WO2022224639A1

WO2022224639A1 - Object recognition device and method for controlling object recognition device

Info

Publication number: WO2022224639A1
Application number: PCT/JP2022/011803
Authority: WO
Inventors: 翔哉村上
Original assignee: オムロン株式会社
Priority date: 2021-04-23
Filing date: 2022-03-16
Publication date: 2022-10-27
Also published as: JP2022167587A

Abstract

Since face recognition takes a lot of time, a user cannot be put to work early. Provided is a face recognition device (10) comprising: a collation unit (14) which sequentially collates a plurality of partially captured images obtained by extracting the images of predetermined portions of a subject from among captured images of the subject with a plurality of partially registered images which are images of predetermined portions of a recognition target stored in a storage unit (12), according to priorities respectively associated with the partially registered images; and a recognition output unit (15) which outputs a recognition result for the subject at the time point when the collation result of the collation unit satisfies a predetermined condition.

Description

OBJECT RECOGNITION DEVICE, CONTROL METHOD OF OBJECT RECOGNITION DEVICE

The present invention relates to an object recognition device.

Conventionally, there are face recognition systems that recognize the user's face and identify whether the user has the authority to perform a specific action. For example, a face recognition system is used as a device for permitting entrance/exit to an area with high security or as a device for issuing permission to use a specific device.

Patent Literature 1 discloses a technique that enables normal face recognition even when the user wears a mask or changes his or her hairstyle. By emphasizing the feature amount of a certain part and reducing the feature amount of other parts, even if there is a difference from the data registered in the database, the face can be recognized normally.

Patent Document 2 is an invention that recognizes a face when a captured image of the user's face differs from the image registered in the database due to a mask, sunglasses, hat, or the like. According to the above example, when there is a difference between the captured image and the registered image in the database, instead of immediately failing face recognition, the captured image is divided into multiple areas, and the area where matching fails is displayed to the user. It is a technology that makes face recognition successful by notifying and prompting re-recognition.

Japanese Patent Application Laid-Open No. 2002-200000 describes an invention relating to a digital camera, and is a technology that enables priority recognition of a subject by defining the priority of the subject for each shooting mode, and performs optical adjustment according to the subject at an early stage. be.

Japanese Patent Laid-Open No. 2004-100002 recognizes a similar pattern for each region divided into a plurality of regions for a pattern image of a living body such as a vein pattern, and saves a number corresponding to the pattern for each region, thereby obtaining a pattern image of the living body. It is an invention that reduces the capacity of

In Patent Document 5, face recognition is performed using a low-resolution image, and if the image is determined to be the same image, face recognition is performed using a portion of a higher-resolution image, thereby gradually increasing the speed while ensuring speed. It is an invention that improves the accuracy of face recognition.

Japanese Patent Application Laid-Open No. 2007-257221 Japanese Patent Application Laid-Open No. 2019-168929 Japanese Patent Application Laid-Open No. 2007-068149 Japanese Patent Application Laid-Open No. 2011-065315 Japanese Patent Application Laid-Open No. 2012-238059

However, in the above-described conventional technology, face recognition takes time due to the size of the face to be matched in face recognition or the large number of face recognition targets. Therefore, the user cannot start work early, which causes stress on the user.

With conventional algorithms, it is possible to speed up processing by roughening the resolution of captured images, but due to the rough resolution, erroneous recognition may occur. In addition to face recognition, there is also demand for object recognition.

Therefore, an object of one aspect of the present invention is to realize a device that can perform object recognition at high speed while maintaining recognition accuracy.

To solve the above problems, an object recognition apparatus according to an aspect of the present invention includes an image acquisition unit that acquires a captured image of a target object; an image extracting unit that creates a plurality of partially captured images obtained by extracting the images of the above, a plurality of registered partial images that are obtained by extracting images of a plurality of predetermined portions of a registered object as a recognition target stored in a storage unit; a matching unit that performs matching with the partially captured images in order according to the priority levels associated with the respective registered partial images; a recognition output unit that outputs a recognition result.

An object recognition apparatus according to another aspect includes an image acquisition step of acquiring a captured image of a target object; an image extracting step for creating; a plurality of partial registration images obtained by extracting images of a plurality of predetermined portions of a registered object as a recognition target stored in a storage unit; and a recognition output step of outputting the recognition result of the target object when the matching in the matching unit satisfies a predetermined condition. include.

According to one aspect of the present invention, a partially captured image extracted from a captured image and a previously registered partially registered image are collated with a priority associated with the partially registered image, and the collation result is a predetermined value. Recognition can be terminated early when the conditions of

2 is a block diagram showing the configuration of the main parts of the face recognition system according to Embodiment 1; FIG. 3A and 3B are model diagrams showing a captured image and a partially captured image according to the first embodiment; FIG. 10 is a flow chart showing a flow of registering a partial registration image; 5 is a flow chart showing the flow of operation of operation example 1 according to the first embodiment. 9 is a flow chart showing the flow of operation of operation example 2 according to the first embodiment; 9 is a flow chart showing the flow of operation of operation example 3 according to the first embodiment; 10 is a flow chart showing the flow of operation of operation example 4 according to the first embodiment; FIG. 11 is a block diagram showing the configuration of the main part of the face recognition system according to Embodiment 2; 13A and 13B are model diagrams showing a captured image and a partially captured image according to Embodiment 3; FIG.

[Embodiment 1]
Hereinafter, an embodiment (hereinafter also referred to as "this embodiment") according to one aspect of the present invention will be described based on the drawings. The same or corresponding parts in the drawings are denoted by the same reference numerals, and the description thereof will not be repeated.

§1. APPLICATION EXAMPLE FIG. 1 is a block diagram showing the configuration of a main part of a face recognition system 100 according to the first embodiment. The face recognition system 100 includes a camera 20 that captures a captured image including a face, a face recognition device (object recognition device) 10 that performs face recognition on the captured image, and an electromagnetic lock 30 that is controlled by the face recognition device 10. .

The face recognition device 10 extracts a partial captured image corresponding to a pre-registered partial registration image from the captured image, and performs matching by comparing the partial captured image and the partial registered image. Here, a plurality of partial registration images are prepared, and a priority is set for each partial registration image. Collation with respect to each partial registration image is performed in the order according to the priority.

The face recognition device 10 can be broadly divided into two modes: a mode that prioritizes personal recognition speed for recognizing that the target person for face recognition is a registrant, and a mode that recognizes that the target person for face recognition is not a registrant. This is a mode that gives priority to other person recognition speed. In the mode that prioritizes personal recognition speed, when a pair of a set of partially captured image and partially registered image is recognized as the same image, it is recognized as substantially the same face. On the other hand, in the mode that prioritizes other person recognition speed, when a pair of a set of partially captured images and partially registered images can be recognized as different images, it is recognized that they are not substantially the same face. When recognizing substantially the same face, the face recognition device 10 releases the electromagnetic lock 30 .

That is, the face recognition device 10 performs matching in the order according to the priority, and can output the result of face recognition when the recognition result of one set of the partially captured image and the partially registered image is obtained. Therefore, the recognition result can be output earlier than when face recognition is performed on the entire captured image.

§2. Configuration example (configuration of face recognition device)
The configuration of the face recognition system 100 will be described based on FIG. The face recognition device 10 includes an image acquisition section 11 , a storage section 12 , an image extraction section 13 , a matching section 14 and a recognition output section 15 .

The camera 20 is a camera that captures a captured image including the face of the target for face recognition. The camera 20 outputs the captured image to the image acquisition section 11 .

The electromagnetic lock 30 is an electronic lock controlled by the face recognition device 10. The electromagnetic lock 30 is not limited to an electronic lock, but may be any program-controlled device or program, and the function of the device or program cannot be used unless it is unlocked.

For example, if the door is provided with an electromagnetic lock 30, the door cannot be opened to enter the interior unless the electromagnetic lock 30 is unlocked by face recognition. Further, in an electronic device such as a smart phone that incorporates a camera 20, the smart phone does not accept an operation unless the electromagnetic lock 30 is unlocked (released) by face recognition.

The image acquisition unit 11 inputs the captured image. The image acquisition unit 11 outputs the input captured image to the image extraction unit 13 .

The storage unit 12 stores a plurality of partial registration images for each registrant who is a target of face recognition. In addition to each partial registration image, the storage unit 12 also stores the priority, and the order of performing face recognition on the partial registration images is determined. The partial registration image and priority information are stored in a storage unit outside the face recognition device 10, and even if the face recognition device 10 reads out the partial registration image and priority information through various communication means, good.

The image extracting unit 13 extracts a portion corresponding to the partial registered image from the input captured image and sets it as a partial captured image. The image extraction unit 13 outputs the plurality of extracted partial captured images to the matching unit 14 .

The matching unit 14 compares the input partial captured image and the partial registered image, and calculates a numerical value called similarity indicating how similar they are to each other. After that, the matching unit 14 compares the degree of similarity with a predetermined threshold to determine whether the images are substantially the same. Also, the partially registered image of a certain priority and the partially captured image corresponding to the partially registered image are collated, and if necessary, the partially registered image of the next priority and the partially captured image corresponding to the partially registered image are matched. Perform matching processing. Collation necessary for face recognition is performed, and when the collation result is determined, the collation unit 14 outputs the collation result to the recognition output unit 15 .

The recognition output unit 15 receives the collation result input and determines whether or not to issue a signal to unlock the electromagnetic lock 30 . The recognition output unit 15 unlocks the electromagnetic lock 30 when the matching result indicates that the target person for face recognition is the registrant. On the other hand, the recognition output unit 15 does not unlock the electromagnetic lock 30 when the matching result indicates that the target person for face recognition is not the registrant.

(Extraction of partial captured image from captured image)
FIG. 2 is a model diagram showing a captured image 111 and a partially captured image 112 according to the first embodiment. The captured image 111 is, as shown in FIG. 2, an image that captures the entire face of the target person to be subjected to face recognition. The partial captured images 112 are, for example, partial images of the five captured images 111, and are partial captured images 112a to 112e. Each partial captured image is an image obtained by partially capturing a part representing features of the face.

For example, the partially captured image 112a is an image of both eyes, the partially captured image 112b is a mouth, the partially captured image 112c is a hair, the partially captured image 112d is a nose, and the partially captured image 112e is an image of a forehead. The ranges of the partial captured images 112 (the partial captured images 112a to 112e are collectively referred to as the partial captured images 112) may overlap each other. For example, in the example of FIG. 2, both the partially captured image 112a and the partially captured image 112e include eyebrows.

The partial captured image 112 is extracted from the captured image 111 by the image extraction unit 13, and the extracted range is the range corresponding to the partial registered image.

That is, the image extracting unit 13 extracts from the captured image 111 the partial captured images 112 respectively corresponding to the registered partial images. At this time, in order to extract the partial captured image 112 , pattern matching may be performed on the captured image 111 to derive the range of the partial captured image 112 .

Also, the method of extracting the partial captured image 112 is not limited to this method, and the position of the partial area to be extracted from the captured image 111 may be determined in advance, and the corresponding partial area may be extracted as the partial captured image 112 .

(Registration method of partially registered image)
FIG. 3 is a flow chart showing the flow of registering a partial registration image. In order to perform face recognition, it is necessary to preliminarily register partial registration images obtained by dividing a face image into parts.

In S11, the camera 20 captures the face image of the registrant to be registered. In S12, the captured face image is extracted for each part of the face. The parts of the face include the eyes, mouth, nose, eyebrows, forehead, hair, etc., and may be extracted as an image that captures the periphery in addition to each part. This operation is performed by the user operating the face recognition device 10, or automatically performed by the face recognition device 10 according to a predetermined rule.

In S13, each divided part is registered in the storage unit 12 as a partial registered image. In S14, a priority, which is the order in which face recognition is performed, is assigned to each partial registration image. Here, priority assignment may be performed manually by the user, or may be performed automatically by the face recognition device 10 according to a predetermined rule.

§3. Operation Examples Hereinafter, four types of operation examples will be described with specific examples.

(Operation example 1: Personal recognition speed priority for one registrant)
FIG. 4 is a flowchart showing the operation flow of operation example 1 according to the first embodiment. Operation example 1 is an operation in which, when a partially registered image to be subjected to face recognition is a face image of one person, priority is given to speed of person recognition for a registrant of the partially registered image. To simplify the explanation, it is assumed that there are five partial registration images.

In S21, the camera 20 captures a face image, and the image acquisition unit 11 acquires the face image as a captured image.

In S22, the recognition output unit 15 performs a matching loop for matching the partially registered images for each part from i=1 to i=5. Here, i is a natural number indicating the priority of the partially registered image.

In S23, the image extraction unit 13 extracts the partial captured image corresponding to the partial registered image with i-th priority.

In S24, the collation unit 14 compares the partial captured image and the partial registered image with the i-th priority for collation. If the partial captured image and the partial registered image substantially match, the process proceeds to S25. If the partial captured image and the partial registered image do not substantially match, the process proceeds to S26.

In S25, the matching unit 14 recognizes that the target person is the same person as the single registrant because the partial captured image and the partial registered image substantially match, and exits the matching loop. After that, the recognition output unit 15 unlocks the electromagnetic lock 30 .

In S26, when i=5, the collation unit 14 exits the collation loop or increments i=i+1 to proceed to the next collation loop.

In S27, the recognition output unit 15 recognizes that the target person is different from the registrant because the target person is not the same person as the registrant. Therefore, the recognition output unit 15 does not unlock the electromagnetic lock 30 .

According to the above processing, when it is determined that one partial captured image substantially matches the partial registered image, it is possible to quickly recognize that the target person is the same person as the registered person. Also, in order to ensure security, it is preferable to set a rather strict similarity threshold for judging that the partially captured image and the partially registered image substantially match each other. By setting the threshold to be rather strict, it is possible to suppress the possibility that a target person other than the registrant is recognized as the same person by face recognition.

Here, in S24, conditional branching is performed depending on whether or not one partial captured image substantially matches the partial registered image, and if the condition is met (Yes in S24), the electromagnetic lock 30 is released (S25 ) does not necessarily have to be canceled based on the result of recognition based on one partially captured image. That is, when a predetermined number of partially captured images substantially match the partially registered images, it may be recognized that the subject is the same person as the single registrant. Security is improved by increasing the number of partially registered images required to determine that the person is the same person. Here, it is preferable that the number of partial registration images necessary for determining that the person is the same person is smaller than the number of partial registration images stored in the storage unit 12 . This makes it possible to output matching results at a higher speed than when matching with all partially registered images.

As an actual example of operation example 1, it is preferably used for face recognition with a device that is basically used only by registrants. For example, in an electronic device such as a smartphone owned by an individual, there is a case where face recognition is used as user recognition.

(Operation example 2: Priority given to other person recognition speed for a single registrant)
FIG. 5 is a flowchart showing the operation flow of Operation Example 2 according to the first embodiment. Operation example 2 is an operation that gives priority to the other person recognition speed for recognizing that the person is not the registrant of the partial registration image when the partial registration image to be subjected to face recognition is the face image of one person. To simplify the explanation, it is assumed that there are five partial registration images.

The processing of S21 to S23 is the same as that of operation example 1, so the description is omitted.

In S34, the collation unit 14 compares the partial captured image and the partial registered image with the i-th priority for collation. If the partial captured image and the partial registered image substantially match, the process proceeds to S36. If the partial captured image and the partial registered image do not substantially match, the process proceeds to S35.

In S35, the matching unit 14 recognizes that the target person is different from the single registrant because the partial captured image and the partial registered image do not substantially match, and exits the matching loop. After that, the recognition output unit 15 does not unlock the electromagnetic lock 30 .

In S36, if i=5, the matching unit 14 exits the matching loop and proceeds to S37, or the matching unit 14 increments i=i+1 and moves to the next matching loop.

In S37, the recognition output unit 15 recognizes that the target person is the same person as the registrant because the partial captured images corresponding to all the partial registration images substantially match. Therefore, the recognition output unit 15 unlocks the electromagnetic lock 30 .

According to the above processing, when it is determined that one partial captured image does not substantially match the partial registered image, it is possible to quickly recognize that the target person is not the same person as the registered person. In addition, in order to suppress erroneous determination that a target person is not a registrant when the target person is a registrant, the similarity threshold for judging that the partially captured image and the partially registered image match each other is set loosely. It is preferable to keep Due to the loose threshold setting, even a subject who is not a registrant may be judged to match substantially in matching per part, but by using a large number of parts in a matching loop, as a result, registration A target person who is not a person can be output as a person different from the registrant.

Here, in S34, conditional branching is performed depending on whether or not one partial captured image substantially matches the partial registered image, and if the condition is not satisfied (No in S34), the recognition process is terminated. , the process does not necessarily end with a recognition result based on one partially captured image. That is, if a predetermined number of partially captured images do not match the partially registered images, it may be recognized that the target person is not the same person as the registered person. Misjudgment can be suppressed by increasing the number of necessary partial registration images. Here, it is preferable that the number of partially registered images required to determine that the person is not the same person is smaller than the number of partially registered images stored in the storage unit 12 . This makes it possible to output matching results at a higher speed than when matching with all partially registered images.

As an actual example of operation example 2, it is preferably used to prevent misuse by others of a device that is basically used only by a registrant. For example, when an electronic device such as a smartphone owned by an individual is lost, it is used to improve security for unlocking by face authentication.

(Operation example 3: Personal recognition speed priority for multiple registrants)
FIG. 6 is a flow chart showing the operation flow of operation example 3 according to the first embodiment. Operation example 3 is an operation that prioritizes the person recognition speed for the person in the partial registration image when facial images of a plurality of persons are registered in the partial registration image to be subjected to face recognition. In order to simplify the explanation, it is assumed that there are three registrants and that each registrant has five partially registered images.

In S41, the camera 20 captures a face image, and the image acquisition unit 11 acquires the face image as a captured image.

In S42, the matching unit 14 performs a matching loop for matching the partially registered images for each part from i=1 to i=5. Here, i is a natural number indicating the priority of the partially registered image.

In S43, the matching unit 14 performs a registrant loop that performs matching for each registrant from j=1 to j=3. Here, j is a natural number representing a registrant.

In S44, the image extraction unit 13 extracts the partial captured image corresponding to the partial registered image with the i-th priority for the j-th registrant.

In S45, the collation unit 14 compares the partially captured image with the partially registered image of the i-th priority for the j-th registrant to perform collation. If the partial captured image and the partial registered image substantially match, the process proceeds to S46. If the partial captured image and the partial registered image do not substantially match, the process proceeds to S47.

In S46, the matching unit 14 recognizes that the target person is the same person as one of the registrants because the partial captured image and the partial registered image substantially match, and exits the matching loop. After that, the recognition output unit 15 unlocks the electromagnetic lock 30 .

In S47, when j=3, the matching unit 14 exits the registrant loop, or increments j=j+1 to move to the next registrant loop.

In S48, when i=5, the collation unit 14 exits the collation loop or increments i=i+1 to proceed to the next collation loop.

In S49, the recognition output unit 15 recognizes that the target person is different from the multiple registrants because the target person does not substantially match all of the multiple registrants. Therefore, the recognition output unit 15 does not unlock the electromagnetic lock 30 .

According to the above-described processing, when it is determined that one partially captured image substantially matches the partially registered image of any one of a plurality of registrants, the target person is the registrant. It recognizes that it is the same person as one of them, and terminates the face recognition. Therefore, the target person's face can be recognized early. Also, in order to ensure security, it is preferable to set a rather strict similarity threshold for judging that the partially captured image and the partially registered image substantially match each other. By setting the threshold to be rather strict, it is possible to suppress the possibility that a target person who is not a plurality of registrants will be recognized as the same person as one of the plurality of registrants by face recognition.

As an actual example of operation example 3, it is a device that is less likely to be matched by a target person other than a registrant, and is preferably used when face recognition is performed by a device used by multiple registrants. For example, it can be used for the purpose of allowing registrants to enter an entrance gate that is basically expected to be entered by related parties.

(Operation example 4: Priority given to other person recognition speed for multiple registrants)
FIG. 7 is a flowchart illustrating the operation flow of Operation Example 4 according to the first embodiment. Operation example 4 is an operation that prioritizes the recognition speed of others other than the person in the partial registration image when facial images of a plurality of persons are registered in the partial registration image to be subjected to face recognition. In order to simplify the explanation, it is assumed that there are three registrants and that each registrant has five partially registered images.

The processing of S41 to S44 is the same as that of Operation Example 3, so the description is omitted.

In S55, the collation unit 14 compares the partially captured image with the partially registered image of the i-th priority for the j-th registrant to perform collation. If the partial captured image and the partial registered image do not substantially match, the process proceeds to S56. If the partial captured image and the partial registered image substantially match, the process proceeds to S57.

In S56, the matching unit 14 determines that the target person is not the j-th registrant. Therefore, in subsequent processing, the j-th registrant is excluded from candidates, and subsequent verification is not performed.

In S57, if j=3, the matching unit 14 exits the registrant loop and proceeds to S58, or increments j=j+1 and moves to the next registrant loop.

In S58, the collation unit 14 determines whether there are any candidate registrants remaining among the plurality of registrants. If there are no candidate registrants left, the process proceeds to S59. If candidate registrants remain, the process proceeds to S60.

In S59, the matching unit 14 recognizes that the target person is a different person from any one of the plurality of registrants because there is no partially registered image that substantially matches the partially captured image, and performs a matching loop. go through. After that, the recognition output unit 15 does not unlock the electromagnetic lock 30 .

In S60, if i=5, the collation unit 14 exits the collation loop and proceeds to S61, or increments i=i+1 and moves to the next collation loop.

In S61, the recognition output unit 15 recognizes that the target person is the same person as one of a plurality of registrants because there is a registrant whose partial captured images corresponding to all of the partial registration images substantially match. do. Therefore, the recognition output unit 15 unlocks the electromagnetic lock 30 .

According to the above-described processing, when it is determined that one partially captured image does not substantially match the partially registered image of any one of the plurality of registrants, the subject becomes the registrant. Since the target person can be recognized as different from the target person, the target person can be recognized as a stranger at an early stage. However, in order not to erroneously recognize a registrant as someone else, it is preferable to loosely set a similarity threshold for judging that the partial captured image and the partially registered image substantially match each other. Since the threshold setting is loose, even a subject who is not a registrant may be judged to match substantially in matching per part, but by using a large number of parts in a matching loop, as a result, the registrant It is possible to output a non-existent target person as a person different from the registrant.

In addition, the partially captured image and the partially registered image are collated, and when the collation does not substantially match, the registrant is excluded from multiple registrant candidates. Therefore, by proceeding with collation, the number of candidates for multiple registrants is narrowed down. As a result, when the target person is the same person as any one of a plurality of registrants, even when the electromagnetic lock is unlocked, it can be unlocked faster than when collating by round-robin.

As an actual example of operation example 4, it is a device that is highly likely to be matched by a target person other than a registrant, and is preferably used when face recognition is performed by a device used by multiple registrants. For example, at an entrance gate where an unspecified number of people are expected to enter, it is conceivable that only registrants are permitted to enter. It is used as a high-security electromagnetic lock for entry/exit gates because it is determined that the person is the same person when all parts are substantially matched.

§4. Actions/Effects The matching unit can determine whether or not the partial registered image and the partial captured image corresponding to the partial registered image substantially match from the calculated degree of similarity.

If there is only one registrant, and if priority is given to the person's recognition speed, the target person is identified as the registrant when a combination in which at least one partial registered image and a partially captured image substantially match is obtained. recognize as a person. On the other hand, when priority is given to the recognition speed of others, it is recognized that the target person is not the same person as the registrant when a combination in which at least one partially registered image and the partially captured image substantially match is obtained. do.

In addition, when there are multiple registrants, when priority is given to the recognition speed of the person, when a combination in which at least one partial registered image and a partially captured image are substantially matched is obtained, the target Recognize the person as the same person as the registrant. On the other hand, when priority is given to the recognition speed of others, the target person is identified as the registered person at the time when a combination of the partially registered image and the partially captured image of at least one part that do not substantially match is obtained. It is determined that the person is not a person, and in subsequent processing, face recognition is performed after removing the registrant from candidates.

As described above, the result of face recognition can be influenced by image recognition of a certain part. Therefore, face recognition can be terminated early.

[Embodiment 2]
In the first embodiment, a method for speeding up face recognition has been described. On the other hand, in the second embodiment, a method of optimizing the priority of partially registered images in order to speed up face recognition will be described.

FIG. 8 is a block diagram showing the configuration of the essential parts of the face recognition system 100a according to the second embodiment. The face recognition system 100a is different from the face recognition system 100 in that the face recognition device 10a is provided with a priority setting unit 16. FIG.

The priority setting unit 16 is a functional block that optimizes the priority of each partial registration image based on the similarity history of past face recognition results. The priority setting unit 16 first learns the similarity tendency based on the similarity history for each partially registered image. As an example, the priority setting unit 16 calculates the average similarity based on the similarity history. When the recognition speed of the person is given priority (operation examples 1 and 3), the priority setting unit 16 assigns priority to each partial registration image in descending order of average similarity. As a result, partially registered images with high priority tend to have a high degree of similarity, and face recognition can be completed early. Similarly, when priority is given to other people's recognition speed (operation examples 2 and 4), the priority setting unit 16 assigns priorities in descending order of average similarity. As a result, a partially registered image with a high priority tends to have a low degree of similarity to others, and face recognition can be completed early.

Also, the priority optimization process may be performed at any time while the face recognition device 10a is not performing the face recognition process, so that the face recognition can always be performed with the optimum priority.

Furthermore, the priority optimization in the priority setting unit 16 may not be performed automatically, and may be manually set in any order of priority by user operation. In this case, the priority setting unit 16 causes the display device provided in the face recognition device 10a to display a list of partial registration images and the images, and the input device provided in the face recognition device 10a allows the user to set the priority of each partial registration image. Setting inputs may be accepted. Alternatively, priority setting information received from a user in an external terminal device may be obtained and set by the priority setting unit 16 via communication means.

As a specific example of assigning priority, for example, since the eyes are a part with strong human characteristics, the priority of the eyes is set to be higher and the mask is often masked, so the priority of the mouth is set lower. In addition, it is possible to set priorities according to one's own characteristics and preferences. For example, it is possible to set a low priority for the eyes because the user often wears sunglasses, or set a low priority for the hair because the user often changes hairstyles. It becomes possible.

[Embodiment 3]
Although the first embodiment focuses on face recognition, the object to be recognized is not limited to the face. In the third embodiment, object recognition using an object as a recognition target will be described.

FIG. 9 is a model diagram showing a captured image 113 and a partially captured image 114 according to the third embodiment. The captured image 113 is an image of an arbitrary object, and may be an image of an automobile as shown in FIG. 9, for example. The partial captured images 114 are, for example, partial captured images 114a to 114d of the four captured images 113, respectively. Each partial captured image represents a feature of the object.

For example, the partially captured image 114a is the side glass, the partially captured image 114b is the windshield, the partially captured image 114c is the headlight, and the partially captured image 114d is the tire. The ranges of the partial captured images 114 (the partial captured images 114a to 114d are collectively referred to as the partial captured images 114) may overlap each other. The method of creating the partial captured image 114 and the method of assigning priority to the partial captured image 114 are the same as those in the flow chart of the face recognition operation shown in FIG.

Also, the flowchart of the operation in object recognition according to the third embodiment is the same as the flowchart of the operation in face recognition shown in FIGS. By the same method as described above, the target object is imaged to obtain a captured image, a partial captured image corresponding to the partially registered image of the registered object is extracted from the captured image, and the partially registered image and the partially captured image are compared.

When priority is given to the speed of recognizing a registered object after receiving the collation result, as shown in the above-described operation example 1, if the collation results in at least one part substantially match, the target object is the same object as the registered object. On the other hand, when priority is given to the speed at which the target object is recognized as not being a registered object, as shown in the above-described operation example 2, the matching result at least at one part does not substantially match, and the target object is recognized as a registered object. is not the same object as

Also, the case where there are multiple registered objects will be explained. When priority is given to the speed of recognition as one of the registered objects, as shown in the above-described operation example 3, in matching the parts for each priority, substantially match any one of a plurality of registered objects. , the target object is recognized as the same object as one of the registered objects. When priority is given to the speed of recognizing none of the registered objects, a plurality of registered objects do not substantially match the target object in the region matching process for each priority, as shown in the operation example 4 described above. As a result, the registered object is removed from the candidates, and when there are no more candidates, it is recognized that the target object is not the same object as all of the multiple registered objects.

〔summary〕
To solve the above problems, an object recognition apparatus according to an aspect of the present invention includes an image acquisition unit that acquires a captured image of a target object; an image extracting unit that creates a plurality of partially captured images obtained by extracting the images of the above, a plurality of registered partial images that are obtained by extracting images of a plurality of predetermined portions of a registered object as a recognition target stored in a storage unit; a matching unit that performs matching with the partially captured images in order according to the priority levels associated with the respective registered partial images; a recognition output unit that outputs a recognition result.

According to the above configuration, matching is repeated until a predetermined condition is met between the partially captured image and the partially registered image in the order of the priority associated with the partially registered image, so that the target object becomes the registered object. It can be determined whether or not Since the recognition result is output when the predetermined condition is satisfied, the recognition result of the target object can be output at an early stage.

The collation unit may perform collation by calculating a degree of similarity between the partial registered image and the partial captured image and comparing it with a predetermined threshold.

According to the above configuration, it is possible to match the partially registered image and the partially captured image by comparing the degree of similarity with the threshold.

When the collation unit performs collation in order according to the priority and determines that the partial registered image substantially matches the partial captured image corresponding to the partial registered image, the recognition output unit may output the recognition result of the target object.

According to the above configuration, when the matching unit determines that there is a partially captured image that substantially matches a partially registered image, it is possible to output a recognition result that the target object is a registered object. .

The storage unit stores the partial registration images and the priorities of the partial registration images with respect to a plurality of the registered objects, and the matching unit performs the partial registration of all the registered objects at a certain priority. After performing matching on the images, matching may be performed on the partially registered images of all the registered objects in the next priority.

According to the above configuration, with respect to all registered objects, partial captured images are collated in order of priority, and when it is determined that they substantially match, the target object is recognized as a registered object. can be output.

The collation unit performs collation in the order according to the priority, and when it is determined that the partial registered image and the partial captured image corresponding to the partial registered image do not substantially match, the recognition output. A unit may output a recognition result of the target object.

According to the above configuration, when the matching unit determines that there is a partial captured image that does not substantially match the partial registered image, it is possible to output a recognition result that the target object is not a registered object. .

The storage unit stores the partial registration images and the priorities of the partial registration images with respect to a plurality of the target objects, and the matching unit performs the partial registration of all the registration objects at a certain priority. After matching the images, matching is performed with respect to the partial registered images of all the registered objects at the next priority, and the partial registered images and the partial captured images corresponding to the partial registered images are substantially If it is determined that they do not match, it is not necessary to collate the corresponding registered object at the next priority.

According to the above configuration, with respect to all registered objects, partial captured images are collated in order of priority, and when it is determined that all registered objects do not substantially match, the target object is not a registered object. It is possible to output the recognition result that there is no

The matching unit may further include a priority setting unit that performs a process of storing the similarity calculation result as a similarity history, and sets a priority of the partially registered image based on the similarity history. .

According to the above configuration, it is possible to automatically set the priority of the partially registered images according to the similarity history. Therefore, by setting the priority of the partial registered images according to the similarity tendency, it is possible to quickly output the recognition result as to whether or not the target object is the registered object.

A priority setting unit may be further provided for setting the priority of the partial registration image based on the user's input.

According to the above configuration, the priority of partially registered images can be arbitrarily set based on user input. Therefore, it is possible to change the priority of the part whose priority is to be intentionally lowered or the priority of the part to be intentionally raised.

The target object may be a person's face.

According to the above configuration, for example, the object may be a person's face, which can be used for early face recognition.

The object recognition device according to each aspect of the present invention may be implemented by a computer. In this case, the object recognition device is implemented by the computer by operating the computer as each part (software element) provided in the object recognition device. An object recognition program for an object recognition device realized by a computer, and a computer-readable recording medium recording it are also included in the scope of the present invention.

[Example of realization by software]
The functions of the

face recognition devices

10 and 10a and the object recognition device (hereinafter referred to as "apparatuses") are programs for causing a computer to function as the devices. 10a) can be implemented by a program for causing a computer to function.

In this case, the device comprises a computer having at least one control device (eg processor) and at least one storage device (eg memory) as hardware for executing the program. Each function described in each of the above embodiments is realized by executing the above program using the control device and the storage device.

The above program may be recorded on one or more computer-readable recording media, not temporary. The recording medium may or may not be included in the device. In the latter case, the program may be supplied to the device via any transmission medium, wired or wireless.

Also, part or all of the functions of the above control blocks can be realized by logic circuits. For example, integrated circuits in which logic circuits functioning as the control blocks described above are formed are also included in the scope of the present invention. In addition, it is also possible to implement the functions of the control blocks described above by, for example, a quantum computer.

Also, each process described in each of the above embodiments may be executed by AI (Artificial Intelligence). In this case, the AI may operate on the control device, or may operate on another device (for example, an edge computer or a cloud server).

[Additional notes]
The present invention is not limited to the above-described embodiments, but can be modified in various ways within the scope of the claims, and can be obtained by appropriately combining technical means disclosed in different embodiments. is also included in the technical scope of the present invention.

10, 10a face recognition device 12 storage unit 13 image extraction unit 14 matching unit 15 recognition output unit 16 priority setting unit 20 camera 30

electromagnetic lock

100, 100a face recognition system 111, 113 captured image 112, 112a to 112e, 114, 114a ~114d Partial image

Claims

an image acquisition unit that acquires a captured image of a target object;
an image extraction unit that creates a plurality of partial captured images by extracting images of predetermined portions of the target object from the captured image;
matching of a plurality of partial registration images obtained by extracting images of a plurality of predetermined portions of a registered object as a recognition target stored in a storage unit and the partial captured image is associated with each of the partial registration images; a matching unit that performs in order according to the priority
and a recognition output unit that outputs a recognition result of the target object when the collation in the collation unit satisfies a predetermined condition.
The object recognition device according to claim 1, wherein the collation unit performs collation by calculating a degree of similarity between the partially registered image and the partially captured image and comparing it with a predetermined threshold value.
When the collation unit performs collation in order according to the priority and determines that the partial registered image substantially matches the partial captured image corresponding to the partial registered image, the recognition output unit 3. The object recognition device according to claim 1, wherein outputs the recognition result of the target object.
the storage unit stores the partial registration image and the priority of the partial registration image with respect to a plurality of the registered objects;
4. The collating unit according to claim 3, wherein after collating the partially registered images of all the registered objects in a certain priority, the collating unit collates the partially registered images of all the registered objects in the next priority. An object recognition device as described.
The collation unit performs collation in the order according to the priority, and when it is determined that the partial registered image and the partial captured image corresponding to the partial registered image do not substantially match, the recognition output. 2. The object recognition device according to claim 1, wherein a unit outputs a recognition result of the target object.
The storage unit stores the partial registration images and the priority of the partial registration images with respect to the plurality of target objects,
The matching unit performs matching of the partially registered images of all the registered objects at a certain priority level, and then performs matching of the partially registered images of all the registered objects at the next priority level,
6. The method according to claim 5, wherein when it is determined that the partial registered image and the partial captured image corresponding to the partial registered image do not substantially match, matching of the corresponding registered object at the next priority is not performed. Object recognition device.
The collation unit performs a process of storing the calculation result of the similarity as a similarity history,
3. The object recognition apparatus according to claim 2, further comprising a priority setting unit that sets priority of the partially registered images based on the similarity history.
The object recognition device according to any one of claims 1 to 6, further comprising a priority setting unit that sets the priority of the partial registration image based on user input.
The object recognition device according to any one of claims 1 to 8, wherein the target object is a person's face.
an image acquisition step of acquiring a captured image of the target object;
an image extracting step of creating a plurality of partial captured images obtained by extracting images of predetermined portions of the target object from the captured image;
matching of a plurality of partial registration images obtained by extracting images of a plurality of predetermined portions of a registered object as a recognition target stored in a storage unit and the partial captured image is associated with each of the partial registration images; a matching step performed in order according to the priority of
and a recognition output step of outputting a recognition result of the target object when the matching in the matching step satisfies a predetermined condition.
2. An object recognition program for causing a computer to function as the object recognition device according to claim 1, wherein the computer functions as the image acquisition unit, the image extraction unit, the storage unit, the collation unit, and the recognition output unit. Object recognition program for.