JP2008211534A - Face detecting device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide high-speed processing based on a reduction in computational complexity because face type template and calculation of image similarity become huge. <P>SOLUTION: A face detection mode switching means for setting a face detection area from an image in a plurality of segmenting modes is provided, and template matching is performed by a face type template of an optimized size in each segmenting mode to thereby detect a face area at a high speed. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a face detection device such as a monitoring system.

  In recent years, face detection processing for classifying a digital image by the face of a person who is a subject has been used in various fields. On the other hand, in face detection processing, it is an important issue to operate at high speed by reducing not only the discrimination accuracy but also the calculation amount.

  As a method for improving the processing speed, for example, Japanese Patent Laid-Open No. 2005-215750 (Patent Document 1) describes, as a problem, “Aiming to shorten processing time while maintaining the accuracy of face detection”. As a solving means, “as the face detecting means 106 for extracting the face area of the subject from the image photographed by the digital camera 101 or the like, the image data and the attached information related to the image data such as the imaging condition are used. Means 107 for acquiring in-focus area information; means 110 for determining a face detection area within the image area based on the in-focus area information; and means for partially executing face detection processing instead of the entire image area 111 ”(see summary).

  Japanese Patent Laid-Open No. 2006-253805 (Patent Document 2) describes as a problem that “a plurality of face regions can be detected from a photographed image with high speed and accuracy with a simple operation”. As a solving means, “a face area is designated by a photographer's zooming operation, and the face area position / size searching unit 26 detects a face area corresponding to the size of the designated face area from the photographed image.” (See summary).

Japanese Patent Laying-Open No. 2005-215750 (see summary) JP 2006-253805 A (see abstract)

By the way, a surveillance system for recording video captured by a surveillance camera on a recorder has been widely used, and is used for identifying a criminal. However, since it takes a lot of labor to confirm all the images of the time zone where the criminal is supposed to be captured with human eyes, it is desired to automatically detect the face area from the images. In such a monitoring system, it is necessary to perform face detection processing for an unspecified number of people, and face detection is performed by template matching.
In face detection processing by template matching, the similarity between a face template and an image is calculated, and if the similarity is higher than a certain threshold, it is determined that there is a face. Then, the entire input image is scanned while shifting the position of the face template, and it can be detected when the size of the face template matches the size of the face shown in the image.

  FIG. 6 is a diagram schematically illustrating a conventional face area detection method. The size of the face shown in the image varies depending on the distance from the camera. Therefore, as shown in FIG. 6, the size of the face template is changed little by little from the maximum size to the minimum size, and matching is repeated over the entire screen for each size. In the conventional method, matching is repeated using a face template over the entire input image. For this reason, the calculation of the similarity between the face template and the image is enormous, and in order to incorporate the template matching face detection function in the monitoring system, it is an issue to reduce the amount of calculation and increase the processing speed. It has become.

  As a method for speeding up the face detection function, Patent Document 1 limits the face detection area by using a distance measuring frame used for autofocus of a digital camera. Further, Patent Document 2 discloses that the size of the face area is limited based on the zoom magnification at the time of zooming up of the digital camera and at the time of actual photographing.

  However, unlike with digital cameras, shooting with a surveillance camera does not precisely set the focus position for the target person, so depending on the face detection range setting or face size setting based on the focus position. The calculation process cannot be reduced. As described above, in the surveillance camera, it is necessary to speed up the face detection based on a new calculation amount reduction process. This point is not considered in either Patent Document 1 or Patent Document 2.

  Accordingly, an object of the present invention is to provide a monitoring system capable of solving the above-described problems and appropriately detecting a face.

  In order to achieve the above object, the present invention provides a face detection device that detects a face included in image data using a face template, and sets a face detection region for detecting a face in the image data region. A face detection mode switching unit that switches between the plurality of face detection modes, and a face template setting unit that sets a size range of the face template according to the face detection mode. A face detection device characterized by this is used.

  ADVANTAGE OF THE INVENTION According to this invention, the monitoring system which can detect a face appropriately can be provided.

  Hereinafter, a monitoring system such as an ATM (automated teller machine) will be described with reference to the drawings as an embodiment of the present invention, but the present invention is not limited to a monitoring system such as an ATM.

  FIG. 12 shows the appearance of an ATM surveillance camera according to this embodiment. B10 in the figure represents ATM, B11 represents a surveillance camera, and B12 represents a user. The ATM is provided with a cash card entry / exit, a bankbook entry / exit, a cash entry / exit, etc. on the user's side, and the user operates each entry / exit on the touch panel. The angle of view of the surveillance camera attached to the ATM is set so as to capture the user's ATM operation position and its surroundings, and the face detection mode can be switched according to the range in which the user's face is detected. Yes.

FIG. 11 shows an example of the face detection mode according to the present invention, which can be switched between three modes of “full screen mode”, “cutout mode 1”, and “cutout mode 2”.
The “full-screen mode” targets the entire screen, which is a screen area of 704 × width 480 in size, and the size of the detected face ranges from the maximum value 387 × length 430 to the minimum value width 90 × length 100. Allows detection of any size face. Here, these sizes represent the number of pixels.
In “Cutout mode 1”, the position of the cutout area on the screen can be arbitrarily moved with the screen area of 312 × vertical 192 as the face detection target, and the size of the detected face is the maximum value of 172 × vertical. 172, it is possible to detect a face of any size within the range of the minimum value 40 × 40.
In “Cutout mode 2”, the position of the cutout area on the screen can be arbitrarily moved with a screen area of 156 × 96 in the face detection target, and the detected face size is the maximum value of 86 × vertical. 86, it is possible to detect a face of any size within the range of the minimum value 20 × 20.

  Thus, the size of the face to be detected is changed according to the size of the detection target in order to shorten the processing time for face detection. For example, when the detection target is “full screen mode” of horizontal 704 × vertical 480, if the detection is performed using the minimum face size 20 × vertical 20 for “cutting mode 2”, a huge calculation time is required. Become.

  In the present invention, it is possible to efficiently detect the face by changing these three modes according to the distance from the camera to the subject, for example.

  FIG. 9 shows an example of face area detection according to the conventional method, and FIG. 10 shows an example of face area detection according to the present invention. It will be described below that a face can be detected appropriately by switching the face detection mode using these.

  In the conventional method, the face detection mode is not switched and the entire screen is always detected, and the size of the template for face detection remains unchanged within a certain range. Therefore, the face of the person A on the screen of FIG. 9 can be detected because it is within the template size range, but cannot be detected because the face of the person B is outside the template size range.

  On the other hand, according to the present invention, since the mode is switched according to the distance from the camera to the subject, the size range of the template for face detection can also be changed. For example, by using the cut-out mode 1 in FIG. 10, the size range of the template changes, and it becomes possible to detect the face of the person B who could not be detected in the full screen mode.

  Thus, FIG. 10 shows an example in which face detection is possible by switching to “cutout mode 1” or “cutout mode 2” even when face detection is difficult in “full-screen mode”. Switching between “full screen mode”, “cutout mode 1”, and “cutout mode 2” is performed according to the distance between the monitoring camera and the subject. Thereby, a face can be detected appropriately.

  FIG. 1 is a configuration diagram of a face area detection device of a monitoring system according to an embodiment of the present invention.

  In FIG. 1, reference numeral 100 denotes an imaging device that captures an image, and 200 sets a face detection area for the captured image, performs face detection using a face template, and uses the matched face area and the image as internal media or An image processing / storing device for storing in an external medium, 300 an image display device for outputting captured images and face detection results, and 400 for storing captured images and face detection results or processing such as issuing an alarm. An external device to perform.

  The image captured by the image capturing apparatus 100 is transmitted to the image processing / storing apparatus 200 every 16 frames / second.

  The image received by the image receiving unit 201 of the image processing / storing apparatus 200 is read by the image reading unit 202 and transmitted to the face detection area setting unit 203.

  The face detection area setting unit 203 sets the face detection area based on the face detection mode specified by the user using the face detection area input unit 207 and the position coordinate information of the cutout area.

  Next, in the face area detection unit 204, a face type template having a size corresponding to the face detection mode set in the face detection area input unit 207 is input from the face type input unit 208 to be gradation, and the face detection area setting is performed. Face coordinates are detected using a face-type template gradationed for the face detection area set by the unit 203.

  Next, the information adding unit 205 adds the face number indicating the number of detected faces to the image, the coordinates of the face detection area, and the face area coordinates, and stores them in the image storage unit 209 and sends them to the image transmission unit 206. The image transmission unit 206 outputs to the external image display device 300 and the external device 400.

  A configuration in which the image reading unit 202, the face detection region setting unit 203, the face region detection unit 204, the face detection region input unit 207, and the face type input unit 208 are built in the imaging apparatus is also possible. This is because space may be saved and control may be facilitated due to the design of the apparatus.

  FIG. 2 is a configuration diagram around the face detection area setting unit 203. FIG. 3 is a diagram illustrating an example of setting a face detection area. A specific method for setting the face detection area will be described with reference to FIGS.

  In FIG. 3, the face detection mode is expressed as “cutout mode 1” for the captured image, and the position designation of the cutout area is expressed by the upper left and lower right vertex coordinates (100,100), (412,292) of the rectangular cutout area. ing.

  In FIG. 2, an image reading unit 202 converts an input image into processable image data, such as converting an analog image into a digital image, and reads the image on the device memory, and 207 refers to a face while viewing the input image. A face detection area input unit that controls movement of the area to be searched and inputs a face detection area, 211 is an area calculation unit that converts the input area into position coordinate information in the X-axis direction and the Y-axis direction, and 210 is an area calculation unit An area determination unit that sets the position coordinate information received from 211 as an input image. The image in which the coordinates are set is transmitted to the area face area detection unit 204. The position coordinate information is specified by marking the upper left and lower right of the rectangle of the face detection area where the face is expected to appear while the user actually sees the image of the imaging device when installing the surveillance camera on ATM etc. Is done. The positions of the two points marked by the user are converted into coordinates for alignment with the input image by the area calculation unit 211 and sent to the area determination unit 210. The area determination unit 210 writes the coordinates of the two points sent from the area calculation unit 211 on the image header and transmits them to the face area detection unit 204. Further, the setting of the face search area is not limited to a rectangle, but can be set using a circular center coordinate or the like.

  Next, setting of the face template and face detection for the face detection area using the face template will be described with reference to a detailed block diagram of the face area detector 204 shown in FIG.

  In FIG. 4, an average face obtained by combining face images of a plurality of people is used as the face template input from the face input unit 208. Further, there are cases where a face is discriminated based on the positional relationship of the facial parts or the similarity of the feature amount using a template of only facial parts such as eyebrows, eyes, nose and mouth. This is because it may be easier to discriminate the face by using the template of these parts. In the case of a surveillance camera, since it is often installed at a high position such as a ceiling, an average face taken from an oblique top may be used. This average face is generated in advance, converted into digital data, and input to the face type input unit 208. Then, the face detection template is set in the face type setting unit 214 of the face area detection unit 204.

The face-type setting unit 214 converts the face-type template into an optimum size for face detection and transmits it to the face area calculation unit 213. For example, in the “cutout mode 2”, the maximum size of the face template is converted to “horizontal 86 × vertical 86”, and a plurality of face templates in which gradation is made between the maximum size and the minimum size of the face size are generated. It is also possible to generate a plurality of scaled-down reduced face templates in advance and input them to the face input unit 208.
The face area calculation unit 213 reads the coordinate information of two points indicating the upper left and lower right of the rectangle of the face detection area from the header of the input image from the face detection area setting unit 203, determines the face detection area range, Perform template matching with face template.

  Next, template matching processing of the present invention will be described using the flowchart of FIG. 7 and the template matching invention method of FIG.

  In step 501, the face detection area is set based on the position coordinates of two points in the cut-out mode from the face detection area setting section 203 in the image development section 212 of the face area detection section 204.

  In step 502, the face detection area is expanded in the memory.

  In step 503, a face template matching process is performed on the face detection area. As shown in FIG. 5, first, after performing a matching operation from the upper left end to the upper right end of the face detection area using the maximum size of the face type template determined by the detection mode setting, the face type template is moved to the face detection area. Return to the top left corner and lower it in the Y-axis direction by the face template, and then perform the matching operation in the X-axis direction again to the right edge of the face detection area. Thereafter, the same processing is repeated until the lowest level of the face detection area.

  If a face is not detected by the matching operation using the face template of the maximum size, the face type of the next size that has been gradated within the range of the maximum size and the minimum size of the face template determined by the detection mode setting Proceed with the matching operation using the template. In the determination of face detection by the matching operation, the similarity between the face template and the image is calculated, and if it is equal to or greater than a predetermined threshold value, YES is determined and the process proceeds to step 504. If it is equal to or less than the predetermined threshold value, the matching operation for the entire face detection area using the face template of the minimum size is performed and NO is determined, and the process proceeds to step 505.

  In step 504, the face area is stored in the memory, and the process proceeds to step 505. An example of information stored in the memory used in step 505 is shown in FIG. In the memory, a face number indicating the number of detected faces, face detection area information, and face area information are written as shown in FIG.

  In step 505, it is determined whether or not the matching operation for the entire face detection area using the face template of the minimum size has been completed. If not, the process returns to step 503. If completed, the process proceeds to step 506 to end the face area detection.

As described above, according to the present embodiment, by limiting the face detection area, the amount of calculation is greatly reduced compared to the conventional template matching, the detection processing speed is increased, and the speed of face detection is increased. . In addition, even in a device with scarce computing resources, such as a monitoring system device, the size of the face image to be detected (distance from the camera to the subject) can be set wide by narrowing the face detection area and performing face detection.
Further, according to the present invention, it is possible to provide a high-speed face detection process particularly suitable for a monitoring system such as an ATM (automated teller machine). Furthermore, there is an advantage that the face area can be detected even by a high-speed detection of the face area and a device having a poor calculation resource.

  Since the face detection device of the present invention can set the detection range according to the calculation power of the installed device, it can detect the face using the maximum capability of the device and contribute to the effective use of calculation resources. .

The block diagram of the face area detection apparatus in the form of the Example of this invention The block diagram around the face detection area setting part of the face area detection apparatus in the form of the Example of this invention Example of setting the face detection area The block diagram of the face area detection part of the face area detection apparatus in the form of the Example of this invention Face area detection method according to the present invention Conventional face area detection method Face area detection flowchart Example of stored information in a memory according to an embodiment of the present invention Example of face area detection by conventional method Example of face area detection according to the present invention Example of face detection mode of the present invention External view of ATM surveillance camera according to this embodiment

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 Image pick-up device 200 Image processing / storing device 201 Image receiving unit 202 Image reading unit 203 Face detection region setting unit 204 Face region detection unit 205 Information addition unit 206 Image transmission unit 207 Face detection region input unit 208 Face type input unit 209 Image storage Unit 210 region determination unit 211 region calculation unit 212 image development unit 213 face region calculation unit 214 face type setting unit 300 image display device 400 external device A01 face detection region (start point)
A02 Face detection area (end point)
B10 ATM
B11 surveillance camera B12 user

Claims (4)

A face detection device that detects a face included in image data using a face template,
A plurality of face detection modes for setting a face detection area for detecting a face in the image data area, and a face detection mode switching means for switching the plurality of face detection modes;
A face template setting means for setting a size range of the face template according to the face detection mode;
A face detection device comprising: The face detection device according to claim 1,
A face detection apparatus comprising a face detection area moving means for moving the face detection area within the area of the image data. The face detection device according to claim 1 or 2,
The face type template setting means sets the maximum size and the minimum size of the face type template which are different for each of the plurality of face detection modes, and performs size setting in which gradation is made between the maximum size and the minimum size. A featured face detection device. The face detection device according to any one of claims 1 to 3,
The image data is captured by an imaging unit,
The face detection mode switching unit switches the plurality of face detection modes according to a distance between the imaging unit and a subject.

Images