Patent · US Active

Machine learning-based method for calibrating a camera with respect to a scene

US12387377B1 · kind B1 · utility

0Cited by
8References
20Claims
0Family size

Assignee

Inventors

Key dates

Filing dateApr 3, 2024
Grant dateAug 12, 2025
Priority date
Expiry dateApr 3, 2044

Classification

  • Technology area (CPC G)Physics
  • CPC primaryG06T2207/20084
  • WIPO fieldComputer technology
  • WIPO sectorElectrical engineering

Abstract

The present invention provides a machine learning-based method for calibrating a camera with respect to a scene, comprising: setting up, in the scene, light sources including: a first group of light sources arranged in a non-linear manner on a first plane and a second group of light sources arranged in a non-linear manner on a second plane orthogonal to the first plane; labelling each light source with a serial number; determining 3D global coordinates of each light source in the scene; configuring the light sources to luminate sequentially; configuring the camera to capture a video of the scene when the light sources sequentially luminate; extracting 2D pixel coordinates of each light source from the video using object detection machine-learning model; matching 3D global coordinates for each light source with respective 2D pixel coordinates to construct a projective relation; and obtaining a projection matrix of the camera from the projective relations.

Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.