Segmentation and representation network for pose discrimination
US10186038B1 · kind B1 · utility
Assignee
Inventors
Key dates
| Filing date | Jul 18, 2017 |
| Grant date | Jan 22, 2019 |
| Priority date | — |
| Expiry date | Jul 18, 2037 |
Classification
- Technology area (CPC G)Physics
- CPC primaryG06V30/422
- WIPO fieldComputer technology
- WIPO sectorElectrical engineering
Abstract
A system and method includes creation of a combined network comprising an image segmentation network and an image representation network, the combined network to generate an image descriptor based on an input query image, training of the combined network based on a plurality of first images and a segmentation mask associated with each of the plurality of first images, reception of a first input query image, use of the combined network to generate an image descriptor based on the first input query image, determination of a matching image descriptor from a plurality of stored image descriptors, determination of a camera pose associated with the matching image descriptor, registration of the first input query image with image data based on the determined camera pose, generation of a composite image based on the registered first input query image and image data, and presentation of the composite image.
Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.