JP2011120076A - Image detection apparatus and image detection method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an image detection apparatus and an image detection method for improving target recognition accuracy. <P>SOLUTION: In an image detection apparatus 100, a cutout region determining section 130 determines a cutout image size in accordance with the image quality of a target candidate. Specifically, in the image detection apparatus 100, an erroneous recognition factor parameter calculating section 120 calculates the erroneous recognition factor parameter of the target candidate that can be the index of the image quality of the target candidates, and the cutout region determining section 130 determines the cutout image size in accordance with a value of the erroneous recognition factor parameter. Accordingly, as the image quality level of the target candidate becomes lower, a margin in image cutout can be made larger, thereby surely including the entire target candidate in a cutout image. As a result, the target recognition accuracy can be improved. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to an image detection apparatus and an image detection method.

  A wide-angle camera such as an omnidirectional camera is widely used in various fields because a single camera can obtain an image with a wide field of view. Wide-angle cameras are used in, for example, surveillance systems. In particular, an omnidirectional camera can obtain an omnidirectional image by using an omnidirectional lens optical system or an omnidirectional mirror optical system. An omnidirectional image obtained by an omnidirectional camera is generally a concentric image (a donut image).

  As a form of displaying a captured image obtained by a wide-angle camera, for example, there is a form (see Patent Document 1) in which a region including a target object (that is, a target) is cut out from the captured image and displayed. In the captured image obtained by the wide-angle camera, the object shape is distorted depending on the position where the object is reflected. For this reason, in the image processing apparatus disclosed in Patent Document 1, distortion correction processing is performed on the cut-out image.

JP 2007-31860 A

  By the way, the image quality after the distortion correction processing varies depending on the set position of the cutout region including the distorted object image. In particular, in the omnidirectional image, the degree of distortion differs between the central portion and the peripheral portion.

  Therefore, when a peripheral image including target candidates is cut out in the first stage and target recognition is further performed using the cut out image in the second stage, the setting method of the cut out area in the first stage is the target recognition. It greatly affects the accuracy. That is, when the image quality of the cutout region is poor, the target may not be included in the cutout region due to an error in setting the cutout region, and target recognition may not be performed. In the case of an omnidirectional image in which the degree of distortion differs between the central part and the peripheral part, this influence is particularly great.

  SUMMARY An advantage of some aspects of the invention is that it provides an image detection apparatus and an image detection method capable of improving target recognition accuracy.

  One aspect of the image detection apparatus of the present invention includes a detection unit that detects a target from an entire image, a determination unit that determines a cut-out image size in accordance with a misrecognition factor of the target, and the cut-out including the target. Clipping means for cutting out an image-sized cut image from the whole image.

  One aspect of the image detection method of the present invention includes a step of detecting a target candidate from an entire image, a step of determining a cut-out image size according to a misrecognition factor of each target candidate, and the cut-out including the target Cutting out a cutout image of the image size from the whole image.

  According to the present invention, it is possible to provide an image detection apparatus and an image detection method that can improve target recognition accuracy.

1 is a block diagram showing a configuration of an image detection apparatus according to an embodiment of the present invention. FIG. 3 is a flowchart for explaining the operation of the image detection apparatus according to the embodiment of the present invention. The figure which shows an example of the whole image The figure which shows an example of the cutting size determination table Diagram showing how to calculate the cropped image size The figure which shows the cut-out image of target candidate TC1 in FIG. 3A The figure which shows the cut-out image of target candidate TC2 in FIG. 3A The figure which contrasts the conventional image cutting and the image cutting of this Embodiment The figure which contrasts the conventional image cutting and the image cutting of this Embodiment FIG. 5 is a diagram for explaining a method for determining a cut-out image size when a misrecognition factor parameter is a target candidate distortion amount. The figure used for explanation of the method for determining the cut-out image size when the misrecognition factor parameter is the separation distance in real space between the wide-angle camera and the target

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

[1] Configuration FIG. 1 is a block diagram showing a configuration of an image detection apparatus 100 according to an embodiment of the present invention. In FIG. 1, the image detection apparatus 100 includes a target candidate detection unit 110, a misrecognition factor parameter calculation unit 120, a cutout region determination unit 130, a cutout unit 140, a distortion correction unit 150, and a template determination unit 160. And a target recognition unit 170 and a target image output unit 180. The image detection apparatus 100 is connected to a wide-angle camera and receives a captured image captured by the wide-angle camera. The wide-angle camera is, for example, an omnidirectional camera. In addition, the image detection apparatus 100 is connected to a display device, and causes the display device to display a cut-out image cut out from the captured image.

  The target candidate detection unit 110 detects target candidates included in the captured image S10. The target is, for example, a person. The target may be an object such as a vehicle. The target candidate detection unit 110 detects a moving object from the captured image S10 as a target candidate based on, for example, a motion difference or background difference (that is, a changed portion) in captured images of the preceding and following captured frames.

  The target candidate detection unit 110 outputs target candidate information S11 indicating the position and size of the detected target candidate. The target candidate position information is, for example, the center coordinates of the target candidate image. The size information of the target candidate is information indicating the spread from the center of the target candidate image, and is indicated by, for example, the height H and the width W of the target candidate. Here, the position information of the target is (x, y) with the camera center coordinates of the wide-angle camera (may be called the center of the captured image) as the origin (0, 0) in the same imaging coordinate system as the captured image S10. ) Information indicated by coordinates.

  The misrecognition factor parameter calculation unit 120 calculates a target misrecognition factor parameter S12 based on the captured image S10 and the target candidate information S11. The misrecognition factor parameter is a target candidate focus value, distortion amount, object size, luminance histogram, or separation distance in real space between the wide-angle camera and the target, and these can be used alone or in combination. The erroneous recognition factor parameter will be described in detail later.

  The cutout area determination unit 130 determines a cutout area based on the target candidate information S11 and the misrecognition factor parameter S12. Based on the target candidate information S11, the cut-out area determination unit 130 determines the position of the cut-out area so that, for example, the center coordinates of the target candidate image coincide with the center of the cut-out area. Further, the cut-out area determination unit 130 determines the size of the cut-out area (that is, the cut-out image size) based on the target candidate information S11 and the misrecognition factor parameter S12. The method for determining the cut-out image size will be described in detail later.

  The cutout unit 140 cuts out an image in the cutout region determined by the cutout region determination unit 130 from the captured image S10 and outputs the cutout image to the distortion correction unit 150.

  The distortion correction unit 150 performs distortion correction processing on the clipped image. When there are a plurality of cutout images, the distortion correction unit 150 performs distortion correction processing for each cutout image. The clipped image after the distortion correction processing is output to the target recognition unit 170.

  The template determination unit 160 determines a template based on the misrecognition factor parameter S12. That is, the template determination unit 160 determines a target template that matches a cut image including a distortion component that has not been completely removed by the distortion correction unit 150.

  The target recognition unit 170 recognizes a target image included in the cut image after distortion correction processing. Here, the target recognition unit 170 recognizes the target by the template matching process using the template determined by the template determination unit 160. For this target recognition, a HoG (Histogram of oriented Gradient) feature value may be used instead of a template, or an Edgelet feature value may be used.

  The target image output unit 180 outputs the cut image whose target is recognized by the target recognition unit 170 to the display device.

[2] Operation The operation of the image detection apparatus 100 having the above configuration will be described.

[2-1] Overview of Processing Flow FIG. 2 is a flowchart for explaining the operation of the image detection apparatus 100 according to an embodiment of the present invention.

  In step ST201, the target candidate detection unit 110 detects target candidates included in the captured image S10.

  In step ST202, the misrecognition factor parameter calculation unit 120 calculates a target misrecognition factor parameter S12 based on the captured image S10 and the target candidate information S11.

  In step ST203, the cut-out area determination unit 130 determines the cut-out image size based on the target candidate information S11 and the misrecognition factor parameter S12. In step ST203, the position of the cutout area is also determined.

  In step ST204, the cutout unit 140 cuts out an image in the cutout area determined by the cutout area determination unit 130 from the captured image.

  In step ST205, the distortion correction unit 150 performs distortion correction processing on the cut image.

  In step ST206, the target recognition unit 170 recognizes the target image included in the cut-out image after the distortion correction processing.

  In step ST207, it is determined whether or not the processing in steps ST202 to ST206 has been performed for all target candidates detected in step ST201. If it is determined that the process has not been performed for all target candidates (step ST207: NO), the processes in steps ST202 to ST206 are performed for the next target candidate to be processed. As described above, the processes in steps ST202 to ST206 are performed for each target candidate detected in ST201. On the other hand, when it is determined that the process has been performed for all target candidates (step ST207: YES), the target detection process ends.

[2-2] Method for Determining Clipped Image Size Next, a method for determining a recognition image parameter according to a misrecognition factor parameter and a misrecognition factor parameter will be described.

[2-2-1] When the misrecognition factor parameter is the focus value of the target candidate:
In step ST202, the misrecognition factor parameter calculation unit 120 calculates the relative focus value of the target candidate based on the captured image S10 and the target candidate information S11.

  Specifically, the misrecognition factor parameter calculation unit 120 first calculates the frequency components of the entire image and the frequency components of each target candidate. This frequency component may be normalized between 0 and 100. The misrecognition factor parameter calculation unit 120 calculates the output value (that is, the magnitude of a predetermined frequency component) of the bandpass filter when the image signal is input to the bandpass filter (not shown). The predetermined frequency component is, for example, a high frequency component indicating an edge portion.

  Next, the misrecognition factor parameter calculation unit 120 calculates the relative focus value of the target candidate. This relative focus value is the ratio of the frequency component of the target candidate to the frequency component of the entire image.

  In step ST203, the cut-out area determination unit 130 determines the cut-out image size based on the target candidate size and the relative focus value indicated by the target candidate information S11.

  Specifically, the cutout region determination unit 130 holds a cutout size determination table in which the relative focus value Fr and the cutout size coefficient Sc are associated with each other. Then, the cut-out area determination unit 130 uses the relative focus value Fr calculated by the misrecognition factor parameter calculation unit 120 as a reading address, and extracts the cut-out size coefficient Sc corresponding to the relative focus value Fr from the cut-out size determination table. Search for. Then, the cut region determination unit 130 calculates the cut image size by multiplying the searched cut size coefficient Sc by the target candidate size indicated by the target candidate information S11.

  FIG. 3 is a diagram for explaining a method for determining the cut-out image size when the misrecognition factor parameter is the focus value of the target candidate. FIG. 3A is a diagram illustrating an example of an entire image. FIG. 3B is a diagram illustrating an example of a cut-out size determination table. FIG. 3C is a diagram illustrating a method for calculating a cut-out image size. FIG. 3D is a diagram showing a cut-out image of the target candidate TC1 in FIG. 3A. FIG. 3E is a diagram showing a cut-out image of the target candidate TC2 in FIG. 3A.

  In FIG. 3, the focus value of the entire image is 80, the focus value of the target candidate TC1 (object 1) is 70, and the focus value of the target candidate TC2 (object 2) is 25. At this time, the relative focus value Fr of the target candidate TC1 is 0.8, and the relative focus value Fr of the target candidate TC2 is 0.3.

  Therefore, the cut region determination unit 130 can determine the cut size coefficient Sc of the target candidate TC1 as 1.2 and the cut size coefficient Sc of the target candidate TC2 as 1.8 from the cut size determination table. .

  As shown in FIG. 3C, if the size of the target candidate is height H and width W, the cut image size is height H × Sc and width W × Sc. Therefore, as shown in FIGS. 3D and 3E, the size of the cut image including the target candidate TC1 is H × 1.2 and the width W × 1.2, and the size of the cut image including the target candidate TC2 is the height. H × 1.8 and width W × 1.8.

  Here, in the extraction size determination table, the corresponding extraction size coefficient Sc increases as the relative focus value Fr decreases. In other words, the cut region determination unit 130 increases the cut image size as the relative focus value Fr decreases. The relative focus value Fr is proportional to the image quality. That is, the higher the relative focus value Fr of the clipped image, the higher the quality level of the clipped image. Therefore, after all, the cut-out area determination unit 130 determines the cut-out image size according to the image quality of the target candidate using the relative focus value Fr as an index.

  When the image quality of the target candidate is poor, a detection error is likely to occur, it is difficult to set a cutout area that is just fitted to the target candidate, and the target candidate may protrude from the set cutout area. That is, as shown in FIG. 4A, even if the image is cut out conventionally, if the image quality of the target candidate is good, a cut-out area that is just fitted to the target candidate can be set. However, when the image quality of the target candidate is poor, there is a problem that the head of the person who is the target protrudes from the cutout region and is missing. There is a high possibility that target recognition using this cut-out image will fail.

On the other hand, as in the present embodiment, the lower the target candidate image quality level, the larger the extracted image size, that is, by increasing the margin at the time of image clipping, the entire target candidate. Can be reliably included in the cut-out image. That is, as shown in FIG. 4B, when the image quality level of the target candidate is low (in FIG. 4B, the focus value is small), a large cut-out size coefficient is selected. Then, since the image is cut out with the cut size obtained by multiplying the cut size coefficient by the reference size, the whole body of the person can be surely included in the cut image.
In addition, when the target candidate image quality level is high, the cut image size is not increased too much, and the cut image size is increased only when the target candidate image quality level is low. In addition, an increase in the processing amount of the target detection process can be prevented.
FIG. 4 shows the case where the focus value of the target candidate is used as a misrecognition factor parameter. Also in the case of the object size and the luminance histogram of the target candidate, the same effect as described above can be obtained.

[2-2-2] When the misrecognition factor parameter is the distortion amount of the target candidate:
In step ST202, the misrecognition factor parameter calculation unit 120 calculates the distortion amount of the target candidate based on the captured image S10 and the target candidate information S11.

  Specifically, since the distance between the center of the captured image and the target candidate is proportional to the amount of distortion, the misrecognition factor parameter calculation unit 120 uses the distance between the center of the captured image and the target candidate as an index of the amount of distortion. r is calculated. This separation distance r is represented by the number of pixels (number of pixels).

  In step ST203, the cut-out area determination unit 130 determines the cut-out image size based on the size of the target candidate indicated by the target candidate information S11 and the separation distance r between the center of the captured image and the target candidate.

  Specifically, the cutout area determination unit 130 holds a cutout size determination table in which the separation distance r between the center of the captured image and the target candidate and the cutout size coefficient Sc are associated with each other. Then, the cutout area determination unit 130 uses the distance r between the center of the captured image calculated by the misrecognition factor parameter calculation unit 120 and the target candidate as a read address, and uses the relative focus value Fr from the cutout size determination table. And a corresponding cut-out size coefficient Sc is retrieved. Then, the cut region determination unit 130 calculates the cut image size by multiplying the searched cut size coefficient Sc by the target candidate size indicated by the target candidate information S11.

  FIG. 5 is a diagram for explaining a method for determining a cut-out image size when the misrecognition factor parameter is the distortion amount of the target candidate. FIG. 5A is a schematic diagram illustrating an example of an entire image. FIG. 5B is a diagram illustrating an example of a cut-out size determination table. FIG. 5C is a diagram showing a cut-out image of the target candidate TC1 in FIG. 5A. FIG. 5D is a diagram showing a cut-out image of the target candidate TC2 in FIG. 5A.

  In FIG. 5, the separation distance r of the target candidate TC1 is 20, and the separation distance r of the target candidate TC2 is 80. Therefore, the cut region determination unit 130 can determine the cut size coefficient Sc of the target candidate TC1 as 1.2 and the cut size coefficient Sc of the target candidate TC2 as 1.8 from the cut size determination table. .

  Here, in the cutout size determination table, the corresponding cutout size coefficient Sc increases as the separation distance r increases. That is, the cut region determination unit 130 increases the cut image size as the separation distance r (that is, the amount of distortion) increases. Since the distortion amount is inversely proportional to the image quality, after all, the cut-out area determination unit 130 determines the cut-out image size according to the image quality of the target candidate using the distortion amount as an index.

[2-2-3] When the misrecognition factor parameter is a separation distance in real space between the wide-angle camera and the target:
In step ST202, the misrecognition factor parameter calculation unit 120 calculates a separation distance in real space between the wide-angle camera and the target based on the captured image S10 and the target candidate information S11. Stereo vision (monocular or binocular) is used to calculate the separation distance. That is, a plurality of images (stereo images) obtained by stereo viewing are used.

  In step ST203, the cut-out area determination unit 130 determines the cut-out image size based on the size of the target candidate indicated by the target candidate information S11 and the separation distance in real space between the wide-angle camera and the target.

  Specifically, the cut-out area determination unit 130 holds a cut-out size determination table in which the separation distance in the real space between the wide-angle camera and the target and the cut-out size coefficient Sc are associated with each other. Then, the cutout region determination unit 130 uses the separation distance in the real space between the wide-angle camera and the target calculated by the misrecognition factor parameter calculation unit 120 as a read address, and uses the relative focus value Fr from the cutout size determination table. The corresponding cut-out size coefficient Sc is searched. Then, the cut region determination unit 130 calculates the cut image size by multiplying the searched cut size coefficient Sc by the target candidate size indicated by the target candidate information S11.

  FIG. 6 is a diagram for explaining a method for determining a cut-out image size when the misrecognition factor parameter is a separation distance in real space between the wide-angle camera and the target. FIG. 6A is a schematic diagram illustrating a positional relationship between target candidates in real space. FIG. 6B is a diagram illustrating an example of a cut-out size determination table. FIG. 6C is a diagram showing a cut-out image of the target candidate TC1 in FIG. 6A. FIG. 6D is a diagram showing a cut-out image of the target candidate TC2 in FIG. 6A. FIG. 6E is a diagram showing a cut-out image of the target candidate TC3 in FIG. 6A.

  In FIG. 6, the separation distance D1 of the target candidate TC1 is 1 [m], the separation distance D2 of the target candidate TC2 is 5 [m], and the separation distance D3 of the target candidate TC3 is 7 [m]. Therefore, the cutout area determination unit 130 sets the cutout size coefficient Sc of the target candidate TC1 to 1.1, the cutout size coefficient Sc of the target candidate TC2 to 1.2, and the cutout of the target candidate TC3 from the cutout size determination table. The output size coefficient Sc can be determined to be 1.5.

  Here, in the cut-out size determination table, the corresponding cut-out size coefficient Sc increases as the separation distance between the wide-angle camera and the target in real space increases. That is, the cut-out area determination unit 130 increases the cut-out image size as the distance between the wide-angle camera and the target in real space increases. Since the separation distance in the real space between the wide-angle camera and the target is inversely proportional to the image quality, after all, the cut-out area determination unit 130 uses the distortion amount as an index and determines the cut-out image size according to the image quality of the target candidate. Is determined.

[2-2-4] When the misrecognition factor parameter is the object size of the target candidate:
In step ST203, the cut-out area determination unit 130 determines a cut-out image size based on the size of the target candidate on the image plane indicated by the target candidate information S11.

  Specifically, the cutout area determination unit 130 holds a cutout size determination table in which the target candidate size and the cutout size coefficient Sc are associated with each other. Then, the cut region determination unit 130 searches the cut size coefficient Sc from the cut size determination table using the size of the target candidate detected by the target candidate detection unit 110 as a read address. Then, the cut region determination unit 130 calculates the cut image size by multiplying the searched cut size coefficient Sc by the target candidate size indicated by the target candidate information S11.

  Here, in the cutout size determination table, the cutout size coefficient Sc may be associated with each of the height H and the width W independently, or an index (for example, an area) that combines the height H and the width W. ) May be associated with the cut-out size coefficient Sc. In the former case, the cut region determination unit 130 determines the cut size coefficient Sc for each of the height H and the width W of the target candidate detected by the target candidate detection unit 110, and determines The cropped image size is calculated by multiplying the cutout size coefficient Sc. In the latter case, one cut-out size coefficient Sc is determined for one target candidate, so that the cut-out area determining unit 130 determines the size of the target candidate detected by the target candidate detecting unit 110 as the determined cut-out size. The cropped image size is calculated by multiplying the output size coefficient Sc.

  Further, when the type of the target is narrowed down to one, that is, for example, when the target is narrowed down to a person, the size of the target falls within a certain range. For this reason, the size of the target image is considered to be substantially proportional to the separation distance in real space between the wide-angle camera and the target. Therefore, the cut-out image size corresponding to the image quality of the target candidate can also be determined by using the target candidate size.

  Note that, when the misrecognition factor parameter is the object size of the target candidate, the size of the target candidate detected by the target candidate detection unit 110 is used as it is as the misrecognition factor parameter. There is no need for processing.

[2-2-5] When the erroneous recognition factor parameter is the luminance histogram of the target candidate:
In step ST202, the misrecognition factor parameter calculation unit 120 calculates a luminance histogram of the target candidate and its peripheral part based on the captured image S10 and the target candidate information S11. Then, the misrecognition factor parameter calculation unit 120 calculates a difference value between the luminance of the target candidate and the luminance of the peripheral portion of the target candidate.

  In step ST203, the cut-out area determination unit 130 determines a cut-out image size based on the target candidate size and the luminance histogram indicated by the target candidate information S11.

  Specifically, the cut region determination unit 130 holds a cut size determination table in which the difference value between the luminance of the target candidate and the luminance of the peripheral portion of the target candidate is associated with the cut size coefficient Sc. ing. Then, the cut-out area determination unit 130 uses the difference value between the luminance of the target candidate calculated by the misrecognition factor parameter calculation unit 120 and the luminance of the peripheral part of the target candidate as a read address, and extracts from the cut-out size determination table. The cut size coefficient Sc is searched. Then, the cut region determination unit 130 calculates the cut image size by multiplying the searched cut size coefficient Sc by the target candidate size indicated by the target candidate information S11.

  Here, the smaller the difference value between the luminance of the target candidate and the luminance of the peripheral portion of the target candidate, the more difficult it is to identify the target candidate. That is, the smaller the difference value, the lower the image quality level. For this reason, it is difficult to set a cutout area that just fits the target candidate, and the target candidate may protrude from the set cutout area. Therefore, the smaller the difference value is, the larger the extracted image size is, so that the entire target candidate can be surely included in the extracted image.

  As described above, according to the present embodiment, in the image detection apparatus 100, the cut region determination unit 130 determines the cut image size according to the image quality of the target candidate.

  By doing so, the lower the image quality level of the target candidate, the larger the margin at the time of image cutting, so that the entire target candidate can be reliably included in the cut-out image. As a result, target recognition accuracy can be improved.

  Specifically, in the image detection apparatus 100, the error recognition factor parameter calculation unit 120 calculates a target candidate error recognition parameter that is an index of the target candidate image quality, and the cutout region determination unit 130 determines that the The cut-out image size is determined according to the value of the recognition factor parameter. This misrecognition factor parameter is at least one of the relative focus value of the target candidate, the amount of distortion, the object size, and the luminance histogram, and the separation distance between the photographing means (here, wide-angle camera) for photographing the entire image and the target. including.

  In the above description, the case where the target candidate detection unit 110 detects a target candidate using a captured image has been described. However, the present invention is not limited to this. For example, the target candidate detection unit 110 may detect a target candidate by receiving a signal from a wireless tag attached to the target and obtaining a position coordinate of the wireless tag based on the received wireless signal.

  In the above description, the case where one cut-out size determination table is prepared for each misrecognition factor parameter has been described. However, the present invention is not limited to this. For example, the darkness level of the imaging target region affects the entire misrecognition factor parameter. Therefore, a plurality of cut-out size determination tables corresponding to the darkness level of the imaging target area may be prepared. In this case, the cut region determination unit 130 determines the cut image size using the cut size determination table suitable for the darkness level by switching the cut size determination table to be used according to the darkness level. be able to.

  In addition, the above-described image detection apparatus 100 can be configured by a computer such as a personal computer including a memory and a CPU. The function of each component constituting the image detection apparatus 100 can be realized by the CPU reading and executing the computer program stored in the memory.

  The image detection apparatus and the image detection method of the present invention are useful for improving the target recognition accuracy.

DESCRIPTION OF SYMBOLS 100 Image detection apparatus 110 Target candidate detection part 120 Error recognition factor parameter calculation part 130 Cutout area determination part 140 Cutout part 150 Distortion correction part 160 Template determination part 170 Target recognition part 180 Target image output part

Claims (7)

Detection means for detecting a target from the entire image;
Determining means for determining a cut-out image size according to the target misrecognition factor;
A cutting means for cutting out the cut image of the cut image size including the target from the whole image;
An image detection apparatus comprising: A calculation means for calculating a target misrecognition factor;
The determining unit determines a cut-out image size according to the misrecognition factor;
The image detection apparatus according to claim 1. The misrecognition factor is at least one of a relative focus value of the target candidate, a distortion amount, an object size, a luminance histogram, and a separation distance between the imaging unit that captures the entire image and the target.
The image detection apparatus according to claim 2. The relative focus value is a relative focus value that is a ratio of the overall image and the frequency component of the target candidate;
The determination unit increases the cut-out image size as the relative focus value decreases.
The image detection apparatus according to claim 2. The misrecognition factor is a distortion amount of the target candidate,
The determining means increases the cut-out image size as the distortion amount increases.
The image detection apparatus according to claim 2. The misrecognition factor is a separation distance between the imaging means for capturing the entire image and the target,
The determination means increases the cut-out image size as the separation distance increases.
The image detection apparatus according to claim 2. Detecting target candidates from the entire image;
Determining a cut-out image size according to a misrecognition factor of each target candidate;
Cutting out the cut image of the cut image size including the target from the whole image;
An image detection method comprising:

Images