JP2018181157A - Person authentication device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a person authentication device favorably selecting data upon additional learning and improving authentication accuracy while canceling learning deviation.SOLUTION: A person authentication device 5 includes: a face authentication part 25 authenticating a person as a target to be traced by use of a face authentication model 36 of a registered person; an additional learning face saving part 26 selecting and saving an additional learning face image 38 from the face image of the person authenticated by the face authentication part 25; and an additional learning execution part 27 executing additional learning of the face authentication model 36 by use of the selected additional learning face image 38. The additional learning face saving part 26 selects, as the additional learning face image 38, the face image whose authentication reliability of the person authenticated by the face authentication part 25 is located between two upper- and lower-limit threshold values.SELECTED DRAWING: Figure 3

Description

  The present invention relates to a person authentication apparatus provided with an additional learning function.

  As background art of this technical field, the person search device which has an additional learning function is disclosed by patent documents 1, and "the person for whom search object was holding different attributes (for example, had different clothes) It is described that "if there is a means for adding the attribute as a search target". Furthermore, “when searching, by having a means for specifying a video such as a time or place where a person specified as a search target can not exist, a video such as a time or place where a specified person can not exist Is excluded from the search target.

JP, 2014-016968, A

  In the technique described in Patent Document 1, by adding an attribute designated as a search target, a search following a change in attribute becomes possible at the time of person search. However, the quality of the data to be added is not sufficiently considered, which may result in additional learning side effects. That is, when additional learning is performed, the quality of data (face image) to be added may cause learning bias. For example, when a face image having a very high degree of similarity at the time of authentication is added, the face recognition model may be biased to the added image. In addition, if a large number of face images in which part of the face parts is missing are added, the threshold value of the authentication determination becomes small, which may cause false detection. As described above, in order to improve the authentication accuracy by the additional learning and to improve the tolerance (robustness) at the time of operation, it is necessary to suitably select data to be additionally learned.

  An object of the present invention is to provide a person authentication apparatus that suitably selects data at the time of additional learning, eliminates bias in learning, and improves authentication accuracy.

  The person authentication apparatus according to the present invention is, for example, a face authentication unit for authenticating a person who is a tracking target using a registered person's face authentication model, and the face authentication unit. An additional learning face storage unit that selects and stores a face image for additional learning from a face image of a person; and an additional learning performing part that performs additional learning of the face recognition model using the selected additional learning face image And the additional learning face storage unit selects a face image having an authentication reliability of a person who has been authenticated by the face authentication unit between two thresholds of upper limit and lower limit as the additional learning face image. It is characterized by

  Alternatively, the additional learning face storage unit is characterized by selecting, as the additional learning face image, a face image whose degree of difference indicating how different it is from the face image added as the face authentication model is larger than a predetermined threshold. Do.

  According to the present invention, the person authentication apparatus has an effect of improving the tolerance at the time of operation while preventing the side effect due to the additional learning.

BRIEF DESCRIPTION OF THE DRAWINGS The figure which shows the structural example of the whole person authentication system (Example 1). FIG. 2 is a diagram showing a hardware configuration of a person authentication apparatus 5; FIG. 2 shows a functional system configuration of a person authentication apparatus 5; FIG. 7 is a view showing a data structure of an initial learning face image 37. The data structure of the face image 38 for additional learning. The figure which shows the sequence of face recognition and the addition learning image accumulation | storage. The figure which shows the sequence of additional learning implementation. The figure which shows the processing flow (step S2) of a person tracking. The figure which shows the data structure of the tracking result preservation | save part 33. FIG. The figure which represented the tracking result shown in FIG. 9 on the map. FIG. 6 is a view showing a data structure of a face image stored in a face image storage unit 34. FIG. 6 is a view showing a data structure of an authentication result stored in an authentication result storage unit 35. The figure which shows the data structure of the additional learning setting information 39. FIG. The figure which shows the flow of an additional learning face image selection process (step S7). The figure which shows the detail of an additional learning face image effectiveness determination flow (step S206). The figure which shows the confirmation screen of the learning result in the additional learning of FIG. The figure which shows the other structural example of a person authentication system (Example 2). The figure which shows the relationship of the additional learning setting information 39 and a face authentication success threshold value (Example 3).

  Hereinafter, an embodiment of a person authentication apparatus according to the present invention will be described with reference to the drawings.

  FIG. 1 is a diagram showing an example of the overall configuration of a person authentication system including a person authentication device according to the present invention. The person authentication system 1 includes a video input device 2 such as a surveillance camera for acquiring a video including the person 7, and a network 6 including a LAN (Local Area Network), a WAN (Wide Area Network), a VPN (Virtual Private Network), and the like. And a person authentication device 5 for authenticating a person, a video recording device 3 for storing a video, and a person authentication management device 4 for managing the person authentication device 5. The person authentication device 5 receives an image from the image input device 2 via the network 6, compares the face image of the person 7 in the picture with the face image of the registered person, and specifies the person 7. The received video is stored in the video recording device 3. The person authentication management device 4 displays the setting of the person authentication device 5 and the authentication result of the person authentication device 5. Next, the internal configuration of the person authentication apparatus 5 will be described.

  FIG. 2 is a diagram showing a hardware configuration of the person authentication apparatus 5. The person authentication device 5 includes a memory 10, an external storage device 11, a CPU (Central Processing Unit) 18, an output device 19, a communication device 20, and a bus 17 connecting these.

  The external storage device 11 is a non-volatile storage device such as a so-called hard disk drive (Hard Disk Drive), an SSD (Solid State Drive), or a flash memory capable of storing digital information. The external storage device 11 stores data such as a face authentication program 12, a face authentication model DB (database) 13, an initial learning face DB 14, an additional learning face DB 15, and an additional learning setting DB 16.

  The face authentication program 12 is a program for executing the face authentication process described below. The face authentication model DB 13 is configured of parameters for identifying each person from other persons. For example, when face recognition is performed using a SVM (Support Vector Machine) identifier, a support vector (Support Vector) is stored that distinguishes the person from others. The initial learning face DB 14 stores a face image used when constructing the face authentication model DB 13. The additional learning face DB 15 stores face images accumulated during operation for additional learning of the face recognition model. The additional learning setting DB 16 stores setting parameters that define selection conditions of face images to be accumulated in the additional learning face DB 15 at the time of operation.

  The CPU 18 is a unit that executes various operations, and executes various processes by executing a predetermined program loaded from the external storage device 11 to the memory 10. That is, the face authentication program 12 in the external storage device 11, the face authentication model DB 13, the initial learning face DB 14, the additional learning face DB 15, and the additional learning setting DB 16 are expanded in the memory 10 by the CPU 18. The memory 10 stores a program executed by the CPU 18 and data necessary for executing the program.

  The output device 19 is a screen generation device such as a graphic board or a video card, and converts the video from the video input device 2 and the authentication result of the person authentication device 5 into a displayable screen. The communication device 20 receives a video from the video input device 2 via the network 6 and transmits / receives an authentication result, an additional learning result and a control command to / from the person authentication management device 4.

  FIG. 3 is a diagram showing a functional system configuration of the person authentication apparatus 5. The person authentication apparatus 5 includes an image input unit 21, a person tracking unit 22, a face extraction unit 23, a face feature amount calculation unit 24, a face authentication unit 25, an additional learning face storage unit 26, an additional learning execution unit 27, and an external cooperation unit 28. , An input receiving unit 29, an output unit 30, and a temporary data storage unit 31 for storing data loaded from the external storage device 11, and a data storage unit 32. The function of each part is explained.

The video input unit 21 decodes the encoded video data received via the network 6 and converts it into a frame unit image.
The person tracking unit 22 calculates the optical flow (the motion vector of the object) of the video input to the video input unit 21 by the block matching method or the gradient method, and calculates the human region in the video. In addition, tracking is performed while calculating the latest location.

The face extraction unit 23 extracts the face area with respect to the area of the person being tracked by the human tracking unit 22, using, for example, the relative position and size of the parts of the face, and the shape of the eyes and nose.
The face feature amount calculation unit 24 is an attribute for distinguishing the face of the person from the face of another person, such as the size and relative position of each part such as the eyes, nose and mouth. Calculated as a feature amount.

  The face authentication unit 25 uses the face authentication model DB 13 to calculate, as a probability value, the degree of similarity with the registered face feature value as a probability value for the face feature value of the face, and identifies the person to which the face belongs.

During the operation of the person authentication apparatus 5, the additional learning face storage unit 26 selects a part of the face image of the person who has already been authenticated as an additional learning image, and stores it in the additional learning face DB 15.
The additional learning execution unit 27 performs additional learning of the face recognition model using the initial learning face DB 14 and the face data accumulated in the additional learning face DB 15. The result of the additional learning of the face recognition model is stored in the face recognition model DB 13.

The external cooperation unit 28 transmits the authentication result of the face authentication unit 25 to an external system, for example, an entrance / exit management system. Also, face recognition processing is performed by a trigger from an external system.
The input receiving unit 29 is a device such as a keyboard or a mouse, recognizes a user operation, and converts it into an operation command.
The output unit 30 converts the authentication result of the face authentication unit 25 and the learning result of the additional learning execution unit 27 into a format that can be confirmed by the operator, and displays the converted result on the output device 19.

  The temporary data storage unit 31 includes a tracking result storage unit 33, a face image storage unit 34, and an authentication result storage unit 35. The tracking result storage unit 33 stores information such as the location of a person and the area on the screen for each tracking target. The face image storage unit 34 stores a face image of each frame to be tracked. The authentication result storage unit 35 stores an authentication result for each tracking target frame.

  The data storage unit 32 includes a face authentication model 36, an initial learning face image 37, an additional learning face image 38, and an additional learning setting information 39. The face authentication model 36 is a face authentication model DB 13 developed on a memory. The initial learning face image 37 is obtained by expanding the initial learning face DB 14 on the memory. The additional learning face image 38 is the additional learning face DB 15 expanded on the memory. The additional learning setting information 39 is obtained by expanding the additional learning setting DB 16 on the memory.

  FIG. 4 is a view showing the data structure of the initial learning face image 37. As shown in FIG. FIG. 5 is a view showing the data structure of the additional learning face image 38. As shown in FIG. In either case, face image data is stored corresponding to a person ID for identifying each person. Using the image data of the initial learning face image 37 and the additional learning face image 38, a face authentication model 36 for each person ID is constructed.

  Hereinafter, the operation of the person authentication process in the person authentication apparatus 5 will be described by dividing it into a face authentication process, an additional learning image storage process, and an additional learning process.

FIG. 6 is a diagram showing a sequence of face recognition and additional learning image storage.
When a video is input from the video input device 2, the video input unit 21 performs video decoding and preprocessing (step S1). The contents of the preprocessing are, for example, white balance adjustment and brightness adjustment. The image for each frame is sent to the person tracking unit 22. The human tracking unit 22 tracks the person (tracking target) in the image, and stores the tracking result in the tracking result storage unit 33 (step S2). Details of step S2 will be described later with reference to FIGS.

  The tracking result of the person is sent to the face extraction unit 23. When there is a person (tracking target) for which the authentication process has not been performed, the following is performed. The face extraction unit 23 extracts a face area for each person being tracked, and stores a face image in the face image storage unit 34 (step S3). Face region extraction methods include face detection methods using cascade classifiers based on Haar feature values. Details of step S3 will be described later with reference to FIG.

  The extracted face image is sent to the face feature amount calculation unit 24. The face feature amount calculation unit 24 calculates a face feature amount from the face image (step S4). The face authentication unit 25 performs face authentication processing from the calculated face feature amount. That is, the ID of the registered person to which the face feature value belongs is identified using the face authentication model 36, and the authentication result is stored in the authentication result storage unit 35 (step S5). Details of step S5 will be described later with reference to FIG.

  If the face authentication process (step S5) is successful, the identified person ID is notified to the external system via the external cooperation unit 28 (step S6). Thus, for example, an operation of automatically opening the door in the entry / exit management system is performed.

  After completing the face authentication process of the frame, the additional learning face storage unit 26 performs a process of selecting a face image for additional learning for additional learning (step S7). Details of step S7 will be described later with reference to FIGS. If there is a face image for additional learning, the face image 38 to be added is stored in the face DB for additional learning 15 (step S8).

FIG. 7 is a diagram showing a sequence of the additional learning implementation.
When receiving the relearning instruction from the operator, the input accepting unit 29 sends an additional learning request to the additional learning performing unit 27. The additional learning execution unit 27 reads a face image from the initial learning face DB 14 and the additional learning face DB 15 (step S9). The face feature amount calculation unit 24 calculates a face feature amount on the read face image (step S10), and returns the result to the additional learning execution unit 27. The additional learning execution unit 27 additionally learns the face authentication model 36 that enables identification of a plurality of individuals by learning the SVM classifier (step S11).

  The additional learning result is sent to the output unit 30, and a screen of the learning result is generated and presented to the operator (step S12). When the operator accepts the additional learning result, the operator updates the face recognition model 36 by pressing a button reflecting the additional learning (step S13). Details of steps S12 and S13 will be described later with reference to FIG.

Hereinafter, the main steps in FIGS. 6 and 7 will be described in detail.
FIG. 8 is a diagram illustrating a process flow (step S2) of the human tracking by the human tracking unit 22.
For the input image, feature points that are easy to track such as edge portions are extracted from the image (step S101). Using the extracted feature points, the motion trajectory of each feature point is calculated by matching the feature points of the previous frame (step S102). Trajectory information on feature points is updated using information on the person being tracked (step S103). In the initial state, since there is no person being tracked, nothing is performed in this step S103. The human region is updated using the information of all feature points and the information of the person being tracked (step S104). The tracking result of the person being tracked is updated using the information on the person area (step S105).

  It is determined whether there is an untracked person area at present (step S106). If there is an untracked human region, the tracking target is added to the tracking result (step S107). After completion of step S107, or when there is no untracked human region, feature points other than the human region are deleted (step S108). The tracking result is output to the face extracting unit 23 and stored in the tracking result storage unit 33 (step S109).

  FIG. 9 shows the data structure of the tracking result storage unit 33. As shown in FIG. The tracking result storage unit 33 includes a frame ID, a tracking target ID, and a tracking target area. The frame ID is a frame number of the image. The tracking target ID is an identification number of the tracking target present in the image of each frame ID. When the tracking target exists over a plurality of frame images, one tracking target ID is included in the plurality of frame IDs. The tracking target area indicates a region (positional coordinates) in which the tracking target is present in the frame.

  FIG. 10 is a diagram showing the tracking result shown in FIG. 9 on a map. The map 40 corresponds to the imaging range of the video input device 2. There are two tracking targets in the map 40, and the tracking target IDs are “1” and “2”, respectively. The tracking target ID = “1” is moving between frames, 41a indicates the tracking target trajectory, and 41b indicates the current position of the tracking target.

  FIG. 11 is a diagram showing the data structure of the face image stored in the face image storage unit 34 as a result of the face area extraction process (step S3) by the face extraction unit 23. The face image storage unit 34 includes a frame ID, a tracking target ID, and face image data. The frame ID and the tracking target ID correspond to the data of the tracking result storage unit 33 of the person whose face image is extracted. The face image data is stored as binary data of the face image.

  FIG. 12 is a diagram showing the data structure of the authentication result stored in the authentication result storage unit 35 as a result of the face authentication process (step S5) by the face authentication unit 25. As shown in FIG. The authentication result storage unit 35 includes a frame ID, a tracking target ID, authentication result details, and a person ID. The frame ID and the tracking target ID are associated with the data of the face image storage unit 34. The authentication result details indicate an individual (person ID) to which the face belongs and its probability (also called similarity or reliability). In the authentication result details, when the probability of belonging to an individual exceeds a predetermined value (authentication success threshold), it is regarded as authentication success and updated to the person ID of the individual. If the authentication fails, the person ID is represented by "-1".

  Since the person being tracked is managed by the same tracking target ID across different frames, if authentication succeeds in one frame, all frames that match the tracking target ID even if authentication fails in another frame It is possible to identify the person ID of the tracking target in. For example, with regard to the tracking target ID = “2”, the authentication fails at the frame ID = 12, but the authentication succeeds with the frame ID = 15 (person ID = 001), so It is possible to trace back and identify the person ID of the tracking target ID = "2".

Next, the additional learning face image selection processing (step S7) by the additional learning face storage unit 26 will be described in detail.
FIG. 13 is a diagram showing the data structure of the additional learning setting information 39 used in the additional learning face image selection process. The additional learning setting information 39 includes a parameter name and a parameter value which are conditions for face image selection. The parameters include a person reliability upper limit T ₁₁ , a person reliability lower limit T ₁₂ , a face completeness lower limit T ₂ , an added person face difference lower limit T ₃ , and a number of waiting frames F for implementation.

Person confidence limit T ₁₁ is the upper limit of confidence belonging to registered person for choosing as the additional learning face data (= similarity). Here, the reliability is represented by a probability value belonging to a registered person obtained as a result of the face authentication process. By setting the upper limit value of the reliability to a value less than 1 (for example, 0.96), there is an effect of preventing the image to be added from being biased to the registered image.
Person confidence limit T ₁₂ is the lower limit of the confidence that belong to the registered person for choosing as the additional learning face data.

Face completeness lower T ₂ are, the lower limit value of the face image completeness for choosing as the additional learning face data. The face image perfection degree is a ratio indicating how many parts such as the mouth, eyes, and nose should be included in the face image. The higher the value, the more effective the face image. If this value is low, some faces may be hidden or may not be face images in the first place. By setting the lower limit value, there is an effect of preventing a decrease in authentication accuracy accompanying the additional learning.

Already added the person face differential degrees lower limit T _3, the face image of the plan to add is, the person face image and how much different or already added (registered), show a lower limit value of the difference degree. By setting the lower limit value to a value larger than 0 (for example, 0.05), it is possible to avoid re-addition of a face image similar to an image already subjected to additional learning, and as a result, it is possible to prevent learning bias. There is.

  The implementation waiting frame number F is the number of waiting frames until the selection of the additional learning face image is performed after the tracking target disappears from the visual field.

FIG. 14 is a diagram showing a flow of additional learning face image selection processing (step S7).
The additional learning face storage unit 26 reads face image data associated with one authentication result from the face authentication result storage unit 35 and the face image storage unit 34 (step S201). That is, the tracking target ID stored in the face authentication result storage unit 35, the authentication result details, the person ID, and the face image data of the face image storage unit 34 related thereto are prepared.

  It is determined whether the “number of disappearing frames” of the tracking target ID matches the number F of pending frames in the additional learning setting information 39 (step S202). Here, the “number of disappearing frames” is the number of frames until the tracking target ID disappears from the frame image and appears again. If the number of disappearing frames matches the number of pending operation frames F, the selection process is continued. If they do not match, the processing of the authentication result is interrupted, and the process returns to step S201 to process the next authentication result.

  The authentication result for the tracking target ID is checked, and whether or not there is a frame for which the authentication target has succeeded for the tracking target ID, that is, there is a frame for which the person ID of the face authentication result storage unit 35 is other than “-1”. It is determined whether or not it is (step S203). If there is a successfully authenticated frame, the selection process is continued. If there is no frame for which authentication has succeeded, the processing of the authentication result is interrupted, and the process returns to step S201 to process the next authentication result.

  Next, for the face image, calculation of the degree of perfection of the face (step S204) and calculation of the degree of human face difference (step S205) are performed. The face image perfection degree is a value indicating how much part of the face such as the mouth, the eyes, and the nose should be included in the face image. The personal face difference degree is a value indicating how much the face image is different from the added personal face image. The data used as the added personal face image may be both the initial learning face image 37 and the additional learning face image 38, or one of them. Using the calculated face image perfection degree and the personal-face difference degree, it is determined whether the face image is effective as an additional learning image (step S206). The details of step S206 will be described with reference to FIG.

  The process branches depending on the determination result of the validity of step S206 (step S207). If it is determined that the face image is valid, the face image is selected as an additional learning target (step S208). If it is determined that the authentication result is not valid, the processing of the authentication result is interrupted, and the process returns to step S201 to process the next authentication result. It is determined whether the processing of all the authentication results has been completed (step S209), and if completed, the processing is terminated. If it has not been completed, the process returns to step S201 to process the next authentication result.

FIG. 15 is a diagram showing details of the additional learning face image validity determination flow (step S206) in FIG. In this determination, the upper user confidence limit T ₁₁ , the lower user confidence degree lower limit T ₁₂ , the lower limit of perfection degree T ₂ , and the added upper limit face difference lower limit T ₃ set in the additional learning setting information 39 of FIG. 13 are used. Do.

First, the person's credibility (probability at the time of authentication) of the face image is determined (step 211). If person reliability is greater than himself reliability lower T _12, and smaller than himself reliability limit T _11, to continue the selection process. If not, it is invalidated (step S215). Next, face image perfection is determined (step S212). If the face image perfection face greater perfection lower T _2, to continue the selection process. If not, it is invalidated (step S215). Further, the determination of the personal-face difference degree is performed (step S213). If principal face difference degree is larger than the already added principal face differential degrees lower T _3, and enable the determination result as an additional learning face images of the face image (step S214). If not, it is invalidated (step S215).

Through the above processing, it is possible to appropriately select a face image to be a target of additional learning using the face authentication result. The selection condition at that time is
(1) To set not only the lower limit T ₁₂ but also the upper limit T ₁₁ as the threshold of the person's credibility,
(2) To set a lower limit T ₂ as a threshold of the perfectness of the face image,
(3) a lower limit T ₃ as a threshold value of the principal face differential degrees,
Since this is done, it is possible to prevent bias during additional learning and to prevent false detection during authentication determination.

  The above selection conditions (1) to (3) do not have to be all satisfied, and if either of (1) or (3) is satisfied, it is possible to prevent bias during additional learning. Therefore, the validity determination flow of FIG. 15 may be used by reassembling the selection conditions appropriately according to the usage environment.

  FIG. 16 is a diagram showing a confirmation screen (steps S12 and S13) of the learning result in the additional learning of FIG. In the confirmation screen 50, an authentication test is performed on the test data of the initial learning face DB and the additional learning face DB as the additional learning result 51, and the authentication failure before the additional learning and after the additional learning Post the probability of). The false rejection rate is the probability that the face image of the person can not be authenticated as the person, and the false acceptance rate is the probability of falsely recognizing the face image of another person as the person. The rejection rate is reduced by additional learning. Although the false acceptance rate has increased by learning, it does not pose a problem because it is originally close to 0%. The operator can confirm the additional learning result 51 and can accept the additional learning by pressing the button 52 reflecting the additional learning. Further, pressing the cancel button 53 can stop the additional learning.

  In the above description, the additional learning is performed according to the instruction of the operator. However, the additional learning may be performed periodically or every time of face authentication.

  According to the first embodiment, when the person authentication system is in operation, by adding a face image having a low degree of reliability of face authentication to the target of additional learning, it is possible to improve the resistance to the aging of the face and the change of illumination. In addition, by excluding a face image with a sufficiently high degree of reliability and a face image that has already been learned in large amounts as a target for additional learning, there is an effect of preventing bias in learning. Further, by excluding a face image with a low degree of perfectness of face image from the target of the additional learning by checking the face completeness, it is possible to prevent the deterioration of the authentication accuracy.

  FIG. 17 is a diagram showing another configuration example of the person authentication system including the person authentication device of the present invention. In the person authentication system 1a, the person authentication device 5 and the person authentication management device 4 are connected via a network 6a including a wide area network (WAN), a virtual private network (VPN), and the like. As a result, the authentication result can be displayed by the person authentication management device 4 at a remote place from the person authentication device 5, and the setting of the person authentication device 5 can be performed.

  On the other hand, as in the first embodiment, the video input device 2 and the person authentication device 5 can be connected via the network 6, receive a video from the video input device 2, and specify the face of the person currently appearing. it can. According to the configuration of the present embodiment, it is possible to maintain the person authentication apparatus from a remote place.

In the third embodiment, preferable setting of the upper limit and the lower limit of the reliability in the additional learning setting information 39 will be described.
FIG. 18 is a view showing the relationship between the additional learning setting information 39 and the face authentication success threshold. The horizontal axis shows the number of image frames, the vertical axis shows the reliability (probability of authentication result) in each frame, and the reliability for a certain tracking target (person) changes as shown by a curve 60. In the face authentication process, the tracked relative reliability for judging the authentication success threshold (face authentication success threshold) When T _0, the reliability of period P ₀ authentication success exceeds the threshold T ₀ It becomes a period.

In contrast conditions of a face image to be used for additional learning, as shown in additional learning setting information 39 in FIG. 13, and an upper limit T ₁₁ and lower T ₁₂ the person reliability of authentication. In particular, in this embodiment, personal confidence limit T ₁₁ is greater than the face authentication succeeds threshold T _0, and identity confidence limit T ₁₂ is set smaller than the face authentication succeeds threshold T _0. As a result, the effectiveness of the determination processing in FIG. 15, the period which is effective as the additional learning face image, a reliability limit T ₁₁ and lower T ₁₂ sandwiched by periods P ₁ and P _2. These periods P ₁ and P ₂ include not only a face image that has been successfully authenticated, but also an image that has failed in authentication as indicated by P ₃ .

However, the image areas of P ₃ is reliability is authentication failed images by the difference of only, but rather beneficial as for additional study. For example, only initial face images are present at the time of initial learning, but additional learning at the time of operation improves resistance to authentication by adding face images to the authentication model when the face orientation changes. It has the effect of

  As mentioned above, although the Example of this invention was described, this invention is not limited to an above-described Example, A various modified example is included. The above-described embodiments are described in detail to explain the present invention in an easy-to-understand manner, and are not necessarily limited to those having all the configurations described. Also, part of the configuration of one embodiment can be replaced with the configuration of another embodiment, and the configuration of another embodiment can be added to the configuration of one embodiment. In addition, with respect to a part of the configuration of each embodiment, it is possible to add, delete, and replace other configurations.

  For example, in the above embodiment, the number of video input devices 2 may be more than one. Also, there may be a plurality of person authentication devices 5 and a plurality of person authentication management devices 4 respectively, and the plurality of devices may be associated with a plurality of video input devices 2 in cooperation with each other. Further, at least two of the video input device 2, the person authentication device 5, the video recording device 3, and the person authentication management device 4 described above may be configured in the same device.

  In addition, each configuration, function, processing unit, etc. described above may be realized by hardware, for example, by designing part or all of them with an integrated circuit. In addition, the technical elements of the above-described embodiment may be applied alone or may be divided and applied to a plurality of parts such as program parts and hardware parts.

1: Person authentication system,
2: Video input device,
4: Person authentication management device,
5: Person authentication device,
21: Video input unit,
22: Person tracking department,
23: face extraction unit,
24: face feature amount calculation unit,
25: Face recognition unit,
26: Additional learning face storage section,
27: Additional Learning Department,
36: Face recognition model,
38: Face image for additional learning,
39: Additional learning setting information.

Claims (4)

In a person authentication apparatus for authenticating a person from a face image of a person,
An image input unit for inputting an image including a person;
A human tracking unit that tracks a movement trajectory of a person from the input video;
A face extraction unit that extracts a face image for the tracked person;
A face feature amount calculation unit that calculates a face feature amount from the extracted face image;
A face authentication unit for authenticating the tracked person from the calculated face feature amount using a registered face authentication model of the person;
An additional learning face storage unit that selects and stores a face image for additional learning from a face image of a person who has been authenticated by the face authentication unit;
And an additional learning unit configured to perform additional learning of the face recognition model using the selected additional learning face image.
The additional learning face storage unit selects, as the additional learning face image, a face image in which the authentication reliability of the person who has been authenticated by the face authentication unit is between two upper and lower thresholds T ₁₁ and T _12. A person authentication apparatus characterized in that. In a person authentication apparatus for authenticating a person from a face image of a person,
An image input unit for inputting an image including a person;
A human tracking unit that tracks a movement trajectory of a person from the input video;
A face extraction unit that extracts a face image for the tracked person;
A face feature amount calculation unit that calculates a face feature amount from the extracted face image;
A face authentication unit for authenticating the tracked person from the calculated face feature amount using a registered face authentication model of the person;
An additional learning face storage unit that selects and stores a face image for additional learning from a face image of a person who has been authenticated by the face authentication unit;
And an additional learning unit configured to perform additional learning of the face recognition model using the selected additional learning face image.
Said additional learning face storage unit, characterized in that the difference degree indicating how much different from the already added the face image as the face authentication model to select a predetermined threshold value T ₃ is greater than the face image as the additional learning face image Person authentication device to be. In the person authentication apparatus according to claim 1 or 2,
The additional learning face storage unit further uses, as the additional learning face image, a face image having a face image perfection degree greater than a predetermined threshold T ₂ indicating how many parts of the face to be originally provided in the face image. A person authentication apparatus characterized by selecting. In the person authentication apparatus according to claim 1,
When a threshold of authentication reliability for determining that the face authentication unit authenticates a person as authentication success is set as an authentication success threshold T ₀ ,
Said additional learning face storage unit is the authentication reliability of the upper T ₁₁ used in selecting as the additional learning face image is larger than the authentication success threshold T _0, the authentication confidence limit T ₁₂ of the authentication success A person authentication apparatus characterized in that it is smaller than a threshold T ₀ .

Images