KR102816948B1 - Apparatus for measuring physical quantity of target object and method thereof - Google Patents

Abstract

A device and method for measuring a physical quantity of a target object according to an embodiment are disclosed. The device for measuring the physical quantity of the target object includes a model learning unit that pre-trains a deep learning model based on a pre-built learning image data set; a feature point extraction unit that inputs an image received in real time for measuring the physical quantity into the pre-trained deep learning model to extract a predetermined number of feature points; and a distance calculation unit that calculates a terminal distance between a heating coil and a tube terminal of the image based on the predetermined number of extracted feature points.

Description

{APPARATUS FOR MEASURING PHYSICAL QUANTITY OF TARGET OBJECT AND METHOD THEREOF}

The invention relates to a device and method for measuring a physical quantity of a target object.

Problems of measuring geometrical physical quantities inside images, such as the Automated Optical Inspection (AOI) problem, are difficult to solve with current deep learning technologies that are mainly used for classification, object detection, and prediction.

At this time, geometric figures within the image, such as distance, length, thickness, height radius, angle, area, position, and center point coordinates, correspond to such physical quantities.

In order to calculate the above physical quantities using current deep learning technology, a regression model must be used. However, this is difficult to implement in reality because it requires a large amount of data and continuous true values to obtain generalization performance.

The embodiment can provide a device and method for measuring a physical quantity of a target object.

The problem to be solved in the embodiment is not limited to this, and it can be said that the purpose or effect that can be understood from the solution or embodiment of the problem described below is also included.

A physical quantity measuring device of a target object according to an embodiment may include a model learning unit that pre-trains a deep learning model based on a pre-built learning image data set; a feature point extraction unit that inputs an image received in real time for measuring a physical quantity into the pre-trained deep learning model to extract a predetermined number of feature points; and a distance calculation unit that calculates a terminal distance between a heating coil and a tube terminal of the image based on the predetermined number of extracted feature points.

The above model learning unit can input learning images from the above pre-built learning image data set, preprocess the input learning images to generate corner maps, label a predetermined number of feature points in the learning images based on the generated corner maps, and train the deep learning model based on the learning images labeled with the predetermined number of feature points.

The above model learning unit can generate a corner map from the learning image using a phase congruency corner algorithm.

The above-mentioned feature point extraction unit can preprocess the image received in real time to generate a corner map, input the generated corner map into the deep learning model, and extract a predetermined number of feature points in the received image in response thereto.

The above distance calculation unit can calculate coordinate values of one feature point of the heating coil and two feature points of the tube terminal, and calculate the terminal distance based on the calculated coordinate values.

The above distance calculating unit can draw a normal line from the feature point of the heating coil to a straight line between the feature points of the tube terminal based on the calculated coordinate values, and calculate the length of the normal line as the terminal distance.

The physical quantity measuring device of the above target object further includes a data construction unit that collects predetermined learning images and constructs a learning image data set including the collected learning images, and the data construction unit can expand the learning image data set by adding an image used for measuring the physical quantity by the deep learning model.

The above data construction unit can generate a virtual learning image using at least one pre-defined data augmentation function, and expand the generated virtual learning image by adding it to the learning image data set.

A method for measuring a physical quantity of a target object according to an embodiment may include a model learning step for pre-learning a deep learning model based on a pre-built learning image data set; a feature point extraction step for inputting an image received in real time for measuring a physical quantity into the pre-learned deep learning model to extract a predetermined number of feature points; and a distance calculation step for calculating a terminal distance between a heating coil and a tube terminal of the image based on the predetermined number of extracted feature points.

The above model learning step may input learning images from the above pre-built learning image data set, preprocess the input learning images to generate corner maps, label a predetermined number of feature points in the learning images based on the generated corner maps, and train the deep learning model based on the learning images labeled with the predetermined number of feature points.

The above model learning step can generate a corner map from the training image using a phase congruency corner algorithm.

The above feature point extraction step can preprocess the image received in real time to generate a corner map, input the generated corner map into the deep learning model, and extract a predetermined number of feature points in the received image in response thereto.

The above distance calculation step can calculate the coordinate values of one feature point of the heating coil and two feature points of the tube terminal, and calculate the terminal distance based on the calculated coordinate values.

The above distance calculation step can calculate the length of the normal line as the terminal distance by drawing a normal line from the feature point of the heating coil to the straight line between the feature points of the tube terminal based on the calculated coordinate values.

The above method for measuring physical quantities of the target object further includes a data construction step of collecting predetermined learning images prior to the model learning step and constructing a learning image data set including the collected learning images, and the data construction step can be expanded by adding an image used for measuring physical quantities by the deep learning model to the learning image data set.

The above data construction step can generate a virtual learning image using at least one pre-defined data augmentation function, and expand the generated virtual learning image by adding it to the learning image data set.

According to an embodiment, it may be possible to precisely measure geometrical physical quantities in an image by calculating the longitudinal distance between a heating coil and a tube terminal using feature points of a heating coil and feature points of a tube terminal extracted by a deep learning model.

The various advantageous and beneficial advantages and effects of the present invention are not limited to the above-described contents, and will be more easily understood in the course of explaining specific embodiments of the present invention.

Figure 1 is a drawing showing a system for measuring physical quantities of a target object according to an embodiment.
Figure 2 is a drawing showing a method for measuring a physical quantity of a target object according to an embodiment.
Figure 3 is a diagram illustrating in detail the deep learning model learning process illustrated in Figure 2.
Figures 4 to 6 are drawings for explaining the principle of calculating the longitudinal distance illustrated in Figure 2.
Figure 7 is a drawing showing a detailed configuration of the physical quantity measuring device illustrated in Figure 1.
Figure 8 is a drawing showing the detailed configuration of the data construction unit illustrated in Figure 7.
Figure 9 is a drawing showing the detailed configuration of the feature point extraction unit illustrated in Figure 7.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the attached drawings.

However, the technical idea of the present invention is not limited to some of the embodiments described, but can be implemented in various different forms, and within the scope of the technical idea of the present invention, one or more of the components among the embodiments can be selectively combined or substituted for use.

In addition, terms (including technical and scientific terms) used in the embodiments of the present invention can be interpreted as having a meaning that can be generally understood by a person of ordinary skill in the technical field to which the present invention belongs, unless explicitly and specifically defined and described, and terms that are commonly used, such as terms defined in a dictionary, can be interpreted in consideration of the contextual meaning of the relevant technology.

Additionally, the terms used in the embodiments of the present invention are for the purpose of describing the embodiments and are not intended to limit the present invention.

In this specification, the singular may also include the plural unless specifically stated otherwise in the phrase, and when it is described as “A and/or at least one (or more) of B, C”, it may include one or more of all combinations that can be combined with A, B, C.

Additionally, in describing components of embodiments of the present invention, terms such as first, second, A, B, (a), (b), etc. may be used.

These terms are only intended to distinguish one component from another, and are not intended to limit the nature, order, or sequence of the component.

In addition, when a component is described as being “connected,” “coupled,” or “connected” to another component, it may include not only cases where the component is directly connected, coupled, or connected to the other component, but also cases where the component is “connected,” “coupled,” or “connected” by another component between the component and the other component.

In addition, when described as being formed or arranged “above or below” each component, above or below includes not only the case where the two components are in direct contact with each other, but also the case where one or more other components are formed or arranged between the two components. In addition, when expressed as “above or below,” it can include the meaning of the downward direction as well as the upward direction based on one component.

Figure 1 is a drawing showing a system for measuring physical quantities of a target object according to an embodiment.

Referring to FIG. 1, a system for measuring a physical quantity of a target object according to an embodiment of the present invention may include an X-ray generator (10) and a physical quantity measuring device (100).

An X-ray generator (10) can generate X-rays and project them onto a target object (1).

A physical quantity measuring device (100) can obtain an image captured by an X-ray and measure a physical quantity of a target object from the obtained image. Here, the physical quantity can include geometric physical quantities of the target object in the image, such as length, area, angle, radius, center point, etc.

In the embodiment, an example is described in which a heating coil is used as a sample to be measured, and the longitudinal distance between the heating coil and the tube terminal in the image is to be measured.

FIG. 2 is a drawing showing a method for measuring a physical quantity of a target object according to an embodiment, FIG. 3 is a drawing showing in detail the deep learning model learning process shown in FIG. 2, and FIGS. 4 to 6 are drawings for explaining the terminal distance calculation principle shown in FIG. 2.

Referring to FIG. 2, a device for measuring a physical quantity of a target object according to an embodiment of the present invention (hereinafter referred to as a physical quantity measuring device) can collect learning images (S100).

Next, the physical quantity measuring device can construct learning image data including the collected learning images (S200).

At this time, the image data for learning may be data indicating predetermined feature points used to use the terminal distance between the heating coil and the tube terminal in the image.

Next, the physical quantity measuring device can train a deep learning model based on training images in the constructed training image data (S300).

Referring to FIG. 3, the physical quantity measuring device can input a learning image from the constructed learning image data and preprocess the input learning image to generate a corner map.

At this time, the learning image may be an image including a heating coil (2) and a tube terminal (3) as shown in Fig. 4.

At this time, the process of preprocessing the learning image may include a filtering process that emphasizes corners that are strong against changes in lighting and contrast. In an embodiment, a corner map may be generated from the learning image using a predetermined algorithm, for example, a phase congruency corner algorithm, but is not necessarily limited thereto.

The physical quantity measuring device can label feature points in the learning image based on the generated corner map. In the embodiment, at least three feature points are required, for example, one feature point of the heating coil and two feature points of the tube terminal are required. Accordingly, the physical quantity measuring device can label one feature point of the heating coil (2) in the learning image as n, and label two feature points of the tube terminal (3) as x and y, as shown in Fig. 5.

A physical quantity measuring device can train a deep learning model based on training images with labeled feature points.

Meanwhile, when the physical quantity measuring device receives an image in real time (S400), it can generate a corner map from the received image.

Next, the physical quantity measuring device can input the generated corner map into a pre-trained deep learning model and extract a number of feature points in the received image in response.

Next, the physical quantity measuring device can calculate the coordinate values of the extracted feature points, and calculate the longitudinal distance between the heating coil and the tube terminal based on the coordinate values of the calculated feature points.

Specifically, as shown in FIG. 6, the physical quantity measuring device calculates the coordinate values (x1, y1), (x2, y2), (x3, y3) of each of the three extracted feature points P1, P2, and P3, and draws a normal line from the feature point P3 of the heating coil (2) to the straight line between the feature points P1 and P2 of the tube terminal (3), and calculates the length of the normal line as the longitudinal distance D between the heating coil (2) and the tube terminal (3).

At this time, since the length of the straight line between the feature points P1 and P2 of the tube terminal corresponds to the diameter of the tube terminal, the feature points P1 and P2 of the tube terminal can be extracted by taking this into consideration.

FIG. 7 is a drawing showing a detailed configuration of the physical quantity measuring device shown in FIG. 1, FIG. 8 is a drawing showing a detailed configuration of the model learning unit shown in FIG. 7, and FIG. 9 is a drawing showing a detailed configuration of the feature point extraction unit shown in FIG. 7.

Referring to FIG. 7, a device for measuring a physical quantity of a target object according to an embodiment of the present invention may include an image acquisition unit (110), a data construction unit (120), a model learning unit (130), a feature point extraction unit (140), and a distance calculation unit (150).

The image acquisition unit (110) can acquire an image captured by an X-ray. Here, the image may be an image captured directly or an image generated by image processing.

The data construction unit (120) can construct a learning image data set including pre-collected learning images. The learning image data set can be continuously updated or expanded.

For example, the data construction unit (120) can construct a learning image data set, and expand the learning image data set by adding an image used to calculate physical quantities by a deep learning model.

As another example, the data construction unit (120) may construct a learning image data set, and may expand the learning image data set by generating a virtual learning image using various pre-determined data augmentation functions and adding the generated virtual learning image to the learning image data set.

At this time, the predefined data increase/decrease functions may include, for example, horizontal flip, vertical flip, random rotate90, radom brightness, guassian noise, Gaussian blur, color jittering, negative image, etc.

The model learning unit (130) can learn a deep learning model based on constructed learning image data.

As shown in Fig. 8, the model learning unit (130) may include a collection unit (131), a preprocessing unit (132), and a learning unit (133).

The collection unit (131) can receive learning images from learning image data.

The preprocessing unit (132) can preprocess the input training image to generate a corner map and label feature points in the training image based on the generated corner map.

The learning unit (133) can learn a deep learning model based on a learning image in which feature points are labeled. Here, the deep learning model can be a model for extracting a predetermined number of feature points to calculate the optimal distance between the heating coil and the tube terminal.

The feature point extraction unit (140) can input an image received in real time to a pre-trained deep learning model to extract a predetermined number of feature points in order to measure physical quantities.

As shown in Fig. 9, the feature point extraction unit (140) may include a collection unit (141), a preprocessing unit (142), and an extraction unit (143).

The collection unit (141) can collect images received in real time to measure physical quantities.

The preprocessing unit (142) can preprocess the received image to generate a corner map. The preprocessing unit (142) can generate a corner map from the image using a predetermined algorithm.

The extraction unit (143) can input the generated corner map into a pre-trained deep learning model and extract a number of predetermined feature points in the received image in response thereto.

The distance calculation unit (150) can calculate the longitudinal distance between the heating coil and the tube terminal based on a number of extracted feature points, i.e., feature points of the heating corner and feature points of the tube corner.

The term '~ part' used in this embodiment means a software or hardware component such as a field-programmable gate array (FPGA) or an ASIC, and the '~ part' performs certain roles. However, the '~ part' is not limited to software or hardware. The '~ part' may be configured to be on an addressable storage medium and may be configured to play one or more processors. Thus, as an example, the '~ part' includes components such as software components, object-oriented software components, class components, and task components, processes, functions, attributes, procedures, subroutines, segments of program code, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, and variables. The functionality provided in the components and '~ parts' may be combined into a smaller number of components and '~ parts' or further separated into additional components and '~ parts'. Additionally, the components and '~parts' may be implemented to regenerate one or more CPUs within the device or secure multimedia card.

Although the present invention has been described above with reference to preferred embodiments thereof, it will be understood by those skilled in the art that various modifications and changes may be made to the present invention without departing from the spirit and scope of the present invention as set forth in the claims below.

10: X-ray generator
100: Physical quantity measuring device
110: Image acquisition section
120: Data Construction Department
130: Model Learning Section
140: Feature extraction section
150: Distance Calculation Unit

Claims (16)

A model learning unit that pre-trains a deep learning model based on a pre-built learning image data set;
A feature point extraction unit that inputs an image received in real time for physical quantity measurement into the pre-trained deep learning model to extract a predetermined number of feature points; and
A distance calculation unit is included for calculating the longitudinal distance between the heating coil and the tube terminal of the image based on the predetermined number of feature points extracted above.
The above distance calculation section
Calculate the coordinate values of one feature point of the above heating coil and two feature points of the above tube terminal,
A physical quantity measuring device for a target object that calculates the terminal distance based on the calculated coordinate values. In the first paragraph,
The above model learning part is,
Input training images from the above pre-built training image data set, preprocess the input training images, and create a corner map.
Based on the generated corner map, a predetermined number of feature points in the training image are labeled,
A physical quantity measuring device for a target object that trains the deep learning model based on a training image labeled with the above-described number of feature points. In the second paragraph,
The above model learning part is,
A physical quantity measuring device of a target object that generates a corner map from the training image using a phase congruency corner algorithm. In the first paragraph,
The above feature extraction unit,
Preprocess the image received in real time to create a corner map,
A physical quantity measuring device for a target object, which inputs the generated corner map into the deep learning model and extracts a predetermined number of feature points in the received image in response thereto. delete In the first paragraph,
The above distance calculation section,
Based on the above-described coordinate values, a normal line is drawn from the feature points of the heating coil to the straight line between the feature points of the tube terminal.
A physical quantity measuring device for a target object that calculates the length of the above normal line as the above terminal distance. In the first paragraph,
Further comprising a data construction unit for collecting pre-determined learning images and constructing a learning image data set including the collected learning images,
The above data construction unit,
A device for measuring physical quantities of a target object, which expands the image used to measure physical quantities by the deep learning model by adding it to the training image data set. In Article 7,
The above data construction unit,
Generate virtual learning images using at least one predefined data augmentation function,
A physical quantity measuring device of a target object, which expands the above-mentioned generated virtual learning image by adding it to the above-mentioned learning image data set. A model training step that pre-trains a deep learning model based on a pre-built training image data set;
A feature point extraction step for extracting a predetermined number of feature points by inputting an image received in real time for physical quantity measurement into the pre-trained deep learning model; and
A distance calculation step for calculating the longitudinal distance between the heating coil and the tube terminal of the image based on the above-determined number of extracted feature points is included.
The above distance calculation step is
Calculate the coordinate values of one feature point of the above heating coil and two feature points of the above tube terminal,
A method for measuring physical quantities of a target object, wherein the terminal distance is calculated based on the calculated coordinate values. In Article 9,
The above model learning step is,
Input training images from the above pre-built training image data set, preprocess the input training images, and create a corner map.
Based on the generated corner map, a predetermined number of feature points in the training image are labeled,
A method for measuring physical quantities of a target object, wherein the deep learning model is trained based on a training image labeled with a predetermined number of feature points. In Article 10,
The above model learning step is,
A method for measuring physical quantities of a target object, which generates a corner map from the training image using a phase congruency corner algorithm. In Article 9,
The above feature extraction step is,
Preprocess the image received in real time to create a corner map,
A method for measuring physical quantities of a target object, wherein the generated corner map is input into the deep learning model and a predetermined number of feature points are extracted from the received image in response thereto. delete In Article 9,
The above distance calculation step is,
Based on the above-described coordinate values, a normal line is drawn from the feature points of the heating coil to the straight line between the feature points of the tube terminal.
A method for measuring physical quantities of a target object, wherein the length of the above normal line is calculated as the above terminal distance. In Article 9,
The method further includes a data construction step of collecting pre-determined learning images prior to the above model learning step and constructing a learning image data set including the collected learning images.
The above data construction step is,
A method for measuring physical quantities of a target object, which expands the training image data set by adding an image used to measure physical quantities by the deep learning model. In Article 15,
The above data construction step is,
Generate virtual learning images using at least one predefined data augmentation function,
A method for measuring physical quantities of a target object, which expands the above-mentioned generated virtual learning image by adding it to the above-mentioned learning image data set.

Images