RGB-D image salient target detection method

A RGB image and target detection technology, applied in the field of computer vision, can solve the problems of expanding depth error, influence, unfavorable detection results of significant targets, etc., and achieve the effect of high precision

Pending Publication Date: 2021-10-22
ANHUI UNIVERSITY
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

In view of the existence of some poor-quality depth images, the use of dual-stream fusion may enlarge the depth error and adversely affect the final salient target detection 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
  • RGB-D image salient target detection method
  • RGB-D image salient target detection method
  • RGB-D image salient target detection method

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0091] The RGB-D image salient target detection method described in this embodiment selects 1485 pictures on the NJU2K data set, selects 700 pictures on the NLPR data set to form a training set, and uses the remaining pictures on the NJU2K data set and the NLPR data set and the entire The STERE, DES and SIP data sets are used as test sets for testing. In addition, for the DUT data set, the same settings as the paper "Depth-induced multiscale recurrent attention network for saliency detection" are used, and the training set is increased with 800 pictures of the DUT training set, and tested on the DUT test set.

[0092] In the training and testing stages, the input RGB-D image is resized to 256*256, and the training set is subjected to data enhancement operations such as random flip, rotation, and border cropping. The Adam optimizer is selected for model training, the initial learning rate is 1e-5, the batch size is 3, the ResNet50 pre-training parameters and the default setting...

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 RGB-D image salient target detection method. The method comprises the following steps: extracting RGB and Depth image features, simultaneously implementing fusion to form RGBD fusion features, and dividing the RGBD fusion features into high-level RGBD fusion features and low-level RGBD fusion features; enhancing the high-level RGBD fusion features by using three Transformers to form high-level RGBD enhanced features; performing three-stream decoding on the high-level RGBD enhanced features, and forming RGBD refined features by combining the high-level RGBD enhanced features with the low-level RGBD fusion features; and fusing the RGBD refined features to form a saliency map. According to the detection method, the high-level features are enhanced by utilizing Transformer, position information of a salient object is accurately obtained, the low-level features are fused by utilizing three-stream fusion, and contour details of the salient object are refined; and through combination of a convolutional neural network and Transformer, and global and local optimization, the saliency map with high precision is generated.

Description

technical field [0001] The invention relates to the field of computer vision, in particular to a salient object detection method of an RGB-D image using three Transformers to enhance high-level features. Background technique [0002] RGB-D image is composed of RGB color image and Depth depth image, expressing the appearance characteristics and three-dimensional information of a certain scene. At present, there are two multi-modal fusion methods in RGB-D image salient target detection methods, one is dual-stream fusion, which treats color and depth images equally, and the other is depth-guided fusion, which focuses on color images and supplements depth images. . In view of the existence of some poor-quality depth images, the use of dual-stream fusion may enlarge the depth error and adversely affect the final salient object detection results. Therefore, depth-guided fusion is a better fusion method. [0003] Image salient object detection is a dense prediction task at the p...

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): G06K9/62G06N3/04G06N3/08
CPCG06N3/08G06N3/045G06F18/253
Inventor 刘政怡汪远张志立檀亚诚姚晟
Owner ANHUI UNIVERSITY
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