WO2021045481A1

WO2021045481A1 - Object recognition system and method

Info

Publication number: WO2021045481A1
Application number: PCT/KR2020/011704
Authority: WO
Inventors: 김보영; 민준홍; 노경식; 최민용
Original assignee: 삼성전자 주식회사
Priority date: 2019-09-04
Filing date: 2020-09-01
Publication date: 2021-03-11
Also published as: KR20210028503A

Abstract

The present disclosure relates to an object recognition system and method. The object recognition system according to an embodiment of the present disclosure comprises: an image acquisition unit for repeatedly acquiring an image of an object at a specified angle or location in the object recognition system; and a processor operatively connected to the image acquisition unit to receive the image of the object acquired by the image acquisition unit, wherein the processor may be configured to calculate a gradient value for the edge of the object from the acquired image of the object and generate a template of the object on the basis of the gradient value.

Description

Object recognition system and method

Various embodiments of the present disclosure relate to an object recognition system and method.

Object tracking is a type of image analysis technology that finds the location of a specific object in a continuous image or a plurality of images, and is used in various fields such as security, surveillance, and traffic. Among the technologies related to objects, the most commonly used method is template matching, which is a method of finding an object of interest that is most similar to a template by comparing a plurality of objects appearing in an image or image with a template.

In recent years, as the smartization of manufacturing plants has accelerated, the cases of recognizing parts or identifying the location of parts by applying a template matching method in the manufacturing process have gradually increased. Specifically, by attaching a camera to a robot arm to acquire an image, and by comparing a part displayed in the image with a template, it is possible to recognize a part necessary for a manufacturing process or to determine the location of the part. Accordingly, manufacturing processes such as assembling, inspecting, or loading parts using only the robot arm without additional manpower input have become possible.

In order to use the robot arm in the manufacturing process, precise recognition of objects (e.g., parts) and accurate positioning are essential. Most of the parts used in the manufacturing process are in a texture-less form. Because it was a product of, it was not easy to recognize or locate the item.

Accordingly, as a method of recognizing parts used in the manufacturing process, a method of creating a template by using edge information of an object acquired through a camera, and recognizing an object by comparing the template with the object has been proposed. . However, since edge information is often distorted by the surrounding environment (eg, light, shadow, etc.), there is a limitation in that it is difficult to accurately recognize an object even by the above method.

As another method, a method of recognizing an object using a template based on the outline information of the CAD design has been proposed, but the object to be recognized is formed in a state having an error different from the CAD design, or in a partially deformed form. When formed, there is a limitation that it is difficult to accurately recognize an object.

Accordingly, the present disclosure provides an object recognition system and method capable of repeatedly driving a robot arm to obtain various images of an object, and to generate a template that minimizes the influence of the surrounding environment based on the obtained various images. , To overcome the above-mentioned limitations.

An object recognition system according to an embodiment is operatively connected to an image acquisition unit that repeatedly acquires an image of an object at a specified angle or position and the image acquisition unit to receive an image of an object acquired by the image acquisition unit. And a processor, wherein the processor calculates a gradient value for an edge of the object from the acquired image of the object, and generates a template of the object based on the gradient value. Can be configured to

An object recognition method according to an embodiment includes an image acquisition step of repeatedly acquiring an image of an object at a specified angle or position through an image acquisition unit, based on the acquired image of the object, an edge of the object A gradient calculation step of calculating a gradient value for, and a template generation step of generating a template of the object based on the gradient value may be included.

An object recognition apparatus and method according to various embodiments of the present disclosure may generate a robust template that minimizes the influence of an object movement or environmental change (eg, light change, shadow), and as a result, an object You can increase the precision of recognition.

1A is a perspective view illustrating a process of acquiring an image of an article from a specific angle according to an exemplary embodiment. 1B is a perspective view illustrating a process of acquiring an image of an article from an angle different from that of FIG. 1A.

2 is a diagram illustrating a configuration of an object recognition system according to an exemplary embodiment.

3 is a flowchart illustrating a process of recognizing an object according to an exemplary embodiment.

4A is a diagram illustrating a process of repeatedly obtaining an image of an object at a specified angle or position. FIG. 4B is a diagram illustrating an edge of an object appearing in the image obtained in FIG. 4A. FIG. 4C is a diagram illustrating a template for an object generated based on the image acquired in FIG. 4A.

5 is a flowchart illustrating a process of generating a template according to an embodiment.

6A is a diagram illustrating a process of quantizing a gradient value of an edge of an object. 6B is a diagram in which gradient values corresponding to respective edge regions of an object are displayed as quantized gradient values. 6C is a diagram showing a histogram generated by accumulating gradient values of the edges of a quantized object. 6D is a diagram illustrating a template generated through the template generation process of FIG. 5.

7 is a flowchart illustrating a process of recognizing an object according to an exemplary embodiment.

8 is a diagram illustrating a process of matching an object with a template generated according to an exemplary embodiment.

Various embodiments of the present document and terms used therein are not intended to limit the technical features described in this document to specific embodiments, and should be understood to include various modifications, equivalents, or substitutes of the corresponding embodiments. In connection with the description of the drawings, similar reference numerals may be used for similar or related components. The singular form of a noun corresponding to an item may include one or more of the above items unless clearly indicated otherwise in a related context. In this document, “A or B”, “at least one of A and B”, “at least one of A or B,” “A, B or C,” “at least one of A, B and C,” and “A Each of the phrases such as "at least one of, B, or C" may include any one of the items listed together in the corresponding one of the phrases, or all possible combinations thereof. Terms such as "first", "second", or "first" or "second" may be used simply to distinguish the component from other Order) is not limited. Some (eg, a first) component is referred to as “coupled” or “connected” to another (eg, a second) component, with or without the terms “functionally” or “communicatively”. When mentioned, it means that any of the above components can be connected to the other components directly (eg by wire), wirelessly, or via a third component.

1A, 1B, and 2, an object recognition system 100 according to an embodiment includes an image acquisition unit 110 and a processor 200 operatively connected to the image acquisition unit 110. Including, it is possible to recognize the object 10 or to determine the location of the object 10.

The image acquisition unit 110 according to an embodiment may acquire an image of the object 10 at a specified angle or position. As an example, the image acquisition unit 110 may be a camera, but is not limited thereto. According to another example, the image acquisition unit 110 may be a 3D scanner.

According to an embodiment, the image acquisition unit 110 may be attached to a robot arm 120. The image acquisition unit 110 moves together according to the motion of the robot arm 120 and may acquire an image of the object 10 at various angles or positions.

According to an embodiment, the robot arm 120 fixes the robot arm 120 to a specific position (eg, the ground, a frame, etc.), and a fixing part 121 and the fixing part capable of rotating based on a fixed position. It may include a plurality of link portions 122 rotatably coupled to 121. According to an embodiment (for example, see FIGS. 1A and 1B ), the plurality of link portions 122 include a first link portion 122a rotatably coupled to the fixing portion 121, a first link portion ( 122a) and a second link portion 122b rotatably coupled to the second link portion 122b, and a third link portion 122c rotatably coupled to the second link portion 122b. Depending on the degree of rotation of the first link portion 122a, the second link portion 122b, and the third link portion 122c, the pose of the robot arm 120 may be variously changed. However, it is only one embodiment of the present disclosure that the plurality of link units 122 include the first link unit 122a, the second link unit 122b, and the third link unit 122c. Accordingly, the plurality of link units 122 may include two link units or four link units.

The image acquisition unit 110 is attached to one end of the third link unit 122c of the robot arm 120 as shown in FIG. 1A or 1B, and various angles according to a change in the pose of the robot arm 120 Alternatively, an image of the object 10 may be obtained from the location. In the present disclosure, the pose of the robot arm 120 refers to the shape of the robot arm according to the rotation of the fixing part 121 and the plurality of link parts 122, and is used in the same meaning hereinafter.

As an example (e.g., FIG. 1A), when the robot arm 120 has a first pose by the first link unit 122a, the second link unit 122b, and the third link unit 122c, an image is acquired. The unit 110 may acquire an image of the object 10 photographed from the upper left. In another example (for example, FIG. 1B), the robot arm 120 is different from the first pose by the rotation of the first link part 122a, the second link part 122b, and the third link part 122c. In the case of having the second pose, the image acquisition unit 110 may acquire an image of the object 10 photographed from the upper right. In this case, the position at which the image acquisition unit 110 is attached to the robot arm 120 is not limited to the illustrated embodiment, and according to the embodiment, the image acquisition unit 110 It may be attached to the center or attached to the side of the third link part 122c.

The processor 200 according to an embodiment is operatively connected to the image acquisition unit 110 and/or the robot arm 120 to which the image acquisition unit 110 is attached, and the image acquisition unit 110 ), or control the motion of the robot arm 120. As an example, the processor 200 may rotate the fixing part 121, and as another example, the first link part 122a, the second link part 122b, and the third link part 122c are driven. By controlling, the rotation angle between the first link part 122a, the second link part 122b, and the third link part 122c may be adjusted.

As described above, the processor 200 may change the pose of the robot arm 120 by changing the rotation angle of the plurality of link units 122 (eg, 122a, 122b, 122c). The movement of the robot arm 120 may be controlled by continuously changing the pose of 120 and/or the rotation angle of the driving unit 121. Accordingly, the processor 200 may allow the image acquisition unit 110 to acquire images of the object 10 from various angles or positions while repeatedly moving the robot arm 120 with a specified movement.

The processor 200 according to an embodiment generates a template for the object 10 based on the image for the object 10 acquired by the image acquisition unit 110, and uses the generated template. Thus, an object of interest can be recognized among a plurality of objects. However, a detailed description of a process of generating a template in the processor 200 and recognizing an object of interest will be described later.

In describing the object recognition process of FIG. 3, reference will be made to the drawings of FIGS. 4A to 4C.

3 and 4A, in step 301, an object 10 (eg, an object of FIGS. 1A and 1B) through an image acquisition unit (eg, the image acquisition unit 110 of FIGS. 1A, 1B, and 2) 10)) can be acquired.

According to an embodiment, the image acquisition unit is attached to at least one area of a robot arm capable of rotational movement (for example, the robot arm 120 of FIGS. 1A and 1B), and an object ( 10) can be obtained.

According to an embodiment, the image acquisition unit is attached to at least one area (eg, one end of the robot arm) of the robot arm (eg, the robot arm 120 of FIGS. 1A and 1B ), and the image acquisition unit The angle or position at which an image of the object 10 is acquired (eg, point A in FIG. 4A) may vary.

The processor (e.g., the processor 200 of FIG. 2) is operatively connected to the image acquisition unit and/or the robot arm to receive an image of the object 10 from the image acquisition unit or to control the movement of the robot arm. have. By controlling the movement of the robot arm, the processor may adjust an angle or position at which the image acquisition unit acquires an image of the object 10. According to an embodiment, the processor may change the robot arm to various poses by driving the robot arm to have a specified motion. Accordingly, the image acquisition unit 10) can be obtained.

According to another embodiment (e.g., see FIG. 4A), the processor may control the robot arm to repeat a designated movement, so that the image acquisition unit acquires images of a plurality of objects 10 at various angles or positions. have. In the process of repeating the movement of the robot arm, the object 10 may move, a minute movement of the robot arm may occur, or the intensity and/or direction of light applied to the object 10 may be changed. The images of the plurality of objects 10 acquired through the above-described process include changes in light generated in the process of repeating the movement of the robot arm, a minute angle change of the object 10, and/or a minute movement of the object 10. Changes in the environment can be reflected.

Referring to FIG. 4B, in step 302, the processor according to an embodiment extracts the edge 11 of the object 10 from the image of the object 10 obtained in step 301, and You can calculate gradient values. In the present disclosure, a gradient value refers to a value for a gradient direction (eg, degrees (°)), and is used in the same meaning hereinafter.

According to an embodiment, the processor extracts the edge 11 of the object 10 using a sobel operator from the image of the object 10 received from the image acquisition unit, and extracts the edge 11 You can calculate the gradient value for each area of. In the present disclosure, the Sobel operator refers to a method used when detecting an edge from an image and detecting the size and direction of the edge.

As the image acquisition unit acquires images for the plurality of objects 10 at various angles or positions in step 301, the processor extracts the edge 11 of the object 10 from each acquired image, and extracts the edge. You can calculate the gradient value for each area of.

Referring to FIG. 4C, in step 303, the processor according to an embodiment of the present invention provides a template 20 of the object 10 based on the gradient value for the edge 11 of the object 10 calculated in step 302. ) Can be formed. A plurality of templates 20 may be generated for one object 10 according to a position and/or angle at which the image acquisition unit captures an image of an object.

Since the template 20 may be differently generated according to the position and/or angle at which the image acquisition unit acquires an image of the object 10, the template 20 not only provides information on the edge shape of the object 10, but also The location information of the object 10 may also be included. In this case, the location information of the object 10 may be obtained based on a pose of the robot arm corresponding to a specific location and/or angle of the image acquisition unit. The location information of the object 10 may be, for example, 6D information (x, y, z, rx, ry, rz) based on the location of the robot arm, but is not limited thereto. The processor may generate a histogram using the gradient value of the edge 11 of the object 10 calculated in step 302, and generate the template 20 of the object 10 using the generated histogram. , A detailed description of this will be described later.

In step 304, the processor according to an embodiment may recognize an object of interest among a plurality of objects by using the template 20 generated through the step 303. The processor may recognize an object corresponding to the template 20 among a plurality of objects by using a template matching method of matching a plurality of objects with the template 20 generated through the process 303. .

As described above, since the template 20 may include location information of the object 10, the processor can recognize not only the object of interest, but also the current location information of the object of interest. As an example, the processor may determine information such as how far the object of interest is from the robot arm and in which direction it is rotated.

After recognizing the object of interest, the processor can control the operation of the robot arm to pick up the object of interest and use it for assembly, inspection, and/or loading, so that the manufacturing process can be smartened through the above-described processes 301 to 304. I can plan.

Hereinafter, a process of generating the template 20 for the object 10 will be described in detail with reference to FIGS. 5 and 6A to 6C.

In describing the process of generating the template of FIG. 5, reference will be made to the drawings of FIGS. 6A to 6D.

5, 6A, and 6B, in step 501, the processor (eg, the processor 200 of FIG. 2) according to an embodiment is an image acquisition unit (eg, the image acquisition unit 110 of FIG. 2). The gradient value for the edge of the object 10 may be calculated from the images of the plurality of objects 10 obtained from. Since the process of calculating the gradient value for the edge of the object 10 by the processor is the same as or similar to the process 302 of FIG. 3 described above, a redundant description will be omitted.

The processor may quantize the gradient value for the edge of the object 10 calculated through the above-described process. According to an embodiment, the processor classifies a gradient value with respect to the edge of the object 10 into 8 bits and quantizes it. As an example (eg, see FIG. 6A), the processor may quantize the gradient value to 0 when the gradient value for the edge of the object 10 is 0° to 22.5°, and the edge of the object 10 When the relative gradient value is 22.5° to 45°, the gradient value can be quantized to 1.

Similarly, when the gradient value for the edge of the object 10 is 180° to 202.5°, the processor can quantize the gradient value back to 0, and the gradient value for the edge of the object 10 is 202.5°. In the case of to 225°, the gradient value may be quantized to 1.

However, the process of quantizing the gradient value for the edge of the object 10 through the processor is not limited to the embodiment shown in FIG. 6A, and the processor is Values can also be quantized by classifying them into 16 bits or 32 bits.

According to an embodiment (for example, FIG. 6B ), the processor may quantize a gradient value for each area of the edge of the object 10 and represent the entire area of the edge of the object 10 as a quantized gradient value. As an example, when the gradient value for one edge area of the object 10 is 25°, the corresponding edge area may be expressed as '1', and the gradient value for the other edge area of the object 10 is 120°. In this case, the corresponding edge area may be represented by '5'.

Referring to FIG. 6C, in step 502, the processor according to an embodiment may generate a histogram by accumulating a gradient value for the edge region of the object 10 quantized in step 501.

The processor according to an embodiment may quantize a gradient value for the edge of the object for each image of a plurality of objects 10 acquired by the image acquisition unit, and calculate the gradient value for the edge of the quantized object for each image. By accumulating, it is possible to generate a histogram having a quantized gradient value as a bin for each edge area of the object.

As an example, a gradient value for a first edge area of an object appearing in a first image among images of a plurality of objects is '1', and a first edge area of an object appearing in a second image among images of a plurality of objects When the gradient value of is also '1', two bins may be accumulated in the quantized gradient value '1'.

As another example, the gradient value for the first edge area of the object appearing in the first image among images of a plurality of objects is '4', and the first edge area of the object appearing in the second image among the images of a plurality of objects If the gradient value of is also '5', one bin may be accumulated in each of the quantized gradient values '4' and '5'.

Referring to FIG. 6D, in step 503, the processor according to an embodiment generates a template 20 (eg, template 20 in FIG. 4C) for an object 10 using the histogram generated in step 502. can do.

As an example (e.g., see FIG. 6C), in the histogram of one edge region of the object 10 generated through the process 502, a bin accumulated in a specific quantized gradient value is specified. ) Or more, the edge region may be stored as having the specific quantized gradient value.

As an example (e.g., see FIG. 6C), when the bin accumulated in the quantized gradient value '5' in the histogram of one edge region of the object 10 is equal to or greater than a specified value (for example, 120), the one edge region is quantized. It can be stored as having a gradient value of '5'.

According to an embodiment, the processor stores the gradient value when the bin accumulated in the quantized gradient value is greater than or equal to the specified value, and does not store the gradient value when the bin accumulated in the quantized gradient value is smaller than the specified value. , It is possible to generate the template 20 by leaving only gradient values with many accumulated beans.

According to another embodiment, when the bin accumulated in the quantized gradient value is equal to or greater than the specified value, the processor stores the gradient value by giving a high weight, and conversely, when the bin accumulated in the quantized gradient value is smaller than the specified value. The template 20 can be generated by assigning a relatively low weight to the gradient value and storing it, so that the weight is differently set according to the accumulated bins.

That is, the processor generates the template 20 from the image of the object 10 repeatedly acquired at various angles or positions, so that the template 20 reflecting environmental changes such as changes in light and fine movement of the object 10 Can be created. As a result, the object recognition system according to an embodiment (for example, the object recognition system 100 of FIG. 2) minimizes the influence of changes in the surrounding environment through the template 20, while strengthening the object of interest 10 ( robust).

Hereinafter, an object recognition process according to an embodiment will be described as a whole with reference to FIG. 7.

Each of the processes of FIG. 7 is the same as or similar to processes 301 to 304 of FIG. 3 and/or processes 501 to 503 of FIG. 5, and thus, redundant descriptions will be omitted hereinafter.

Referring to FIG. 7, in step 701, a processor (eg, the processor 200 of FIG. 2) of an object recognition system (eg, the object recognition system 100 of FIG. 2) according to an embodiment is a robot arm (eg: The movement of the robot arm 120 of FIGS. 1A and 1B and/or the number of times the movement is repeated may be set. The processor sets the movement of the robot arm in various ways, so that the image acquisition unit (eg, the image acquisition unit 110 of FIGS. 1A and 1B) attached to at least one area of the robot arm receives an image of the object 10. You can adjust the angle or position to acquire.

In step 702, the processor according to an embodiment may control the driving of the robot arm based on the number of movements and/or repetitions set in step 701, and the image acquisition unit is attached to the robot arm and moves set together with the robot arm. While moving according to (motion), images of objects can be acquired from various angles and/or positions. Images of objects at various angles and/or positions acquired through the image acquisition unit may be transmitted to the processor. As will be described later, the processor may generate a template (eg, the template 20 of FIG. 4C) based on the image of the received object, but a detailed description thereof will be described later.

In step 703, the processor may determine whether the number of times the robot arm repeats the specified movement matches the number of repetitions set in step 701.

If it is determined that the number of times the robot arm repeats the specified movement in step 703 matches the number of repetitions set in step 701, the processor according to an embodiment in step 704 may perform images of a plurality of objects acquired in step 702. From, we can calculate the gradient value for the edge of the object.

Conversely, if it is determined that the number of times the robot arm repeats the specified movement in step 703 matches the number of repetitions set in step 701, the processor repeats step 702, and the image acquisition unit displays images for a plurality of objects. Can be obtained.

In step 704, the processor according to an embodiment may extract an edge of an object from an image of a plurality of objects using a Sobel operator, and calculate a gradient value for each edge region. The processor may quantize the calculated gradient value for the edge of the object. As an example, the processor may quantize a gradient value with respect to an edge of an object into 8 bits, but is not limited thereto.

In step 705, the processor according to an embodiment may generate a histogram having the quantized gradient value as a bin by accumulating the gradient value of the quantized object.

According to an embodiment, the processor may quantize a gradient value for the edge of the object for each image of a plurality of objects acquired in step 702, and accumulate the gradient value for the edge of the quantized object for each image, A histogram can be generated for each area at the edge of an object.

In step 706, the processor according to an embodiment determines whether or not the bin accumulated in the specific quantized gradient value in the histogram for one edge region of the object generated in step 705 is equal to or greater than a specified value. I can.

In step 706, when it is determined that the bin accumulated in the specific gradient value in the histogram for the specific edge area of the object is equal to or greater than the specified value, the processor proceeds through step 707, and the specific edge area of the object is the specific gradient value. Can be stored as having Conversely, in step 706, if it is determined that the bin accumulated in the specific gradient value in the histogram for the specific edge area of the object is smaller than the specified value, the processor does not store the gradient value for the specific edge area of the object as a specific gradient value. May not.

In step 708, the processor according to an embodiment may generate a template corresponding to the object based on a gradient value stored for each edge area of the object in step 707. That is, in step 708, the processor may generate a template using only the gradient values that remain strong despite changes in the surrounding environment (eg, changes in light intensity, fine movement of objects, etc.).

In step 709, the processor according to an embodiment may recognize an object of interest among the plurality of objects by matching the template generated in step 708 with a plurality of objects. The template includes location information of the object (eg, 6D location information), and the processor can recognize the object of interest and also the current location information of the object of interest.

Referring to FIG. 8, an object recognition system (eg, the object recognition system 100 of FIG. 2) according to an embodiment generates a template (eg, the template 20 of FIG. 4C), and generates a template and an object. An object can be recognized using a matching template matching method. According to an embodiment, the template may be generated through processes 501 to 503 of FIG. 5, and redundant descriptions will be omitted.

The template may be generated based on images of a plurality of objects, and the surrounding environment (eg, change in light intensity, fine movement of objects, etc.) may be reflected in images of a plurality of objects. The template generated based on the image for the object can minimize the influence of the surrounding environment in the process of recognizing the object.

According to an embodiment, the object recognition system may accumulate (eg, refer to FIG. 6C) gradient values for edges of an object acquired from a plurality of images. As an example, the object recognition system may generate a template by comparing an accumulated gradient value with a specified threshold, leaving only a gradient value in which the accumulated gradient value is equal to or greater than a specified value.

The object recognition system can reduce the capacity of template data by creating a template by leaving only a gradient value in which an accumulated gradient value is equal to or greater than a specified value.

However, the method of generating a template of the object recognition system is not limited thereto, and according to an embodiment, the object recognition system may generate a template by assigning a different weight according to the accumulated gradient value.

The template generated through the above-described process in the object recognition system may have a strong outline, unlike a conventional template generated by simply accumulating the edges of objects. As a result, the object recognition system according to an exemplary embodiment can robustly recognize an object even when the direction of light changes or some deformation of the object occurs during the object recognition process using the template.

An object recognition system according to an embodiment of the present disclosure, in the object recognition system, is an image acquisition unit that repeatedly acquires an image of an object at a specified angle or position, and an image acquisition unit that is operatively connected to the image acquisition unit, And a processor that receives an image of an object acquired by an acquisition unit, wherein the processor calculates a gradient value for an edge of the object from the acquired image of the object, and based on the gradient value , It may be configured to generate a template of the object.

According to an embodiment, the image acquisition unit may be a camera.

According to an embodiment, the image acquisition unit may be attached to at least one area of a robot arm capable of rotational movement.

According to an embodiment, the robot arm includes a fixing part, a first link part rotatably coupled to the fixing part, a second link part rotatably connected to the first link part, and the second link part. It may include a third link that is rotatably coupled with.

According to an embodiment, the processor may be configured to control a pose or an angle of the robot arm.

According to an embodiment, the processor may be configured to control the driving of the robot arm so that the robot arm repeats a designated motion.

According to an embodiment, the processor may be configured to quantize a gradient value for an edge of the calculated object.

According to an embodiment, the processor may be configured to accumulate the quantized gradient values and generate a histogram having the quantized gradient values as bins.

According to an embodiment, when a gradient value for an edge of an object accumulated in the histogram is equal to or greater than a specified value, the processor may be configured to store the edge of the object as having the gradient value.

According to an embodiment, the processor may be configured to generate a template for the object based on a gradient value corresponding to each area of the edge of the stored object.

According to an embodiment, the processor may be configured to recognize the object by matching the generated template with at least one object appearing in an image acquired through the image acquisition unit.

An object recognition method according to an embodiment of the present disclosure includes, in an object recognition method, an image acquisition step of repeatedly obtaining an image of an object at a specified angle or position through an image acquisition unit, based on the acquired image of the object. Thus, a gradient calculation step of calculating a gradient value for an edge of the object, and a template generation step of generating a template of the object based on the gradient value may be included.

According to an embodiment, in the image acquisition step, as the robot arm is repeatedly driven by a specified motion by a processor, images of a plurality of objects may be acquired at a specified angle or position through the image acquisition unit. have.

According to an embodiment, prior to the step of acquiring the image, a setting step of setting a movement of the robot arm or a number of times to repeat the movement may be further included.

According to an embodiment, a quantization step of quantizing the calculated gradient value of the edge of the object may be further included.

According to an embodiment, a histogram generation step of generating a histogram having the quantized gradient values as bins by accumulating the quantized gradient values may be further included.

According to an embodiment, when the gradient value for the edge of the object accumulated in the histogram is equal to or greater than a specified value, the step of storing the gradient value as storing the edge of the object as having the gradient value may be further included.

According to an embodiment, in the generating of the template, a template for the object may be generated based on a gradient value corresponding to each area of an edge of the stored object.

According to an embodiment, an object recognition step of recognizing the object by matching the generated template with at least one object appearing in an image acquired through the image acquisition unit may be further included.

In the above-described specific embodiments of the present disclosure, components included in the disclosure are expressed in the singular or plural according to the presented specific embodiments. However, the singular or plural expression is selected appropriately for the situation presented for convenience of description, and the present disclosure is not limited to the singular or plural constituent elements, and even constituent elements expressed in plural are composed of the singular or Even the expressed constituent elements may be composed of pluralities.

Meanwhile, although specific embodiments have been described in the detailed description of the present disclosure, various modifications are possible without departing from the scope of the present disclosure. Therefore, the scope of the present disclosure is limited to the described embodiments and should not be defined, and should be determined by the scope of the claims and equivalents as well as the scope of the claims to be described later.

Claims

In the object recognition system,

An image acquisition unit that repeatedly acquires an image of an object at a specified angle or position; And

Includes; a processor that is operatively connected to the image acquisition unit and receives an image of an object acquired by the image acquisition unit,

The processor,

Calculate a gradient value for an edge of the object from the acquired image of the object,

Based on the gradient value, configured to generate a template of the object.
The method of claim 1,

The image acquisition unit is an object recognition system that is a camera.
The method of claim 1,

The image acquisition unit is attached to at least one area of a robot arm capable of rotational movement.
The method of claim 3,

The robot arm,

Fixed part;

A first link portion rotatably coupled to the fixing portion;

A second link portion rotatably coupled to the first link portion; And

Containing, an object recognition system; a third link unit rotatably coupled to the second link unit.
The method of claim 3,

The processor,

An object recognition system configured to control a pose or angle of the robotic arm.
The method of claim 5,

The processor,

The object recognition system, configured to control the driving of the robotic arm so that the robotic arm can repeat a specified motion.
The method of claim 3,

The processor,

An object recognition system configured to quantize the calculated gradient value for the edge of the object.
The method of claim 7,

The processor,

And accumulating the quantized gradient values to generate a histogram having the quantized gradient values as bins.
The method of claim 8,

The processor,

The object recognition system, configured to store the edge of the object as having the gradient value when the gradient value for the edge of the object accumulated in the histogram is equal to or greater than a specified value.
The method of claim 8,

The processor,

The object recognition system, configured to generate a template for the object based on a gradient value corresponding to each area of the edge of the stored object.
The method of claim 1,

The processor,

The object recognition system, configured to recognize the object by matching the generated template with at least one object appearing in the image acquired through the image acquisition unit.
In the object recognition method,

An image acquisition step of repeatedly acquiring an image of an object at a specified angle or position through an image acquisition unit;

A gradient calculation step of calculating a gradient value for an edge of the object based on the acquired image of the object; And

Including, object recognition method comprising; a template generation step of generating a template (template) of the object based on the gradient value.
The method of claim 12,

The image acquisition unit is attached to at least one area of a robot arm capable of rotational movement.
The method of claim 13,

The image acquisition step,

As the robot arm is repeatedly driven by a specified motion by a processor, images of a plurality of objects are acquired at a specified angle or position through the image acquisition unit.
The method of claim 13,

Prior to the image acquisition step, a setting step of setting a movement of the robot arm or a number of times to repeat the movement; further comprising, object recognition method.