 10/243 (2022.01); G06V 10/764 (2022.01); G06V 10/82 (2022.01); G06V 20/13 (2022.01); G06V 20/17 (2022.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01)] 12 Claims
OG exemplary drawing
 
1. A computer-implemented method for detecting and correcting an orientation of a target object in an image, the method comprising:
training a deep neural network using a set of training images;
providing the image as input to the deep neural network, wherein the image is a segmented image;
utilizing the deep neural network to determine an initial orientation prediction for the target object; and
utilizing the deep neural network to obtain a rectified image from the segmented image based at least in part on the initial orientation prediction, wherein an orientation of the target object in the rectified image is within a threshold value of a target orientation,
wherein utilizing the deep neural network to obtain the rectified image comprises:
determining that the initial orientation prediction deviates from the target orientation by more than the threshold value; and
generating an aligned image from the segmented image.