JP2008146132A - Image detection device, program, and image detection method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an image detection technique for accurately detecting the shape of a symmetrical object such as a human. <P>SOLUTION: From an image taken by an imaging device 1, an image detection device detects a candidate for detection through template matching at a detection candidate detection means 3. A symmetrical image creation means 4 divides an image of the candidate for detection into two images at the central vertical line of the image and inverts one of the images. A symmetry detection means 6 compares the inverted image with the other of the images, calculates the similarity between the symmetric images, compares the similarity with a threshold value, and detects an image that has symmetry. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to an image detection technique for performing predetermined processing in response to input of a captured image.

  2. Description of the Related Art Conventionally, an image processing apparatus using a template matching process has been proposed as an apparatus for detecting a region having a symmetrical shape in a captured image (see, for example, Patent Document 1).

The image processing apparatus employs a technique of calculating a similarity by superimposing a captured image and a template to be detected, and searching for the same pattern as the template from the captured image.
Japanese Patent Laid-Open No. 11-284997

  Since the conventional template matching process described above simply searches for an image similar to a template, there is a problem that false detection increases. For example, if a person is photographed from directly above and matching is attempted with a round template, the shoulder portion is also detected in addition to the head.

  The present invention has been made in view of the problems of the prior art, and an object thereof is to provide a technique for accurately detecting a symmetrical object shape such as a person.

  In order to achieve the above object, the invention described in claim 1 is an image detection apparatus that captures an image and performs a predetermined process in response to the input of the image. Detection target candidate detection means for detecting a detection target candidate from a captured image by template matching, and an image obtained by dividing the image of the detection target candidate into two by a central vertical line of the image and inverting one image symmetrically And a symmetry detecting means for calculating the degree of similarity between the other image created by the cutout and the inverted image.

  The invention described in claim 2 is a program for causing a computer configured to input a captured image to detect the detection target candidate from the image by template matching. A detection means; a symmetrical image generation means for generating a reverse image of one of the images by dividing the image of the detection target candidate into two by a central vertical line of the image; and the reverse of the other image created by the two division It is characterized by functioning as a symmetry detecting means for calculating the degree of similarity with an image.

  The invention described in claim 3 is an image detection method for performing predetermined processing in response to input of an image, the step of matching the image with a template image as an image to be detected, and the matching A step of dividing the image detected in step 2 into two parts, a step of generating an image obtained by inverting one image created by the cut-out symmetrically, and a step of inverting the other image created by the cut-out The method includes a step of calculating the similarity of images, and a step of determining that the region has left-right symmetry when the similarity is equal to or greater than a threshold value.

  According to the first to third aspects of the invention, if the object to be detected has left-right symmetry, all can be detected accurately. For example, it accurately detects all the symmetry such as the front face of a person, the head from above the person, the whole image of the front of the person, the image from above the car, the front image of the car, and the front face of the animal without false detection. it can.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. FIG. 1 is a diagram illustrating a configuration of an image detection apparatus. The apparatus includes a photographing apparatus 1, a template image database 2 that stores template images, a detection target candidate detection unit 3, a symmetric image generation unit 4 that generates a bilaterally symmetric image and stores it in a memory 5, and a memory 5. It comprises a symmetry detecting means 6 for calculating the similarity of the images and comparing it with a threshold value, and a display monitor 7 for the detection target area.

  The photographing apparatus 1 may be a camera capable of photographing in color or a camera capable of photographing in black and white. The template image database 2 stores images having left-right symmetry that are images to be detected. The detection target candidate detection means 3 detects the image of the detection target candidate by performing a matching process between the image captured by the imaging apparatus 1 and the template image stored in the template image database 2. In general, detection target candidate images include images that are not detection targets. The symmetrical image generation means 4 divides the detection target candidate image into two by the central vertical line of the image, inverts one image, and stores it in the memory 5. The symmetry detection means 6 calculates the similarity between the two images stored in the memory 5, that is, the right image and the left image, and outputs an object having symmetry as a detection target region. The detection target area display monitor 7 is used, for example, to highlight and display the detection target area from the image captured by the photographing apparatus.

  Details of the operation will be described below. FIG. 2 is a flowchart of image detection. This flowchart is an example of a processing procedure defined in a program executed by a computer.

  First, in order to facilitate template matching described later on the image obtained by the photographing apparatus 1, for example, edge enhancement processing is performed as processing for removing noise (step S1). If the image does not contain noise, this process may be omitted.

  Next, matching processing (step S2) is performed on the image from which noise has been removed using the image stored in the template image database 2. By this processing, all images similar to the template image are detected. In general, detected images include images that are not detected.

  Next, the image detected by the template matching is divided into two with the central vertical line as an axis and then cut out (step S3).

  Next, a process of storing one image of the images divided and divided into two in the right side image of the memory 5 and storing an image obtained by inverting the other image symmetrically in the left side image in the memory 5 (Step S4) is performed. Note that the image reversed in the left-right symmetry may be stored in the right image of the memory 5, and one of the cut out images may be stored in the left image of the memory 5.

  Next, a process of calculating the similarity between the right image and the left image in the memory 5 (step S5) is performed. For example, the similarity may be calculated by calculating the difference between the right image and the left image in the memory 5.

  Next, a process of comparing the calculated similarity with a threshold value (step S6) is performed. An appropriate value may be set in advance as the threshold value.

  Next, if the similarity is higher than the threshold, it is determined that the region has left-right symmetry (step S7). If the similarity is lower than the threshold, the left-right symmetry is obtained. It is determined that there is no area (step S8).

  Next, a case where the present invention is applied to tracking of a person's head will be described. This image detection apparatus is installed in a store and aims at investigating a customer's purchasing behavior by tracking human behavior. The imaging device uses a CCD camera that can capture black and white images. The camera is installed on the ceiling in the store, and the shooting direction is the vertical downward direction. A computer is used to detect symmetry from the captured image. The computer may be provided with a video capture circuit for capturing a moving image, for example.

  A method for detecting the head of a person in the image detection apparatus will be described with reference to FIG. When a person is photographed vertically downward with a camera installed on the ceiling and subjected to edge enhancement processing, an image (a) is obtained. In the image (a), the head and shoulder portions of one person are displayed.

  A matching process is performed on the image (a) using the template image (b). As a result of the matching process, three regions of the images (c) to (e) are detected. In addition to the head, the shoulder portion is also detected because it resembles the shape of the template.

  Each detected area is divided into two by a central vertical line, and the left image (c1) (d1) (e1) and the right image are reversed symmetrically to generate a generated image (c2) (d2) (e2) Create Next, the images (c1) and (c2), the images (d1) and (d2), the images (e1) and (e2) are compared, and the similarity is calculated and calculated. Since the similarity between the images (d1) and (d2) is high, it is determined that the region has left-right symmetry and the human head is located. By such processing, the position of the head can be tracked and the purchase behavior of the person can be grasped. The number of people can also be counted by the same processing.

  In this embodiment, the target image to be photographed is not slanted. However, when the target image in the photographed image is slanted, such as when the target image is not symmetrical, matching is performed by rotating the template image at an appropriate angle. If the same processing is performed thereafter, an image having left-right symmetry independent of the angle can be detected.

  Further, in this embodiment, a person is targeted, but if it is symmetrical, for example, the front face of the person, the head from above the person, the whole image of the front of the person, the image from above the car, the front image of the car, the animal It is possible to accurately detect all people, objects and animals such as the front face.

It is a block diagram of an image detection apparatus. It is a flowchart of an image detection process. It is explanatory drawing which shows the method of detecting a person's head.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Image pick-up device 2 Template image database 3 Detection target candidate detection means 4 Symmetric image generation means 5 Memory 6 Symmetry detection means 7 Display monitor of detection target area

Claims (3)

An image detection apparatus that performs predetermined processing in response to input of a captured image,
Detection target candidate detection means for detecting a detection target candidate from the image by template matching;
A symmetric image generating means for dividing the detection target candidate into two at a central vertical line and generating a reverse image of one image;
An image detection apparatus comprising: symmetry detection means for calculating a similarity between the other image generated by the two divisions and the inverted image. In a computer configured to input captured images,
The computer,
Detection target candidate detection means for detecting a detection target candidate from the image by template matching;
A symmetric image generating means for dividing the detection target candidate into two at a central vertical line and generating a reverse image of one image;
A program that functions as a symmetry detection unit that calculates the degree of similarity between the other image generated by the two divisions and the inverted image. An image detection method for performing predetermined processing in response to input of a captured image,
Matching the image with a template;
Dividing the image detected by the matching into two;
Generating a reverse image of one of the images created by the two divisions;
Calculating the similarity between the other image created by the two divisions and the inverted image;
And comparing the similarity with a threshold to determine a region having left-right symmetry.

Images