Unsupervised monocular depth estimation method based on generative adversarial network

Abstract

The invention discloses an unsupervised monocular depth estimation method based on a generative adversarial network, and the method comprises the following steps: 1, obtaining a left and right image pair with strict time synchronization through a binocular camera, building a binocular color image data set, and correcting a binocular color image; 2, establishing an unsupervised generative adversarial network model, inputting the corrected binocular color image into the network, and performing training and iterative regression on the network model; 3, inputting the monocular color image into thetrained network model to generate a disparity map corresponding to the monocular color image; and 4, converting the disparity map into depth information through a binocular disparity depth conversionformula, and synthesizing a depth map. According to the depth estimation method provided by the invention, the monocular color image is converted into the depth map containing the depth information by using the unsupervised network model, and complex real depth data is not needed.

Description

technical field [0001] The invention belongs to the technical field of robot vision, and relates to an unsupervised monocular depth estimation method based on a generative confrontation network. Background technique [0002] Depth information is the core issue in the fields of visual SLAM, 3D scene reconstruction, and medical imaging in computer vision. In the field of robotics, accurate depth estimation is very important for computer vision to understand the three-dimensional environment for machine motion planning, navigation positioning, motion obstacle avoidance, and control decision-making. [0003] Generally speaking, there are two main methods for depth estimation: direct measurement by 3D measurement sensors and depth restoration of image information. Three-dimensional measurement sensors mainly rely on various direct measurement sensors such as Velodyne designed lidar to create a three-dimensional environment by emitting laser beams at a certain frequency to scan t...

AI Technical Summary

Problems solved by technology

[0007] The purpose of the present invention is to provide an unsupervised monocular depth estimation method based on generative confrontation networks, which can solve the problems of high cost of current depth estimation sensors and inaccurate camera pose estimation, and at the same time perform environment perception and three-dimensional scene reconstruction for the camera , providing the basis for autonomous driving and other related tasks

Images