JP2005215899A - Object detector and its method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an object detector and its method capable of providing high-speed object detection without decreasing detection accuracy. <P>SOLUTION: Detection processes for an object are provided with a process for identifying the peripheral region of the object by an object peripheral region identification section 43 as a pre-step for the process identifying the object by an object identification section 45. The object peripheral region identification section 43 roughly detects a region having a high possibility of the presence of the object. Then, an object identification section 45 investigates minutely whether or not the picture of the object is present only within the region. As a result, the object peripheral region identification section 43 limits a search range to the peripheral region of the picture of the object found at a high speed. The object identification section 45 searches for the picture of the object. Therefore, the processing amount of the object detection processing for the entire object detector 10 is reduced, providing the high speed. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to an object detection apparatus and method for detecting an object included in an image.

  An object detection apparatus that uses a human face as an object is disclosed in Patent Document 1. In the conventional object detection device, as shown in FIG. 8, a collation local region to which a region model that functions as a filter for detecting an image of the object is applied is set, and the position of the collation local region is set as a position of the input image. Specify and move sequentially throughout. Then, luminance normalization is performed for each designated local region to be checked, and image processing such as edge detection is performed. Further, the feature amount of each determination element acquisition region in the region model in which the face region model is applied to the image processed image is extracted, and it is determined whether each face image is a face image. Detection is performed.

Further, Non-Patent Document 1 discloses a method of detecting an object used in an object detection apparatus using a human face as an object. In a conventional object detection method, a feature pattern of an object (hereinafter referred to as an object feature pattern) is extracted to form a filter, and the object is detected from an image using this filter.
JP-A-11-306348 SECOND INTERNATIONAL WORKSHOP ON STATISTICAL AND COMPUTATIONAL THEORIES OF VISION-MODELING, LEARNING, COMPUTING, AND SAMPLING VANCOUVER, CANADA, JULY 13, 2001.

  However, in the conventional object detection device and the object detection method, the processing amount becomes enormous due to the sequential movement of the filter, etc. There is a problem that object detection cannot be performed.

  The present invention has been made in view of this point, and an object of the present invention is to provide an object detection apparatus and method that can increase the speed of object detection without reducing the detection accuracy.

  The object detection device of the present invention is an object detection device that detects an image area of a detection object included in an image, and detects a peripheral image area of the detection object using a first detection filter. Object surrounding area detection means, and object detection means for detecting an image of the detection object for only the peripheral image area of the detection object detected using the second detection filter. Take.

  According to this configuration, the first detection filter grasps the image area around the detection target, and then uses the second detection filter to detect only the image area detected by the first detection filter. Since the image of the object is detected, the processing amount of the object detection process can be reduced and the speed can be increased.

  The object detection device of the present invention employs a structure in which the first detection filter is a feature pattern of the detection object extracted from an image sample including the offset image of the detection object in the above configuration.

  According to this configuration, since the feature pattern of the detection target is extracted from the image sample including the detection target image offset by the first detection filter, the first detection filter is offset from the detection target image. Since it can be identified that it is the periphery of the image of the detection target even if it is at a location shifted by this amount, it is possible to detect a wide range by the offset at a time.

  The object detection device according to the present invention is configured such that, in the above configuration, the first detection filter includes an eye and an eyebrow as a feature pattern of the detection object when the detection object is a human face. A first rectangular area corresponding to the first rectangular area, and a second rectangular area located below the first rectangular area and corresponding to an area including a cheek and a nose, wherein the first rectangular area and Luminance in both areas when the size of the second rectangular area, the position of the first rectangular area and the second rectangular area, and both rectangular areas are applied to an image including the detection target And a structure defined as a parameter.

  According to this configuration, whether the image to be identified is an image of the face of the person that is the detection target based on the luminance relationship of the images corresponding to both areas, based on the feature points of the face of the person that is the detection target. Therefore, it is possible to detect the peripheral region of the image of the detection object without complicated processing.

  The object detection apparatus according to the present invention includes an area designating unit that designates an area to which the first detection filter is applied in the image including the detection object in the configuration described above. A configuration is adopted in which the region is designated at an interval corresponding to the offset amount of the image sample.

  According to this configuration, the area designating unit designates the area to which the first detection filter is applied at an interval corresponding to the offset amount of the image sample from which the feature pattern of the detection object is extracted. The amount of processing required to detect the image area can be reduced. As a result, the image of the object can be detected at high speed.

  The object detection method of the present invention is an object detection method for detecting an image area of a detection object included in an image, and detects a peripheral image area of the detection object using a first detection filter. And a step of detecting an image of the detection object only for a peripheral image region of the detected detection object using a second detection filter.

  According to this method, the first detection filter grasps the peripheral image region of the detection target, and then uses the second detection filter to detect only the image region captured by the first detection filter. Since the image of the object is detected, the amount of processing for detecting the image of the object can be reduced, and the speed can be increased.

  The object detection program of the present invention is a program for detecting an image area of a detection object included in an image, and a step of detecting a peripheral image area of the detection object using a first detection filter. And a step of causing the computer to execute the step of detecting the image of the detection object for only the peripheral image region of the detected detection object using the second detection filter.

  According to this configuration, the peripheral image region of the detection target is grasped by the first detection filter, and then the second detection filter is used only in the image region captured by the first detection filter. Since the image of the object is detected, the amount of processing for detecting the image of the object can be reduced, and the speed can be increased.

  ADVANTAGE OF THE INVENTION According to this invention, the processing amount of a target object detection process can be reduced without reducing detection accuracy, and the target object detection apparatus and method which can aim at speed-up can be provided.

  The essence of the present invention is to provide a process for identifying a peripheral area of an object as a pre-stage of the object identification process in the object detection process, and after roughly detecting an area where the object is likely to exist, It is to examine whether or not the object exists only in the area.

  Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings.

  FIG. 1 is a block diagram showing a configuration of an object detection apparatus 10 according to an embodiment of the present invention.

  As shown in FIG. 1, the object detection apparatus 10 includes two systems, an online execution environment system and an offline execution environment system. The object detection apparatus 10 prepares two filters having different functions in an offline execution environment system. In the online execution environment system, first, an image area around the detection object (hereinafter referred to as an object periphery area) is detected by the first filter. ) Is roughly grasped, and then the second filter is used to examine whether there is an object image in the image area captured by the first filter and detect the object. .

  The offline execution environment system of the object detection device 10 includes an object peripheral area image storage unit 21, an object peripheral area feature calculation unit 22, an object peripheral area feature dictionary 23, an object image storage unit 31, and an object An object feature calculation unit 32 and an object feature dictionary 33 are included.

  The online execution environment system of the object detection apparatus 10 includes an image input unit 41, a surrounding area detection window designation unit 42, an object surrounding area identification unit 43, an object detection window designation unit 44, and an object identification. Unit 45 and a detection result output unit 46.

  First, the offline execution environment system will be described.

  The object peripheral area image storage unit 21 stores a sample image for extracting a peripheral area feature pattern used for identifying the object peripheral area.

  FIG. 2 is a diagram illustrating an example of a sample image stored in the object peripheral area image storage unit 21. FIG. 2 shows a sample image when the object is a human face.

  The center image shown in FIG. 2 is obtained by photographing the face of a person from the front and extracting only the inside of the face based on the corner of the eye. This central image has been artificially processed so that the face fits into the determined frame.

  The images in the left and right columns are shifted by one pixel to the left and right with respect to the image in the center column, that is, offset is applied in the left and right direction. In addition, the images in the upper and lower rows are each tilted 5 degrees to the left and right with respect to the image in the center row.

  The sample image shown in FIG. 2 is prepared as follows. First, a central image is prepared as described above, and then the central image is processed to create an image that is offset by one pixel to the left and right. Thus, all the images in the center row are prepared. Next, by tilting the image in the center row by 5 degrees to the left and right, the images in the upper and lower rows are also prepared.

  The object peripheral area image storage unit 21 prepares such sample images for a number of different people. In FIG. 2, nine image samples are prepared for one person, but the number is not limited to this.

  The size of the left and right offsets affects how the window designated by the surrounding area detection window designating unit 42 is moved. Specifically, when the pixel is offset by one pixel as in the example shown in FIG. 2, the window specified by the surrounding area detection window specifying unit 42 described later is an interval corresponding to the offset amount (when the pixel is offset by one pixel). Is at least 2 pixels), there is no high probability that the object peripheral area identification unit 43 described later misses the peripheral area where the video of the object actually exists.

  The object peripheral region feature calculation unit 22 calculates a peripheral region feature pattern of the face of a person who is a detection target, stochastically and statistically, based on the sample image stored in the object peripheral region image storage unit 21. And is output to the object peripheral region feature dictionary 23. The extraction of the peripheral area feature pattern will be described later.

  The object peripheral area feature dictionary 23 stores the peripheral area feature pattern output by the object peripheral area feature calculation unit 22. This peripheral area feature pattern is used as a filter used when the object peripheral area identification unit 43 identifies the object peripheral area.

  The object image storage unit 31 stores a sample image for extracting an object feature pattern used for identifying an object.

  FIG. 3 is a diagram illustrating an example of a sample image stored in the object image storage unit 31. FIG. 3 shows a sample image when the object is a human face.

  The center image shown in FIG. 3 is obtained by photographing a human face from the front and extracting only the inside of the face with the eye corners as a reference. This central image has been artificially processed so that the face fits into the determined frame.

  The upper and lower images are obtained by tilting the center image left and right by 5 degrees. For the upper and lower images, only the inside of the face with respect to the corners of the eyes is extracted. Such a tilted sample image is prepared in order to form an object detection filter that allows the object identification unit 45 to identify the object even if the object is slightly inclined.

  The object feature calculation unit 32 calculates and extracts an object feature pattern of a person's face that is a detection object stochastically and statistically based on the sample image stored in the object image storage unit 31. And output to the object feature dictionary. The extraction of the object feature pattern will be described later.

  The object feature dictionary 33 stores the object feature pattern extracted and output by the object feature calculation unit 32. This object feature pattern is used as an object detection filter used when the object identifying unit 45 identifies the object.

  Next, the online execution environment system will be described.

  The image input unit 41 outputs an image including a detection target imaged by an imaging unit such as a camera to the surrounding area detection window designating unit 42.

  The peripheral area detection window designating unit 42 performs image input while moving an initial window having a predetermined size in a jumping manner with a pixel interval corresponding to the offset amount of the sample image stored in the object peripheral area image storage unit 21. The entire area of the image input by the unit 41 is specified, and partial area image information that is image information of the specified area is output to the target peripheral area identification unit 43. At this time, the image input unit 41 in the designated area outputs the image in association with the position information in the captured image.

  When the peripheral area detection window designating unit 42 finishes designating the entire area of the image using the initial window, it enlarges the window size by a predetermined ratio and repeats the same process as the process performed in the initial window. . This process is repeated as long as the window does not become larger than the size of the entire image area.

  The processing performed by the peripheral area detection window designating unit 42 using the initial window is shown in FIG. As shown in FIG. 4, the surrounding area detection window designating unit 42 first places an initial window at the start position in the image area, and places the initial window in the right direction at any time according to the pixel interval (here, the offset amount). 3 pixel intervals) and scan the entire image area. When the initial window reaches the right end, the surrounding area detection window designating unit 42 returns the initial window to the left end, but this time shifts the initial window slightly upward and similarly moves it to the right.

  When the initial window comes to the position of the stop in FIG. 4 by repeating this, the surrounding area detection window designating unit 42 enlarges the window at a predetermined ratio, places the enlarged window at the start position, and sets the initial window. Processing similar to that performed in the window is performed. As described above, the surrounding area detection window designating unit 42 repeats this process in a range where the window does not become larger than the size of the entire image area.

  In addition, when the surrounding area detection window designating unit 42 shifts the window to 1 pixel, the object surrounding area identifying unit 43 described later can most accurately identify the surrounding area of the object. However, if the image sample stored in the object peripheral area image storage unit 21 includes an image offset to the left and right by N pixels, there is a window at a position shifted by N pixels from the image of the object. However, since the object surrounding area identification unit 43 identifies the position as the object surrounding area, the object surrounding area identification unit 43 recognizes the pixel by covering N pixels on the left and right sides and covering a wide range of 2N pixels. Equivalent effect is obtained. Therefore, the peripheral area detection window designating unit 42 can shift the window in the left-right direction by at least 2N pixels. Note that 2N + 1 pixels may be used as long as it is possible to identify the peripheral area of the object on either the left or right side of the image of the object. In addition, even when the peripheral area detection window designating unit 42 moves the window for each of the plurality of pixels in the left-right direction in this way, the probability that the target peripheral area identifying unit 43 misses the peripheral area where the target image actually exists Will not be high.

  The object peripheral area identification unit 43 uses a filter stored in the object peripheral area feature dictionary 23 (that is, a filter used to identify an area around the object) to detect the peripheral area. It is identified whether or not the image represented by the partial region image information input from the window designating unit 42 is an image of the peripheral region of the object.

  The object peripheral area identification unit 43 is the size of the image represented by the partial area image information before the identification, that is, when the peripheral area detection window designation unit 42 designates the partial area image information. Enlarge appropriately according to the size of the window.

  When the identification result indicates that the object is a peripheral region of the target object, the target object peripheral region identifying unit 43 outputs the partial region image information to the target object detection window designating unit 44 as the peripheral region image information. At this time, the position information in the image captured by the image input unit 41 associated with the partial area image information is also output together. This identification method will be described later.

  The object detection window designating unit 44 represents the surrounding area image information that the object surrounding area identifying unit 43 has identified as the surrounding area of the object while moving the object detection initial window having a predetermined size. The entire region of the image to be processed is designated, and the peripheral region partial image information that is the image information of the designated region is output to the object identifying unit 45.

  At this time, the position information in the image captured by the image input unit 41 associated with the partial region image information and the periphery representing the position in the image represented by the peripheral region image information of the region specified by the object detection window specifying unit 44 The area position information is also output.

  When the object detection window specifying unit 44 finishes specifying the entire area of the image represented by the peripheral area image information using the object detection initial window, the window size of the object is enlarged by a predetermined ratio, The same processing as that performed in the initial object detection window is repeated. This process is repeated as long as the window does not become larger than the size of the entire image area represented by the peripheral area image information.

  The object identification unit 45 uses the object filter stored in the object feature dictionary 33 to display the image represented by the peripheral area partial image information input from the object detection window designating unit 44 as an image of the object. Identify whether or not there is.

  The object identifying unit 45 specifies the size of the object filter represented by the peripheral area partial image information before the identification, that is, the object detection window designating unit 44 has designated the peripheral area partial image information. Enlarge appropriately according to the size of the object detection window at the time.

  When the identification result indicates that the identification result is an image of the object, the object identification unit 45 outputs the peripheral area partial image information as the object image information to the detection result output unit. At this time, the position information in the image captured by the image input unit 41 associated with the partial area image information and the peripheral area position information corresponding to the peripheral area partial image information are also output together. This identification method will be described later.

  The detection result output unit 46, based on the position information and the peripheral region position information, shows the outline of the object filter when the object identification unit 45 identifies the image of the object on the video captured by the image input unit. Display at a fixed position. The area of the image identified as the object can be recognized by the outline of the object filter.

  Next, a method in which the object peripheral region feature calculation unit 22 extracts a peripheral region feature pattern of the face of the person who is the detection target will be described.

  First, in order to extract a peripheral region feature pattern, it is necessary to consider feature points of a human face that is a detection target. What should be noted as a feature of the face is that the horizontally long rectangular area including the eyes and eyebrows is dark, and the horizontally long rectangular area including the cheek and nose below it is bright. If the size of the horizontally long rectangular area is defined and the difference between the average luminance values of the horizontally long rectangular areas can be extracted as a feature amount, this can be extracted from the peripheral area feature pattern (position and size of both horizontally long rectangular areas). And the average luminance difference), that is, as a filter, it is possible to identify whether or not the image to be identified is an image of the face of a person who is the detection target.

  Focusing on the above trend, the facial feature points are reflected in a rectangle (hereinafter referred to as an outer frame rectangle) having the same size as the outer frame of the sample image stored in the object peripheral area image storage unit 21. FIG. 5 shows two horizontally long rectangular areas defined. Here, it is assumed that the outer frame rectangle has a size of 20 × 20 pixels.

  In FIG. 5, two horizontally long rectangular regions are defined using an orthogonal coordinate system in which the vertical axis with the lower left vertex of the outer frame rectangle as the origin is the Y axis and the horizontal axis is the X axis. In FIG. 5, the horizontally long rectangular area located on the lower side is defined as a first rectangular area, and the horizontally elongated rectangular area located on the upper side is defined as a second rectangular area.

  The first rectangular area is defined as a rectangle having coordinates of the lower left point as (X1, Y1) and having a size of vertical N1 and horizontal M1. Further, the parameters are defined on the assumption that the coordinates of the lower left point of the second rectangular area are (X2, Y2), and the rectangle has a size of vertical N2 and horizontal M2.

  In consideration of the symmetry of the left and right faces, the relational expressions X1 × 2 + M1 = 20 and X2 × 2 + M2 = 20 hold. Considering this objectivity, the number of forms that can be taken by the first rectangular area and the second rectangular area can be reduced, so that the calculation time described below can be reduced.

  First, for each of all combinations of two horizontally long rectangular areas defined by the above parameters, a true image sample (including a face image sample that is offset to the left and right) and a fake image sample ( An average luminance difference between two horizontally long rectangular regions that can best separate the image sample other than the face) is obtained.

  Here, “best separation” means that the true image sample image is correctly identified as a face image, and the false image sample image is not mistakenly identified as a face image. is there. Note that the true image sample image used here is a sample image (including an offset image) stored in the object peripheral region image storage unit 21 as shown in FIG.

  In this way, it is possible to obtain the most appropriate average luminance difference (referred to as the optimal average luminance difference) between the two horizontally long rectangular regions for each of all combinations of the two horizontally long rectangular regions. Then, each of the combinations of the two horizontally long rectangular areas is applied with the optimum average luminance difference of each of them to include a true image sample (including a face image sample that is offset left and right) and a fake image sample ( A combination of two horizontally long rectangular areas that can be best separated is used as a peripheral area feature pattern of a human face. The optimum average luminance difference of this combination is a threshold value used when the object peripheral area identifying unit 43 identifies whether or not the object peripheral area is a face peripheral area.

  In addition, although the case where the average brightness difference between both horizontally long rectangular areas is used has been described here, the present invention is not limited to this, and the relationship between the brightness of both horizontally long rectangular areas such as the brightness ratio of both horizontally long rectangular areas is used. be able to.

  The object peripheral area feature calculation unit 22 outputs the peripheral area feature pattern to the object peripheral area feature dictionary. FIG. 6 shows an example of the position and size of both horizontally long rectangular regions of the peripheral region feature pattern obtained by such a method. The peripheral area feature pattern differs somewhat depending on what sample is prepared, but the tendency of the form is common.

  In addition, about the target object feature pattern calculation method of the detection target object which the target object feature calculation part 32 performs, the conventional technique can be utilized as it is. For example, the object feature pattern extraction method as described in Patent Document 1 mentioned in the background section above can be used as it is.

  Next, a method for identifying whether or not the target object peripheral area identifying unit 43 is a peripheral area of the target object will be described.

  The target object peripheral area identifying unit 43 adds the peripheral area feature of the target object obtained by the target object peripheral area feature calculating unit 22 to the image represented by the partial area image information input from the peripheral area detection window specifying unit 42. A pattern is applied as a filter, and an average luminance difference between images corresponding to two horizontally long rectangular regions is calculated and compared with a threshold value corresponding to the feature pattern.

  As a result of the comparison, when the average luminance difference is larger than the threshold value, the object peripheral area identification unit 43 outputs the partial area image information as the peripheral area image information to the object detection window specifying unit 44.

  Note that the filter used is different in the method of identifying whether or not the image represented by the peripheral area partial image information input from the object detection window designating unit 44 performed by the object identifying unit 45 is an image of the object. However, this is basically the same as the method for identifying whether or not the object peripheral area identifying unit 43 is the peripheral area of the object.

  Next, the flow of processing operations in the online execution environment system of the object detection apparatus 10 will be described with reference to FIG.

  FIG. 7 is a flowchart showing the flow of processing operations in the online execution environment system.

  First, in ST701, an image input process is performed. That is, the image input unit 41 inputs the captured image including the detection target object to the surrounding area detection window designating unit 42.

  In ST702, an object peripheral area identification process is performed.

  That is, the surrounding area detection window designating unit 42 designates the entire area of the image input by the image input unit 41 while moving the initial window having a predetermined size little by little, and the image information of the designated area is used. Certain partial region image information is output to the object peripheral region identification unit 43.

  The object surrounding area identification unit 43 uses the filter stored in the object surrounding area feature dictionary 23 to display an image represented by the partial area image information input from the surrounding area detection window designating unit 42 of the object. Whether or not the image is a peripheral area is identified.

  Then, when the identification result indicates that the region is the peripheral region of the target, the target peripheral region identification unit 43 outputs the partial region image information to the target detection window specifying unit 44 as the peripheral region image information.

  Next, in ST703, an object image identification process is performed.

  That is, the object detection window designating unit 44 moves the object detection initial window having a predetermined size, and the object surrounding area identification unit 43 identifies the surrounding area image information identified as the object surrounding area. The entire area of the image represented by is designated, and the peripheral area partial image information of the designated area is output to the object identifying unit 45.

  The object identification unit 45 uses the object filter stored in the object feature dictionary 33 to display the image represented by the peripheral area partial image information input from the object detection window designating unit 44 as an image of the object. Identify whether or not there is.

  Next, in ST704, an object area specifying process is performed.

  That is, when the identification result indicates that the identification result is an image of the target, the target object identification unit 45 outputs the peripheral area partial image information as the target image information to the detection result output unit.

  Then, the detection result output unit 46 displays the outline of the object filter when the object identifying unit 45 identifies the image of the object on the image captured by the image input unit, and displays the image of the object Is identified. The area of the image identified as the object can be recognized by the outline of the object filter.

  As described above, according to the present embodiment, the object peripheral area identification unit 43 uses the peripheral area feature pattern extracted by the object peripheral area feature calculation unit 22 from the image sample including the offset image as a filter. Since the image represented by the partial region image information input from the peripheral region detection window specifying unit 42 is an image of the peripheral region of the target object, the peripheral region detection window specifying unit 42 determines the offset. Since scanning can be performed at intervals corresponding to the size, the peripheral region of the object image can be found at high speed.

  In addition, for only the peripheral region of the target image found by the target peripheral region identification unit 43, an image represented by the peripheral region partial image information input by the target object identification unit 45 from the target object detection window designating unit 44 is displayed. Since it is discriminated whether or not it is an image of the object, the area where the object identification unit 45 searches for the image of the object is limited, so that the image of the object can be searched at high speed. As a result, the processing amount of the object detection process of the entire object detection apparatus 10 can be reduced and the processing speed can be increased.

  In other words, in the object detection process, as a pre-stage of the process in which the object identification unit 45 identifies the object, a process in which the object peripheral area identification unit 43 identifies the peripheral area of the object is provided. Since the peripheral area identification unit 43 roughly detects an area where there is a high possibility that the object exists, the object identification unit 45 examines whether or not the image of the object exists only in the area. The object peripheral area identifying unit 43 limits the search range to the peripheral area of the object image searched for at high speed, and the object identifying unit 45 searches for the image of the object. The processing amount of the object detection process can be reduced and the processing speed can be increased.

  In addition, the object detection process for the object surrounding area identification unit 43 roughly detects the object detection process, and the object identification unit 45 examines whether the image of the object exists only within the area. Therefore, even if the detection of the surrounding area of the object by the object surrounding area discriminating unit 43 is rough, the object is detected by only one stage of the process of examining whether or not the image of the object exists. Compared to a conventional detection device that detects an image of an object, the time required for detection is short, and the detection accuracy can be expected to be comparable.

  INDUSTRIAL APPLICABILITY The present invention is useful as an object detection apparatus and method that can reduce the processing amount of the object detection process and reduce the processing speed without reducing the detection accuracy.

The block diagram which shows the structure of the target object detection apparatus which concerns on one embodiment of this invention. The figure which showed an example of the sample image which an object surrounding area image storage part stores The figure which showed an example of the sample image which a target object image storage part stores The figure for demonstrating the process which a window designation | designated part performs using an initial window A figure that defines two horizontally long rectangular areas that reflect facial feature points in the outer frame rectangle The figure which showed an example about the position and size of both horizontally long rectangular areas of the peripheral area feature pattern Flow diagram showing the flow of processing operations in the online execution environment The figure for demonstrating the scanning of the filter for the target object detection in the conventional target object detection apparatus

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Object detection apparatus 21 Object periphery area | region image storage part 22 Object periphery area | region feature calculation part 23 Object periphery area | region feature dictionary 31 Object object storage part 32 Object object feature calculation part 33 Object object characteristic dictionary 41 Image input part 42 Peripheral area detection window designation unit 43 Object peripheral area identification unit 44 Object detection window designation part 45 Object identification part 46 Detection result output part

Claims (6)

An object detection device for detecting an image area of a detection object included in an image,
An object surrounding area detecting means for detecting a surrounding image area of the detection object using a first detection filter;
Object detection means for detecting an image of the detection object for only the peripheral image region of the detection object detected using the second detection filter;
An object detection apparatus comprising:   The target detection apparatus according to claim 1, wherein the first detection filter is a feature pattern of the detection target extracted from an image sample including an image of the detection target offset. When the detection target is a human face, the first detection filter is a feature pattern of the detection target,
A first rectangular area corresponding to an area including eyes and eyebrows;
A second rectangular area located below the first rectangular area and corresponding to an area including a cheek and a nose;
Comprising
The size of the first rectangular area and the second rectangular area;
Positions of the first rectangular area and the second rectangular area;
The relationship between the luminance in both areas when both rectangular areas are applied to the image including the detection object;
The object detection device according to claim 1, wherein the object detection device is defined as a parameter. In the image including the detection object, comprising an area designating unit for designating an area to which the first detection filter is applied,
4. The object detection apparatus according to claim 1, wherein the area designating unit designates the area at an interval corresponding to an offset amount of the image sample. 5. An object detection method for detecting an image area of a detection object included in an image, comprising:
Detecting a peripheral image region of the detection object using a first detection filter;
Detecting the image of the detection object only for the peripheral image area of the detected detection object using a second detection filter;
The object detection method characterized by comprising. A program for detecting an image area of a detection target included in an image,
Detecting a peripheral image region of the detection object using a first detection filter;
Detecting an image of the detection object only for the peripheral image region of the detected detection object using a second detection filter;
An object detection program that causes a computer to execute the above.

Images