JP2004046399A - Face recognition device and method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To acquire a front image of a face with high possibility in face recognition. <P>SOLUTION: A human body front detection part 32 executes matching processing on image data captured by a camera 17 to detect the human body front, a movement control part 33 then controls a camera moving part 31 to direct the camera 17 to the human body front, and in this state the camera 17 continuously captures image data, from which a face recognition processing part 34 recognizes the face. <P>COPYRIGHT: (C)2004,JPO

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to an apparatus and a method for recognizing a face.
[0002]
[Prior art]
In the face recognition processing, unless a face image of a human face photographed from the front is used, the recognition accuracy is significantly reduced. Therefore, as a conventional technique, a technique disclosed in Japanese Patent Application Laid-Open No. 2001-243466, “a frontal image of a face is acquired by detecting a face direction and moving a camera in the direction” is known.
[0003]
[Problems to be solved by the invention]
However, since the head of the human is more likely to move left and right due to the rotation of the neck than the body, it is necessary to change the direction of the camera in real time and always follow the face direction. Therefore, when the above-described conventional technology is applied to a device that cannot move the camera quickly, such as a mobile robot, there is a problem that it becomes difficult to obtain a front face image after all.
[0004]
An object of the present invention is to provide a technique capable of acquiring a frontal image of a face even with a device such as a mobile robot that cannot move a camera quickly, in view of the above conventional technology.
[0005]
[Means for Solving the Problems]
A first invention is a face recognition device for recognizing a face from image data acquired by a camera, a camera moving means for moving at least one of a position and a direction of the camera, and detecting a front direction of a human body from the image data. Human body front direction detection means, movement control means for controlling the camera moving means so that the camera faces the human body front direction detected by the human body front direction detection means, and a camera recognizing a face from image data obtained continuously. And a face recognizing processing means.
[0006]
In a second aspect based on the first aspect, the human body front direction detecting means includes:
A database for storing a plurality of two-dimensional projected image silhouettes having different rotation angles created from a three-dimensional human body model, a human body silhouette extracting means for extracting a human body silhouette from the image data, and a human body silhouette extracting means. Processing means for performing a matching process between a silhouette of a human body and a plurality of two-dimensional projected image silhouettes stored in the database, and a two-dimensional pattern that best matches the extracted human body silhouette obtained from the processing result of the matching processing means A face recognition apparatus comprising: a human body front direction estimating unit that estimates a front direction of a human body from a rotation angle of a projected image silhouette.
[0007]
A third invention is the face recognition apparatus according to the second invention, wherein the human body silhouette extracting means extracts a human body silhouette from the image data by a difference from a background image or by edge detection.
[0008]
A fourth invention is the face recognition device according to the second invention, wherein the database stores two-dimensional projected image silhouettes having the same rotation angle and different postures.
[0009]
A fifth invention is a face recognition method for recognizing a face from image data acquired by a camera, detecting a front direction of a human body from the image data, moving the camera so as to face the detected front direction of the human body, It is characterized in that image data is continuously acquired in a state, and a face is recognized from these image data.
[0010]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, embodiments of the present invention will be described with reference to the drawings.
[0011]
In the present embodiment, the face recognition device is mounted on a mobile robot. Therefore, prior to the description of the face recognition device, an example of a mobile robot to which the present invention is applied will be described with reference to FIG.
[0012]
The robot 1 shown in FIG. 3 simulates a human having a robot body 11 having a head 12, a chest 13, a torso (cart unit) 14, and left and right arms 15. Is rotatable by a drive mechanism (not shown), and the left and right traveling wheels 16 mounted on the body 14 can be steered and run by a drive mechanism (not shown), and move autonomously in the work space by battery driving. It is configured to The robot 1 coexists with a human in an indoor space such as a general home as a work space, and is used, for example, to assist, support, and care for a user's life in the general home. Therefore, the robot 1 has a function of talking with the user, watching the user's action, assisting the user's action, and acting together with the user by using the built-in CPU (computer) and various sensors. ing. Various shapes of the robot 1 are conceivable, such as an animal model for pets.
[0013]
Specifically, the robot 1 shown in FIG. 3 has a control unit (not shown) using a CPU, two cameras 17 and two microphones 18 are mounted on the head 12, and a volume / sound source direction sensor is provided at the center of the chest 13. And a human detection sensor 19 composed of a pyroelectric infrared sensor, and speakers 20a on the left and right. An image display 20 a is mounted on the center of the chest 13, and an ultrasonic obstacle sensor 21 and a laser obstacle sensor 22 are mounted on the body 4. A keyboard and a touch sensor may be mounted.
[0014]
The camera 17 captures an image of the user or the room and outputs the image to the control unit, and the microphone 18 captures the user's voice and life sounds such as a telephone bell, a doorbell, and a television sound and outputs the captured image to the control unit. The sensor 19 detects the presence or absence of a user and outputs the result to the control unit. The speaker 20a is used for conversation with the user together with the microphone 18, and the image display 20b is used for providing information to the user. The keyboard and the touch sensor are data input devices for the user, and are used for inputting the life pattern data of the user and inputting the intention of the user. The speaker 20a is also used for providing information to a user. The control unit includes a calendar for a predetermined period and a calendar clock for measuring the current date and time.
[0015]
The control unit includes a data input device such as a keyboard, an external sensor that constantly monitors a user's life pattern such as a camera 17, a microphone 18, a human detection sensor 19, and an obstacle sensor 21, and an internal sensor such as a calendar clock. , Information on the position of the robot 1, information on the position of the user, and information on the behavior of the user are input.
[0016]
The control unit has a self-position recognition function, a user recognition function, a database (data storage unit), a life pattern data processing function, and a drive control function.
[0017]
The self-position recognition function recognizes the position and azimuth (direction, posture) of the robot 1 itself based on image information captured by the camera 17. The user recognition function recognizes the content of the user's utterance (voice recognition) from the voice captured by the microphone 18, and performs personal recognition of the user from the image information captured by the camera 17 and the voice captured by the microphone 18. , The user's position, orientation, posture, and activity amount are recognized from the detection results of the camera 17 and the human detection sensor 19.
[0018]
The database includes voice data for conversation with the user and talking to the user, data of a gesture motion at the time of talking to the user and talking to the user, user's living information (room layout and furniture related to the user's residence, etc.). Location, personal daily life pattern of the user, user's hobby, health condition, etc.). A keyword such as a character string, video, or audio is added to the user's life information as needed. As the life pattern data, events related to activities such as getting up, going to bed, eating, taking medicine, taking a walk, free time, and taking a bath are listed, and these events are stored as time schedules together with time data.
[0019]
The life pattern data processing function stores life information directly input by a user from a data input device in a database with a keyword attached thereto, and various types of user recognition such as voice recognition results obtained by a camera 17, a microphone 18, and a human detection sensor 19. By processing the information, each piece of life information is attached to the database with a keyword and its date and time. Further, the life pattern data processing function selects data corresponding to the command from the database in response to an instruction or speech from the user, or according to time, and issues a command corresponding to the selected data to the operation control unit. Give and control.
[0020]
The drive control function drives the running wheels 16 to execute running and steering of the robot 1 in accordance with a command from the life pattern data processing function, and drives the head 12 and the arm 13 to rotate the head 12 and the arm 13. Perform the action. Further, the speaker 20a and the image display 20b are driven to execute a conversation with the user.
[0021]
Next, an image recognition apparatus and method according to an embodiment of the present invention will be described with reference to FIGS.
[0022]
The above-described robot 1 has a face recognition device illustrated in FIG. 1. The face recognition device includes a camera 17, a camera moving unit 31, a human body front direction detection unit 32, a movement control unit 33, And a face recognition processing unit 34. The human body front direction detection unit 32 includes a database 35, a human body silhouette extraction unit 36, a matching processing unit 37, and a human body front direction estimation unit 38.
[0023]
The camera 17 is mounted on the head 12 of the robot 1. The camera moving unit 31 is a unit that moves at least one of the position and the azimuth of the camera 17, and a mechanism that rotates the traveling wheel 16 and the head 12 that travels and steers the trunk 14 of the robot 1 corresponds thereto. I do.
[0024]
The human body front direction detection unit 32 is a unit that detects the front direction of the human body from the image data acquired by the camera 17, and
(1) In the database 35, a plurality of two-dimensional projected image silhouettes having different rotation angles are created in advance from a three-dimensional human body model and stored. The database 35 stores two-dimensional projected image silhouettes having the same rotation angle but different postures.
(2) The human body silhouette extracting unit 36 is a means for extracting a human body silhouette from the image data acquired by the camera 17, and extracts a human body silhouette by a difference from a background image or by edge detection. .
(3) The matching processing unit 37 is a unit that performs a matching process between the silhouette of the human body extracted by the human body silhouette extraction unit 36 and a plurality of two-dimensional projected image silhouettes stored in the database 35.
(4) The human body front direction estimating unit 38 can determine the two-dimensional projected image silhouette that best matches the extracted human body silhouette from the processing result of the matching processing unit 37. Means for estimating the front direction of the human body. That is, the front direction of the human body can be detected.
[0025]
The movement control unit 33 is a unit that controls the camera movement unit 31 so that the camera 17 faces in the front direction of the human body detected by the front direction detection unit 32.
[0026]
The face recognition processing unit 34 performs a process of recognizing a face from image data continuously acquired by the camera 17. In the face recognition processing, the eyes, nose, mouth, ears, and the like, which are characteristic parts of the face, are extracted from the image data and their sizes and shapes are used, or their positional relationships are used. It can be performed by various methods such as checking the color.
[0027]
Next, the face recognition procedure will be described with reference to FIG.
(1) First, the human body front direction detector 32 detects the front direction of the human body from the image data acquired by the camera 17 (step S1).
(2) The position and / or azimuth of the camera 17 is moved by the movement control unit 33 so that the camera 17 faces the front of the detected human body (step S2).
(3) With the camera 17 facing the front of the human body, image data is continuously acquired by the camera 17 and face recognition processing is performed by the face recognition processing unit 34 from these image data (step S3).
[0028]
As described above, by performing face recognition processing from image data acquired in a state where the camera 17 faces the front of the human body, even if the mobile robot cannot move the camera 17 quickly, even if the person turns his head, The probability of obtaining a frontal face image increases. That is, the accuracy of face recognition is improved.
[0029]
In the above embodiment, the detection of the front direction of the human body is performed by matching the human body silhouette extracted from the image data with a plurality of two-dimensional projected image silhouettes prepared in advance. By defining and learning the characteristics according to the direction of the human body, the frontal direction of the human body can be detected.
[0030]
This human body front direction detection focusing on clothes is performed when the face recognition target person is limited to some extent, for example, in the case of a mobile robot used to assist, support, and care for the life of the user in a general home, Especially effective.
[0031]
In the above embodiment, the face recognition device has been described for use in a robot. However, the face recognition device is not limited to this and can be used arbitrarily.
[0032]
【The invention's effect】
A first aspect of the present invention is a face recognition apparatus for recognizing a face from image data acquired by a camera, a camera moving means for moving at least one of a position and a direction of the camera, and a human body front direction for detecting a front direction of a human body from the image data. Detecting means; moving control means for controlling the camera moving means so that the camera faces the frontal direction of the human body detected by the human frontal direction detecting means; and face recognition processing for recognizing a face from image data continuously obtained by the camera. Therefore, even if the camera cannot be moved quickly, even if the person turns his / her head, the probability of obtaining a frontal face image increases. Therefore, the accuracy of face recognition is improved.
[0033]
According to a second aspect, in the first aspect, as a human body front direction detecting means, a database storing a plurality of two-dimensional projected image silhouettes having different rotation angles created from a three-dimensional human body model, and a human body based on the image data. A human body silhouette extracting unit for extracting a silhouette, a matching processing unit for performing a matching process between a human body silhouette extracted by the human body silhouette extracting unit and a plurality of two-dimensional projected image silhouettes stored in the database, and a matching processing unit. A human body front direction estimating means for estimating the front direction of the human body from the rotation angle of the two-dimensional projected image silhouette most matching the extracted human body silhouette obtained from the processing result, so that the human body front direction is accurately detected. be able to.
[0034]
In a third aspect based on the second aspect, the human body silhouette extracting means extracts the human body silhouette from the image data by a difference from the background image or by edge detection, so that the human body silhouette can be accurately extracted. .
[0035]
According to a fourth aspect, in the second aspect, the database stores two-dimensional projected image silhouettes having the same rotation angle and different postures. The direction can be accurately detected.
[0036]
A fifth aspect of the present invention is a face recognition method for recognizing a face from image data acquired by a camera, wherein the front direction of a human body is detected from the image data, and the camera is moved so as to face the detected front direction of the human body. Even when a camera that continuously obtains image data and recognizes a face from these image data cannot be moved quickly, even if a person turns his / her head, the probability of obtaining a frontal face image increases. Therefore, the accuracy of face recognition is improved.
[Brief description of the drawings]
FIG. 1 is a diagram showing a configuration of a face recognition device according to an embodiment of the present invention.
FIG. 2 is a diagram showing an example of a face recognition procedure.
FIG. 3 is a diagram illustrating an example of a robot.
[Explanation of symbols]
Reference Signs List 1 robot 12 head 14 torso 16 running wheel 17 camera 31 camera moving unit 32 human body front direction detection unit 33 movement control unit 34 face recognition processing unit 35 database 36 human body silhouette extraction unit 37 matching processing unit 38 human body front direction estimation unit

Claims (5)

In a face recognition device that recognizes a face from image data acquired by a camera,
Camera moving means for moving at least one of the position or orientation of the camera, human body front direction detecting means for detecting the front direction of the human body from the image data,
Movement control means for controlling the camera movement means so that the camera faces the human body front direction detected by the human body front direction detection means,
A face recognition apparatus comprising: a face recognition processing unit that recognizes a face from image data obtained continuously by a camera. In claim 1, as the human body front direction detecting means,
A database storing a plurality of two-dimensional projected image silhouettes having different rotation angles created from a three-dimensional human body model;
Human body silhouette extraction means for extracting a human body silhouette from the image data,
Matching processing means for performing matching processing between the silhouette of the human body extracted by the human body silhouette extraction means and a plurality of two-dimensional projected image silhouettes stored in the database;
And a human body front direction estimating means for estimating the front direction of the human body from the rotation angle of the two-dimensional projected image silhouette most matching the extracted human body silhouette obtained from the processing result of the matching processing means. Recognition device. 3. The face recognition device according to claim 2, wherein the human body silhouette extracting means extracts a human body silhouette from the image data by a difference from a background image or by edge detection. 3. The face recognition apparatus according to claim 2, wherein the database stores two-dimensional projected image silhouettes having the same rotation angle and different postures. In a face recognition method for recognizing a face from image data obtained by a camera, a frontal direction of a human body is detected from the image data, and the camera is moved so as to face the detected human frontal direction. A face recognition method characterized by acquiring data and recognizing a face from the image data.

Images