| CPC G06T 7/50 (2017.01) [G06V 10/75 (2022.01); G06V 10/98 (2022.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01)] | 20 Claims |
|
1. A method for obtaining depth images for improved driving safety comprising:
obtaining first images and second images captured by a camera;
processing the first images by a deep learning network model, and obtaining first predicted depth maps of the first images;
processing the second images by the deep learning network model, and obtaining second predicted depth maps of the second images;
determining a transformation matrix of the camera between the first images and the second images;
converting the first predicted depth maps into first point cloud maps, and converting the second predicted depth maps into second point cloud maps;
converting the first point cloud maps into third point cloud maps according to the transformation matrix of the camera, and converting the second point cloud maps into fourth point cloud maps according to the transformation matrix of the camera;
matching the first point cloud maps with the fourth point cloud maps, and calculating first error values between the first point cloud maps and the fourth point cloud maps;
matching the second point cloud maps with the third point cloud maps, and calculating second error values between the second point cloud maps and the third point cloud maps;
adjusting the deep learning network model according to the first error values and the second error values and obtaining a target deep learning network model; and
inputting images to be detected into the target deep learning network model, and obtaining depth images corresponding to the images to be detected.
|