| CPC G06T 7/55 (2017.01) [G06T 7/75 (2017.01); G06T 19/00 (2013.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01); G06T 2207/30244 (2013.01); G06T 2210/56 (2013.01)] | 17 Claims |

|
1. A three-dimensional reconstruction method, comprising:
obtaining an image of a first object and a camera pose of the image;
determining a first normalized object location field (NOLF) image of the first object in the image by using a first deep learning network, wherein the first NOLF image indicates a normalized three-dimensional point cloud of the first object at a photographing angle of view of the image;
determining a first relative pose of the first object based on the first NOLF image, wherein the first relative pose is a relative pose between a pose of the first object and the camera pose of the image, wherein the determining the first relative pose of the first object based on the first NOLF image comprises:
determining pixel coordinates of at least four feature points of the first object in the image by using a second deep learning network, wherein four object points indicated by the four feature points are not coplanar in a three-dimensional space;
determining three-dimensional coordinates of the at least four feature points in the first NOLF image; and
determining the first relative pose based on the pixel coordinates and the three- dimensional coordinates of the at least four feature points;
determining a plurality of NOLF images of a plurality of three-dimensional models at an angle of view corresponding to the first relative pose;
determining, from the plurality of three-dimensional models in a model database based on the first NOLF image, a first model corresponding to the first object;
determining the pose of the first object based on the first model and the camera pose of the image; and
performing a three-dimensional reconstruction on the first object in the image based on the first model and the pose of the first object.
|