JP2007156541A - Person recognition apparatus and method and entry/exit management system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a person recognition apparatus and method and an entry/exit management system capable of recognizing subjects without depending on speeds at which the subjects pass and the number of subjects. <P>SOLUTION: The person recognition apparatus, for recognizing previously registered persons by means of the persons' face feature information, controls the accuracy of a recognition process and the subjects depending on the degree of load applied to the recognition process according to information about speeds at which the subjects pass, the number of passengers, distances to the passengers, and the time elapsed since the start of the recognition process. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention acquires, for example, biometric feature information such as a subject's face, retina, mouth, iris, fingerprint, palm print, ear, and part of a human body vein, and at least one biometric feature information registered in advance. The person recognition apparatus and the person recognition method for determining whether or not the target person is a person registered in advance.

  The present invention also relates to a person exit management system for managing entrance / exit to an entrance / exit area such as a room or facility that requires security using the person recognition apparatus and person recognition method.

  Recently, for example, biometric feature information such as the subject's face, retina, mouth, iris, fingerprint, palm print, ear, and part of the human body's vein has been acquired and collated with at least one pre-registered biometric feature information As a result, a technology has been developed to determine whether or not the target person is a pre-registered person, and to manage entrance and exit and output an alarm according to the determination result. . These technologies include various methods such as allowing the recognition target person to stop each time and intentionally perform the recognition process, or performing a recognition process while the recognition target person is walking without knowing the person. Application examples can be seen in various combinations.

  However, in particular, in a person recognition device or a person exit management system that performs person recognition using biometric feature information such as a face image of a target person, a case where a large number of target persons pass continuously or a target person (passers) ) Has a problem that the processing time is not enough and high-accuracy recognition processing cannot be performed.

Therefore, recently, as a similar technique for such a problem, for example, the degree of similarity between registered data is obtained by brute force, and the order of persons to be searched next is changed based on the degree of similarity with input collation data. There exists a technique (for example, refer patent document 1).
JP 2002-163652 A

  However, the technique of Patent Document 1 is mainly a technique for increasing the priority of similar persons within a registrant, and is effective in speeding up recognition processing for a plurality of subjects who pass continuously. It is not something that can be done. Therefore, the technique of Patent Document 1 cannot solve the conventional problems.

  SUMMARY OF THE INVENTION An object of the present invention is to provide a person recognition device, a person recognition method, and an entrance / exit management system that can recognize a subject without depending on the speed of the subject and the number of people.

  The person recognition device of the present invention checks whether or not the subject is a pre-registered person by collating biometric feature information obtained from the subject with at least one or more biometric feature information registered in advance. In the person recognition device to be determined, an image acquisition unit that acquires an image including biological information of the subject, a biological detection unit that detects a biological region from the image acquired by the image acquisition unit, and the biological detection unit Biometric feature extraction means for extracting personal biometric feature information from the biometric area, registered biometric feature management means in which personal biometric feature information previously obtained from the biometric information of the subject is registered, and the subject The information is determined by the recognition processing accuracy determination means for determining the recognition processing accuracy according to the information related to the determined target person, and the recognition processing accuracy determination means. The target person is a person registered in advance by collating the biometric feature information extracted by the biometric feature extraction unit with the biometric feature information registered in the registered biometric feature management unit based on the recognition processing accuracy. Recognizing means for determining whether or not there is.

  According to the person recognition method of the present invention, it is determined whether or not the target person is a pre-registered person by collating biometric feature information obtained from the target person with at least one or more pre-registered biometric feature information. In the person recognition method to be determined, an image acquisition step for acquiring an image including the biological information of the subject person, a biological detection step for detecting a biological region from the image acquired by the image acquisition step, and the biological detection step A biometric feature extraction step for extracting personal biometric feature information from the biometric region, a recognition processing accuracy determination step for obtaining information relating to the subject, and determining a recognition processing accuracy in accordance with the obtained information relating to the subject, The biometric features extracted by the biometric feature extraction step based on the recognition processing accuracy determined by the recognition processing accuracy decision step. The subject is registered in advance by collating the information with the biometric feature information registered in the registered biometric feature management means in which the individual biometric feature information previously obtained from the subject's biometric information is registered. And a recognition step for determining whether the person is a person.

  In the entrance / exit management system according to the present invention, is the entrance / exit person a pre-registered person by collating biometric feature information obtained from the entrance / exit with at least one pre-registered biometric feature information? In an entrance / exit management system that controls whether or not the entrance / exit person enters / exits the entrance / exit area based on the determination result, an image acquisition unit that acquires an image including biometric information of the entrance / exit person, and Biological detection means for detecting a biological area from the image acquired by the image acquisition means, biological feature extraction means for extracting personal biological feature information from the biological area detected by the biological detection means, and biological bodies of the entrance and exit persons in advance The registered biometric feature management means in which the individual biometric feature information obtained from the information is registered, and information related to the entry / exit person is obtained, and the obtained entry / exit person is obtained. Recognition processing accuracy determination means for determining the recognition processing accuracy according to the information, and the biometric feature information extracted by the biometric feature extraction means and the registered biometric feature based on the recognition processing accuracy determined by the recognition processing accuracy determination means Recognizing means for determining whether or not the entry / exit person is a person registered in advance by collating with the biometric feature information registered in the management means, and the entry / exit based on the search result of the recognition means Entry / exit control means for controlling entry / exit of the person to / from the entry / exit target area.

  According to the present invention, the accuracy of the recognition process and the target person are determined based on the degree of load applied to the recognition process from the information on the speed of the target person, the number of people passing, the distance from the target person, and the elapsed time from the start of the recognition process. A person recognition device, a person recognition method, and an entrance / exit management system that enable recognition of a target person without depending on the speed and number of people of the target person by completing recognition processing as quickly as possible by controlling Can provide.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
In the following description, it is determined whether or not the target person is a pre-registered person by comparing the face characteristic information obtained from the target person with at least one or more pre-registered face characteristic information. The present invention can be applied to a method for recognizing a target person by inputting an image including various biological feature information other than the face, and includes a retina, mouth, iris, The present invention can also be applied as a person recognition apparatus using biometric feature information such as fingerprints, palm prints, ears, and some veins of the human body.

  In addition, as a target to which the present invention is applied, for example, a place where only a designated person is allowed to pass or a case where a person requiring attention is searched from passers-by are assumed. It is especially effective in places where there are a large number of people who pass through where you want to pass, such as for entrance / exit management systems that manage people entering and exiting buildings and corporate buildings, entertainment facilities, and transportation facilities. An application such as an entrance / exit management system for managing entrance / exit is assumed.

First, the first embodiment will be described.
FIG. 1 schematically shows a configuration of a person recognition apparatus according to the first embodiment. This person recognition apparatus captures images of a face image (an image including at least a face) of a target person (hereinafter also referred to as a person or an entrance / exit) M as an image acquisition unit that captures and inputs the image. An image input unit 101 that performs image processing such as A / D conversion and digitization of the captured image, and biometric detection that detects the position of a facial part such as the face, eyes, and nose from the image obtained from the image input unit 101 The face detection unit 102 as a means, the face feature extraction unit 103 as a biometric feature extraction means for extracting personal face feature information from the image of the face area detected by the face detection unit 102, and the recognition result from the face detection result Recognition processing accuracy determination that determines the accuracy of recognition processing based on the relationship between the estimated processing time for recognition processing and the time limit for which recognition results must be obtained by obtaining the number of people and the speed of traffic. A calculation load determination unit 104 as a means, a registered face feature management unit 105 as a registered biometric feature management means that registers and manages personal face feature information obtained beforehand from the face image of the subject, a calculation load determination unit Whether the target person is a person registered in advance by collating the input facial feature information with the facial feature information registered in the registered facial feature management unit 105 based on the recognition processing accuracy determined in 104. A recognition unit 106 that determines whether or not, an output unit that outputs a control signal in accordance with the recognition result of the recognition unit 106 or the recognition result itself by sound or image, and an output unit 107 as an entrance / exit control unit. ing.

Hereinafter, each part will be described in detail.
The camera 100 captures and inputs a face image of the subject M, and is mainly composed of, for example, a television camera using an image sensor such as a CCD sensor. Note that if the calculation load determination unit 104 uses the auxiliary means other than the image when obtaining the travel speed of the subject M, an input means such as a distance sensor or a speed measuring device is provided along with the camera 100. Is also possible. In this case, an image captured by the camera 100 is output to the image input unit 101, and information acquired by an input unit such as a distance sensor or a speed measuring device is output to the calculation load determination unit 104.
The image input unit 101 performs A / D conversion on the image captured by the camera 100 and digitizes the image, and then sends the image to the face detection unit 102.

  The face detection unit 102 first detects the face area of the subject M from the image input by the camera 100. For example, in the input image from the camera 100, a correlation value is obtained while moving a template prepared in advance in the image, so that the position that gives the highest correlation value is set as the face region. In addition, face detection means such as a face region extraction method using an eigenspace method or a subspace method may be used.

  Next, the face detection unit 102 detects the position of a face part such as eyes, nose, and mouth from the detected face area. As the detection method, for example, Document [1] (Kazuhiro Fukui, Osamu Yamaguchi: “Face feature point extraction by combination of shape extraction and pattern matching”, IEICE Transactions (D), vol. J80-D- II, No. 8, pp 2170-2177 (1997)) may be used.

  In addition to the above eye and nose detection, the detection of the mouth area is described, for example, in the literature (Mayumi Yuasa, Akiko Nakajima: “Digital Make System Based on High-Precision Facial Feature Point Detection” Proc. It can be easily realized by using the technology of pp219-224 (2004).

  As for fingerprints and vein patterns, for example, reference [2] (Optical Industry Technology Promotion Association (http://www.oitda.or.jp/index-j.html): FY2003 Optical Technology Trend Survey This can be realized by using the input means and technology described in “15-003-1 Optical Technology Trend Survey Report”, Chapter 5 “Human Interface” (2003)). In any case, it is possible to acquire information that can be handled as a two-dimensional array image and detect a region of the biometric feature from the acquired information.

  In addition, in these processes, in order to extract only one piece of biometric feature information from one image, the correlation value with the template is obtained for all images, and the maximum position and size may be output. To extract multiple pieces of biometric feature information, find the local maximum correlation value for the entire image, narrow down the candidate face positions in consideration of the overlap in one image, and finally enter the past It is also possible to finally find a plurality of pieces of biological feature information at the same time in consideration of the relationship (temporal transition) with the image. In the present embodiment, an effect is likely to appear particularly when a plurality of pieces of biological feature information are detected at the same time.

  The face feature extraction unit 103 cuts out a face area into a certain size and shape based on the position of the component detected by the face detection unit 102, and uses the density information as a feature amount. Here, the gray value of an area of m pixels × n pixels is used as information as it is, and m × n-dimensional information is used as a feature vector. These are normalized so that the vector and the length of each vector are set to “1” by a method called a simple similarity method, and a similarity indicating the similarity between feature vectors is obtained by calculating an inner product.

  If it is a process that produces a recognition result with a single image, the feature extraction is completed in this way, but a more accurate recognition process can be performed by calculating with a moving image using a plurality of consecutive images. For this reason, in this embodiment, this method will be described as an example. Similar to the feature extraction unit 103, images of m × n pixels are cut out from images continuously obtained from the camera 100, a correlation matrix of feature vectors is obtained from these images, and an orthonormal vector by a well-known KL expansion is obtained. To calculate a subspace indicating the facial features obtained from successive images.

The subspace calculation method calculates a subspace by obtaining a correlation matrix (or covariance matrix) of feature vectors and obtaining an orthonormal vector (eigenvector) by KL expansion. In the subspace, k eigenvectors corresponding to eigenvalues are selected in descending order of eigenvalues and expressed using a set of the eigenvectors. In the present embodiment, the correlation matrix Cd is obtained from the feature vector, and the correlation matrix is obtained.

And a matrix Φ of eigenvectors is obtained. This information becomes a partial space indicating the characteristics of the face of the person currently recognized. In addition, regarding the partial space registered in advance in the registered face feature management unit 105, it is only necessary to register the partial space itself calculated here.

The calculation load determination unit 104 calculates the calculation load by integrating the number of faces searched from the input image, the position where the current face is supposed to be present, the passing speed of each person, and the like, and the recognition unit 106 In what processing accuracy should be calculated for each face, the processing method for each face or the entire processing method is determined and output. The output value here is a coefficient that indicates the length of recognition processing time allowed per person. The larger the number, the more processing time is available and the more accurate calculation can be done, and vice versa. If it is small, it indicates that there is no margin in processing time and only simple processing with accuracy should be performed. Basically, how to get these numbers
・ Coefficient adjustment to shorten the processing time per person as the number of people passing at the same time increases.
・ The faster the passerby's walking speed, the faster the coefficient adjustment to shorten the processing time per person.
-The coefficient adjustment is performed so that the longer the elapsed time from the start of the same person recognition process, the longer the processing time per person.
It is sufficient to obtain a calculation formula so as to comply with the above, and a function with the above three elements as coefficients may be obtained. For example, by creating a table as shown in FIG. A coefficient of processing time is obtained from the “number of people N” and “passing speed V”, and the processing time to be used for face recognition is obtained by multiplying this coefficient by “elapsed time T from the start of recognition processing”.

  Here, FIG. 3 will be specifically described as an example. In the state of the image [1], everyone is still a recognition target, and at the same time there are four persons A to D on the screen, and the front side of the screen. The situation where everyone is walking toward. Here, since there are many people, it outputs so that processing with short processing time may be selected. If the persons A and C can be recognized with good reliability, they are excluded from the recognition targets. Here, “sufficient reliability is obtained” indicates that the degree of similarity in recognition is sufficiently high and there is a margin enough to say that the person is sure.

  Subsequently, at the time of the image [2], there are also four persons A to D. At this time, the persons A and C are already recognized by the tracking process (tracking) from the previous image. Therefore, the persons A and C are excluded from the processing target of the similarity calculation, and the similarity calculation is performed only for the remaining two persons B and D.

Here, since the two persons A and C have already been excluded from the target, the processing of the remaining two persons B and D is performed, and therefore, similarity calculation that requires more processing time than the image of [1] is performed. You may set it to do. However, when the similarity calculation with high accuracy that takes processing time is performed, the reliability of the calculation itself becomes high. Therefore, it is not necessary to relax the threshold processing for the similarity calculation compared to the case of the image [1]. There is no.
In the case of the image [2], if it is not possible to determine “who” with sufficient certainty in the recognition process for the remaining two persons, the process continues with the next image.

  Subsequently, in the image [3], since the person A disappeared from the field of view, the remaining three persons may be processed. However, since the person C has already been recognized, the remaining two persons are similar to the image [2]. What is necessary is just to perform the similarity calculation of the persons B and D. However, since the time elapsed since the start of recognition is different for the image [3] compared to the image [2], and the time for the [3] image has elapsed, Change the processing time to a slightly longer one. Here, it is assumed that the person B can be recognized.

  When shifting to the image of [4], the person B also disappears from the field of view, and the person C has already been recognized, so that there is only one person of the remaining person D. In this case, since there is only one person, the processing time can be sufficiently longer than the image up to [3]. If the person D can also be recognized, the image [5] does not have to be processed.

  In this way, when a large number of people appear in a traffic place, it is ideal to shift to a process with a long calculation time in accordance with the elapsed time or the number of people. As for “traffic speed” that did not appear in the example here, if the same number of people are passing, the faster the passing speed, the shorter the time the face is reflected in the screen. In order to be able to recognize as reliably as possible before disappearing from the screen, the processing is switched to a processing with a longer processing time.

  Whether the distance from the camera 100 to the person M is far or near can also be seen in the image. If it is close to the camera 100, the time until it disappears from the screen is short. Judge whether or not. The travel speed can be calculated by image processing because the amount of movement per unit time can be estimated from the face tracking results, and it is also possible to actually calculate and reflect the travel speed using a distance sensor or speed meter. .

  Similarly, the distance can be estimated based on the size of the face and the position in the screen, but the distance can be surely understood by using the distance sensor. The distance sensor and speed measuring instrument described here may be provided in the camera 100 and the output information may be sent to the measurement load determination unit 104.

  In the example shown in the image of [1] in FIG. 3, it can be seen that the person A is close to the camera 100 and the person D is far from the camera 100. For these, the actual distance may be measured using a distance sensor, and since the proximity of the face can be calculated based on the size of the face in the image and the Y coordinate, this method may be used.

  Here, an example is shown in which the length of processing time is simultaneously switched for the entire person on the screen, but the distance from the camera, the elapsed time from the start of recognition processing, and the resolution for obtaining a face image are different for each person. Therefore, a person who has failed in the recognition calculation until the end can recognize whether it is “unregistered” or “registrant” by performing a certain process with a long processing time before it seems to disappear from the screen.

  In FIG. 3, (a) shows a processing content description, (b) shows an input image, and (c) shows the positional relationship of the subject with respect to the camera 100.

  The registered face feature management unit 105 stores and manages the same face feature information as the face feature information input from the camera 100 in order to identify a person. In the present embodiment, the same feature extraction as the face feature information obtained from the camera 100 is used as an m × n feature vector, but a face image before feature extraction may be used or used. A correlation matrix immediately before subspace or KL expansion may be used. These store identification information using an ID number as a key. The face feature information held here may be one per person, or a plurality of registered face feature information may be held so that they can be switched depending on the situation or used for recognition at the same time.

  The recognizing unit 106 obtains a similarity between the input subspace obtained by the feature extracting unit 103 and one or a plurality of subspaces registered in advance in the registered face feature managing unit 105, so that a person registered in advance can be obtained. It is possible to determine whether the current image is present.

As a calculation method for obtaining the similarity between subspaces, a method such as a subspace method or a compound similarity method may be used. The calculation method in the present embodiment is described in, for example, Document [3] (Kenichi Maeda, Sadaichi Watanabe: “Pattern Matching Method Introducing Local Structure”, IEICE Transactions (D), vol. D, No. 3, pp 345-352 (1985)). In this method, recognition data and input data in registration information stored in advance are expressed as partial spaces, and an “angle” formed by two partial spaces is defined as a similarity. The subspace input here is referred to as an input subspace. Similarly, a correlation matrix Cin is obtained for the input data string,

And the eigenvector Φin is obtained. The similarity between subspaces (0.0 to 1.0) of the subspaces represented by two Φin and Φd is obtained and used as the similarity for recognizing this.

  When there are a plurality of faces in the input image, if the similarity calculation with the face image (dictionary) stored in the registered face feature management unit 105 is calculated in a round-robin order, the results for all persons are obtained. Can be obtained. For example, when a person with X names walks and a dictionary with Y names exists, the result of all X names can be output by performing the similarity calculation X * Y times.

  Also, if the recognition result cannot be output as the calculation result when m images are input (when the next frame image is acquired and calculated without being determined by any registrant), it is input to the partial space. It is possible to update the input side subspace by adding one of the frames to the sum of the correlation matrices created in the past multiple frames, calculating the eigenvectors, and creating the subspace again. Become. In other words, when the face images of passersby are photographed continuously and collation is performed, the collation calculation is performed while acquiring the images one by one and updating the partial space. It becomes possible.

  Here, since the calculation load determination unit 104 is notified of the coefficient of the length that can be applied to all or one person individually, the person with a large coefficient is surely authenticated over time. Do processing. On the other hand, a person with a small coefficient performs a calculation with a slightly higher recognition threshold in a process with a short processing time.

Specifically, regarding changes in the length of processing time,
(1) Increasing the resolution of the face region obtained from the input image (see FIG. 4)
(2) Widen the processing target area (see FIG. 5),
(3) Change the number of subspaces using similarity calculation (see reference [3]),
It becomes possible to adjust by such things. The above (1) is often selected when the face is in a closer state, and it is easy to obtain a face image with a higher resolution for a predetermined area of the face. With regard to (2) above, it is possible to initially set only some parts in the face, etc., and gradually increase the number of face parts to be processed or increase the area of the processing target area itself. The number of faces in (3) above is the number of vectors used to calculate the similarity in the mutual subspace method, and there is a strong tendency to increase the number of faces precisely and easily if it decreases. However, if the number of surfaces is increased too much, the method of increasing the processing time changes in a square order or more, so if the number of surfaces is simply increased too much, the recognition accuracy may also decrease.

  4A shows the case of low resolution, FIG. 4B shows the case of medium resolution, and FIG. 4C shows the case of high resolution. 5A shows a case where the target area is small, FIG. 5B shows a case where the target area is intermediate, and FIG. 5C shows a case where the target area is large.

  From the result of the calculation load determination unit 104 and the result of the recognition unit 106, when there is no allowance for the processing time and a short-time collation is required, the reliability as the calculation processing tends to decrease. Basically, the threshold for determining the recognition result is basically set strictly. In other words, the narrowing-down process is performed when a “similarity that can be asserted to be a specific person without fail even when a calculation with low reliability” is obtained.

In this way, it is possible to speed up the processing without lowering the security level, but if there are many other people, if it is allowed to simply lower the threshold, such means can also speed up the processing. Is feasible.
In addition, there is a method of changing the recognition processing accuracy while sequentially changing the input images, but the narrowing processing is performed while sequentially changing the accuracy of one same image. It is also applicable.

  The output unit 107 outputs information itself such as an input image input by the camera 100, a face area and feature information obtained from the face detection unit 102, and a recognition result of the recognition unit 106 to a device serving as an output destination. Processing such as changing the contents of the control signal separately according to the recognition result of the unit 106 is performed.

In the simplest example, it may be a monitoring screen output in which the input face image is displayed side by side as to which person among the persons registered in the registered face feature management unit 105. If there is a similarity that indicates whether there is a similarity, the example that an alarm is sounded if the similarity of the recognized person with the highest similarity is greater than or equal to a predetermined threshold, or entrance / exit management In the system, if the degree of similarity with one of the registered persons is equal to or greater than a predetermined threshold, a gate opening signal is output to the gate device 108 (see FIG. 1) that controls entry / exit for the entry / exit area. Can be easily realized.
As described above, according to the first embodiment, in the person recognition apparatus for recognizing a person registered in advance using the facial feature information of the person, the speed of the target person, the number of people passing, By completing the recognition process at the earliest possible time by controlling the accuracy of the recognition process and the target person according to the degree of load applied to the recognition process from the distance information and the elapsed time from the start of the recognition process, It becomes possible to perform recognition and traffic control of the subject regardless of speed and number of traffic. A summary of these images is shown in FIG.

Next, a second embodiment will be described.
FIG. 7 schematically shows the configuration of the person recognition apparatus according to the second embodiment. This person recognition device captures images of a face image (an image including at least a face) of a target person (hereinafter also referred to as a person or an entry / exit) M as an image acquisition unit that captures and inputs the image. An image input unit 201 that performs image processing such as A / D conversion and digitization of the captured image, and biometric detection that detects the position of a facial part such as the face, eyes, and nose from the image obtained from the image input unit 201 The face detection unit 202 as a means, the face feature extraction unit 203 as a biometric feature extraction means for extracting personal face feature information from the image of the face area detected by the face detection unit 202, and the recognition result from the face detection result The number of persons and the speed of traffic are determined, and the target person and the number of persons to be recognized are determined from the relationship between the estimated processing time for the recognition process and the time limit for which the recognition result must be obtained. A calculation load determination unit 204 as a recognition target determination unit to be registered, a registered face feature management unit 205 as a registered biometric feature management unit that registers and manages personal face feature information obtained beforehand from the face image of the target person, Based on the content determined by the calculation load determination unit 204, the target person is a person registered in advance by collating the input facial feature information with the facial feature information registered in the registered facial feature management unit 205. A recognition unit 206 that determines whether or not there is, an output unit that outputs a control signal in accordance with the recognition result of the recognition unit 206 or the recognition result itself by sound or image, an output unit 207 as an entrance / exit control unit, an entry / exit target The gate device 208 is configured to control entry / exit to / from the area.

  Hereinafter, each unit will be described in detail. The camera 200, the image input unit 201, the face detection unit 202, the face feature extraction unit 203, the registered face feature management unit 205, the output unit 207, and the gate device 208 are described above. Since it is the same as that in the embodiment, the description is omitted.

  Similar to the first embodiment, the calculation load determination unit 204 determines the person to be recognized from the time allowed for the recognition process according to the number of persons and the speed of the target person, the elapsed time from the start of the recognition process, and the distance from the target person. Determine the number of people and the number of people and output. For example, when changing the narrowing target based on the usage frequency of the user, as shown in FIG. 8, the usage frequency of the target person registered in advance is managed, and when the processing time is shortened, the narrowing down is performed. Only people who have a usage frequency of level 1 or higher are recognized. Here, first, recognition processing is performed with only three persons of D, H, and M at the narrowing level 1, and in the narrowing level 2, A, B, D, E, F, G, H, J, K, It shows that M, N, O, P, R and the target person are increased.

  Similarly, when the target person is switched according to the date and place instead of the frequency of use, the target person can be narrowed down according to the contents. In addition, a method of changing the priority level for a person who wants to increase the priority in the case of a person who needs attention or a VIP as viewed from the manager side can be used.

  The recognition unit 206 can output the recognition result using the same recognition method as in the first embodiment. However, in the second embodiment, the person to be recognized is limited by the calculation load determination unit 204. The similarity calculation is performed only for the person who has been selected. If it is determined from the recognition result that there is no corresponding person among the persons registered in the registered special work management unit 206, the recognition target person is narrowed down using the same image or the next input image. Raise the level, increase the number of subjects, and repeat the recognition process. In other words, at first, the recognition process is performed quickly with a small number of people, focusing only on the more frequently used people, and if the recognition result does not appear, the number of people to be recognized is gradually increased. The recognition processing time can be reduced. Similarly, it is possible to apply a method of performing VIP with high priority on subject information by date and place or a person who needs attention.

  As described above, according to the second embodiment, the average time of the recognition process can be effectively shortened without changing the security level even when the recognition process is switched. This can be applied to the method of switching from time to time for continuous input images for pedestrians, but the same applies to the case where the target person recognizes stationary and processing for a single image. Can be applied.

Next, a third embodiment will be described.
FIG. 9 schematically shows a configuration of a person recognition apparatus according to the third embodiment. The person recognition apparatus captures images of a face image (an image including at least a face) of a target person (hereinafter, also referred to as a person or an entry / exit person) M as an image acquisition unit that captures and inputs the image. An image input unit 301 that performs image processing such as A / D conversion and digitization of the captured image, and biometric detection that detects the position of a facial part such as the face, eyes, and nose from the image obtained from the image input unit 301 The face detection unit 302 as a means, the face feature extraction unit 303 as a biometric feature extraction means for extracting personal face feature information from the image of the face area detected by the face detection unit 302, and the recognition result from the face detection result A recognition target that determines the number of people and the speed of travel, and determines the recognition target person's confirmation method based on the relationship between the estimated processing time required for recognition processing and the time limit for which recognition results must be obtained. A calculation load determination unit 304 as a confirmation method determination unit, a registered face feature management unit 305 as a registered biometric feature management unit that registers and manages personal facial feature information obtained from a face image of the subject in advance, a calculation Based on the confirmation method determined by the load determination unit 304, the input facial feature information and the facial feature information registered in the registered facial feature management unit 305 are collated, so that the target person is a person registered in advance. A recognition unit 306 that determines whether or not there is, an output unit that outputs a control signal according to the recognition result of the recognition unit 306 or the recognition result itself by sound or image, an output unit 307 as an entrance / exit control unit, an entry / exit target Gate device 308 for controlling entry / exit to / from the area, history storage unit 309 as storage means for recording the face image of the subject as necessary, recognition as necessary A server 310 as the installed validator to a remote location to perform the management of the interpolation operation is constituted.

  Hereinafter, each unit will be described in detail. The camera 300, the image input unit 301, the face detection unit 302, the face feature extraction unit 303, the registered face feature management unit 305, the output unit 307, and the gate device 308 are described above. Since it is the same as that in the embodiment, the description is omitted.

Similar to the first embodiment, the calculation load determination unit 304 determines the confirmation method from the time allowed for the recognition process according to the number of persons, the traveling speed, the elapsed time from the start of the recognition process, and the distance from the target person. Determine and output. In the present embodiment, when the recognition unit 106 determines that normal recognition processing cannot be performed,
(1) Remote notification,
(2) Save history,
(3) Calculation on another server,
(4) Thinning process of recognition process,
Switch to the confirmation method. One of these methods may be used, and switching may be performed by a combination of two or more methods.

Hereinafter, the confirmation methods (1) to (4) will be described in order, assuming that only switching is performed alone.
In the case of (1), if it is determined that the time required for the recognition process is short, there is a possibility that the recognition process may not be in time, so that only the person who could not perform the recognition process is provided in the output unit 307. Using the communication device for the user, the remote manager is notified that the face image obtained from the camera 300 has not been transmitted and recognized, and that visual confirmation is desired. Here, by performing visual confirmation by the administrator, it is possible to cope with a person whose processing is not in time without confirmation.

  In the case of (2), if it is determined that the time required for the recognition process is short, there is a possibility that the recognition process may not be in time, so only the person who could not perform the recognition process is stored in the history storage unit 309 in the date / time and location. At the same time, the face image obtained from the camera 300 is stored. This can be confirmed later by the administrator, even if it cannot be confirmed on the spot, and can be confirmed even in the event of an emergency, or the face image is newly registered or additional learning of the face is performed. Therefore, the accuracy of recognition processing can be improved.

  In the case of (3), if it is determined that the time required for the recognition process is short, the recognition process may not be in time, so the necessary facial feature information of the face is transmitted to the remote server 310. Then, the similarity calculation in the server 310 is performed. In this case, there is no particular limitation on how to use the cluster personal computer or the method of realizing the client server format. Therefore, there is a confirmation device (server 310) that can perform similar similarity calculation in addition to the recognition unit 306 of the person recognition device. It is assumed that. By returning the result of similarity calculation in the server 310 to the person recognition device, output control by the output unit 307 becomes possible.

  In the case of (4), if it is determined that the time required for the recognition process is short, the recognition process may not be in time, so the recognition unit 306 recognizes from the recognition target person as necessary. By performing decimation that is not performed or by performing decimation processing in units of frames of the input image, it is possible to perform recognition processing while distributing the load as necessary.

  As described above, according to the third embodiment, the average time of the recognition process can be effectively shortened without changing the security level even if the confirmation method is switched. This can be applied to a method of switching from time to time for continuous input images intended for pedestrians, but the same applies to recognition processing for a single image when the subject is stationary. Can be applied.

1 is a block diagram schematically showing the configuration of a person recognition device according to a first embodiment of the present invention. The figure for demonstrating the process example of the calculation load determination part in 1st Embodiment. The figure which shows the relationship between the position of a passerby and the content of a process in 1st Embodiment. The figure for demonstrating the determination method of the recognition process precision in the calculation load determination part in 1st Embodiment. The figure for demonstrating the determination method of the recognition process precision in the calculation load determination part in 1st Embodiment. The schematic diagram explaining the flow of the process in 1st Embodiment. The block diagram which shows roughly the structure of the person recognition apparatus which concerns on the 2nd Embodiment of this invention. The figure for demonstrating the determination method of the recognition process precision in the calculation load determination part in 2nd Embodiment. The block diagram which shows roughly the structure of the person recognition apparatus which concerns on the 3rd Embodiment of this invention.

Explanation of symbols

  M: Target person, person, entry / exit, 100, 200, 300 ... camera (image acquisition means), 101, 201, 301 ... image input section, 102, 202, 302 ... face detection section (biological detection means), 103 , 203, 303... Facial feature extraction unit (biological feature extraction unit), 104, 204, 304... Calculation load determination unit (recognition processing accuracy determination unit, recognition target determination unit, recognition target person confirmation method determination unit), 105, 205 , 305 ... registered face feature management unit (registered face feature management unit), 106, 206, 306 ... recognition unit (recognition unit), 107, 207, 307 ... output unit (output unit, entrance / exit control unit), 108, 208 308 ... Gate device (entrance / exit control means) 309 ... History storage unit (storage means) 310 ... Server (confirmation device).

Claims (26)

In a person recognition device that determines whether or not the target person is a pre-registered person by comparing biometric feature information obtained from the target person with at least one or more pre-registered biometric feature information,
Image acquisition means for acquiring an image including biological information of the subject;
A living body detecting means for detecting a living body region from the image acquired by the image acquiring means;
Biometric feature extraction means for extracting personal biometric feature information from the biometric area detected by the biometric detection means;
Registered biometric feature management means in which individual biometric feature information obtained from the subject's biometric information in advance is registered;
Recognizing process accuracy determining means for determining information related to the target person, and determining a recognition process precision according to the information related to the determined target person;
Based on the recognition processing accuracy determined by the recognition processing accuracy determination means, by comparing the biometric feature information extracted by the biometric feature extraction means with the biometric feature information registered in the registered biometric feature management means, Recognizing means for determining whether or not the target person is a person registered in advance;
A person recognition device comprising:   The person recognition apparatus according to claim 1, wherein the recognition processing accuracy determined by the recognition processing accuracy determination means is a resolution of biometric feature information used for recognition processing.   The person recognition apparatus according to claim 1, wherein the recognition processing accuracy determined by the recognition processing accuracy determination means is a processing target region of biometric feature information used for recognition processing.   2. The person recognition apparatus according to claim 1, wherein the recognition processing accuracy determined by the recognition processing accuracy determination means is calculation accuracy of biometric feature information used for recognition processing.   The recognizing means performs a collation process based on the recognition processing accuracy determined by the recognition processing accuracy determining means, so that when the target person's passage speed is high or when the number of the target persons is large, the recognition means can be started early The collation processing is completed by simple processing, and thereafter, the target person is excluded from the processing target as being verified, and the load of the collation processing is assigned to another target person to be processed. The person recognition device described. In a person recognition device that determines whether or not the target person is a pre-registered person by comparing biometric feature information obtained from the target person with at least one or more pre-registered biometric feature information,
Image acquisition means for acquiring an image including biological information of the subject;
A living body detecting means for detecting a living body region from the image acquired by the image acquiring means;
Biometric feature extraction means for extracting personal biometric feature information from the biometric area detected by the biometric detection means;
Registered biometric feature management means in which individual biometric feature information obtained from the subject's biometric information in advance is registered;
Recognizing target determining means for determining information related to the target person, and determining a target person to be recognized and the number of persons according to the information related to the determined target person;
Based on the target person to be recognized determined by the recognition target determination means and the number of persons, the biometric feature information extracted by the biometric feature extraction means and the biometric feature information registered in the registered biometric feature management means Recognizing means for determining whether or not the target person is a person registered in advance by collating;
A person recognition device comprising:   For the target person to be recognized and the number of persons to be determined by the recognition target determining means, a high priority is set for the person who is frequently used according to the date and place obtained from the past matching history etc. The person recognition apparatus according to claim 6.   The target person to be recognized and the number of persons to be determined by the recognition target determining means set a high priority to be a person to be recognized according to a date and place predetermined by an administrator or the like. Item 7. The person recognition apparatus according to Item 6.   For the target person to be recognized and the number of persons to be determined by the recognition target determining means, the priority level to be recognized is set high for a person who needs to be recognized as a target person or a target person such as VIP. The person recognition apparatus according to claim 6.   The recognition means starts the recognition when the speed of the target person is high or the number of the target person is large by performing a collation process based on the target person and the number of target persons to be recognized determined by the recognition target determining means. The verification process is completed by a simple process that narrows down the number of people to be recognized to those with high priority at an early time, and then the target person is excluded from the target of processing as verified and other target persons to be processed The person recognition apparatus according to claim 6, wherein a load of a collation process is assigned to. In a person recognition device that determines whether or not the target person is a pre-registered person by comparing biometric feature information obtained from the target person with at least one or more pre-registered biometric feature information,
Image acquisition means for acquiring an image including biological information of the subject;
A living body detecting means for detecting a living body region from the image acquired by the image acquiring means;
Biometric feature extraction means for extracting personal biometric feature information from the biometric area detected by the biometric detection means;
Registered biometric feature management means in which individual biometric feature information obtained from the subject's biometric information in advance is registered;
By comparing the biometric feature information extracted by the biometric feature extraction unit with the biometric feature information registered in the registered biometric feature management unit, it is determined whether the target person is a person registered in advance. Recognition means to
A recognition target person confirmation method for obtaining information on the target person and determining a confirmation method of the recognition target person according to the information on the obtained target person when it is determined that the recognition means cannot perform normal recognition processing. A determination means;
Confirmation means for confirming whether or not the subject person is a pre-registered person based on the recognition subject person confirmation method determined by the recognition subject person confirmation method decision means;
A person recognition device comprising:   The confirmation means enables visual confirmation by an administrator by transmitting at least the image acquired by the image acquisition means to a confirmation apparatus installed at a remote place when there is no allowance for recognition time for the subject. The person recognition apparatus according to claim 11.   The confirmation means, when there is no allowance for the recognition time for the target person, at least adds information that the recognition processing is not performed on the image acquired by the image acquisition means, and stores it as history information in the storage means The person recognition device according to claim 11.   The confirmation means transmits the biometric feature information extracted by the biometric feature extraction means to a separately prepared recognition processing device to perform recognition processing when there is no allowance for recognition time for the subject. Item 12. The person recognition device according to Item 11.   The confirmation means controls the recognition means to reduce the number of recognition target persons to reduce the number of persons to be recognized as only a part of the target persons when the recognition time for the target person is not sufficient. Item 12. The person recognition device according to Item 11.   The biometric feature information uses features such as a person's face, retina, mouth, iris, fingerprint, palm print, ear, and part of a human body vein. The person recognition device according to any one of the above.   The person recognition apparatus according to claim 1, wherein the information related to the subject is a walking speed of the subject to be recognized.   The person recognition apparatus according to claim 1, wherein the information related to the target person is the number of target persons to be recognized.   The person recognition apparatus according to claim 1, wherein the information related to the target person is a distance from the target person to be recognized.   The person recognition apparatus according to claim 1, wherein the information related to the target person is an elapsed time from the start of recognition processing of the target person to be recognized. In the person recognition method for determining whether or not the target person is a pre-registered person by comparing the biometric characteristic information obtained from the target person with at least one or more pre-registered biometric characteristic information,
An image acquisition step of acquiring an image including biological information of the subject;
A living body detecting step of detecting a living body region from the image acquired by the image acquiring step;
A biometric feature extraction step of extracting personal biometric feature information from the biometric region detected by the biometric detection step;
A recognition processing accuracy determination step for determining information related to the target person and determining a recognition processing accuracy according to the information related to the determined target person,
Based on the recognition processing accuracy determined in the recognition processing accuracy determination step, biometric feature information extracted in the biometric feature extraction step and personal biometric feature information obtained in advance from the subject's biometric information are registered. A recognition step of determining whether or not the target person is a pre-registered person by collating with biometric feature information registered in the registered biometric feature management means;
A person recognition method comprising: In the person recognition method for determining whether or not the target person is a pre-registered person by comparing the biometric characteristic information obtained from the target person with at least one or more pre-registered biometric characteristic information,
An image acquisition step of acquiring an image including biological information of the subject;
A living body detecting step of detecting a living body region from the image acquired by the image acquiring step;
A biometric feature extraction step of extracting personal biometric feature information from the biometric region detected by the biometric detection step;
A recognition target determination step for determining information related to the target person, and determining a target person to be recognized and the number of persons according to the information related to the determined target person,
The biometric feature information extracted by the biometric feature extraction step and the personal biometric feature information obtained in advance from the biometric information of the subject based on the target subject to be recognized determined by the recognition target determining step and the number of subjects. A step of recognizing whether or not the target person is a person registered in advance by comparing biometric feature information registered in the registered biometric feature management means.
A person recognition method comprising: In the person recognition method for determining whether or not the target person is a pre-registered person by comparing the biometric characteristic information obtained from the target person with at least one or more pre-registered biometric characteristic information,
An image acquisition step of acquiring an image including biological information of the subject;
A living body detecting step of detecting a living body region from the image acquired by the image acquiring step;
A biometric feature extraction step of extracting personal biometric feature information from the biometric region detected by the biometric detection step;
The biometric feature information extracted in the biometric feature extraction step is compared with biometric feature information registered in registered biometric feature management means in which individual biometric feature information obtained in advance from the subject's biometric information is registered. A recognition step for determining whether the target person is a person registered in advance;
When it is determined that normal recognition processing cannot be performed in this recognition step, information relating to the target person is obtained, and a confirmation method for the person to be recognized is determined according to the information relating to the obtained subject person. A decision step;
A confirmation step for confirming whether or not the subject person is a pre-registered person based on the recognition subject person confirmation method determined in the recognition subject person confirmation method decision step;
A person recognition method comprising: The biometric feature information obtained from the entry / exit person is checked against at least one or more pre-registered biometric feature information to determine whether the entry / exit person is a pre-registered person, and this determination result In the entrance / exit management system for controlling entrance / exit to the entrance / exit area of the entrance / exit based on
An image acquisition means for acquiring an image including biometric information of the entering / exiting person;
A living body detecting means for detecting a living body region from the image acquired by the image acquiring means;
Biometric feature extraction means for extracting personal biometric feature information from the biometric area detected by the biometric detection means;
Registered biometric feature management means in which individual biometric feature information obtained from the biometric information of the entrance / exit is registered in advance;
Recognizing process accuracy determining means for determining information related to the entry / exit, and determining the recognition process accuracy according to the information related to the determined entry / exit,
Based on the recognition processing accuracy determined by the recognition processing accuracy determination means, by comparing the biometric feature information extracted by the biometric feature extraction means with the biometric feature information registered in the registered biometric feature management means, Recognizing means for determining whether or not the entry / exit is a person registered in advance;
Entrance / exit control means for controlling entrance / exit of the entrance / exit area for the entrance / exit area based on the search result of the recognition means,
An entrance / exit management system characterized by comprising: The biometric feature information obtained from the entry / exit person is checked against at least one or more pre-registered biometric feature information to determine whether the entry / exit person is a pre-registered person, and this determination result In the entrance / exit management system for controlling entrance / exit to the entrance / exit area of the entrance / exit based on
An image acquisition means for acquiring an image including biometric information of the entering / exiting person;
A living body detecting means for detecting a living body region from the image acquired by the image acquiring means;
Biometric feature extraction means for extracting personal biometric feature information from the biometric area detected by the biometric detection means;
Registered biometric feature management means in which individual biometric feature information obtained from the biometric information of the entrance / exit is registered in advance;
Recognizing target determining means for determining information related to the entrance / exit, and determining the number of entrance / exit persons to be recognized and the number of persons according to the information related to the determined entrance / exit,
Based on the entry / exit persons to be recognized and the number of persons determined by the recognition target determination means, the biometric feature information extracted by the biometric feature extraction means and the biometric feature information registered in the registered biometric feature management means Recognizing means for determining whether or not the entry / exit is a person registered in advance,
Entrance / exit control means for controlling entrance / exit of the entrance / exit area for the entrance / exit area based on the search result of the recognition means,
An entrance / exit management system characterized by comprising: The biometric feature information obtained from the entry / exit person is checked against at least one or more pre-registered biometric feature information to determine whether the entry / exit person is a pre-registered person, and this determination result In the entrance / exit management system for controlling entrance / exit to the entrance / exit area of the entrance / exit based on
An image acquisition means for acquiring an image including biometric information of the entering / exiting person;
A living body detecting means for detecting a living body region from the image acquired by the image acquiring means;
Biometric feature extraction means for extracting personal biometric feature information from the biometric area detected by the biometric detection means;
Registered biometric feature management means in which individual biometric feature information obtained from the biometric information of the entrance / exit is registered in advance;
By comparing the biometric feature information extracted by the biometric feature extraction unit with the biometric feature information registered in the registered biometric feature management unit, it is determined whether or not the entry / exit person is a pre-registered person. A recognition means for determining;
When it is determined that normal recognition processing cannot be performed in this recognition means, information regarding the entry / exit person is obtained, and a recognition target person is determined according to the information related to the obtained entry / exit person. Confirmation method determining means;
Confirmation means for confirming whether or not the entry / exit is a person registered in advance based on the recognition target person confirmation method determined by the recognition target person confirmation method determination means;
Entrance / exit control means for controlling entrance / exit of the entrance / exit area of the entrance / exit based on the confirmation result of the confirmation means or the recognition result of the recognition means,
An entrance / exit management system characterized by comprising:

Images