Training method, device and image positioning method for image positioning model

An image positioning and training method technology, applied in the field of image processing and computer vision, which can solve the problems of slow algorithm speed, low precision, and large positioning error.

Active Publication Date: 2019-02-26
SHENZHEN UNIV
View PDF4 Cites 7 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0007] In view of the deficiencies in the prior art above, the purpose of the present invention is to provide a training method, device and image positioning method for an image positioning model to solve the problems of slow speed, low precision and large positioning errors caused by similar scenes in the current algorithm. problems, providing technical support for applications such as pedestrian navigation and autonomous driving

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
  • Training method, device and image positioning method for image positioning model
  • Training method, device and image positioning method for image positioning model
  • Training method, device and image positioning method for image positioning model

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0067] The first implementation provided by the present invention is a training method of an image positioning model, such as figure 1 shown, including:

[0068] Step S11, extracting an image set from the video, selecting training images from the image set, and selecting paired images for each training image, the training images and their paired images form a training image pair.

[0069] First extract the image set from the video, and select training images for model training from the image set, and select a paired image for each training image. Preferably, the step of selecting paired images for each training image described in this step includes :

[0070] Select the image at the next moment of the training image as the paired image of the training image;

[0071] And, select the first image at the beginning as the paired image of the last training image.

[0072] If multiple image sets are used to select paired images, then the paired image randomly selects an unpaired ...

Embodiment 2

[0131] The second embodiment provided by the present invention is a training device for an image positioning model, such as figure 2 As shown, the image positioning model is obtained by the training method as described, and is constructed based on a two-stream neural convolutional network;

[0132] Contains: two ResNet50 convolutional network branches and an inverted Y-shaped structure branch composed of three fully connected layers, and each network branch includes: feature extraction module, absolute value calculation module, and the second half of the two branches The branches of the inverted Y-shaped structure are connected, and the branches of the inverted Y-shaped structure include: a loss calculation module and a relative value calculation module;

[0133] The feature extraction module 210 is used to extract the feature value of the image in the ResNet50 convolutional network model of input construction;

[0134] The absolute value calculation module 220 is used to ca...

Embodiment 3

[0139] The third embodiment provided by the present invention is a method for positioning a single image using the image positioning model, such as Figure 4 and Figure 5 shown, including:

[0140] Step S51, preprocessing the single image: planning the gray value of the single image within a preset range, calculating the mean and standard deviation of the gray value of each color channel of the planned training image, and calculating the planned After subtracting the mean value from the gray value, divide by the standard deviation, and finally obtain an image with normalized pixel values;

[0141] Step S52 , inputting the preprocessed single image into a single ResNet50 convolutional network branch of the image positioning model to obtain the absolute value of the position and attitude of the single image positioning.

[0142]The following is an application of image localization for single image input.

[0143] 1. Image preprocessing

[0144] Scale the training image to 2...

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 a training method, a device and an image positioning method of an image positioning model. The invention designs an image relative position consistency loss function, an imagerelative position prediction loss function and an adaptive image characteristic difference loss function according to the position and posture differences between two images. The loss function based on two images can keep the relative position and attitude between images and improve the accuracy of image positioning together with the global position and attitude loss function. The method providedby the invention can realize real-time positioning based on images, and has the advantages of high precision, good reliability and the like.

Description

technical field [0001] The present invention relates to the technical fields of image processing and computer vision, in particular to a training method and device for an image positioning model based on a two-stream convolutional neural network, and an image positioning method. Background technique [0002] Image-based localization methods have important applications in robotics, automatic navigation, AR and VR games and other fields. The traditional methods are mainly divided into two types, the image location method based on image retrieval technology and the image location method based on 3D model. [0003] Based on the image retrieval method, by comparing the currently captured image with the existing database with geographic location tags, the position of the database image most similar to the currently captured image is taken as the position of the currently captured image, that is, the location of the person. This type of method is mainly divided into three steps, 1...

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
CPCG06N3/08G06N3/045G06F18/214
Inventor 李庆朱家松李清泉邱国平
Owner SHENZHEN 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