Multi-scale enhanced monocular depth estimation method

Abstract

The invention provides a multi-scale enhanced monocular depth estimation method, which comprises the following steps: step 1, inputting a single RGB image, and then performing multi-scale feature extraction on the RGB image by adopting a context and receptive field enhanced high-resolution network CRE-HRNet to obtain a high-resolution first image; step 2, using a residual error expansion convolution unit of a receptive field enhancement module to carry out expansion convolution on the first depth image to obtain a second image; and step 3, capturing long-distance pixel points of the second depth image by using a weighted non-local neighborhood module to obtain a depth image. According to the method provided by the invention, the monocular depth estimation precision is high on the basis of obtaining the feature information of the intermediate layer.

Description

technical field [0001] The invention belongs to the field of computer vision of deep learning, and in particular relates to a multi-scale enhanced monocular depth estimation method. Background technique [0002] Image-based depth information estimation refers to learning the three-dimensional information of the scene from a single or multiple two-dimensional images, aiming to predict the pixel depth of the image, and the estimated depth map can be applied to intelligent robots and scene reconstruction , semantic segmentation, unmanned driving and other fields, has important research significance and application value, and is an important research issue in the field of computer vision. Among them, estimating the depth information from a single image is also called monocular depth estimation, because it only needs a single image to achieve depth estimation, which is more portable than the multiple images required by the multi-view method, but because The single image may be t...

AI Technical Summary

Problems solved by technology

However, most of the monocular depth estimation methods based on deep learning, in order to improve the receptive field of the monocular depth estimation network, most of the CNNs used are obtained through repeated stacking of long-range dependency capture and backpropagation. When transmitting information back and forth over a long distance, such local operations are difficult to achieve, and it is easy to lose the feature information of the middle layer, resulting in the consequence of low monocular depth estimation accuracy, such as through literature [1], literature [2] and literature [3] The accuracy of the depth map obtained by the monocular depth estimation method involved in

Images