Apparatus and method of learning pose of moving object
AssigneeElectronics and Telecommunications Research Institute
InventorJin Woo KIM
Life of this patent
Estimating the term…
Abstract
Provided is a method of learning a pose of a moving object. The method includes determining 3D feature points in the 3D mesh model obtained by previously modeling a general shape of a moving object, fitting the 3D mesh model to the 2D learning image obtained by previously photographing the real shape of the moving object with respect to the determined 3D feature points, obtaining learning data associated with pose estimation of the moving object from the 2D learning image with the 3D mesh model fitted thereto, and learning a pose estimation model estimating a pose of a target moving object included in one real image obtained by the camera by using the learning data.
Claims20
A method of learning a pose of a moving object, performed by a pose learning apparatus including a camera, an input interface, a storage unit storing a three-dimensional (3D) mesh model and a two-dimensional (2D) learning image, and a processor module processing the 2D learning image on basis of the 3D mesh model, the method comprising:
- by using the processor module,
- determining 3D feature points in the 3D mesh model obtained by previously modeling a general shape of a moving object;
- fitting the 3D mesh model to the 2D learning image obtained by previously photographing the real shape of the moving object with respect to the determined 3D feature points;
- obtaining learning data associated with pose estimation of the moving object from the 2D learning image with the 3D mesh model fitted thereto;
- and
- learning a pose estimation model estimating a pose of a target moving object included in one real image obtained by the camera by using the learning data.
An apparatus for learning a pose of a moving object, the apparatus comprising:
- a camera;
- an input interface;
- a storage unit configured to store a three-dimensional (3D) mesh model obtained by previously modeling a general shape of a moving object and a two-dimensional (2D) learning image captured by photographing a real shape of the moving object;
- and
- a processor module configured to process the 2D learning image, based on the 3D mesh model,
- wherein, based on a user input received through the input interface, the processor module fits the 3D mesh model to the 2D learning image with respect to 3D feature points determined in the 3D mesh model, obtains learning data associated with pose estimation of the moving object from the 2D learning image with the 3D mesh model fitted thereto, and learns a pose estimation model estimating a pose of the moving object included in one real image captured by the camera by using the obtained learning data.
An apparatus for learning a pose of a moving object, the apparatus comprising:
- a camera;
- an input interface;
- a storage unit configured to store a three-dimensional (3D) mesh model obtained by previously modeling a general shape of a moving object and a two-dimensional (2D) learning image captured by photographing a real shape of the moving object;
- a feature point determination module configured to determine 3D feature points in the 3D mesh model, based on a user input received through the input interface;
- a fitting module configured to fit the 3D mesh model to the 2D learning image;
- a data extraction module configured to extract learning data associated with pose estimation of the moving object from the 2D learning image with the 3D mesh model fitted thereto and to store the extracted learning data in the storage unit;
- a learning module configured to learn a pose estimation model estimating a pose of a target moving object by using the learning data stored in the storage unit;
- and
- a pose estimation module configured to estimate the pose of the target moving object included in one image captured by the camera by using the pose estimation model.
Dependencies are inferred from the claim text.