An image depth estimation system and method based on hole convolution and semi-supervised learning

A semi-supervised learning and image depth technology, applied in the field of computer vision, can solve problems such as ignoring the spatial information dependence of depth estimation tasks, reducing spatial resolution, and limiting prediction accuracy, achieving large spatial perception range, strong spatial dependence, The effect of removing noise

Pending Publication Date: 2019-05-10
XIDIAN UNIV
View PDF3 Cites 29 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

But there is no effective true depth as a constraint, and the method is also limited in prediction accuracy
[0006] At the same time, the existing image depth estimation methods based on deep learning ignore the dependence of the depth estimation task on spatial information.
When using a deep convolutional network, a large number of pooling operations reduces the spatial resolution of the features, which leads to the loss of spatial information and has a non-negligible impact on the output results.

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • An image depth estimation system and method based on hole convolution and semi-supervised learning
  • An image depth estimation system and method based on hole convolution and semi-supervised learning
  • An image depth estimation system and method based on hole convolution and semi-supervised learning

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0026] Existing image depth estimation methods based on deep learning ignore the dependence of depth estimation tasks on spatial information. When using a deep convolutional network, a large number of pooling operations reduces the spatial resolution of the features, which leads to the loss of spatial information and has a non-negligible impact on the output results.

[0027] The present invention conducts research on the above-mentioned status quo, and proposes an image depth estimation system based on atrous convolution and semi-supervised learning, including a decoder module and a decoder module symmetrically connected to it. The decoder module inputs the image to be estimated, decodes The module obtains and outputs the image after depth estimation, see figure 1 , the present invention adds a dilated convolution module between the decoder module and the decoder module. The main body of the dilated convolution module of the present invention is a dilated convolution pyramid...

Embodiment 2

[0030] The image depth estimation system based on hole convolution and semi-supervised learning is the same as embodiment 1, see figure 1 , the main body of the atrous convolution module in the present invention is an atrous convolution pyramid model containing four parallel-connected convolution kernels with different expansion rates, the output of the encoder module in the present invention is used as the input of the atrous convolution module, and the atrous convolution The output of the product module is used as the input of the decoder module. The encoder module of the present invention is composed of 4 convolution blocks, each convolution block contains two convolution layers, the decoder module is composed of 4 deconvolution layers, the decoder module and the decoder module have a symmetrical structure, and Make skip connections.

[0031] The encoder module of the present invention reduces the spatial resolution of the input step by step through operations of different...

Embodiment 3

[0035] The present invention is also an image depth estimation method based on atrous convolution and semi-supervised learning, which utilizes deep learning to estimate the depth of an input single image. Realized on the system, the image depth estimation system based on hole convolution and semi-supervised learning is the same as embodiment 1-2, see figure 2 , including the following steps:

[0036] Step 1: Build a deep learning network model based on hole convolution: use hole convolution in the model to form a network structure composed of a decoder module, a hole convolution module and a decoder module; the network structure is characterized by The atrous convolution module, encoder module and decoder module have a symmetrical structure and are skip-connected.

[0037]Step 2: Obtain training set and test set: the training set and test set are composed of stereo image pairs and corresponding depth maps, and the number of training sets is expanded; the acquisition of train...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention discloses an image depth estimation system and method based on cavity convolution and semi-supervised learning, and solves the problem of estimating scene depth from a single image. According to the invention, a conventional encoder is adopted; the network structure model of the decoder is improved, and a cavity convolution module is added between an encoder module and a decoder module. The image depth estimation method specifically comprises the steps of obtaining a training set and a test set; using a semi-supervised learning strategy to train a model; testing the precision ofthe model by using a test set; estimating image depth using a model. According to the method, the spatial perception capability of the network is improved by using hole convolution, a semi-supervisedlearning strategy is adopted, and the output depth map is optimized through the depth map smoothing error. The method has the characteristics of small parameter model, high prediction precision and complete detail information. The method is applied to the fields of image three-dimensional reconstruction and automatic driving.

Description

technical field [0001] The present invention belongs to the field of computer vision technology, and mainly relates to image depth estimation, specifically, an image depth estimation system and method based on atrous convolution and semi-supervised learning, which are used for automatic driving, three-dimensional scene modeling, and image understanding tasks, such as Semantic segmentation, object tracking, etc. Background technique [0002] Image depth estimation is one of the important topics of computer vision. It can effectively help the computer understand the three-dimensional structural relationship of the scene, thus providing the possibility for more advanced image processing. Image depth estimation plays a very important role in various applications in the field of computer vision. For example, robot navigation and car driverless driving need to rely on the depth information of the scene to accurately locate themselves in real time and judge surrounding obstacles; ...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
Patent Type & Authority Applications(China)
IPC IPC(8): G06T7/50G06N3/04G06N3/08G06K9/62
Inventor 王伟刘逸颖王川功许琳珊丁泽赵雯倩
Owner XIDIAN UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap