Control device, control method, and program

Abstract

To reduce the possibility that a situation in which privacy, human rights or the like will be infringed.SOLUTION: A control device includes acquisition means which acquires an image including a person, estimation means which performs estimation processing for estimating information on the person from the image, and output means which outputs a result of the estimation processing in the estimation means. The control device further includes determination means which determines a use state of the estimation processing in the estimation means, and control means which restricts at least one of the estimation processing in the estimation means and output from the output means on the basis of a determination result of the use state from the determination means.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a control technology for estimation processing.

Background Art

[0002] In recent years, many technologies for highly processing images to extract useful information have been proposed. In particular, among them, technologies for recognizing a person's face using a multi-layer neural network called a deep net (or also referred to as a deep neural net or deep learning) have been actively researched and developed. Further, the face recognition technology using a deep net is also used for person authentication that compares an input face image with a pre-registered face image and determines whether the input face image is of the same person as the registered face image. Furthermore, the technology for extracting information from an image using a deep net is not limited to application to face recognition, and research for estimating a person's emotion from a face image is also progressing.

[0003] Further, Patent Document 1 discloses a technology that enables execution of both card authentication for identifying a person and face authentication for identifying a person based on the person's face image information, and enables the face authentication to be arbitrarily enabled or disabled by an operation on a prohibition button. Furthermore, Patent Document 2 discloses a technology for collating data that is difficult to visually identify a person image extracted from a captured image with registered person data pre-registered for a detection target person to collate whether or not the subject of the data that is difficult to visually identify is the detection target person.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0005] On the other hand, it can also be said that various estimation processes including face recognition technology are technologies that may infringe on people's privacy, human rights, etc. Therefore, when using the estimation process, it is necessary to give full consideration to privacy and human rights violations. In addition, there is a discussion that it is desirable to impose a certain level of restrictions on the use of various estimation processes including face recognition technology. For example, in the EU (European Union), regarding AI (artificial intelligence) that conducts large-scale surveillance activities using face recognition technology in public places, a law to prohibit its use is being considered. However, after the face recognition technology is sold, etc., it is not possible to confirm how the technology is being used, so it may be used in a situation that infringes on people's privacy, human rights, etc.

[0006] Therefore, an object of the present invention is to reduce the possibility of a situation occurring that infringes on privacy, human rights, etc.

Means for Solving the Problem

[0007] The control device of the present invention includes an acquisition means for acquiring an image including a person, an estimation means for executing an estimation process for estimating information of the person from the image, an output means for outputting a result of the estimation process by the estimation means, a determination means for determining a usage situation of the estimation process by the estimation means, and a control means for restricting at least one of the estimation process by the estimation means and the output by the output means based on a determination result of the usage situation by the determination means.

Effects of the Invention

[0008] According to the present invention, the possibility of a situation occurring that infringes on privacy, human rights, etc. can be reduced.

Brief Description of the Drawings

[0009]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Embodiments for Carrying Out the Invention

[0010] Hereinafter, embodiments of the present invention will be described with reference to the drawings. Each of the following embodiments does not limit the present invention, and not all combinations of the features described in this embodiment are essential for the solution means of the present invention. The configuration of the embodiment can be appropriately modified or changed according to the specifications of the device to which the present invention is applied and various conditions (usage conditions, usage environment, etc.). Also, in the following embodiments, the same or similar configurations and processing steps are denoted by the same reference numerals, and redundant explanations are omitted.

[0011] <First Embodiment> In the first embodiment, as an application example of a control device that performs person authentication (biometric authentication) using face recognition technology, an AI (artificial intelligence) system will be described. In this embodiment, person authentication using face recognition technology refers to a person authentication method in which a face image is acquired as biometric information, information (biometric identifier) representing an individual is extracted from the face image, and the person identification information (person ID) of that person is specified using this information.

[0012] However, depending on how face recognition technology is used, there is a risk of privacy infringement, human rights infringement, etc. For this reason, in the AI system which is an application example of the control device according to this embodiment, by estimating the usage situation of face recognition technology, it is determined whether or not the technology is being used in a way that causes such risks. And in the AI system of this embodiment, when it is determined that the technology is being used in a way that causes risks, control is performed to limit at least one of face recognition processing and the output of face recognition processing results. In this embodiment, for example, when face recognition technology is used in a public place and it is determined that there is a risk of privacy infringement or human rights infringement due to the use of this face recognition technology, an example of stopping face recognition processing will be described.

[0013] FIG. 1 is a block diagram showing a configuration example of an AI system 100 using face recognition technology, which is an example of a control device according to the first embodiment. In FIG. 1, an image acquisition unit 101 acquires an image captured by a camera or the like. Note that the image acquisition unit 101 may include an image capturing device such as a camera. The image acquisition unit 101 acquires an image that is the target for information extraction using AI. In this embodiment, it is assumed that an image including the face of a person for whom face recognition is to be performed is acquired by the image acquisition unit 101. The images acquired by the image acquisition unit 101 vary depending on the installation conditions (installation location and viewing angle) of the image acquisition unit 101. In this embodiment, an example is given in which the image acquisition unit 101 acquires an image captured by a surveillance camera installed at a street corner or the like, and it is assumed that the image acquisition unit 101 has acquired an image 400 in which a plurality of people walking on the road or the like are shown, as shown in FIG. 2.

[0014] The person estimation unit 102 is a functional unit that executes an estimation process for estimating person information from the image acquired by the image acquisition unit 101, and has a configuration as shown in FIG. 3, for example. In FIG. 3, the face detection unit 301 detects a face region from the image input from the image acquisition unit 101. Any known method may be used as the method for detecting the face region. For example, a method of extracting shapes corresponding to components inside the face such as the nose, mouth, and eyes, estimating the size of the face from the size of both eyes and the distance between them, and detecting as the face region a region surrounded by a region of the estimated size with the position corresponding to the center of the nose as a reference can be mentioned. The image of the face region detected by the face detection unit 301 is normalized to a predetermined size and output as a face region image. Any known method may be used as the normalization method. In the case of the example of the image 200 in FIG. 2, since a plurality of persons are shown, the face detection unit 301 will detect a plurality of faces.

[0015] The identifier extraction unit 302 extracts information (information called a biometric identifier) effective for identifying each person from the face region image sent from the face detection unit 301. The extraction process of the biometric identifier in the identifier extraction unit 302 is performed for each input face region image. It is assumed that the identifier extraction unit 302 of the present embodiment is realized by a deep net in which parameters suitable for extracting the biometric identifier are set. Parameters suitable for extracting the biometric identifier can be obtained by performing machine learning on the deep net. That is, the identifier extraction unit 302 is composed of a learned deep net. Since the biometric identifier extracted by the deep net is generally expressed as a multi-dimensional vector, it is also called a feature vector. The identifier extraction unit 302 outputs the biometric identifier extracted from the input face region image to the person identification unit 304.

[0016] The identifier management unit 303 manages the registered biometric identifiers calculated in advance from the face region images of registered persons. A registered person refers to a person for whom a person ID is to be specified by face recognition. The method of extracting a registered feature vector from the image of a registered person may be performed in the same manner as the procedure shown so far. That is, for an image including a registered person, face detection is performed, and a biometric identifier is extracted from the resulting face region image obtained from the detection result, and this becomes the registered biometric identifier. The identifier management unit 303 manages the person ID of the registered person in association with the registered biometric identifier.

[0017] FIG. 4 is a diagram showing an example of a management table 401 used for managing the person ID and its registered biometric identifier in the identifier management unit 303. As shown in FIG. 4, the management table 401 is information representing the correspondence between the person ID for a registered person and the registered biometric identifier. For example, it is shown that the person with the person ID of ID_1 has a registered biometric identifier of Feature_1. Similarly, the registered biometric identifier of the person with the person ID of ID_2 is managed as Feature_2, and the registered biometric identifier of the person with the person ID of ID_3 is managed as Feature_3.

[0018] The person identification unit 304 collates the biometric identifier sent from the identifier extraction unit 302 with the registered biometric identifiers managed by the identifier management unit 303 to calculate a similarity score. In the present embodiment, the cosine distance between these two biometric identifiers is used as the similarity score. That is, as described above, since the biometric identifier is represented as a multi-dimensional vector, the cosine distance can be calculated. Therefore, the larger the similarity score, the more similar these two biometric identifiers are. The person identification unit 304 compares the maximum similarity score among the plurality of similarity scores calculated from the biometric identifier extracted by the identifier extraction unit 302 and the plurality of registered biometric identifiers managed by the identifier management unit 303 with a preset score threshold. When the maximum similarity score exceeds the score threshold, the person identification unit 304 identifies the person ID of the person in the face region image from which the biometric identifier was extracted by the identifier extraction unit 302, assuming that it is the same thing as the registered person of the registered biometric identifier when the similarity score was calculated.

[0019] For example, when three feature vectors as illustrated in FIG. 4 are managed in the identifier management unit 303, the person identification unit 304 calculates three similarity scores. If the maximum similarity score among the three similarity scores exceeds a preset threshold value, the person ID corresponding to the feature vector used for calculating the maximum similarity score is specified as the ID of the person in the face region image input to the identifier extraction unit 302.

[0020] In the person estimation unit 102, biometric authentication processing using the face recognition technology as described above is realized. Here, in the case of the present embodiment, the person estimation unit 102 switches between a mode (normal operation mode) for executing face recognition processing and a mode (estimation stop mode) for stopping face recognition processing under the control from the estimation control unit 104. That is, the person estimation unit 102 can execute or stop face recognition processing under the control from the estimation control unit 104. Further, the estimation stop mode for stopping face recognition processing may include a case where the estimation processing for face recognition is executed but the output of the result of the estimation processing is restricted. That is, the person estimation unit 102 can also switch between outputting the result of face recognition processing and not outputting the result according to the control from the estimation control unit 104.

[0021] An image is supplied from the image acquisition unit 101 to the situation monitoring unit 103 in FIG. 1. FIG. 5 is a block diagram showing a configuration example of the situation monitoring unit 103. In FIG. 5, the environment estimation unit 500 estimates the environment in which the input image is taken, in other words, the environment such as the place and situation where the image used for face recognition is acquired, from the input image. That is, in the case of the present embodiment, the environment to be estimated means a predetermined place such as a public place and the situation of that place. In the present embodiment, as an implementation form of the environment estimation unit 500, for example, a case where scene recognition technology is applied will be described.

[0022] In FIG. 5, the scene recognition unit 501 estimates the scene shown in the input image. In the case of this embodiment, the scene recognition technology means a technology for estimating the type of place where the image was acquired. For example, when a bench, a swing, etc. are reflected in the image, in scene recognition, the place where the image was acquired is estimated as a "park". As a method for realizing scene recognition in the scene recognition unit 501, any existing method may be used. As an example, object detection may be performed on the image to estimate "what" and "where" is shown, and then the scene may be estimated from the positional relationship of the detected objects. Alternatively, in recent years, many scene recognition technologies using deep nets have been researched and developed, and these may also be used. In this embodiment, the scene recognition unit 501 shall use a scene recognition technology using a deep net.

[0023] Also, in this embodiment, as described above, it is assumed that face recognition technology is used in a public place. When face recognition is performed in such a public place, there is a risk of privacy infringement and human rights infringement. Therefore, the scene recognition unit 501 performs scene recognition as to whether the environment where the image to be the target of face recognition was acquired is a public place or not. Note that the scene recognition unit 501 that can estimate whether it is a public place or not can be realized by preparing a large number of images of public places and having them learned by a deep net. Alternatively, the scene recognition unit 501 may use a more general scene recognition technology and make a determination that it is a public place when a scene related to a public place such as "park", "coast", "square", etc. is obtained as an estimation result.

[0024] In the scene recognition unit 501, in this way, a determination is made as to whether the environment where the image was acquired by the image acquisition unit 101 is a public place or not, and the result is output. For example, in a scene where a large number of people are walking on a road like the image 200 shown in FIG. 2, the scene recognition unit 501 determines that it is a public place.

[0025] Based on the estimation result of the environment estimation unit 500, the environment determination unit 502 determines whether the environment in which the image acquisition unit 101 captures an image is inappropriate for face recognition. In the present embodiment, when an estimation result indicating that it is a public place is input as a result of scene recognition, the environment determination unit 502 determines that the environment is inappropriate for face recognition and outputs the determination result to the estimation control unit 104 as usage status information. On the other hand, when an estimation result indicating that it is outside a public place is input as a result of scene recognition, the environment determination unit 502 determines that the environment is appropriate for face recognition and outputs the determination result to the estimation control unit 104 as usage status information. In this way, based on the estimation result of the environment estimation unit 500, the environment determination unit 502 outputs, as usage status information, the determination result as to whether the environment in which the image acquisition unit 101 captures an image is appropriate or inappropriate for face recognition to the estimation control unit 104.

[0026] Based on the usage information input from the situation monitoring unit 103, the inference control unit 104 determines whether to stop the operation of the person estimation unit 102 and controls the operation mode of the person estimation unit 102. For example, when the usage information input from the situation monitoring unit 103 is information on the determination result that the environment is inappropriate for face recognition, the inference control unit 104 controls the person estimation unit 102 to transition to or maintain the estimation stop mode in which the face recognition process is stopped. Conversely, when the usage information from the situation monitoring unit 103 is a determination result that the environment is appropriate for face recognition, the inference control unit 104 controls the person estimation unit 102 to transition to or maintain the normal mode in which the person estimation unit 102 performs face recognition processing. Also, even when the usage information from the situation monitoring unit 103 is a determination result that the environment is inappropriate for face recognition, the inference control unit 104 can perform control to limit the output of the estimation processing result without restricting the execution of the estimation processing in the person estimation unit 102. That is, when face recognition is performed in a public place, there is a risk of privacy infringement and human rights infringement. Therefore, the inference control unit 104 controls the person estimation unit 102 to limit at least one of the face recognition process and the output of the face recognition processing result. Conversely, when face recognition is performed in a place that is not a public place, since there is almost no risk of privacy infringement and human rights infringement, the inference control unit 104 controls the person estimation unit 102 not to limit the face recognition process and the output of the face recognition processing result.

[0027] Subsequently, the operation sequence of the AI system 100 according to the present embodiment will be described using the flowchart of FIG. 6. In the following flowchart, it is assumed that the symbol S represents a processing step (processing process).

[0028] First, as the process of S600, the image acquisition unit 101 acquires an image. Subsequently, as the process of S601, the situation monitoring unit 103 estimates the environment in which the face recognition process is performed based on the determination of the usage situation for the image acquired by the image acquisition unit 101, that is, based on the result of the scene recognition process by the above-described scene recognition unit 501. Furthermore, as the process of S602, the situation monitoring unit 103 determines whether it is estimated that the environment where the face recognition process is performed based on the scene recognition result is a public place or not. And, when it is estimated in the situation monitoring unit 103 that it is a public place, as the process of S603, the estimation control unit 104 controls to shift the operation mode of the person estimation unit 102 to the estimation stop mode. On the other hand, when it is estimated in the situation monitoring unit 103 that it is not a public place, as the process of S604, the estimation control unit 104 controls to shift the operation mode of the person estimation unit 102 to the normal operation mode. After these S603 or S604, the process of the AI system 100 shifts to S605.

[0029] When shifting to S605, the person estimation unit 102 confirms the operation mode controlled by the estimation control unit 104. In the case of the process of S605, the person estimation unit 102 determines whether the operation mode controlled by the estimation control unit 104 is the normal operation mode or not. And, when the operation mode is the normal operation mode, as the process of S606, the person estimation unit 102 executes the face recognition process on the image input from the image acquisition unit 101 and outputs the face recognition process result. On the other hand, when the operation mode is not the normal operation mode, that is, when it is the estimation stop mode, as the process of S607, the person estimation unit 102 stops the execution of the face recognition process on the image input from the image acquisition unit 101 or stops the output of the face recognition process result.

[0030] As described above, in the AI system 100 of the present embodiment, based on the acquired image, the environment where the image was acquired, that is, the environment where the estimation process for face recognition is performed, is estimated. Further, when the AI system 100 estimates that the environment where the image was acquired is a public place, it determines that the use of face recognition is inappropriate and stops the face recognition process. Alternatively, even in an environment inappropriate for performing face recognition, the AI system 100 does not limit the estimation process for face recognition, but limits the output of the result of the estimation process for face recognition. That is, the AI system 100 of the present embodiment limits at least one of the face recognition process and the output of the face recognition process result when the environment where the estimation process for face recognition is performed is an environment inappropriate for performing face recognition, such as a public place. Thereby, according to the AI system of the present embodiment, it is possible to prevent the use of face recognition processing that may cause infringement of people's privacy, human rights, etc.

[0031] In the first embodiment described above, scene recognition processing was exemplified as an estimation method of the usage environment in the situation monitoring unit 103, but it is not limited thereto. For example, based on the detection information of the GPS (Global Positioning System) and the direction sensor mounted on the camera, it is determined from which position and in which direction the acquired image of the image acquisition unit 101 was taken, and based on the determination result, the usage environment (such as whether a public place was photographed) may be estimated.

[0032] Also, the environment determined to be inappropriate for the use of face recognition processing is not limited to public places. In addition, in any place where privacy should be protected, or in any place where there is a risk of human rights infringement if there is no anonymity, it can be said that the use of face recognition processing is inappropriate. For example, when there is a possibility of human rights infringement or the like due to face recognition in a certain country or region, it is also possible to perform control to stop the face recognition processing by estimating the usage environment using GPS or the like.

[0033] <Second Embodiment> In the first embodiment described above, as an inappropriate usage pattern of face recognition, usage in public places and the like was assumed, and it was determined whether an image was acquired in a public place using scene recognition technology, and an example of preventing inappropriate use of face recognition based on the determination result was shown. In the second embodiment, as an inappropriate usage pattern of face recognition processing, it is determined whether face recognition is being performed on an unspecified number of people. That is, even if it is not a public place, if face recognition is performed on an unspecified number of people, there is a risk of privacy infringement, human rights infringement, etc. Therefore, in the second embodiment, an example of determining that it is an inappropriate use of face recognition when it is determined that face recognition is being performed on an unspecified number of people will be described. Regarding the control device of the second embodiment as well, similar to the example of the first embodiment, an application example to an AI system that performs face recognition processing will be described.

[0034] FIG. 7 is a block diagram showing a configuration example of an AI system 700 that performs face recognition as an application example of the control device of the second embodiment. In FIG. 7, components having the same functions as those in FIG. 1 described above are assigned the same reference numerals as in FIG. 1, and detailed descriptions thereof are omitted. In FIG. 7, the person estimation unit 702 performs face recognition processing in the same manner as the person estimation unit 102 in FIG. 1. However, in the case of the second embodiment, the person estimation unit 702 in addition to the face region detection function realized by the person estimation unit 102 in FIG. 1, also acquires information on how many faces are detected from the input image, and notifies the situation monitoring unit 703 of the information on the number of face detections. The configuration of the person estimation unit 702 is the same as that in FIG. 3 described above. However, in the case of the second embodiment, the face detection unit 301 performs face detection as described above and also has a function of counting the number of detected faces. Then, the face detection unit 301 in the second embodiment notifies the situation monitoring unit 703 of the information on the counted number of face detections.

[0035] FIG. 8 is a block diagram showing a configuration example of the situation monitoring unit 703 in the second embodiment. In FIG. 8, the state estimation unit 800 estimates the state of a person who is the target of face recognition processing in the person estimation unit 702. In the case of this embodiment, the state estimation unit 800 estimates whether or not face recognition processing is being performed on an unspecified number of persons based on the number of face detections input from the person estimation unit 702. Since the number of face detections input from the person estimation unit 702 is the number of face detections per image, a large number of such detections means that an unspecified number of people are shown in the image. In this case, it can be estimated that the person estimation unit 702 is performing face recognition on an image showing an unspecified number of persons. For example, a large number of faces will be detected from an image such as the image 200 shown in FIG. 2. Detecting a large number of faces in a single image like this can be said to be an important basis for determining whether face recognition for an unspecified number of persons is being performed. Therefore, when the number of face detections input from the person estimation unit 702 is equal to or greater than a predetermined number threshold, the state estimation unit 800 estimates that face recognition is being performed on an image showing an unspecified number of persons. Then, the state estimation unit 800 outputs the estimation result to the state determination unit 802.

[0036] Based on the estimation result input from the state estimation unit 800, the state determination unit 802 determines whether the image acquired by the image acquisition unit 101 is an inappropriate target for face recognition. In the case of this embodiment, when an estimation result that an unspecified number of persons are shown in the image is input from the state estimation unit 800, the state determination unit 802 determines that the image is an inappropriate target for face recognition and outputs the determination result to the estimation control unit 104.

[0037] Based on the determination result input from the situation monitoring unit 703, the estimation control unit 104 of this embodiment controls the operation mode of the person estimation unit 702 to either the normal operation mode or the estimation stop mode. That is, when an estimation result is input from the situation monitoring unit 703 that the image is inappropriate for face recognition because an unspecified number of persons are shown, the estimation control unit 104 performs control to limit at least one of face recognition processing and the output of face recognition processing results.

[0038] Figure 9 is a flowchart of the AI system 700 according to the second embodiment. In Figure 9, the same reference numerals as those in Figure 6 are given to the same processing steps (processing processes) as in Figure 6, and the description thereof is omitted. In the flowchart of the first embodiment shown in Figure 6, the environment estimation unit 500 determines whether the face recognition is being used inappropriately based on the estimation of the usage location (whether it is a public place or not). In contrast, in the flowchart of the second embodiment shown in Figure 9, the situation monitoring unit 703 determines whether the face recognition is being used inappropriately by estimating the state of the person in the image (whether an unspecified number of people are shown in the image).

[0039] In the case of the flowchart in Figure 9, after the process of S600, the process of the AI system 700 proceeds to S901. In S901, the person estimation unit 702 performs face detection on the image input from the image acquisition unit 101, further counts the number of detected faces, and outputs the face detection count to the situation monitoring unit 703.

[0040] Subsequently, as the process of S902, the situation monitoring unit 703 determines whether it is estimated that an unspecified number of people are shown in the image acquired by the image acquisition unit 101 based on the face detection count counted by the person estimation unit 702. If it is estimated that an unspecified number of people are shown, as the process of S603, the estimation control unit 104 controls the operation mode of the person estimation unit 702 to the estimation stop mode as the process of S603. On the other hand, if it is not estimated that an unspecified number of people are shown, as the process of S604, the estimation control unit 104 controls the operation mode of the person estimation unit 702 to the normal operation mode. The subsequent processes are the same as those in the flowchart of Figure 6 described above, so the description is omitted.

[0041] In the case of the AI system 700 of the second embodiment, in order to determine whether or not to perform face recognition processing using the face detection result, once face detection is performed on the input image, the face recognition processing stops. Then, in the AI system 700, the situation monitoring unit 703 determines whether or not an unspecified number of people are shown, and according to the determination result, the operation mode is controlled by the estimation control unit 104. That is, in the AI system 700 of the second embodiment, when it is determined that face recognition is to be performed on an unspecified number of people, it is considered an inappropriate use of face recognition, and the execution of face recognition processing (processing after the extraction of biometric identifiers) is stopped, or the output of the face recognition processing result is stopped.

[0042] In the second embodiment, as an example of the determination criterion for whether or not face recognition processing is to be performed on an unspecified number of people, the number of face detections in the image is used, but the method is not limited to using the number of face detections. For example, it is possible to use a crowd estimation technique for estimating the degree of congestion of people in the image. In this example, when a certain degree of congestion or more is detected, it can be determined that face recognition processing is to be performed on an unspecified number of people. As exemplified above, the number of face detections in the image and the crowd estimation technique are examples of means for detecting that an unspecified number of people are the target of face recognition processing, and the techniques that can be used for estimating the target state in the state estimation unit 800 are not limited to these. Any method can be used as long as it can detect that an unspecified number of people are the target of face recognition processing.

[0043] <The Third Embodiment> Next, as a third embodiment, a configuration example in which the situation monitoring unit has both functions of those described in the first embodiment and those described in the second embodiment will be described. Since the configuration of the AI system of the third embodiment is basically the same as that of FIG. 7 described above, its illustration and description are omitted. The situation monitoring unit in the case of the third embodiment is the situation monitoring unit 1003 having the configuration as shown in FIG. 10. In FIG. 10, the same reference numerals as those in FIGS. 3 and 8 described above are assigned to the same components, and their detailed descriptions are omitted. The situation monitoring unit 1003 according to the third embodiment includes an environment estimation unit 500 similar to that described in the first embodiment, a state estimation unit 800 similar to that described in the second embodiment, and a result integration unit 1001.

[0044] In FIG. 10, the result integration unit 1001 uses the two estimation results by the environment estimation unit 500 and the state estimation unit 800 described above to determine whether the face recognition process for the image acquired by the image acquisition unit 101 is appropriate use or inappropriate use. That is, the result integration unit 1001 integrates these two estimation results and determines whether it is appropriate to perform face recognition on the person shown in the image acquired by the image acquisition unit 101. As a method for integrating the two estimation results, various methods can be considered depending on the criteria for determining whether the face recognition is inappropriately used or not. As an example, a method may be used in which it is finally determined as inappropriate use only when it is estimated that it is inappropriate to perform face recognition in both of the two estimation results. In addition, a method may be used in which it is finally determined as inappropriate use when it is estimated that it is inappropriate to perform face recognition in either one of the two estimation results.

[0045] FIG. 11 is a flowchart of the AI system according to the third embodiment including the situation monitoring unit 1003 configured as shown in FIG. 10. The flowchart of FIG. 11 shows an operation sequence in the case where face recognition is actually stopped when it is determined that there is inappropriate use of face recognition based on both of the two estimation results of the environment estimation unit 500 and the state estimation unit 800. That is, the flowchart of FIG. 11 shows the flow in the case where face recognition processing is stopped or the output and stop of face recognition results are performed when face recognition is performed in a public place and for an unspecified number of people. FIG. 11 is a flowchart combining the flowcharts of FIGS. 6 and 9. The same reference numerals as those in FIG. 6 are assigned to the processing steps similar to those in FIG. 6, and the same reference numerals as those in FIG. 9 are assigned to the processing steps similar to those in FIG. 9, and detailed descriptions are omitted as appropriate.

[0046] As shown in the flowchart of FIG. 11, the processing of the AI system according to the third embodiment proceeds to the processing of S901 and then to the processing of S902 after the processing of S600. Then, in S902, when it is estimated that an unspecified number of people are shown in the image, the processing of the AI system proceeds to the processing of S601 to execute scene recognition, and then proceeds to the processing of S602. Also, in S902, when it is estimated that an unspecified number of people are not shown in the image, the processing of the AI system proceeds to the processing of S604.

[0047] Also, in the processing of the AI system, when it is estimated in S602 that it is a public place, the processing proceeds to the processing of S603, while when it is estimated that it is not a public place, the processing proceeds to the processing of S604. Then, after the processing of S603 or S604, the processing of the AI system proceeds to S605. Since the processing after S605 is the same as described above, the description is omitted.

[0048] <Fourth Embodiment> In the first to third embodiments, an example was described in which the face recognition process is stopped or the output of the face recognition process result is stopped when it is determined that face recognition is being used inappropriately. In the following fourth embodiment, an example will be described in which, when it is determined that face recognition is being used inappropriately, instead of immediately stopping the face recognition process, first, a warning to that effect is displayed. Also, in the fourth embodiment, an example will be given in which an action is requested from the user (the user of the AI system) in response to the warning display, and it is determined whether or not to stop the face recognition process according to the action from the user. Further, in the fourth embodiment, an example will also be described in which, even when the situation monitoring unit determines that face recognition is being used inappropriately, face recognition can be performed if it is a truly unavoidable situation. Note that a truly unavoidable situation is assumed to be, for example, a case where face recognition is permitted by a formal warrant or the like by public authority or the like.

[0049] FIG. 12 is a block diagram showing a configuration example of the AI system 1200 according to the fourth embodiment. In FIG. 12, the same reference numerals as those in FIG. 1 are given to the same components as those in FIG. 1, and the description thereof is omitted.

[0050] In addition to the operation of the estimation control unit 104 in the first embodiment, when an estimation result that the environment is inappropriate for performing face recognition is input, the estimation control unit 1204 in the fourth embodiment outputs that information to the input / output IF unit 1205. Then, the estimation control unit 1204 instructs the person estimation unit 102 whether or not to perform face recognition (which operation mode to transition to / maintain) according to the user action information input via the input / output IF unit 1205.

[0051] When the input / output IF unit 1205 receives information from the inference control unit 1204 indicating that the environment is inappropriate for face recognition, it issues a warning notice to the user (the user of the AI system) to that effect. In the case of this embodiment, examples of the warning notice include a warning message such as "It has been detected that face detection may be taking place in a public place". Further, the input / output IF unit 1205 requests the user to input whether or not they will comply with the warning notice. Examples of the request to the user in response to the above-mentioned warning notice (warning message) include presenting the user with two options, such as "Instruct face recognition as this is not a public place" and "Stop face recognition as this is a public place", and asking the user to make a selection.

[0052] This enables recovery, for example, in the case where the estimation result of the situation monitoring unit 103 is incorrect. That is, even if the estimation result of the situation monitoring unit 103 is incorrect, it is not necessary to immediately stop the face recognition process, and it is possible to prevent unnecessary confusion from being given to the user. Also, even if the user accidentally tries to execute face recognition in a public place, displaying a warning makes it possible for the user to notice the accidental mistake, and it can be expected that this will lead to suppression of repeating the same kind of failure in the future.

[0053] The input / output IF unit 1205 transmits the above-described action information from the user to the inference control unit 1204. When the action information input from the user to the inference control unit 1204 is "Instruct face recognition as this is not a public place", the inference control unit 1204 controls the person estimation unit 102 to transition to or maintain the normal operation mode.

[0054] In addition, the input / output IF unit 1205 may, for example, display three options to the user, namely, "Instruct face recognition because it is not a public place", "Stop face recognition because it is a public place", and "Instruct face recognition because it is a truly unavoidable situation", and allow the user to make a selection. In this case, if the user selects the third option, the user will be further requested to input an emergency code. After the emergency code is input, it will be transmitted to the inference control unit 1204 to perform face recognition processing. Note that the emergency code is assumed to be a code (password) etc. issued by the developer of the AI system of this embodiment when there is an official warrant of public authority etc. When a formal code issued by the developer of the AI system is input, it can be confirmed that face recognition processing in a public place is permitted through formal procedures. Therefore, the AI system operates to perform face recognition processing even in a public place.

[0055] Also, in this embodiment, the case where the situation monitoring unit 103 monitors "whether face recognition is being performed in a public place" has been described, but it is not limited to this. For example, similarly, when the situation monitoring unit 103 monitors "whether face recognition is being performed on an unspecified number of people", the operation mode of the person estimation unit 102 using the input / output IF unit 1205 can be controlled.

[0056] <Other Embodiments> In each of the above-described embodiments, an example has been described in which the operation mode is switched in response to each estimation result sent from the situation monitoring unit 103. However, the switching control of the operation mode does not need to be performed every time an image is input from the image acquisition unit 101. For example, the above-described situation monitoring unit 103 outputs usage status information based on the scene recognition result and sends it to the estimation control unit 104 every time an image is input from the image acquisition unit 101. On the other hand, when an estimation result that the location is a public place is input from the situation monitoring unit 103 for a number of images equal to or more than a predetermined threshold among a predetermined number of images determined in advance, the estimation control unit 104 may control the person estimation unit 102 to transition to or maintain the estimation stop mode. In other words, only when the ratio of the number of images determined to be public places reaches or exceeds a predetermined ratio threshold, the face recognition process or the output of the face recognition process result is stopped. Thereby, the tolerance for misrecognition (incorrectly estimating the scene) that may occur during scene recognition can be increased.

[0057] Also, as described above, when the situation monitoring unit 103 determines the usage environment of face recognition for a plurality of images, the determination of "whether to perform face recognition for an unspecified number of people" described in the second embodiment can also be performed as follows. That is, in the second embodiment, the number of face detections in the image is used as a reference, but the "ratio of the number of correct recognitions to the number of face detections" can also be used as a reference. Here, a correct recognition means that as a result of collation based on face recognition, it is determined that the person is a registered person. For example, when searching for a person requiring attention (such as a wanted criminal) on the street by face recognition, it is expected that the number of correct recognitions is extremely small compared to the number of face detections. That is, most of the people shown in the image are not the person requiring attention, and it is extremely rare for the person requiring attention to appear in the image. Thus, when the "ratio of the number of correct recognitions to the number of face detections" is less than a predetermined ratio threshold determined as, for example, an extremely small value, it can be said to be useful information as a basis for estimating that face recognition is being performed for an unspecified number of people.

[0058] In addition, in the above-described embodiments, an example of determining whether the use is appropriate or inappropriate has been described with respect to face recognition technology. However, the object of use determination is not limited to face recognition technology. For example, the object of use determination may be emotion estimation technology. Even in the case of emotion estimation technology, depending on the usage method, there is a risk of privacy infringement or human rights infringement, which is the same as in the case of face recognition technology. Therefore, it may also be an object for which it is necessary to determine whether the use is appropriate or inappropriate.

[0059] FIG. 13 is a diagram showing an example of the hardware configuration of an information processing apparatus to which the control apparatus according to each of the above-described embodiments can be applied. The control apparatus 1300 includes a CPU 1301, a ROM 1302, a RAM 1303, a large-capacity memory 1304, a network IF 1306, an input device 1307, a display device 1308, and the like. The network IF 1306 is connected to a network 1311. The imaging device that captures the image acquired by the image acquisition unit 101 may be connected to the control apparatus 1300 via the network 1311 or may be included in the control apparatus 1300.

[0060] The CPU 1301 controls the control apparatus 1300 in an overall manner. The ROM 1302 stores a control program for the CPU 1301 to control the control apparatus 1300, a control program for performing control processing related to each functional unit of the control apparatus shown in FIGS. 1, 7, and 12 described above, and the like. The RAM 1303 is a memory in which the program read from the ROM 1302 is expanded and the CPU 1301 executes the program. Further, the RAM 1303 is also used as a temporary storage area for temporarily storing data to be subjected to various processes.

[0061] The network IF 1306 is a circuit that performs communication via the network 1311. The CPU 1301 performs processing related to each functional unit of the control apparatus described above on the image received by the network IF 1306. The large-capacity memory 1304 is an HDD, an SSD, or the like, and can store images and the like acquired by the image acquisition unit 101. Therefore, the CPU 1301 can also read the images stored in the large-capacity memory 1304 and perform the above-described processing and the like.

[0062] The display device 1308 is a display device that displays images, text, and the like, and displays notifications and the like to the user. The input device 1307 is a device including at least any one of an input keyboard, a pointing device for screen display on the display device 1308, a mouse, a touch panel, and the like. The user of the control device 1300 can input the above-described action information and the like via the input device 1307, and the CPU 1301 performs processing according to the action information from the user.

[0063] As described above, the hardware configuration of the control device 1300 has components similar to the hardware components mounted on an information processing device such as a general PC (personal computer). Therefore, the various functions realized by the control device 1300 can be implemented as software (program) operating on a PC. The CPU 1301 can realize the processing related to each functional unit shown in FIGS. 1, 7, 12, etc. described above by executing the control program according to this embodiment. Of course, each functional unit of the control device shown in FIGS. 1, 7, 12, etc. may be realized as a circuit configuration.

[0064] The present invention can also be implemented by supplying a program that realizes one or more functions of the above-described embodiments to a system or apparatus via a network or a storage medium, and causing one or more processors in a computer of the system or apparatus to read and execute the program. It can also be implemented by a circuit (for example, ASIC) that realizes one or more functions. The above-described embodiments are merely examples of implementation in carrying out the present invention, and the technical scope of the present invention should not be construed in a limited manner by these. That is, the present invention can be implemented in various forms without departing from its technical idea or its main features.

[0065] The disclosure of each embodiment includes the following configurations, methods, and programs. (Configuration 1) An acquisition means for acquiring an image including a person, An estimation means for executing an estimation process for estimating information of the person from the image, An output means for outputting a result of the estimation process by the estimation means, A determination means for determining a usage status of the estimation process by the estimation means, A control means for restricting at least one of the estimation process by the estimation means and the output by the output means based on a determination result of the usage status by the determination means, A control device characterized by comprising the above. (Configuration 2) The determination means An environment estimation means for estimating an environment in which the image used for the estimation process is acquired, An environment determination means for determining whether the environment estimated by the environment estimation means is an appropriate environment for use of the estimation process, and has The control device according to Configuration 1, characterized in that a determination result by the environment determination means is used as a determination result of the usage status. (Configuration 3) The environment estimation means includes a scene recognition means for recognizing a type of a place where the image is acquired as an environment in which the image is acquired, The environment determination means determines whether the type of location recognized by the scene recognition means is a location appropriate for use of the estimation process, according to Configuration 2 of the control device characterized by this. (Configuration 4) When the type of location recognized by the scene recognition means is a location of a predetermined type, the environment determination means determines that it is not a location appropriate for use of the estimation process. Based on the determination result that it is not a location appropriate for use of the estimation process, the control means restricts at least one of the estimation process by the estimation means and the output by the output means, according to Configuration 3 of the control device characterized by this. (Configuration 5) The location of the predetermined type is a public location, according to Configuration 4 of the control device characterized by this. (Configuration 6) The determination means has state estimation means for estimating the state of the person estimated by the estimation means, and state determination means for determining whether the state estimated by the state estimation means is a state appropriate for use of the estimation process, and uses the determination result by the state determination means as the determination result of the usage situation, according to any one of Configurations 1 to 5 of the control device characterized by this. (Configuration 7) The state estimation means counts the number of persons included in the image as the state of the person. Based on the number of persons counted by the state estimation means, the state determination means determines whether it is a state appropriate for use of the estimation process, according to Configuration 6 of the control device characterized by this. (Configuration 8) When the number of persons counted by the state estimation means is equal to or more than a predetermined threshold value, the state determination means determines that it is not a state appropriate for use of the estimation process. Based on the determination result that it is not a state appropriate for use of the estimation process, the control means restricts at least one of the estimation process by the estimation means and the output by the output means, according to Configuration 7 of the control device characterized by this. (Configuration 9) The state estimation means estimates the emotion of the person as the state of the person, The state determination means determines whether it is an appropriate state for use of the estimation process based on the emotion of the person estimated by the state estimation means, according to the control device described in Configuration 6. (Configuration 10) When a determination result is obtained that the determination means determines that the use of the estimation process is not appropriate for images of a number equal to or greater than a predetermined threshold among the number of images acquired by the acquisition means, the control means restricts at least one of the estimation process by the estimation means and the output by the output means, according to the control device described in any one of Configurations 1 to 9. (Configuration 11) When the ratio of images for which the determination means determines that the use of the estimation process is not appropriate among the plurality of images acquired by the acquisition means is equal to or greater than a predetermined threshold, the control means restricts at least one of the estimation process by the estimation means and the output by the output means, according to the control device described in any one of Configurations 1 to 9. (Configuration 12) The estimation means has a first mode for executing the estimation process and a second mode for not executing the estimation process or not outputting the result of the estimation process, The control means switches the estimation means to either the first mode or the second mode based on the determination result of the usage situation, according to the control device described in any one of Configurations 1 to 11. (Configuration 13) The estimation means face detection means for performing face detection of a person from the image, identifier extraction means for extracting a biometric identifier used for identifying the person, specific identification means for identifying the person whose face is detected from the image based on the result of comparing the biometric identifier extracted by the identifier extraction means with the biometric identifier of a registered person, The control device according to any one of Configurations 1 to 12, characterized by having (Configuration 14) The face detection means counts the number of faces of the person detected from the image, The determination means determines the usage status of the estimation process by the estimation means based on the ratio between the counted number of faces and the number of persons specified as the registered persons by the specifying means. The control device according to Configuration 13, characterized by this. (Configuration 15) When the ratio is less than a predetermined threshold value, the control means restricts at least one of the estimation process by the estimation means and the output by the output means. The control device according to Configuration 14, characterized by this. (Configuration 16) The control device according to any one of Configurations 1 to 15, characterized by having a notification means for notifying the user of the determination result of the usage status of the estimation process by the determination means. (Configuration 17) The control device according to Configuration 16, characterized by having an input means for inputting user action information regarding the notification by the notification means. (Configuration 18) When action information for instructing the execution of the estimation process is input from the user through the input means, the control means executes the estimation process by the estimation means or the output by the output means. The control device according to Configuration 17, characterized by this. (Method 1) An acquisition step of acquiring an image including a person, An estimation step of executing an estimation process for estimating information of the person from the image, An output step of outputting the result of the estimation process by the estimation step, A determination step of determining the usage status of the estimation process by the estimation step, A control step of restricting at least one of the estimation process by the estimation step and the output by the output step based on the determination result of the usage status by the determination step, A control method, characterized by having (Program 1) A program that causes a computer to function as the control device described in any one of Configurations 1 to 18.

Explanation of Signs

[0066] 100: AI system, 101: Image acquisition unit, 102: Person estimation unit, 103: Situation monitoring unit, 104: Estimation control unit

Claims

1. An acquisition means for acquiring an image including a person, An estimation means for executing an estimation process for estimating information of the person from the image, An output means for outputting the result of the estimation process by the estimation means, A determination means for determining the usage status of the estimation process by the estimation means, A control means for restricting at least one of the estimation process by the estimation means and the output by the output means based on the determination result of the usage status by the determination means, A control device characterized by comprising the above.

2. The determination means, An environment estimation means for estimating the environment in which the image used for the estimation process is acquired, An environment determination means for determining whether the environment estimated by the environment estimation means is an appropriate environment for the use of the estimation process, and has, The control device according to claim 1, characterized in that the determination result by the environment determination means is used as the determination result of the usage status.

3. The environment estimation means includes a scene recognition means for recognizing the type of the place where the image is acquired as the environment in which the image is acquired, The control device according to claim 2, characterized in that the environment determination means determines whether the type of the place recognized by the scene recognition means is an appropriate place for the use of the estimation process.

4. The environment determination means determines that when the type of the place recognized by the scene recognition means is a predetermined type of place, it is not an appropriate place for the use of the estimation process, The control device according to claim 3, characterized in that the control means restricts at least one of the estimation process by the estimation means and the output by the output means based on the determination result that it is not an appropriate place for the use of the estimation process.

5. The control device according to claim 4, characterized in that the predetermined type of place is a public place.

6. The determination means, A state estimation means for estimating the state of the person estimated by the estimation means, A state determination means for determining whether the state estimated by the state estimation means is an appropriate state for the use of the estimation process, and has, The control device according to any one of claims 1 to 5, characterized in that the determination result by the state determination means is used as the determination result of the usage status.

7. The state estimation means counts the number of persons included in the image as the state of the person, The control device according to claim 6, wherein the state determination means determines whether the state is appropriate for using the estimation process based on the number of persons counted by the state estimation means.

8. When the number of the persons counted by the state estimation means is equal to or greater than a predetermined threshold value, the state determination means determines that the state is not appropriate for using the estimation process. The control device according to claim 7, wherein the control means restricts at least one of the estimation process by the estimation means and the output by the output means based on a determination result that the state is not appropriate for using the estimation process.

9. The state estimation means estimates the emotion of the person as the state of the person. The control device according to claim 6, wherein the state determination means determines whether the state is appropriate for using the estimation process based on the emotion of the person estimated by the state estimation means.

10. When a determination result is obtained that the determination means determines that the use of the estimation process is not appropriate for images of a number equal to or greater than a predetermined threshold value among the number of images acquired by the acquisition means, the control means restricts at least one of the estimation process by the estimation means and the output by the output means. The control device according to claim 1.

11. When the ratio of the images determined by the determination means that the use of the estimation process is not appropriate among the plurality of images acquired by the acquisition means is equal to or greater than a predetermined threshold value, the control means restricts at least one of the estimation process by the estimation means and the output by the output means. The control device according to claim 1.

12. The estimation means has a first mode for executing the estimation process and a second mode for not executing the estimation process or not outputting the result of the estimation process. The control device according to claim 1, wherein the control means switches the estimation means to either the first mode or the second mode based on the determination result of the usage situation.

13. The estimation means a face detection means for detecting a face of a person from the image, an identifier extraction means for extracting a biometric identifier used for identifying the person, Identification means for identifying the person whose face has been detected from the image based on the result of comparing the biometric identifier extracted by the identifier extraction means with the biometric identifier of the registered person; The control device according to claim 1, characterized by comprising the same. **Claim 14** The face detection means counts the number of faces of the person detected from the image; The determination means determines the usage status of the estimation process by the estimation means based on the ratio between the counted number of faces and the number of persons identified as the registered persons by the identification means. The control device according to claim 13, characterized by the above. **Claim 15** When the ratio is less than a predetermined threshold value, the control means restricts at least one of the estimation process by the estimation means and the output by the output means. The control device according to claim 14, characterized by the above. **Claim 16** The control device according to claim 1, characterized by comprising notification means for notifying the user of the determination result of the usage status of the estimation process by the determination means. **Claim 17** The control device according to claim 16, characterized by comprising input means for inputting user action information regarding the notification by the notification means. **Claim 18** When action information for instructing the execution of the estimation process is input from the user through the input means, the control means executes the estimation process by the estimation means or the output by the output means. The control device according to claim 17, characterized by the above. **Claim 19** An acquisition step of acquiring an image including a person; An estimation step of executing an estimation process for estimating information of the person from the image; An output step of outputting the result of the estimation process by the estimation step; A determination step of determining the usage status of the estimation process by the estimation step; A control step of restricting at least one of the estimation process by the estimation step and the output by the output step based on the determination result of the usage status by the determination step; A control method, characterized by comprising the same. **Claim 20** A computer, An acquisition means for acquiring an image including a person; An estimation means for executing an estimation process for estimating information of the person from the image; An output means for outputting the result of the estimation process by the estimation means; A determination means for determining the usage status of the estimation process by the estimation means; Based on the determination result of the usage status by the determination means, a control means for restricting at least one of the estimation process by the estimation means and the output by the output means; A program for functioning as a control device having the above.

Images