Monocular scene depth prediction method based on deep learning

Abstract

The invention discloses a monocular scene depth prediction method based on deep learning. The method is suitable for monocular pictures or videos, a calibrated binocular color image pair is used for training a depth prediction model, a DenseNet convolution module is used for extracting a feature space in a network architecture, and dense blocks and transition layers in the feature space are used,so that each layer in a network is directly connected with a front layer of the network, and repeated utilization of features is realized; binocular matching loss improvement, a depth prediction problem is regarded as an image reconstruction problem; the method comprises the steps of sampling an input left viewpoint color image and a disparity image to generate a virtual color image and a disparity image, and restraining the consistency of the generated virtual view and a correspondingly input right viewpoint image on an RGB level and a disparity level by utilizing a stereo matching algorithmof a binocular image pair, thereby obtaining a better depth; the depth smooth loss is improved, the high-quality dense depth map can be generated, the artifact problem caused by shielding in monocularscene depth prediction is effectively solved, and the 2D-to-3D conversion requirements of multiple indoor and outdoor real scenes can be met.

Description

technical field [0001] The invention relates to a monocular scene depth prediction method based on deep learning, belonging to the fields of computer vision and image processing. Background technique [0002] Monocular depth prediction is a research topic that has attracted much attention in computer vision, and has a wide range of application values ​​in areas such as autonomous driving, VR game production, and film and television production. However, there are still many unsolved problems in this field, such as: [0003] 1) The process of using radar laser to collect depth data is costly and greatly affected by weather; [0004] 2) The generated depth image produces artifacts and smears due to light shadows or occlusions in the original image; [0005] 3) The method based on the depth information restored by the sparse depth map has the problem of discontinuous edge depth; [0006] 4) The depth model is not completely differentiable, and the computability of the gradien...

AI Technical Summary

Problems solved by technology

[0005] 3) The method based on the depth information restored by the sparse depth map has the problem of discontinuous edge depth;

[0006] 4) The depth model is not completely differentiable, and the computability of the gradient in optimization is lost, making the training suboptimal;

[0007] 5) The image generation model does not have the ability to scale images to large output resolutions;

[0008] 6) The generalization ability of the model is generally restricted by the training data

Images