Data construction and learning system and method based on method of splitting and arranging multiple images

Patent metadata

Field	Value
Publication number	US-12482268-B2
Application number	US-202217981933-A
Country	US
Kind code	B2
Filing date	Nov 7, 2022
Priority date	Nov 9, 2021
Publication date	Nov 25, 2025
Grant date	Nov 25, 2025

Abstract

The present disclosure relates to a data construction and learning system and method based on a method of splitting and arranging multiple images. The data construction and learning system based on a method of splitting and arranging multiple images includes an input unit configured to receive images captured by a plurality of cameras disposed in a vehicle, a memory in which a program for merging the images into a single image and estimating information on a road situation and an object has been stored, and a processor configured to execute the program. The processor merges and recognizes, as one situation, road situations and objects redundantly included in the images.

First claim

What is claimed is:

1 . A data construction and learning system based on a method of splitting and arranging multiple images, the system comprising:
a processor having an input that receives images captured by a plurality of cameras disposed in a vehicle; and a memory that stores a program for merging the images into a single image and estimating information on a road situation and an object; and wherein the processor executes the program, wherein the processor merges and recognizes, as one situation, road situations and objects redundantly included in the images, wherein the input receives images having different field of views (FOVs) from the plurality of cameras, wherein the processor constructs the single image in which the images are disposed for each section by rearranging and merging the images having the different FOVs, and wherein the processor constructs data to be delivered in a learning level in a way to merge and manage information of an object that is partially displayed for each section and whose region of interest (ROI) is fully displayed and becomes relatively small by performing annotation on the object within the single image.

2 . The system of claim 1 , wherein:
the processor constructs structural data of objects within images by using the single image and calibrated information based on a distance sensor and a position sensor, and the structural data comprises information of an object, a model, a class, a position, heading, a size, and a target.

3 . The system of claim 2 , wherein the processor
merges the images into the single image comprising an image main sector and an image sub-sector, and outputs final estimation information based on only a merged image input upon real-time execution by using learning results using the structural data of the objects within the images.

4 . A data construction and learning method based on a method of splitting and arranging multiple images, which is performed by a data construction and learning system based on a method of splitting and arranging multiple images, the method comprising steps of:
(a) obtaining image data photographed by a plurality of cameras; (b) merging the image data into a single frame having a tile array; (c) converting, into an image domain, object information labeled on an object within the single frame; and (d) generating learning data by merging an image converted into the image domain and labeling information of the image domain, wherein the step (b) comprises constructing the data capable of being delivered in a learning level by merging and managing an object partially displayed in the single frame for each section, information of a region of interest (ROI) displayed in a complete form, and information of an object that is disposed at a long distance and that has a relatively small ROI.

5 . The method of claim 4 , wherein the step (a) comprises obtaining the image data from the plurality of cameras whose photographing information has been set so that the cameras have different field or views (FOVs) and resolution.

6 . The method of claim 4 , wherein the step (c) comprises converting, into the image domain, the object information comprising object coordinate information in a birds' eye view, form modeling information corresponding to an object size, and moving direction information.

7 . The method of claim 4 , wherein the step (d) comprises performing learning using structural data of objects within images comprising information of an object, a model, a class, a position, heading, a size, and a target.

Assignees

• Electronics & Telecommunications Res Inst

Inventors

• Kim Jin Woo
• Kim Ki Tae

Classifications

• G06T5/50

  using two or more images, e.g. averaging or subtraction · CPC title

• G06V2201/07

  Target detection · CPC title

• G06T2207/20221

  Image fusion; Image merging · CPC title

• G06V10/759

  Region-based matching · CPC title

• G06V10/803

  of input or preprocessed data · CPC title