KR20220169445A - Electronic device and method for detecting an object in an image - Google Patents

Abstract

An electronic device and a method for detecting an object in an image are provided. An object detection method for detecting an object in an image using an artificial intelligence model according to an embodiment of the present disclosure comprises: a step of obtaining an image including an object including a first object and a second object which is part of the first object; a step of generating a plurality of first bounding boxes corresponding to the first object included in the image; a step of creating a first sub-bounding box corresponding to the second object included in the image; a step of obtaining a first overlap score related to the degree of overlap between the first sub-bounding box and the first bounding boxes; a step of removing at least one first bounding box having the first overlap score with the first sub-bounding box exceeding a first threshold value, among a plurality of first bounding boxes; and a step of detecting the first object through the first bounding box which is not removed among a plurality of first bounding boxes. Accordingly, the present invention provides an electronic device and method for more accurately recognizing an object using the relationship between an object and a part of the object.

Description

Electronic device and method for detecting an object in an image {ELECTRONIC DEVICE AND METHOD FOR DETECTING AN OBJECT IN AN IMAGE}

Various embodiments of the present disclosure relate to an electronic device and method for detecting an object in an image.

An image recognition or object detection system that detects an object within an image may detect a single object or multiple objects from a digital image or video frame. Here, object detection may mean estimating the location and size of an object in an image in the form of a bounding box and classifying a specific object in a given image.

However, when using a lightweight deep learning model for fast processing speed, there are often many cases in which objects in a given image are misclassified. For example, in an image including a person, there was inconvenience in misclassifying a part of a person's body as a person by recognizing a hand, face, etc. as the whole of the person's body. .

Various disclosed embodiments may provide an electronic device and method for more accurately recognizing an object by using a relationship between an object and a part of the object.

According to an embodiment of the present disclosure, an object detection method for detecting an object in an image using an artificial intelligence model includes an object including a first object and a second object that is a part of the first object. Acquiring an image; generating a plurality of first bounding boxes corresponding to a first object included in the image; generating a first sub-bounding box corresponding to a second object included in the image; Obtaining a first overlap score related to the degree of overlap between the bounding boxes and the first bounding boxes, wherein, among a plurality of first bounding boxes, a first overlap score with a first sub-bounding box exceeds a first threshold value. The method includes removing a first bounding box of and detecting a first object through a first bounding box that is not removed among a plurality of first bounding boxes.

In one embodiment of the present disclosure, the object further includes a third object that is a part of the first object, and generating a second sub-bounding box corresponding to the third object included in the image; and obtaining a second overlap score related to the degree of overlap between first bounding boxes; and, among a plurality of first bounding boxes, a second overlap score with a second sub-bounding box exceeds a second threshold value. 1 The step of removing the bounding box may be further included.

In one embodiment of the present disclosure, the plurality of first bounding boxes include a first reference bounding box, and the degree of overlap between one of the plurality of first bounding boxes excluding the first reference bounding box and the first reference bounding box obtaining a third overlap score for the first bounding box; removing at least one first bounding box whose third overlap score with the first reference bounding box exceeds a third threshold value, among the plurality of first bounding boxes; can include

In an embodiment of the present disclosure, a plurality of first bounding boxes may be labeled as a first class specifying a first object, and a first sub-bounding box may be labeled as a second class specifying a second object. .

In one embodiment of the present disclosure, the first class may mean a person class, and the second class may mean at least one of a face class, an ear class, and a hand class. .

In addition, an object detection method for detecting an object in an image using an artificial intelligence model according to an embodiment of the present disclosure obtains an image including an object including a first object and a second object that is a part of the first object. generating a plurality of first bounding boxes corresponding to a first object included in the image, generating a first sub-bounding box corresponding to a second object included in the image, an area inside the first bounding box Obtaining a reliability of a probability that an object included in corresponds to a first object, obtaining a reference value corresponding to the first bounding box according to whether a first sub-bounding box is disposed in the first bounding box, removing at least one first bounding box, the reliability of which does not exceed a reference value corresponding to the at least one first bounding box, and a first among a plurality of first bounding boxes not removed. 1 detecting the first object through the bounding box.

In one embodiment of the present disclosure, the reference value includes a first reference value and a second reference value, and obtaining the reference value includes a first sub-bounding box corresponding to at least one first bounding box including a first sub-bounding box therein. The method may include obtaining a reference value and obtaining a second reference value corresponding to at least one first bounding box that does not contain a first sub-bounding box, wherein the first reference value may be lower than the second reference value. .

In one embodiment of the present disclosure, a plurality of first bounding boxes may be labeled as a first class specifying a first object, and a first sub-bounding box may be labeled as a second class specifying a second object. .

In one embodiment of the present disclosure, the first class may mean a person class, and the second class may mean at least one of a face class, an ear class, and a hand class. .

In one embodiment of the present disclosure, the object further includes a third object that is a part of the first object, and generating a second sub-bounding box corresponding to the third object included in the image, the second sub-bounding box inside the Obtaining a third reference value corresponding to at least one first bounding box included in ; Obtaining a fourth reference value corresponding to at least one first bounding box not included in a second sub-bounding box; The method further includes removing at least one first bounding box whose reliability of the at least one first bounding box does not exceed one of a third reference value and a fourth reference value corresponding to the at least one first bounding box; The third reference value may be lower than the fourth reference value.

In addition, an electronic device for detecting an object in an image using an artificial intelligence model according to an embodiment of the present disclosure includes a memory for storing one or more instructions, and at least one processor for executing the one or more instructions stored in the memory. and, by executing one or more instructions, the at least one processor acquires an image including an object including a first object and a second object that is part of the first object, and corresponds to the first object included in the image. A plurality of first bounding boxes are generated, a first sub-bounding box corresponding to a second object included in the image is generated, and a first overlap score regarding the degree of overlap between the first sub-bounding boxes is obtained. and removing at least one first bounding box whose first overlap score with the first sub-bounding box exceeds a first threshold value from among the plurality of first bounding boxes, and which is not removed from among the plurality of first bounding boxes. The first object is detected through the first bounding box that does not exist.

In an embodiment of the present disclosure, the object further includes a third object that is a part of the first object, and the at least one processor executes one or more instructions to generate a second sub-boundary corresponding to the third object included in the image. A box is generated, a second overlap score related to an overlapping degree between the second sub-bounding box and the first bounding box is obtained, and a second overlap score with the second sub-bounding box among the plurality of first bounding boxes is the first. At least one first bounding box exceeding 2 thresholds may be removed.

In an embodiment of the present disclosure, the plurality of first bounding boxes include first reference bounding boxes, and at least one processor executes one or more instructions, thereby executing the plurality of first bounding boxes excluding the first reference bounding boxes. A third overlap score related to an overlap between one of the plurality of first bounding boxes and the first reference bounding box is obtained, and a third overlap score with the first reference bounding box among the plurality of first bounding boxes exceeds a third threshold value. One first bounding box may be removed.

In an embodiment of the present disclosure, a plurality of first bounding boxes may be labeled as a first class specifying a first object, and a first sub-bounding box may be labeled as a second class specifying a second object. .

In one embodiment of the present disclosure, the first class may mean a person class, and the second class may mean at least one of a face class, an ear class, and a hand class. .

In addition, an electronic device for detecting an object in an image using an artificial intelligence model according to an embodiment of the present disclosure includes a memory for storing one or more instructions, and at least one processor for executing the one or more instructions stored in the memory. and, by executing one or more instructions, the at least one processor acquires an image including an object including a first object and a second object that is part of the first object, and corresponds to the first object included in the image. A plurality of first bounding boxes are generated, a first sub-bounding box corresponding to a second object included in the image is created, and confidence in the probability that an object included in the inner area of the first bounding box coincides with the first object Obtain a reference value corresponding to the first bounding box according to whether the first sub-bounding box is disposed in the first bounding box, and determine whether the reliability of the at least one first bounding box corresponds to at least one first bounding box. At least one first bounding box that does not exceed a reference value corresponding to is removed, and a first object is detected through a first bounding box that is not removed among a plurality of first bounding boxes.

In an embodiment of the present disclosure, the reference value includes a first reference value and a second reference value, and at least one processor executes one or more instructions to determine at least one first bounding box including a first subbounding box therein. A first reference value corresponding to a box is obtained, and a second reference value corresponding to at least one first bounding box that does not contain a first sub-bounding box is obtained, and the first reference value may be lower than the second reference value. .

In one embodiment of the present disclosure, a plurality of first bounding boxes may be labeled as a first class specifying a first object, and a first sub-bounding box may be labeled as a second class specifying a second object. .

In one embodiment of the present disclosure, the first class may mean a person class, and the second class may mean at least one of a face class, an ear class, and a hand class. .

In an embodiment of the present disclosure, the object further includes a third object that is a part of the first object, and the at least one processor executes one or more instructions to generate a second sub-boundary corresponding to the third object included in the image. A box is generated, a third reference value corresponding to at least one first bounding box in which a second sub-bounding box is included is obtained, and a third reference value corresponding to at least one first bounding box in which the second sub-bounding box is not included is obtained. A fourth reference value corresponding to is obtained, and the reliability of the at least one first bounding box does not exceed one of the third reference value and the fourth reference value corresponding to the at least one first bounding box. , and the third reference value may be lower than the fourth reference value.

1 is a schematic block diagram of an electronic device according to an embodiment of the present disclosure.
2 is a conceptual diagram for explaining an operation of detecting an object according to an embodiment of the present disclosure.
3 is a flowchart illustrating a method of detecting an object according to an embodiment of the present disclosure.
4 is a diagram for explaining an operation of detecting an object according to an embodiment of the present disclosure.
5 is a flowchart illustrating a method of detecting an object according to an embodiment of the present disclosure.
6 is a diagram for explaining an operation of detecting an object according to an embodiment of the present disclosure.
7 is a flowchart illustrating a method of detecting an object according to an embodiment of the present disclosure.
8 is a diagram for explaining an operation of detecting an object according to an embodiment of the present disclosure.
9 and 10 are flowcharts for describing a method of detecting an object according to an embodiment of the present disclosure.
11 is a diagram for explaining an operation of detecting an object according to an embodiment of the present disclosure.
12 and 13 are flowcharts for describing a method of detecting an object according to an embodiment of the present disclosure.
14 is a diagram for explaining an operation of detecting an object according to an embodiment of the present disclosure.

Hereinafter, various embodiments of the present disclosure will be described with reference to the accompanying drawings. However, this is not intended to limit the technology described in this disclosure to specific embodiments, and it should be understood that it includes various modifications, equivalents, and/or alternatives of the embodiments of this disclosure. In connection with the description of the drawings, like reference numerals may be used for like elements.

In the present disclosure, expressions such as “has,” “can have,” “includes,” or “can include” indicate the presence of a corresponding feature (eg, numerical value, function, operation, or component such as a part). , which does not preclude the existence of additional features.

In this disclosure, expressions such as “A or B,” “at least one of A and/and B,” or “one or more of A or/and B” may include all possible combinations of the items listed together. . For example, “A or B,” “at least one of A and B,” or “at least one of A or B” (1) includes at least one A, (2) includes at least one B, Or (3) may refer to all cases including at least one A and at least one B.

Expressions such as “first,” “second,” “first,” or “second,” as used in the present disclosure may modify various components regardless of order and/or importance, and refer to one component as It is used only to distinguish it from other components and does not limit the corresponding components. For example, a first user device and a second user device may represent different user devices regardless of order or importance. For example, a first element may be named a second element without departing from the scope of rights described in the present disclosure, and similarly, the second element may also be renamed to the first element.

Terms such as "module", "unit", and "part" used in the present disclosure are terms for referring to components that perform at least one function or operation, and these components are implemented in hardware or software. or may be implemented as a combination of hardware and software. In addition, a plurality of "modules", "units", "parts", etc. are integrated into at least one module or chip, except for cases where each of them needs to be implemented with separate specific hardware, so that at least one processor can be implemented as

A component (e.g., a first component) is "(operatively or communicatively) coupled with/to" another component (e.g., a second component); When referred to as "connected to", it should be understood that the certain component may be directly connected to the other component or connected through another component (eg, a third component). On the other hand, when an element (eg, a first element) is referred to as being “directly connected” or “directly connected” to another element (eg, a second element), the element and the above It may be understood that other components (eg, a third component) do not exist between the other components.

Terms used in the present disclosure are only used to describe a specific embodiment, and may not be intended to limit the scope of other embodiments. Singular expressions may include plural expressions unless the context clearly dictates otherwise. Terms used herein, including technical or scientific terms, may have the same meaning as commonly understood by one of ordinary skill in the art described in this disclosure. Among the terms used in the present disclosure, terms defined in general dictionaries may be interpreted as having the same or similar meanings as those in the context of the related art, and unless explicitly defined in the present disclosure, ideal or excessively formal meanings. not be interpreted as In some cases, even terms defined in the present disclosure cannot be interpreted to exclude embodiments of the present disclosure.

1 is a schematic block diagram of an electronic device according to an embodiment of the present disclosure.

Referring to FIG. 1 , an electronic device 100 may include a controller (processor) 110 and a memory 120 . However, it is not limited thereto. For example, the electronic device 100 may further include components or may not include some components to perform the functions of the electronic device 100 . For example, the electronic device 100 may further include a communication interface for establishing communication with an external device.

The processor 110 (or control unit) may be composed of one or a plurality of processors. In addition, the processor 110 stores a non-volatile memory including a ROM in which a control program for controlling the electronic device 100 is stored and signals or data input from the outside of the electronic device 100, or the electronic device 100 ) may include volatile memory including RAM used as a storage area for various tasks performed in The RAM may be used as a storage area for control information received from the outside, operation information of the electronic device 100, or state information of the electronic device 100.

The processor 110 controls overall operation of the electronic device 100 and signal flow between internal components of the electronic device 100 and processes data. The processor 110 uses a power supply unit to control power supplied to internal components. For example, the processor 110 may control the memory 120 to execute a program stored in the memory 120 and to load or store necessary information.

According to an embodiment, the processor 110 obtains an image including an object including a first object and a second object that is a part of the first object using an artificial intelligence model by executing one or more instructions, and the image includes the image. A plurality of first bounding boxes corresponding to the included first object is generated, a first sub bounding box corresponding to the second object included in the image is created, and the degree of overlap between the first sub bounding boxes and the first bounding boxes is generated. obtaining a first overlap score for , removing at least one first bounding box having a first overlap score with a first sub-bounding box exceeding a first threshold value, among a plurality of first bounding boxes; The first object may be detected through a first bounding box that is not removed among the first bounding boxes.

According to an embodiment, the processor 110 obtains an image including an object including a first object and a second object that is a part of the first object by executing one or more instructions, and the first object included in the image A plurality of corresponding first bounding boxes are generated, a first sub-bounding box corresponding to a second object included in the image is generated, and a probability that an object included in an area inside the first bounding box coincides with the first object is determined. A reliability of the at least one first bounding box is obtained, and a reference value corresponding to the first bounding box is obtained according to whether the first sub-bounding box is disposed in the first bounding box, and the reliability of the at least one first bounding box is determined to be at least one first sub-bounding box. At least one first bounding box that does not exceed a reference value corresponding to the bounding box may be removed, and the first object may be detected through the first bounding box that is not removed among the plurality of first bounding boxes.

Functions related to artificial intelligence according to the present disclosure are operated through a processor and a memory. A processor may consist of one or a plurality of processors. In this case, the one or more processors may be a general-purpose processor such as a CPU, an AP, or a digital signal processor (DSP), a graphics-only processor such as a GPU or a vision processing unit (VPU), or an artificial intelligence-only processor such as an NPU. One or more processors control input data to be processed according to predefined operating rules or artificial intelligence models stored in a memory. Alternatively, when one or more processors are processors dedicated to artificial intelligence, the processors dedicated to artificial intelligence may be designed with a hardware structure specialized for processing a specific artificial intelligence model.

A predefined action rule or an artificial intelligence model is characterized in that it is created through learning. Here, being made through learning means that a basic artificial intelligence model is learned using a plurality of learning data by a learning algorithm, so that a predefined action rule or artificial intelligence model set to perform a desired characteristic (or purpose) is created. means burden. Such learning may be performed in the device itself in which artificial intelligence according to the present disclosure is performed, or through a separate server and/or system. Examples of learning algorithms include supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning, but are not limited to the above examples.

An artificial intelligence model may be composed of a plurality of neural network layers. Each of the plurality of neural network layers has a plurality of weight values, and a neural network operation is performed through an operation between an operation result of a previous layer and a plurality of weight values. A plurality of weights possessed by a plurality of neural network layers may be optimized by a learning result of an artificial intelligence model. For example, a plurality of weights may be updated so that a loss value or a cost value obtained from an artificial intelligence model is reduced or minimized during a learning process. The artificial neural network may include a deep neural network (DNN), for example, a Convolutional Neural Network (CNN), a Deep Neural Network (DNN), a Recurrent Neural Network (RNN), a Restricted Boltzmann Machine (RBM), A deep belief network (DBN), a bidirectional recurrent deep neural network (BRDNN), or deep Q-networks, but is not limited to the above examples.

According to an embodiment, the memory 120 is a flash memory type, a hard disk type, a multimedia card micro type, or a card type memory (eg SD or XD memory, etc.), RAM (RAM, Random Access Memory) SRAM (Static Random Access Memory), ROM (ROM, Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), PROM (Programmable Read-Only Memory) , a magnetic memory, a magnetic disk, and an optical disk may include at least one type of storage medium.

According to an embodiment, the memory 120 obtains an image including an object including a first object and a second object that is a part of the first object by using the artificial intelligence model, and the processor 110 obtains an image included in the image. Generating a plurality of first bounding boxes corresponding to the first object, generating a first sub-bounding box corresponding to the second object included in the image, and generating an overlap between the first sub-bounding box and the first bounding box. obtaining a first overlap score, removing at least one first bounding box having a first overlap score with a first sub-bounding box exceeding a first threshold value, among a plurality of first bounding boxes; At least one instruction configured to detect a first object through a first bounding box that is not removed among the bounding boxes may be stored.

According to an embodiment, the processor 110 obtains an image including an object including a first object and a second object that is a part of the first object, and corresponds to the first object included in the image. A plurality of first bounding boxes are generated, a first sub-bounding box corresponding to a second object included in the image is created, and confidence in the probability that an object included in the inner area of the first bounding box coincides with the first object Obtain a reference value corresponding to the first bounding box according to whether the first sub-bounding box is disposed in the first bounding box, and determine whether the reliability of the at least one first bounding box corresponds to at least one first bounding box. At least one command configured to remove at least one first bounding box that does not exceed a reference value corresponding to , and to detect a first object through a first bounding box that is not removed among a plurality of first bounding boxes may be stored. there is.

Although not shown, in one embodiment, the electronic device 100 may further include a communication interface, a display unit, a user input unit, and a camera unit.

The electronic device 100 may be functionally connected to an external device using a communication interface. The external device may include, for example, at least one of a user terminal, a platform server, an artificial intelligence server, or a home appliance.

The communication interface may be connected to an external device through a mobile communication network, a wireless LAN communication network, or a local area network using one or more antennas under the control of the processor 110 . Wireless LAN communication may be wirelessly connected to an AP (access point) under the control of the processor 110 at a place where an access point (AP) is installed. For example, it may include Wi-Fi communication. Short-range communication includes Bluetooth communication, Bluetooth low energy communication, infrared data association (IrDA) communication, ultra-wideband (UWB) communication, magnetic secure transmission (MST) communication, and/or NFC communication. can include In various embodiments of the present disclosure, the term 'communication interface' may connect with an external device through mobile communication, wireless LAN communication, and/or short-range communication.

The display unit may provide (or display) a graphical user interface (GUI) corresponding to various services (eg, data transmission, broadcast reception, photo taking, video content capture, etc.). In an embodiment of the present disclosure, the display unit may include a touch screen.

The display unit may display an operation of detecting an object in an image under the control of the processor 110 .

The user input unit means a means through which a user inputs data for controlling the electronic device 100 . For example, the user input unit includes a key pad, a dome switch, and a touch pad (contact capacitance method, pressure resistive film method, infrared sensing method, surface ultrasonic conduction method, integral tension measurement method) , piezo effect method, etc.), a jog wheel, a jog switch, and the like, but are not limited thereto.

The camera unit may capture a still image or a video under the control of the processor 110 .

The camera unit may capture a still image or a video under the control of the processor 110 . The camera unit may obtain a still image or a moving image of a space including the object in order to detect the object in the image.

According to an embodiment, the camera unit may be at least one of a telephoto camera, a wide-angle camera, and a general camera, but is not limited thereto.

It is easy for those skilled in the art that at least one component of the aforementioned electronic device 100 can be added, deleted, or changed in response to the performance of the electronic device 100. can be understood

2 is a conceptual diagram for explaining an operation of detecting an object according to an embodiment of the present disclosure.

Referring to FIG. 2 , the electronic device 100 according to an embodiment may detect an object in an image using an artificial intelligence model. The artificial intelligence model may be composed of a deep learning-based object recognition model.

The electronic device 100 may obtain an image 200 including an object. The electronic device 100 may capture an image 200 including an object through a camera unit and receive the image 200 including an object through a communication interface. A method of acquiring the image 200 by the electronic device 100 is only an example, and the technical spirit of the present disclosure is not limited thereto.

Objects included in the image 200 may include the first object 300 and the second object 310 . The second object 310 may be a part of the first object 300 . For example, as shown, the first object 300 may be a person, and the second object 310 may be a face that is part of a person. However, this is only an example, and the types of the first object 300 and the second object 310 do not limit the technical spirit of the present disclosure. The first object 300 may be a car, and the second object 310 may be a wheel that is part of the car.

The electronic device 100 may create a plurality of first bounding boxes 401 , 402 , and 403 corresponding to the first object 300 included in the image 200 . The plurality of first bounding boxes 401 , 402 , and 403 may classify an object positioned inside each of the first bounding boxes 401 , 402 , and 403 as the first object 300 . For example, the plurality of first bounding boxes 401 , 402 , and 403 may classify an object positioned inside each of the first bounding boxes 401 , 402 , and 403 as a person.

The electronic device 100 may create a first sub-bounding box 411 corresponding to the second object 310 included in the image 200 . The first sub-bounding box 411 may classify objects located inside the first sub-bounding box 411 as the second object 310 . For example, the first sub-bounding box 411 may classify an object located inside the first sub-bounding box 411 as a face.

For reference, the electronic device 100 may acquire a degree of reliability regarding a probability that an object included in the inner region of the bounding box coincides with a certain type of object. The electronic device 100 may label the bounding box with a class specifying an object in an area inside the bounding box as an object type corresponding to the highest level of reliability among the levels of reliability. That is, the electronic device 100 labels a class specifying the type of object included in the inner area of the bounding box and a reliability level, which is a probability that the type of the specified object matches the type of object actually included in the inner area of the bounding box, A bounding box can be created.

In other words, the bounding box can classify objects located in the inner area of the bounding box into a specific class. For example, if the bounding box classifies an object located inside the bounding box into a class corresponding to a person, the bounding box may be a bounding box that recognizes the object located inside the bounding box as a person. there is. Also, the bounding box may include reliability for a specific class. Reliability may relate to the accuracy with which a real object included in the inner region of the bounding box matches an object of a type corresponding to a class classified by the bounding box.

For example, as shown in FIG. 2 , the 1_1 bounding box 401 may be labeled as a class specifying the first object 300 as a person. The reliability of the probability that an object included in the inner region of the 1_1 bounding box 401 coincides with the first object 300 may be 0.91.

Also, the first sub-bounding box 411 may be labeled with a class specifying the second object 310 as a face. The reliability of the probability that an object included in the inner area of the first sub-bounding box 411 coincides with the second object 310 may be 0.94.

The electronic device 100 compares the plurality of first bounding boxes 401 , 402 , and 403 with the first sub-bounding box 411 , and thereby detects at least one of the plurality of first bounding boxes 401 , 402 , and 403 . can be removed.

In an embodiment, the electronic device 100 may obtain a first overlap score regarding the degree of overlap between the first sub-bounding box 411 and one of the plurality of first bounding boxes. The electronic device 100 selects at least one first bounding box whose first overlap score with the first sub-bounding box 411 exceeds a first threshold value among the plurality of first bounding boxes 401 , 402 , and 403 . can be removed That is, the electronic device 100 may remove at least one first bounding box having a high degree of overlap with the first sub-bounding box 411 from among the plurality of first bounding boxes 401 , 402 , and 403 .

For example, the electronic device 100 may obtain a first overlap score related to the degree of overlap between the first sub bounding box 411 and the 1_1st bounding box 401 . A first overlap score between the first sub-bounding box 411 and the 1_1-th bounding box 401 may be evaluated as, for example, 0.9. The electronic device 100 may set the first threshold to, for example, 0.8. The electronic device 100 may remove the 1_1 bounding box 401 since the first overlap score (ie, 0.9) for the 1_1 bounding box 401 exceeds the first threshold value (ie, 0.8). .

In another embodiment, the electronic device 100 may obtain a degree of reliability regarding a probability that an object included in the inner region of the first bounding box coincides with the first object 300 . The electronic device 100 may obtain a reference value corresponding to the first bounding box according to whether the first sub-bounding box 411 is disposed within the predetermined first bounding box. The electronic device 100 may remove the first bounding box when the reliability of the first bounding box does not exceed a reference value corresponding to the first bounding box.

For example, the electronic device 100 may obtain a degree of reliability regarding a probability that an object included in the inner region of the 1_2 bounding box 402 coincides with the first object 300, that is, a person. The electronic device 100 may obtain a reference value corresponding to the 1_2 bounding box 402 according to whether the first sub bounding box 411 is disposed within the 1_2 bounding box 402 .

As shown in FIG. 2 , the electronic device 100 obtains a first reference value corresponding to the 1_2 bounding box 402 because the first sub bounding box 411 is disposed within the 1_2 bounding box 402 . can The electronic device 100 may remove the 1_2nd bounding box 402 when the reliability of the 1_2nd bounding box 402 does not exceed the first reference value corresponding to the 1_2nd bounding box 402, and If the reliability of 402 exceeds the first reference value corresponding to the 1_2 bounding box, the 1_2 bounding box 402 may not be removed.

For reference, unlike what is shown in FIG. 2 , when the first sub-bounding box 411 is not disposed within the 1_2 bounding box 402, the electronic device 100 corresponds to the 1_2 bounding box 402. A second reference value may be obtained. When the reliability of the 1_2 bounding box 402 does not exceed the second reference value corresponding to the 1_2 bounding box, the electronic device 100 may remove the 1_2 bounding box 402, and the 1_2 bounding box 402 If the reliability of the 1_2 bounding box 402 exceeds the second reference value corresponding to the 1_2 bounding box, the 1_2 bounding box 402 may not be removed.

The electronic device 100 may detect the first object 300 through the first bounding box that is not removed among the plurality of first bounding boxes 401 , 402 , and 403 . That is, the electronic device 100 may detect a person as the first object 300 through the 1_3 bounding boxes 403 that are not removed among the plurality of first bounding boxes 401 , 402 , and 403 . there is.

3 is a flowchart illustrating a method of detecting an object according to an embodiment of the present disclosure. 4 is a diagram for explaining an operation of detecting an object according to an embodiment of the present disclosure. Hereinafter, an operating method of the electronic device 100 will be described with reference to FIGS. 3 and 4 together.

For convenience of description, overlapping descriptions with those described with reference to FIG. 2 are simplified or omitted.

Referring to FIGS. 3 and 4 together, in step S310, the electronic device 100 according to an embodiment may obtain an image 200 including an object. Objects included in the image 200 may include a first object 300 and a second object 310 , and the second object 310 may be a part of the first object 300 .

The types of the first object 300 and the second object 310 are only examples, and the technical idea of the present disclosure is not limited thereto. Although the first object 300 is shown as a person and the second object 310 is shown as a face, the first object and the second object may correspond to other objects.

For example, the first object may be an airplane and the second object may be a wing that is part of the airplane. As another example, the first object may be a person, and the second object may be a hand that is part of the person.

In step S320, the electronic device 100 according to an embodiment may generate a plurality of first bounding boxes 401, 402, and 403 corresponding to the first object 300 included in the image 200. .

The electronic device 100 assigns a first class that specifies the type of object included in the inner area of the plurality of first bounding boxes 401, 402, and 403 as the first object 300 to the plurality of first bounding boxes ( 401, 402, 403) can be labeled. In addition, the electronic device 100 determines reliability, which is a probability that the type of object specified as the first object 300 coincides with the type of object actually included in the inner region of the plurality of first bounding boxes 401, 402, and 403. may be labeled on the plurality of first bounding boxes 401, 402, and 403.

As shown in FIG. 4 , the electronic device 100 may create a 1_1 bounding box 401 , a 1_2 bounding box 402 , and a 1_3 bounding box 403 .

For example, the 1_1 bounding box 401 may be labeled with a first class that specifies an object included therein as a person, and the number of objects actually included in the inner region of the 1_1 bounding box 401 0.61 can be labeled as the reliability, which is the probability that the type and person match.

Also, for example, the 1_2 bounding box 402 may be labeled as a first class that specifies an object included therein as a person, and an object actually included in the inner region of the 1_2 bounding box 402 0.53 can be labeled as the reliability, which is the probability that the type of and the person match.

Also, for example, the 1_3 bounding box 403 may be labeled with a first class that specifies an object included therein as a person, and an object actually included in the inner area of the 1_3 bounding box 403 0.91 can be labeled as the reliability, which is the probability that the type of and person match.

Among the plurality of first bounding boxes 401 , 402 , and 403 , the first_third bounding box 403 may be a bounding box that more appropriately recognizes a person. Accordingly, the reliability of the 1_3 bounding box 403 may be higher than that of the 1_1 bounding box 401 and the 1_2 bounding box 402 .

For example, the 1_1 bounding box 401 may be a bounding box limitedly recognized for a face, which is a part of a person, and the 1_2 bounding box 402 may be a part of a chair as a person. It may be a bounding box that is mistakenly recognized as a person.

In step S330, the electronic device 100 according to an embodiment may create a first sub-bounding box 411 corresponding to the second object 310 included in the image 200.

The electronic device 100 may label the first sub-bounding box 411 with a second class that specifies the type of object included in the inner area of the first sub-bounding box 411 as the second object 310 . In addition, the electronic device 100 assigns a reliability level, which is a probability that the type of object specified as the second object 310 coincides with the type of object actually included in the inner area of the first sub-bounding box 411 , to the first sub-bounding box 411 . (411) can be labeled.

For example, the second class may mean at least one of a face class, an ear class, and a hand class, but this is only an example, and the technical idea of the present disclosure is not limited thereto.

As shown in FIG. 4 , the electronic device 100 may create a first sub-bounding box 411 . For convenience of description, only one sub-bounding box is shown, and the technical idea of the present disclosure does not limit the number of sub-bounding boxes.

For example, the first sub-bounding box 411 may be labeled with a second class that specifies an object included therein as a face, and the first sub-bounding box 411 may be labeled with a second class that is actually included in the inner area. 0.94 may be labeled as the reliability, which is the probability that the type of object matches the face.

In operation S340, the electronic device 100 according to an embodiment may obtain a first overlap score regarding the degree of overlap between the first sub bounding box 411 and the first bounding box.

The overlap score may be, for example, an intersection over union (IOU) score. However, this is only an example, and the technical spirit of the present disclosure is not limited thereto. The overlap score may be calculated as a value obtained by dividing the area of the intersection area by the area of the union area of the two bounding boxes.

For example, in step S410 of FIG. 4 , the electronic device 100 may compare the 1_1 bounding box 401 and the first sub bounding box 411 . The electronic device 100 may obtain a first overlap score regarding the degree of overlap between the first sub bounding box 411 and the 1_1st bounding box 401 .

로 계산될 수 있다.Specifically, as shown in FIG. 4 , the area of the inner area of the 1_1 bounding box 401 can be expressed as A+B, and the area of the inner area of the first sub-bounding box 411 is B It can be expressed as +C. In this case, the area of the intersection area of the first sub-bounding box 411 and the 1_1st bounding box 401 may be B, and the area of intersection of the first sub-bounding box 411 and the 1_1st bounding box 401 is The area may be A+B+C. Accordingly, the first overlap score for the degree of overlap between the first sub bounding box 411 and the 1_1 bounding box 401 is

can be calculated as

For example, when the ratio of A, B, and C is 1:8:1, the electronic device 100 provides a first overlap regarding the degree of overlap between the first sub-bounding box 411 and the 1_1st bounding box 401. You can get a score of 0.8.

As another example, in step S420 of FIG. 4 , the electronic device 100 may compare the 1_2 bounding box 402 and the first sub bounding box 411 . The electronic device 100 may obtain a first overlap score regarding the degree of overlap between the first sub-bounding box 411 and the 1_2nd bounding box.

Specifically, as shown in FIG. 4 , the area of the inner area of the 1_2 bounding box 402 can be expressed as D, and the area of the inner area of the first sub-bounding box 411 is B+C. can be expressed as In this case, the width of the intersection area of the first sub-bounding box 411 and the 1_2nd bounding box 402 may be 0, and the intersection area of the first sub-bounding box 411 and the 1_2nd bounding box 402 may be The area may be B+C+D. Accordingly, the first overlap score regarding the degree of overlap between the first sub-bounding box 411 and the 1_2-th bounding box 402 may be calculated as 0.

The electronic device 100 may obtain 0 as a first overlap score regarding the degree of overlap between the first sub-bounding box 411 and the 1_2nd bounding box 402 .

In step S350, the electronic device 100 according to an embodiment determines that a first overlap score with a first sub-bounding box 411 among a plurality of first bounding boxes 401, 402, and 403 meets a first threshold value. At least one first bounding box that crosses over may be removed.

The first threshold value may be, for example, an arbitrarily set value. The electronic device 100 may set the first threshold value according to a user input or may receive threshold values set according to cases by the server, but this is only an example and the technical spirit of the present disclosure is not limited thereto.

For example, in step S410 of FIG. 4 , the electronic device 100 may compare a first overlap score calculated between the 1_1 bounding box 401 and the first sub-bounding box 411 with a first threshold value. The electronic device 100 may remove the 1_1 bounding box 401 when the calculated first overlap score between the 1_1 bounding box 401 and the 1st sub bounding box 411 exceeds a first threshold value. there is.

As another example, in step S420 of FIG. 4 , the electronic device 100 may compare the first overlap score calculated between the 1_2 bounding box 402 and the first sub-bounding box 411 with a first threshold value. The electronic device 100 may remove the 1_2 bounding box 402 when the calculated first overlap score between the 1_2 bounding box 402 and the 1st sub bounding box 411 exceeds a first threshold value. there is.

In step S360, the electronic device 100 according to an embodiment may detect the first object 300 through the first bounding box that is not removed among the plurality of first bounding boxes 401, 402, and 403. .

For example, a first overlap score between the 1_3 bounding box 403 and the first sub bounding box 411 may not exceed a first threshold value. Accordingly, the electronic device 100 may not remove the 1_3 bounding box 403 . The electronic device 100 may detect the first object (person; 300) through the 1_3 bounding box 403 that is not removed.

5 is a flowchart illustrating a method of detecting an object according to an embodiment of the present disclosure. 6 is a diagram for explaining an operation of detecting an object according to an embodiment of the present disclosure. Hereinafter, the operating method of the electronic device 100 will be described with reference to FIGS. 5 and 6 together.

For convenience of description, overlapping descriptions with those described with reference to FIGS. 3 and 4 are simplified or omitted.

Referring to FIGS. 5 and 6 together, in step S510, the electronic device 100 according to an embodiment may obtain an image 200 including an object. Objects included in the image 200 may include a first object 300 , a second object 310 , and a third object 320 . The second object 310 and the third object 320 may be part of the first object 300 .

The types of the first object 300 , the second object 310 , and the third object 320 are only examples, and the technical idea of the present disclosure is not limited thereto. Although the first object 300 is shown as a person, the second object 310 as a face, and the third object 320 as an ear, the first to third objects are shown respectively. It can correspond to other objects.

For example, the first object may be a car, the second object may be a wheel that is part of the car, and the third object may be a window that is part of the car.

In step S520, the electronic device 100 according to an embodiment generates a plurality of first bounding boxes 601, 602, 603, and 604 corresponding to the first object 300 included in the image 200. can

The electronic device 100 assigns a first class that specifies, as the first object 300, the type of object included in the inner area of the plurality of first bounding boxes 601, 602, 603, and 604 to a plurality of first bounding boxes. Fields 601, 602, 603, and 604 may be labeled. In addition, the electronic device 100 determines the probability that the type of object specified as the first object 300 coincides with the type of object actually included in the inner area of the plurality of first bounding boxes 601, 602, 603, and 604. A plurality of first bounding boxes 601 , 602 , 603 , and 604 may be labeled with the reliability of .

As shown in FIG. 6 , the electronic device 100 may create a 1_1 bounding box 601, a 1_2 bounding box 602, a 1_3 bounding box 603, and a 1_4 bounding box 604. .

Each of the 1_1st to 1_4th bounding boxes 601, 602, 603, and 604 may be labeled with a first class that specifies an object included therein as a person, and is actually included in each inner area. Reliability, which is the probability that the type of object and person match each other, may be labeled.

For example, the 1_1 bounding box 601 may be a bounding box limitedly recognized for a face that is a part of a person, and the 1_2 bounding box 602 may be an ear that is a part of a person. It may be a bounding box limitedly recognized in (ear), and the 1st_3rd bounding box 603 may be a bounding box in which a part of a chair is mistakenly recognized as a person.

For reference, for convenience of description, operations described in steps S530, S540, and S550 are the same as operations described in steps S330, S340, and S350 of FIG. will be explained based on

In step S531, the electronic device 100 according to an embodiment may create a second sub-bounding box 612 corresponding to the third object 320 included in the image 200.

The electronic device 100 may label the second sub bounding box 612 with a third class that specifies the type of object included in the inner area of the second sub bounding box 612 as the third object 320 . In addition, the electronic device 100 assigns a reliability level, which is a probability that the type of object specified as the third object 320 coincides with the type of object actually included in the inner area of the second sub-bounding box 612, to the second sub-bounding box 612. (612) can be labeled.

The third object 320 may be different from the second object 310 specified by the second class labeled in the first sub-bounding box 611 . For example, unlike the second object 310 which is a face, the third object 320 may be an ear.

For example, the second sub-bounding box 612 may be labeled with a third class that specifies an object included therein as an ear, and the second sub-bounding box 612 actually included in the inner area may be labeled as an ear. 0.95 may be labeled as the reliability, which is the probability that the type of object and the ear match.

In step S541, the electronic device 100 according to an embodiment may obtain a second overlap score related to the degree of overlap between the second sub bounding box 612 and the first bounding box.

For reference, in step S610 of FIG. 6 , the electronic device 100 may compare the 1_1 bounding box 601 and the first sub bounding box 411 . The electronic device 100 may obtain a first overlap score regarding the degree of overlap between the first sub bounding box 411 and the 1_1st bounding box 601 .

Since the description of the first overlap score regarding the degree of overlap between the first sub-bounding box 411 and the 1_1-th bounding box 601 is the same as that of step S410 described in FIGS. 3 and 4 , overlapping descriptions are omitted.

For example, in step S620 of FIG. 6 , the electronic device 100 may compare the 1_1 bounding box 601 and the second sub bounding box 612 . The electronic device 100 may obtain a second overlap score regarding the degree of overlap between the second sub bounding box 612 and the 1_1st bounding box 601 .

Since the method of obtaining the second overlap score is the same as the method of obtaining the first overlap score described in FIGS. 3 and 4, duplicate descriptions are omitted. However, the second overlap score regarding the degree of overlap between the 1st_1st bounding box 601 and the 2nd sub bounding box 612 may be somewhat low. For example, the electronic device 100 may obtain a second overlap score of 0.2 regarding the degree of overlap between the 1_1 bounding box 601 and the second sub bounding box 612 .

As another example, in step S630 of FIG. 6 , the electronic device 100 may compare the 1_2 bounding box 602 and the second sub bounding box 612 . The electronic device 100 may obtain a second overlap score regarding the degree of overlap between the second sub bounding box 612 and the 1_2 th bounding box 602 .

Since the method of obtaining the second overlap score is the same as the method of obtaining the first overlap score described in FIGS. 3 and 4, duplicate descriptions are omitted. However, the second overlap score regarding the degree of overlap between the 1st_2nd bounding box 602 and the 2nd sub bounding box 612 may be somewhat high. For example, the electronic device 100 may obtain a second overlap score of 0.9 regarding the degree of overlap between the 1_2 bounding box 602 and the second sub bounding box 612 .

In step S551, the electronic device 100 according to an embodiment determines that a second overlap score with a second sub-bounding box 612 among the plurality of first bounding boxes 601, 602, 603, and 604 is a second threshold. At least one first bounding box exceeding the value may be removed.

The second threshold may be, for example, an arbitrarily set value. The electronic device 100 may set the second threshold according to a user input or may receive thresholds set according to cases by the server, but this is only an example and the technical idea of the present disclosure is not limited thereto.

For reference, in step S610 of FIG. 6 , the electronic device 100 may compare the calculated first overlap score between the 1_1 bounding box 601 and the first sub-bounding box 611 with a first threshold value. The electronic device 100 may remove the 1_1 bounding box 601 when the calculated first overlap score between the 1_1 bounding box 601 and the 1 st sub bounding box 611 exceeds a first threshold value. there is.

In step S620 of FIG. 4 , the electronic device 100 may compare the calculated second overlap score between the 1_1 bounding box 601 and the second sub-bounding box 612 with a second threshold. The electronic device 100 does not remove the 1_1 bounding box 601 when the calculated second overlap score between the 1_1 bounding box 601 and the second sub bounding box 612 does not exceed the second threshold. may not be

As another example, in step S630 of FIG. 4 , the electronic device 100 may compare the calculated second overlap score between the 1_2 bounding box 602 and the second sub-bounding box 612 with the second threshold. The electronic device 100 may remove the 1_2 bounding box 602 when the calculated second overlap score between the 1_2 bounding box 602 and the second sub bounding box 612 exceeds the second threshold. there is.

In step S560, the electronic device 100 according to an embodiment detects the first object 300 through the first bounding box that is not removed among the plurality of first bounding boxes 601, 602, 603, and 604. can

For example, a first overlap score between the 1_4th bounding box 604 and the first sub bounding box 611 may not exceed a first threshold value. Accordingly, the electronic device 100 may not remove the 1_3 th bounding box 603 . Also, the second overlap score between the 1_4th bounding box 604 and the second sub bounding box 612 may not exceed the second threshold. Accordingly, the electronic device 100 may not remove the 1_4th bounding box 604 .

The electronic device 100 may detect the first object (person; 300) through the 1_4 bounding box 604 that is not removed.

7 is a flowchart illustrating a method of detecting an object according to an embodiment of the present disclosure. 8 is a diagram for explaining an operation of detecting an object according to an embodiment of the present disclosure. Hereinafter, a method of operating the electronic device 100 will be described with reference to FIGS. 7 and 8 together.

For convenience of description, overlapping descriptions with those described with reference to FIGS. 7 and 8 are simplified or omitted.

Referring to FIGS. 7 and 8 together, in step S710, the electronic device 100 according to an embodiment may obtain an image 200 including an object. Objects included in the image 200 may include a first object 300 , a second object 310 , and a third object 320 . The second object 310 and the third object 320 may be part of the first object 300 .

In step S720, the electronic device 100 according to an embodiment may generate a plurality of first bounding boxes corresponding to the first object 300 included in the image 200. As shown in FIG. 8 , the electronic device 100 may create a 1_1 bounding box 801 , a 1_2 bounding box 802 , and a 1_3 bounding box 803 .

Each of the 1_1st to 1_3rd bounding boxes 801, 802, and 803 may be labeled as a first class specifying an object included therein as a person, and an object actually included in each inner area. Reliability, which is the probability that the type and person match each other, may be labeled.

For reference, for convenience of description, operations described in steps S730, S740, and S750 are the same as operations described in steps S330, S340, and S350 of FIG. Since it is the same as the operation described in steps S531, S541, and S551, duplicate descriptions are omitted. Hereinafter, operations of steps S732, S742 and S752 will be mainly described.

In step S732, the electronic device 100 according to an embodiment determines a first reference bounding box R1 corresponding to the first object 300 included in the image 200 from among a plurality of first bounding boxes. can

The electronic device 100 may determine a first bounding box having the highest labeled reliability among the plurality of first bounding boxes as the first reference bounding box R1. For example, the reliability of the 1_1st bounding box 801 may be 0.61. The reliability of the 1_2 bounding box 802 may be 0.84. The reliability of the 1_3 bounding box 803 may be 0.91. In this case, the electronic device 100 may determine the 1_3 bounding box 803 as the first reference bounding box R1. The 1_3 bounding box 803 may be a first reference bounding box R1.

In step S742, the electronic device 100 according to an embodiment includes one of the plurality of first bounding boxes 801, 802, and 803 excluding the first reference bounding box R1 and the first reference bounding box R1. A third overlap score regarding the degree of overlap between the livers may be obtained.

For example, in step S810 of FIG. 8 , the electronic device 100 may compare the 1_1 bounding box 801 and the first reference bounding box R1. The electronic device 100 may obtain a third overlap score regarding the degree of overlap between the first reference bounding box R1 and the 1_1st bounding box 801 .

Since the method for obtaining the third overlap score is the same as the method for obtaining the first overlap score described in FIGS. 3 and 4, duplicate descriptions are omitted. However, the third overlap score regarding the degree of overlap between the 1_1st bounding box 801 and the first reference bounding box R1 may be somewhat low. For example, the electronic device 100 may obtain a third overlap score of 0.4 regarding the degree of overlap between the 1_1st bounding box 801 and the first reference bounding box R1.

As another example, in step S820 of FIG. 8 , the electronic device 100 may compare the 1_2nd bounding box 802 and the first reference bounding box R1. The electronic device 100 may obtain a third overlap score regarding the degree of overlap between the first reference bounding box R1 and the 1_2nd bounding box 802 .

For example, the third overlap score regarding the degree of overlap between the 1_2nd bounding box 802 and the first reference bounding box R1 may be somewhat high. The electronic device 100 may obtain a third overlap score of 0.8 regarding the degree of overlap between the 1st_2nd bounding box 802 and the first reference bounding box R1.

In step S752, the electronic device 100 according to an embodiment determines at least one first bounding box whose third overlap score with the first reference bounding box R1 exceeds a third threshold among a plurality of first bounding boxes. The box can be removed.

The third threshold may be, for example, an arbitrarily set value. The electronic device 100 may set the third threshold according to user input and may receive thresholds set according to cases by the server, but this is only an example and the technical spirit of the present disclosure is not limited thereto.

For example, in step S810 of FIG. 8 , the electronic device 100 may compare the calculated third overlap score between the 1_1 bounding box 801 and the first reference bounding box R1 with a third threshold. The electronic device 100 does not remove the 1_1 bounding box 801 when the calculated third overlap score between the 1_1 bounding box 801 and the first reference bounding box R1 does not exceed the third threshold. may not be

As another example, in step S820 of FIG. 8 , the electronic device 100 may compare the calculated third overlap score between the 1_2 bounding box 802 and the first reference bounding box R1 with a third threshold. The electronic device 100 does not remove the 1_2 bounding box 802 when the calculated third overlap score between the 1_2 bounding box 802 and the first reference bounding box R1 does not exceed the third threshold. may not be

In step S760, the electronic device 100 according to an embodiment may detect the first object 300 through the first bounding box that is not removed among the plurality of first bounding boxes 801, 802, and 803. .

For example, the third overlap score between the 1_1st bounding box 801 and the first reference bounding box R1 may exceed the third threshold value. Accordingly, the electronic device 100 may remove the 1_1 bounding box 801 . Also, a third overlap score between the 1_2 bounding box 802 and the first reference bounding box R1 may exceed a third threshold value. Accordingly, the electronic device 100 may remove the 1_2 bounding box 802 . The electronic device 100 may detect a person as the first object 300 through the first_third bounding box 803 that is not removed among the plurality of first bounding boxes 801 , 802 , and 803 .

9 and 10 are flowcharts for describing a method of detecting an object according to an embodiment of the present disclosure. 11 is a diagram for explaining an operation of detecting an object according to an embodiment of the present disclosure. Hereinafter, an operating method of the electronic device 100 according to the flowcharts of FIGS. 9 and 10 will be described with reference to FIG. 11 .

For convenience of explanation, overlapping descriptions with those described with reference to FIG. 3 are simplified or omitted.

Referring to FIGS. 9 and 11 together, in step S910, the electronic device 100 according to an embodiment may acquire an image 200 including an object. Objects included in the image 200 may include a first object 300 and a second object 310 . The second object 310 may be a part of the first object 300 .

In step S920, the electronic device 100 according to an embodiment may generate a plurality of first bounding boxes corresponding to the first object 300 included in the image 200. As shown in FIG. 11 , the electronic device 100 may create a 1_1 bounding box 1101 , a 1_2 bounding box 1102 , and a 1_3 bounding box 1103 .

Each of the 1_1 to 1_3 bounding boxes 1101, 1102, and 1103 may be labeled with a first class that specifies an object included therein as a person, and an object actually included in each inner area. Reliability, which is the probability that the type and person match each other, may be labeled.

In step S930, the electronic device 100 according to an embodiment may create a first sub-bounding box 1110 corresponding to the second object 310 included in the image 200.

As shown in FIG. 11 , the electronic device 100 may create a first sub-bounding box 1110 . For convenience of description, only one sub-bounding box is shown, and the technical idea of the present disclosure does not limit the number of sub-bounding boxes.

In step S940, the electronic device 100 according to an embodiment may obtain a degree of reliability regarding a probability that an object included in the inner area of the first bounding box coincides with the first object 300.

The electronic device 100 may label the first bounding box with a first class specifying the type of object included in the inner region of the first bounding box as the first object 300 . In addition, the electronic device 100 may label the first bounding box with reliability, which is a probability that the type of object specified as the first object 300 matches the type of object actually included in the inner region of the first bounding box. .

For example, the first bounding box may be labeled as a first class that specifies an object included therein as a person, and the type of object actually included in the first bounding box and the person 0.91 can be labeled as the reliability, which is the probability of matching.

The electronic device 100 may obtain the reliability level labeled for each first bounding box. For example, the electronic device 100 may obtain the reliability of the 1_1 bounding box 1101, and as shown in FIG. 11, the reliability of the 1_1 bounding box 1101 may be 0.91. The electronic device 100 may acquire the reliability of the 1_2 bounding box 1102 , and as shown in FIG. 11 , the reliability of the 1_2 bounding box 1102 may be 0.37. The electronic device 100 may acquire the reliability of the 1_3 bounding box 1103, and as shown in FIG. 11, the reliability of the 1_3 bounding box 1103 may be 0.31.

In step S950, the electronic device 100 according to an embodiment may obtain a reference value corresponding to the first bounding box according to whether the first sub bounding box 1110 is disposed in the first bounding box.

The reference value may be, for example, an arbitrarily set value. The electronic device 100 may set a reference value according to a user input or may receive reference values set according to a case by a server, but this is only an example and the technical idea of the present disclosure is not limited thereto.

The reference value may be, for example, an indicator related to the validity of the first bounding box. When the reliability of the first bounding box exceeds the reference value, the electronic device 100 according to an embodiment may not remove the first bounding box. When the reliability of the first bounding box does not exceed the reference value, the electronic device 100 may remove the first bounding box.

In step S960, when the reliability of at least one first bounding box does not exceed the reference value, the electronic device 100 according to an embodiment may remove the corresponding first bounding box.

Referring to step S1110 of FIG. 11 together, the electronic device 100 according to an embodiment may obtain a reliability of 0.91 for the 1_1 bounding box 1101 . Also, the electronic device 100 may obtain a reference value corresponding to the 1_1 bounding box 1101 according to whether the first sub bounding box 1110 is disposed within the 1_1 bounding box 1101 . The reference value corresponding to the 1_1st bounding box 1101 may be, for example, 0.5. Since the reliability (0.91) of the 1_1 bounding box 1101 exceeds the reference value (0.5) of the 1_1 bounding box 1101, the electronic device 100 may not remove the 1_1 bounding box 1101. .

Referring to step S1120 of FIG. 11 together, the electronic device 100 may acquire the reliability of the 1_2 bounding box 1102 as 0.37. Also, the electronic device 100 may obtain a reference value corresponding to the 1_2 bounding box 1102 according to whether the first sub bounding box 1110 is disposed within the 1_2 bounding box. The reference value corresponding to the 1_2 bounding box 1102 may be, for example, 0.5. Since the reliability (0.37) of the 1_2 bounding box 1102 does not exceed the reference value (0.5) of the 1_2 bounding box 1102, the electronic device 100 may remove the 1_1 bounding box 1101. .

Referring to step S1130 of FIG. 11 together, the electronic device 100 may acquire the reliability of the 1_3 bounding box 1103 as 0.31. In addition, the electronic device 100 may obtain a reference value corresponding to the 1_3 bounding box 1103 according to whether the first sub bounding box 1110 is disposed within the 1_3 bounding box. The reference value corresponding to the 1st_3rd bounding box 1103 may be, for example, 0.5. Since the reliability (0.31) of the 1_3 bounding box 1103 does not exceed the reference value (0.5) of the 1_3 bounding box 1103, the electronic device 100 may remove the 1_3 bounding box 1103. .

In step S970, the electronic device 100 according to an embodiment may detect the first object 300 through a first bounding box that is not removed among the plurality of first bounding boxes 1101, 1102, and 1103. .

For example, the reliability of the 1_1 bounding box 1101 may exceed the reference value of the 1_1 bounding box 1101 . Accordingly, the electronic device 100 may not remove the 1_1 bounding box 1101 . The electronic device 100 may detect the first object (person; 300) through the 1_1 bounding box 1101 that is not removed.

Referring to FIGS. 10 and 11 together, step S950 may include steps S951 and S952. In step S951, the electronic device 100 according to an embodiment may obtain a first reference value corresponding to at least one first bounding box in which the first sub-bounding box 1110 is included. Also, in step S952, the electronic device 100 according to an embodiment may obtain a second reference value corresponding to at least one first bounding box in which the first sub-bounding box 1110 is not included. The first reference value may be lower than the second reference value.

In FIG. 10 , the sequence of steps S951 and S952 is shown as specified, but this does not limit the technical spirit of the present disclosure. For example, it may be possible that step S951 is performed after step S952 is performed.

Referring to step S1110 of FIG. 11 , a first sub-bounding box 1110 may be disposed in an inner area of the 1_1 bounding box 1101 . The electronic device 100 may obtain a first reference value corresponding to the 1_1 bounding box 1101 in which the first sub-bounding box 1110 is included.

Referring to steps S1120 and S1130 of FIG. 11 , the first sub-bounding box 1110 may not be disposed in the inner area of the 1_2 bounding box 1102 . The electronic device 100 may obtain a second reference value corresponding to the 1_2 bounding box 1102 in which the first sub-bounding box 1110 is not included. The first sub-bounding box 1110 may not be disposed in the inner region of the 1_3 bounding box 1103 . The electronic device 100 may obtain a second reference value corresponding to the 1_3 bounding box 1103 in which the first sub-bounding box 1110 is not included.

Step S960 may include steps S961 to S963. In step S961, the electronic device 100 according to an embodiment may obtain information about whether the first sub-bounding box 1110 is included in the first bounding box.

In step S962, when the first sub-bounding box 1110 is included in the first bounding box, the electronic device 100 determines the reliability of the first bounding box including the first sub-bounding box 1110 therein. If does not exceed the first reference value, the corresponding first bounding box may be removed.

For example, referring to step S1110 of FIG. 11 , when the first sub-bounding box 1110 is included in the 1_1 bounding box 1101, the electronic device 100 corresponds to the 1_1 bounding box 1101. A first reference value may be obtained. The reliability of the 1_1 bounding box 1101 may be, for example, 0.91, and the first reference value may be, for example, 0.5. Since the reliability (0.91) of the 1_1st bounding box 1101 including the first sub-bounding box 1110 therein exceeds the first reference value (0.5), the electronic device 100 corresponds to the 1_1st bounding box ( 1101) may not be removed.

However, although not shown, the electronic device 100, if the reliability of the 1_1 bounding box 1101 including the first sub bounding box 1110 therein does not exceed the first reference value, corresponds to the corresponding 1_1 The bounding box 1101 can be removed.

As another example, referring to step S1120 of FIG. 11 , when the first sub-bounding box 1110 is not included in the 1_2 bounding box 1102, the electronic device 100 corresponds to the 1_2 bounding box 1102. A second reference value may be obtained. The reliability of the 1_2 bounding box 1102 may be, for example, 0.37, and the second reference value may be, for example, 0.7. Regarding the 1_2 bounding box 1102 that does not include the first sub bounding box 1110 therein, the electronic device 100 has a reliability level (0.37) that does not exceed the second reference value (0.7), and thus corresponds to the 1_2 bounding box 1102. The bounding box 1102 can be removed.

However, although not shown, the electronic device 100, if the reliability of the 1_2 bounding box 1102 that does not include the first sub-bounding box 1110 therein exceeds the second reference value, corresponds to the 1_2 bounding box 1102. The bounding box 1102 may not be removed.

As another example, referring to step S1130 of FIG. 11 , when the first sub-bounding box 1110 is not included in the 1_3 bounding box 1103, the electronic device 100 corresponds to the 1_3 bounding box 1103. A second reference value may be obtained. The reliability of the 1_3 bounding box 1103 may be, for example, 0.31, and the second reference value may be, for example, 0.7. Regarding the 1_3 bounding box 1103 that does not include the first sub bounding box 1110 therein, the electronic device 100 has a reliability level (0.31) that does not exceed the second reference value (0.7), and thus corresponds to the 1_3 bounding box 1103. The bounding box 1103 can be removed.

However, although not shown, the electronic device 100, if the reliability of the 1_3 bounding box 1103 that does not include the first sub bounding box 1110 therein exceeds the second reference value, corresponds to the 1_3 bounding box 1103. The bounding box 1103 may not be removed.

12 and 13 are flowcharts for describing a method of detecting an object according to an embodiment of the present disclosure. 14 is a diagram for explaining an operation of detecting an object according to an embodiment of the present disclosure. Hereinafter, an operating method of the electronic device 100 according to the flowcharts of FIGS. 12 and 13 will be described with reference to FIG. 14 .

For convenience of description, overlapping descriptions with those described with reference to FIGS. 3 and 9 to 11 are simplified or omitted.

Referring to FIGS. 12 and 14 together, in step S1210, the electronic device 100 according to an embodiment may obtain an image 200 including an object. Objects included in the image 200 may include the first object 300 , the second object 310 , and the third objects 330 and 340 . The second object 310 and the third objects 330 and 340 may be parts of the first object 300 .

In operation S1220, the electronic device 100 according to an embodiment may generate a plurality of first bounding boxes 1101, 1102, and 1103 corresponding to the first object 300 included in the image 200. .

The electronic device 100 assigns a first class that specifies the type of object included in the inner area of the plurality of first bounding boxes 1101, 1102, and 1103 as the first object 300 to the plurality of first bounding boxes ( 1101, 1102, 1103) can be labeled. In addition, the electronic device 100 determines the probability that the type of object specified as the first object 300 coincides with the type of object actually included in the inner region of the plurality of first bounding boxes 1101, 1102, and 1103. may be labeled on the plurality of first bounding boxes 1101, 1102, and 1103.

As shown in FIG. 14 , the electronic device 100 may create a 1_1 bounding box 1101 , a 1_2 bounding box 1102 , and a 1_3 bounding box 1103 .

Each of the 1_1 to 1_3 bounding boxes 1103 may be labeled with a first class that specifies an object included therein as a person, and the type of object actually included in each inner area and the person Reliability, which is the probability that (person) matches, may be labeled respectively.

In step S1230, the electronic device 100 according to an embodiment corresponds to the first sub-bounding box 1110 corresponding to the second object 310 included in the image 200 and the third objects 330 and 340. second sub-bounding boxes 1120 and 1121 may be created. As shown in FIG. 14 , the electronic device 100 may create a first sub-bounding box 1110 and second sub-bounding boxes 1120 and 1121 .

The first sub-bounding box 1110 may be labeled with a second class that specifies an object included therein as a face, and a type of object actually included in the inner area of the first sub-bounding box 1110. Reliability, which is the probability that the face coincides with , may be labeled respectively. The second sub-bounding boxes 1120 and 1121 may be labeled with a third class that specifies an object included therein as a hand, and is actually included in an inner area of the second sub-bounding boxes 1120 and 1121. Confidence, which is the probability that the type of object and hand match, may be labeled respectively.

For example, the second class may mean at least one of a face class, an ear class, and a hand class, but this is only an example, and the technical idea of the present disclosure is not limited thereto.

In step S1240, the electronic device 100 according to an embodiment may obtain a degree of reliability regarding a probability that an object included in the inner area of the first bounding box coincides with the first object 300.

The electronic device 100 may label the first bounding box with a first class specifying the type of object included in the inner region of the first bounding box as the first object 300 . In addition, the electronic device 100 may label the first bounding box with reliability, which is a probability that the type of object specified as the first object 300 matches the type of object actually included in the inner region of the first bounding box. . The electronic device 100 may obtain the reliability level labeled for each first bounding box.

For reference, for convenience of description, operations described in steps S1250 and S1260 are the same as those described in steps S950 and S960 of FIG.

In step S1251, the electronic device 100 according to an embodiment may obtain a reference value corresponding to the first bounding box according to whether the second sub bounding boxes 1120 and 1121 are disposed in the first bounding box. .

The reference value may be, for example, an arbitrarily set value. The electronic device 100 may set a reference value according to a user input or may receive reference values set according to a case by a server, but this is only an example and the technical idea of the present disclosure is not limited thereto.

The reference value may be, for example, an indicator related to the validity of the first bounding box. When the reliability of the first bounding box exceeds the reference value, the electronic device 100 according to an embodiment may not remove the first bounding box. When the reliability of the first bounding box does not exceed the reference value, the electronic device 100 may remove the first bounding box.

In operation S1261, when the reliability of at least one first bounding box does not exceed the reference value, the electronic device 100 according to an embodiment may remove the corresponding first bounding box.

Referring to step S1410 of FIG. 14 together, the electronic device 100 according to an embodiment may obtain a reliability of 0.91 for the 1_1 bounding box 1101 . Also, the electronic device 100 may obtain a reference value corresponding to the 1_1 bounding box 1101 according to whether the second sub bounding box 1120 is disposed within the 1_1 bounding box 1101 . The reference value corresponding to the 1_1st bounding box 1101 may be, for example, 0.5. Since the reliability (0.91) of the 1_1 bounding box 1101 exceeds the reference value (0.5) of the 1_1 bounding box 1101, the electronic device 100 may not remove the 1_1 bounding box 1101. .

Referring to step S1420 of FIG. 14 together, the electronic device 100 may acquire the reliability of the 1_2 bounding box 1102 as 0.37. Also, the electronic device 100 may obtain a reference value corresponding to the 1_2 bounding box 1102 according to whether the second sub bounding box 1121 is disposed within the 1_2 bounding box. The reference value corresponding to the 1_2 bounding box 1102 may be, for example, 0.5. Since the reliability (0.37) of the 1_2 bounding box 1102 does not exceed the reference value (0.5) of the 1_2 bounding box 1102, the electronic device 100 may remove the 1_1 bounding box 1101. .

Referring also to step S1430 of FIG. 14 , the electronic device 100 may acquire the reliability of the 1_3 bounding box 1103 as 0.31. Also, the electronic device 100 may obtain a reference value corresponding to the 1_3 bounding box 1103 according to whether the second sub bounding boxes 1120 and 1121 are disposed within the 1_3 bounding box. The reference value corresponding to the 1st_3rd bounding box 1103 may be, for example, 0.5. Since the reliability (0.31) of the 1_3 bounding box 1103 does not exceed the reference value (0.5) of the 1_3 bounding box 1103, the electronic device 100 may remove the 1_3 bounding box 1103. .

In step S1270, the electronic device 100 according to an embodiment may detect the first object 300 through a first bounding box that is not removed among a plurality of first bounding boxes.

For example, the reliability of the 1_1 bounding box 1101 may exceed the reference value of the 1_1 bounding box 1101 . Accordingly, the electronic device 100 may not remove the 1_1 bounding box 1101 . The electronic device 100 may detect the first object (person; 300) through the 1_1 bounding box 1101 that is not removed.

Referring to FIGS. 13 and 14 together, step S1251 may include steps S1252 and S1253. In step S1252, the electronic device 100 according to an embodiment may obtain a third reference value corresponding to at least one first bounding box in which the second sub-bounding boxes 1120 and 1121 are included. Also, in step S1253, the electronic device 100 according to an embodiment may obtain a fourth reference value corresponding to at least one first bounding box in which the second sub-bounding boxes 1120 and 1121 are not included. there is. The third reference value may be lower than the fourth reference value.

Although the order of steps S1252 and step S1253 is illustrated in FIG. 13 as specified, this does not limit the technical spirit of the present disclosure. For example, it may be possible that step S1252 is performed after step S1253 is performed.

Referring to step S1410 of FIG. 14 , a second sub bounding box 1120 may be disposed in an inner area of the 1_1 bounding box 1101 . The electronic device 100 may obtain a third reference value corresponding to the 1_1 bounding box 1101 in which the second sub bounding box 1120 is included.

Referring to step S1420 of FIG. 14 , similarly, a second sub bounding box 1121 may be disposed in an inner area of the 1_2 bounding box 1102 . The electronic device 100 may obtain a third reference value corresponding to the 1_2 bounding box 1102 in which the second sub bounding box 1121 is included.

Referring to step S1430 of FIG. 14 , the second sub bounding boxes 1120 and 1121 may not be disposed in the inner region of the 1_3 bounding box 1103 . The electronic device 100 may obtain a fourth reference value corresponding to the 1_3 bounding box 1103 in which the second sub-bounding boxes 1120 and 1121 are not included.

Step S1261 may include steps S1262 to S1264. In step S1262, the electronic device 100 according to an embodiment may obtain information about whether the second sub-bounding boxes 1120 and 1121 are included in the first bounding box.

In step S1263, when the first bounding box includes the second sub-bounding box, the electronic device 100 determines that the reliability of the first bounding box including the second sub-bounding box exceeds the third reference value. If not, the corresponding first bounding box may be removed.

For example, referring to step S1410 of FIG. 14 , when the second sub-bounding box 1120 is included inside the 1_1 bounding box 1101, the electronic device 100 corresponds to the 1_1 bounding box 1101. A third reference value may be obtained. The reliability of the 1_1 bounding box 1101 may be, for example, 0.91, and the third reference value may be, for example, 0.4. Since the reliability (0.91) of the 1_1st bounding box 1101 including the second sub-bounding box 1120 therein exceeds the third reference value (0.4), the electronic device 100 corresponds to the 1_1st bounding box ( 1101) may not be removed.

Also for reference, considering steps S1250 and S1260 of FIG. 12 , when the first sub-bounding box 1110 is included inside the 1_1 bounding box 1101 as shown in FIG. may obtain a first reference value corresponding to the 1_1 bounding box 1101 . The reliability of the 1_1 bounding box 1101 may be, for example, 0.91, and the first reference value may be 0.5. Since the reliability (0.91) of the 1_1st bounding box 1101 including the first sub-bounding box 1110 therein exceeds the first reference value (0.5), the electronic device 100 corresponds to the 1_1st bounding box ( 1101) may not be removed. That is, the electronic device 100 may determine whether to remove the first bounding box by considering positions where the first sub-bounding box 1110 and the second sub-bounding boxes 1120 and 1121 are disposed.

However, although not shown, the electronic device 100, when the reliability of the 1_1 bounding box 1101 including the second sub bounding box 1120 does not exceed the third reference value, corresponds to the 1_1 bounding box 1101. The bounding box 1101 can be removed.

Referring to step S1420 of FIG. 14 , when the second sub-bounding box 1121 is included inside the 1_2 bounding box 1102, the electronic device 100 sets the third reference value corresponding to the 1_2 bounding box 1102. can be obtained. The reliability of the 1_2 bounding box 1102 may be, for example, 0.37, and the third reference value may be, for example, 0.4. Regarding the 1_2 bounding box 1102 including the second sub bounding box 1121 therein, the electronic device 100 has a reliability level (0.37) that does not exceed the third reference value (0.4), and thus corresponds to the corresponding 1_2 bounding box. (1102) can be eliminated.

Also for reference, considering steps S1250 and S1260 of FIG. 12 , as shown in FIG. 14 , when the first sub-bounding box 1110 is not included inside the 1_2 bounding box 1102, the electronic device ( 100) may obtain a second reference value corresponding to the 1_2 bounding box 1102. The reliability of the 1_2 bounding box 1102 may be, for example, 0.37, and the second reference value may be 0.7. Regarding the 1_2 bounding box 1102 that does not include the first sub bounding box 1110 therein, the electronic device 100 has a reliability level (0.37) that does not exceed the second reference value (0.7), so the corresponding 1_2 bounding box Box 1102 can be removed. That is, the electronic device 100 may determine whether to remove the first bounding box by considering positions where the first sub bounding box 1110 and the second sub bounding box 1121 are disposed.

However, although not shown, if the reliability of the 1_2 bounding box 1102 including the second sub bounding box 1121 therein exceeds the third reference value, the corresponding 1_2 bounding box Box 1102 may not be removed.

Referring to step S1430 of FIG. 14 , when the second sub-bounding boxes 1120 and 1121 are not included in the 1_3 bounding box 1103, the electronic device 100 corresponds to the 1_3 bounding box 1103. A fourth reference value may be obtained. The reliability of the 1_3 bounding box 1103 may be, for example, 0.31, and the fourth reference value may be, for example, 0.9. Regarding the 1_3 bounding box 1103 that does not include the second sub bounding boxes 1120 and 1121 therein, the electronic device 100 has a reliability level (0.31) that does not exceed the second reference value (0.9), and thus corresponds to the second sub bounding box 1103. The 1_3 bounding box 1103 can be removed.

Also for reference, considering steps S1250 and S1260 of FIG. 12 , as shown in FIG. 14 , when the first sub-bounding box 1110 is not included in the 1_3 bounding box 1103, the electronic device ( 100) may obtain a second reference value corresponding to the 1_3 bounding box 1103. The reliability of the 1_3 bounding box 1103 may be, for example, 0.31, and the second reference value may be 0.7. With respect to the 1_3 bounding box 1103 that does not include the first sub bounding box 1110 therein, the electronic device 100 has a reliability level (0.31) that does not exceed the second reference value (0.7), and thus the corresponding 1_3 bounding box Box 1103 can be removed. That is, the electronic device 100 may determine whether to remove the first bounding box by considering positions where the first sub-bounding box 1110 and the second sub-bounding boxes 1120 and 1121 are disposed.

However, although not shown, the electronic device 100 determines that if the reliability of the 1_3 bounding box 1103 that does not include the second sub bounding boxes 1120 and 1121 therein exceeds the fourth reference value, the corresponding The 1_3 bounding box 1103 may not be removed. That is, the electronic device 100 may determine whether to remove the first bounding box by considering positions where the first sub-bounding box 1110 and the second sub-bounding boxes 1120 and 1121 are disposed.

The device-readable storage medium may be provided in the form of a non-transitory storage medium. Here, 'non-temporary storage medium' only means that it is a tangible device and does not contain signals (e.g., electromagnetic waves), and this term refers to the case where data is stored semi-permanently in the storage medium and temporary It does not discriminate if it is saved as . For example, a 'non-temporary storage medium' may include a buffer in which data is temporarily stored.

According to one embodiment, the method according to various embodiments disclosed in this document may be provided by being included in a computer program product. Computer program products may be traded between sellers and buyers as commodities. A computer program product is distributed in the form of a device-readable storage medium (eg compact disc read only memory (CD-ROM)), or via an application store or between two user devices (eg smartphones). It can be distributed (e.g., downloaded or uploaded) directly or online. In the case of online distribution, at least a part of a computer program product (eg, a downloadable app) is stored on a device-readable storage medium such as a memory of a manufacturer's server, an application store's server, or a relay server. It can be temporarily stored or created temporarily.

Claims (20)

As an object detection method for detecting an object in an image using an artificial intelligence model,
acquiring the image including the object including a first object and a second object that is part of the first object;
generating a plurality of first bounding boxes corresponding to the first object included in the image;
generating a first sub-bounding box corresponding to the second object included in the image;
obtaining a first overlap score related to an overlapping degree between the first sub-bounding box and the first bounding box;
removing at least one first bounding box whose first overlap score with the first sub-bounding box exceeds a first threshold value, among the plurality of first bounding boxes; and
and detecting the first object through a first bounding box that is not removed among the plurality of first bounding boxes. According to claim 1,
The object further includes a third object that is part of the first object,
generating a second sub-bounding box corresponding to the third object included in the image; and
obtaining a second overlap score related to an overlap between the second sub-bounding box and the first bounding box; and
The method further comprising removing at least one first bounding box whose second overlap score with the second sub-bounding box exceeds a second threshold value, among the plurality of first bounding boxes. According to claim 1,
The plurality of first bounding boxes include a first reference bounding box,
obtaining a third overlap score related to an overlapping degree between one of the plurality of first bounding boxes excluding the first reference bounding box and the first reference bounding box;
The method further comprises removing the at least one first bounding box, of the plurality of first bounding boxes, for which the third overlap score with the first reference bounding box exceeds a third threshold. According to claim 1,
The plurality of first bounding boxes are labeled with a first class specifying the first object,
The first sub-bounding box is labeled with a second class specifying the second object. According to claim 4,
The first class means a person class,
The second class means at least one of a face class, an ear class, and a hand class. As an object detection method for detecting an object in an image using an artificial intelligence model,
acquiring the image including the object including a first object and a second object that is part of the first object;
generating a plurality of first bounding boxes corresponding to the first object included in the image;
generating a first sub-bounding box corresponding to the second object included in the image;
obtaining a degree of reliability about a probability that an object included in the inner region of the first bounding box coincides with the first object;
obtaining a reference value corresponding to the first bounding box according to whether the first sub-bounding box is disposed within the first bounding box;
removing the at least one first bounding box whose reliability does not exceed the reference value corresponding to the at least one first bounding box; and
and detecting the first object through a first bounding box that is not removed among the plurality of first bounding boxes. According to claim 6,
The reference value includes a first reference value and a second reference value,
The step of obtaining the reference value,
obtaining a first reference value corresponding to at least one first bounding box in which the first sub-bounding box is included; and
obtaining a second reference value corresponding to at least one first bounding box in which the first sub-bounding box is not included;
The first reference value is lower than the second reference value. According to claim 6,
The plurality of first bounding boxes are labeled with a first class specifying the first object,
Wherein the first sub-bounding box is labeled with a second class specifying the second object. According to claim 8,
The first class means a person class,
The second class means at least one of a face class, an ear class, and a hand class. According to claim 6,
The object further includes a third object that is part of the first object,
generating a second sub-bounding box corresponding to the third object included in the image;
obtaining a third reference value corresponding to at least one first bounding box in which the second sub-bounding box is included;
obtaining a fourth reference value corresponding to at least one first bounding box in which the second sub-bounding box is not included;
removing the at least one first bounding box whose reliability of the at least one first bounding box does not exceed one of the third reference value and the fourth reference value corresponding to the at least one first bounding box; contain more,
The third reference value is lower than the fourth reference value. An electronic device for detecting an object in an image using an artificial intelligence model,
a memory that stores one or more instructions; and
and at least one processor to execute the one or more instructions stored in the memory, wherein the at least one processor executes the one or more instructions to:
obtaining the image including the object including a first object and a second object that is part of the first object;
Creating a plurality of first bounding boxes corresponding to the first object included in the image;
Creating a first sub-bounding box corresponding to the second object included in the image;
Obtaining a first overlap score related to the degree of overlap between the first sub-bounding box and the first bounding box;
Among the plurality of first bounding boxes, at least one first bounding box having the first overlap score with the first sub-bounding box exceeding a first threshold value is removed;
The electronic device detects the first object through a first bounding box that is not removed among the plurality of first bounding boxes. According to claim 11,
The object further includes a third object that is part of the first object,
The at least one processor, by executing the one or more instructions,
Creating a second sub-bounding box corresponding to the third object included in the image;
Obtaining a second overlap score related to the degree of overlap between the second sub-bounding box and the first bounding box;
Among the plurality of first bounding boxes, at least one first bounding box whose second overlap score with the second sub-bounding box exceeds a second threshold is removed. According to claim 11,
The plurality of first bounding boxes include a first reference bounding box,
The at least one processor, by executing the one or more instructions,
Obtaining a third overlap score related to an overlap between one of the plurality of first bounding boxes excluding the first reference bounding box and the first reference bounding box;
Among the plurality of first bounding boxes, the at least one first bounding box having the third overlap score with the first reference bounding box exceeding a third threshold is removed. According to claim 11,
The plurality of first bounding boxes are labeled with a first class specifying the first object,
The first sub-bounding box is labeled with a second class specifying the second object. According to claim 14,
The first class means a person class,
The second class means at least one of a face class, an ear class, and a hand class. An electronic device for detecting an object in an image using an artificial intelligence model,
a memory that stores one or more instructions; and
and at least one processor to execute the one or more instructions stored in the memory, wherein the at least one processor executes the one or more instructions to:
obtaining the image including the object including a first object and a second object that is part of the first object;
Creating a plurality of first bounding boxes corresponding to the first object included in the image;
Creating a first sub-bounding box corresponding to the second object included in the image;
Obtaining a degree of confidence in a probability that an object included in the inner region of the first bounding box coincides with the first object;
Obtaining a reference value corresponding to the first bounding box according to whether the first sub-bounding box is disposed within the first bounding box;
removing the at least one first bounding box whose reliability of the at least one first bounding box does not exceed the reference value corresponding to the at least one first bounding box;
The electronic device detects the first object through a first bounding box that is not removed from among the plurality of first bounding boxes. According to claim 16,
The reference value includes a first reference value and a second reference value,
The at least one processor, by executing the one or more instructions,
Obtaining a first reference value corresponding to at least one first bounding box in which the first sub-bounding box is included;
Obtaining a second reference value corresponding to at least one first bounding box in which the first sub-bounding box is not included;
The first reference value is lower than the second reference value, the electronic device. According to claim 16,
The plurality of first bounding boxes are labeled with a first class specifying the first object,
The first sub-bounding box is labeled with a second class specifying the second object. According to claim 18,
The first class means a person class,
The second class means at least one of a face class, an ear class, and a hand class. According to claim 16,
The object further includes a third object that is part of the first object,
The at least one processor, by executing the one or more instructions,
Creating a second sub-bounding box corresponding to the third object included in the image;
obtaining a third reference value corresponding to at least one first bounding box in which the second sub-bounding box is included;
Obtaining a fourth reference value corresponding to at least one first bounding box in which the second sub-bounding box is not included;
removing the at least one first bounding box whose reliability of the at least one first bounding box does not exceed one of the third reference value and the fourth reference value corresponding to the at least one first bounding box;
The third reference value is lower than the fourth reference value, the electronic device.

Images