Method and apparatus for detecting dimension error

Patent metadata

Field	Value
Publication number	US-12112468-B2
Application number	US-202117163051-A
Country	US
Kind code	B2
Filing date	Jan 29, 2021
Priority date	Jan 30, 2020
Publication date	Oct 8, 2024
Grant date	Oct 8, 2024

Abstract

An apparatus for detecting a dimension error obtains an image of a target object, estimates dimensional data for a region of interest (ROI) for which dimensions are to be measured from the image of the target object using a learned dimensional measurement model, and determines whether there is a dimension error in the ROI from the estimated dimension data using a learned dimension error determination model.

First claim

What is claimed is:

1. A method for detecting a dimension error of a target object in an apparatus for detecting a dimensional error, the method comprising:
obtaining an image of the target object;
estimating dimensional data for a region of interest (ROI) for which dimensions are to be measured from the image of the target object using a learned dimensional measurement model; and
determining whether there is a dimension error in the ROI from the estimated dimension data using a learned dimension error determination model,
wherein the method further comprising learning the dimension measurement model using learning data including an image of each analyzed object and ground truth (GT) dimension data of each analyzed object,
wherein the learning includes:
converting the estimated dimension data of the dimension measurement model to the image of each analyzed object and the GT dimension data corresponding to the image of each analyzed object by using an objective function that converts data values within a set target value range, respectively;
calculating differences between the estimated dimension data and the ground truth dimension data converted, respectively, by the objective function; and
updating weights set in the dimension measurement model based on the differences,
wherein the learning further includes:
classifying the estimated dimension data of the dimension measurement model into data corresponding to the range of the target value and data outside the range of the target value;
substituting the data within the range of the target value and ground truth (GT) data as an input value of the objective function, respectively.

2. The method of claim 1 , wherein the learning further includes:
in the case of data outside the range of the target value, calculating a difference between the data outside the range of the target value and the corresponding dimension data without conversion using the target function.

3. The method of claim 1 , wherein the objective function includes a function of Equation 1,
wherein Equation 1 is z=log(c1+1−x), and
the c1 is the target value, and x is the estimated dimension data or the GT dimension data.

4. The method of claim 1 , wherein the objective function includes a function of Equation 2,
wherein Equation 2 is z=log(abs(log(c1+1−x))), and
the c1 is the target value, while x is the estimated dimension data or the GT dimension data.

5. The method of claim 1 , further comprising:
learning the dimension measurement model using first learning data including an image of each analyzed object and GT dimension data of each analyzed object; and
learning the dimension error determination model using second learning data including the image of each analyzed object, the estimated dimension data of the dimension measurement model for the image of each analyzed object, and information necessary for determining the dimension error of each analyzed object.

6. The method of claim 5 , wherein the information necessary for determining the dimension error includes at least one of information on objects with a dimensional error, information on objects without a dimensional error, a shape of the region of interest; and a position of the ROI.

7. An apparatus for detecting a dimension error of a target object, the apparatus comprising:
an image capturing unit that acquires an image of the target object; and
a dimension error analyzer that determines whether or not there is a dimension error in the region of interest for which dimensions are to be measured from the image of the target object using a learned dimension error determination model,
wherein the apparatus further comprising a three-dimensional (3D) image acquirer that estimates dimension data for the ROI from the image of the target object using a learned dimension measurement model,
wherein the dimension error determination model receives dimensional data estimated by the dimensional measurement model as an input and detects whether the dimension error has occurred,
wherein the 3D image acquirer includes a dimensional measurement model learning unit that learns the dimension measurement model using learning data including an image of each analyzed object and ground truth (GT) dimension data of each analyzed object, and
the dimension measurement model learning unit includes:
a data converter that converts the estimated dimension data of the dimension measurement model to the image of each analyzed object and the GT dimension data corresponding to the image of each analyzed object by using an objective function that converts data values within a set target value range, respectively; and
a weight updater that updates weights set in the dimension measurement model based on differences between the estimated dimension data and the ground truth dimension data converted, respectively, by the objective function,
wherein the dimension measurement model learning unit classifies the estimated dimension data of the dimension measurement model into data corresponding to the range of the target value and data outside the range of the target value and substitutes the data within the range of the target value and ground truth (GT) data as an input value of the objective function, respectively.

8. The method of claim 7 , wherein the objective function includes a function of Equation 1 or Equation 2,
wherein Equation 1 is z=log(c1+1−x),
Equation 2 is z=log(abs(log(c1+1−x))), and
the c1 is the target value, while x is the estimated dimension data or the GT dimension data.

9. The apparatus of claim 7 , further comprising, when a 2D image of the target object is acquired from the image capturing unit, a three-dimensional (3D) image acquirer that generates a depth image including distance information between the image capturing unit and the target object from the 2D image,
wherein the dimension error analyzer determines whether the dimension error has occurred from the 2D image and the depth image of the target object using the dimension error determination model.

10. The apparatus of claim 9 , wherein the dimension error analyzer includes a dimension error determination model learning unit that learns dimension error determination model using the 2D image of each object, a depth image generated from each 2D image, a 3D image of each object, dimension data corresponding to the 2D image for each object, and information necessary to determine the dimension error of each object.

11. The apparatus of claim 9 , wherein the 3D image acquirer that generates the depth image from the 2D image of the target object through a learned depth estimation model.

12. The apparatus of claim 11 , wherein the 3D image acquirer that learns a depth estimation model using the image of each object and the depth image including corresponding distance information of each 2D image.

Assignees

• Electronics & Telecommunications Res Inst

Inventors

• Kim Hye-Jin
• Chi Suyoung

Classifications

• G06N20/00

  Machine learning · CPC title

• G06T2207/20081

  Training; Learning · CPC title

• G06T2207/30108

  Industrial image inspection · CPC title

• G06T7/50

  Depth or shape recovery · CPC title

• G06T7/0006 Primary

  using a design-rule based approach · CPC title