JP2000123185A - Object recognizing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an object recognizing device by which constantly stable and highly reliable object recognizing characteristics is obtained with minimum calculation quantity even when photographing magnification is different from image magnification of template data. SOLUTION: The object recognizing device is constituted by providing a characteristic extracting means 103 to form an image from which the characteristics are extracted by extracting a characteristic part of an image pickup field, a magnification calculating means 107 to calculate the image magnification of the image from which the characteristics are extracted formed by the characteristic extracting means and a variable magnification means 103 to perform a variable magnification processing for either of the image from which the characteristics are extracted or an object standard image so that the image magnification calculated by the magnification calculating means approximately coincides with the image magnification of the object standard image and object recognition is performed based on the image from which the characteristics are extracted and the object standard image with the approximately coincident image magnification.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an object recognizing device capable of recognizing objects having different imaging magnifications.

[0002]

2. Description of the Related Art A conventional object recognition apparatus will be described with reference to FIG. In this figure, reference numeral 1 denotes a photometric circuit for capturing a scene image, 2 a feature extraction circuit for extracting a characteristic portion for recognizing a subject from the scene image, and 3 a memory 4 for performing subject recognition in advance. This is a comparator for comparing the set template information with the feature extraction image.

A photographed image is obtained by a photometric circuit 1 comprising a CCD or the like via a lens, a diaphragm, an optical filter, and the like. The obtained captured image is input to the feature extraction circuit 2.
The feature extraction circuit 2 selectively extracts image information considered to be a subject from the obtained image. At this time, it is generally known to use the brightness, color information, contour information, and the like of the subject. In addition, a feature area is extracted by dividing the area into the same luminance area or the same color area, or by combining with the contour. Feature extraction circuit 2
Is input to the comparator 3.

The comparator 3 compares and determines the correlation between the template information and each other, and outputs the result as subject information. In the template information, information on a subject to be recognized is set in advance under an assumed condition. This information is generally image information (subject standard image)
Is set as Generally, the correlation can be obtained by calculating the sum of the absolute values of the differences between the pixels and calculating the correlation position or the correlation based on the sum. Various other methods are known for the correlation.

Further, the concept of the operation will be described with reference to FIG. The input image is an image obtained by recognizing a subject obtained by photographing or the like. Now the person is extracted, 640x480
A pixel feature extraction image is output. By performing a correlation operation between the feature extraction image and a template image for recognizing a person, and comparing the similarities, subject recognition can be performed.

This correlation operation is obtained by shifting the positional relationship between the extracted image and the template image little by little, and calculating the sum of the absolute values of the differences between the pixels at each of the positional relationships. As described above, there are various other methods for obtaining the correlation, and the method is not limited to this.

If the obtained value (correlation) is larger than a predetermined threshold value, it is determined that the contents of the feature extraction image and the contents of the template image relate to the same subject, and output as subject information. At this time, if the magnification of the template image is equal to the magnification of the photographed and extracted image, good recognition is performed. However, if the magnification is different, the recognition characteristics are deteriorated depending on the degree.

This will be described with reference to FIG. The upper graph shows the recognition rate and the lower graph shows the erroneous recognition rate. The horizontal axis indicates the image magnification, and the vertical axis indicates the recognition rate or the erroneous recognition rate. Regarding the recognition rate, the maximum recognition rate is shown when the magnification of the preset template image and the input image are equal, and the recognition rate decreases as the difference between them increases.

On the contrary, the erroneous recognition rate is lowest when the image magnifications are equal, and increases as the difference increases. The false recognition rate is the rate at which different things are erroneously recognized as the same, so that a lower one is better.

As can be seen from the graph of FIG. 11, if the template image magnification and the input image magnification are not equal, good recognition characteristics cannot be obtained. In order to improve the recognition rate, there is a method of preparing a plurality of templates having different magnifications in advance and maintaining the recognition rate even when the input image magnification changes.

FIG. 12 shows the concept. A plurality of template images are prepared in advance by changing the magnification to T1, T2, T3,
A correlation operation is performed between each template image and the input image to recognize a subject.

FIG. 13 shows the recognition / erroneous recognition characteristics at that time. The recognition rate is the logical sum of the three template characteristics,
It can be seen that stable recognition is possible even when the magnification changes over a wide range.

[0013]

However, since the erroneous recognition rate is the same logical sum of the three template images, the erroneous recognition rate always has a high characteristic. With such characteristics, stable recognition characteristics with low reliability cannot be obtained.

Furthermore, the correlation calculation requires processing for the number of templates. For example, if three template images are used, about three times the calculation processing is required.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a subject recognizing apparatus capable of always obtaining a stable and reliable subject recognizing characteristic with a minimum amount of calculation even when the photographing magnification is different from the image magnification of the template data.

[0016]

In order to achieve the above object, a subject recognizing device according to the first aspect of the present invention includes a feature extracting means for extracting a feature portion of an imaged object scene to form a feature extraction image. A magnification calculating means for calculating an image magnification of the feature extracted image formed by the characteristic extracting means, and an image magnification calculated by the magnification calculating means and a predetermined image magnification of the standard image of the subject are substantially matched. Is provided with scaling means for performing scaling processing on either the feature extraction image or the subject standard image, so that the subject can be recognized based on the feature extraction image and the subject standard image having substantially the same image magnification.

In the object recognition apparatus according to the second aspect of the present invention, a feature extraction unit for extracting a feature portion of an imaged object scene to form a feature extraction image, and a feature extraction image of the feature extraction image formed by the feature extraction unit Magnification calculation means for calculating the image magnification, and image selection means for selecting a subject standard image that substantially matches the image magnification calculated by the magnification calculation means from a plurality of subject standard images having different preset image magnifications,
The subject recognition is performed based on the feature extraction image and the subject standard image whose image magnifications substantially match. As a result, only the feature extracted image and the subject standard image having substantially the same magnification are always compared for subject recognition, so that the subject can be recognized stably and erroneous recognition can be performed without performing complicated arithmetic processing. It is possible to provide a less reliable object recognition device.

[0018]

(First Embodiment) A subject recognition apparatus according to a first embodiment of the present invention will be described with reference to FIG. FIG. 1 is a block diagram showing the configuration of the object recognition device. In this figure, 101 is a photometric circuit, 102
Is a distance measurement circuit, 103 is a feature extraction circuit (feature extraction means),
104 is a comparator, 105 is a memory storing a template image (subject standard image), 106 is a memory storing scaling (image magnification) information of template image information, 107 is a magnification calculator (magnification) for calculating an imaging magnification Computing means) and 108 are a magnification (magnifying means).

The captured image obtained by the photometry circuit 101 is input to a feature extraction circuit 103. The distance measuring circuit 10
2, the distance measurement data of each part of the captured image is input to the feature extraction circuit 103. The distance measurement data corresponds to each part of the captured image of the photometry circuit 102.

A feature extracting circuit 103 extracts a feature portion considered to be a subject based on data of the photometric circuit 101 and the distance measuring circuit 102, and outputs the extracted portion to the comparator 104 as feature extracted image data. The feature extraction circuit 103 extracts a subject from brightness, color information, contour information, distance information, etc., and has the same brightness, the same color, the area divided by the same distance and the contour, and a combination thereof. And so on.

On the other hand, the template image stored and set in the memory 105 is image information that is the basis of object recognition.

The feature extraction image is sent from the feature extraction circuit 103 to the comparator 104, and the comparator 104 performs comparison processing. In this embodiment, first, the imaging magnification of the feature extraction image is obtained. Specifically, the magnification calculator 107 calculates the magnification of the feature extraction image from the feature extraction image extracted by the feature extraction circuit 103 and the distance measurement information at the time of imaging, as well as lens information and the like.

Then, the magnification calculator 107 obtains a difference between the calculated image magnification of the feature extraction image and the scaling information (image magnification of the template image) stored in the memory 106, and calculates the magnification of the template image as the characteristic extraction image. The value for making the value substantially equal to the magnification (approximately to the extent that it can be regarded as a perfect match or a match in the object recognition) is used as a magnifier 108
Output to

The scaling unit 108 scales the template image stored in the memory 105 based on the value from the magnification calculator 107 and outputs the scaled image to the comparator 104. Thereby, the comparator 104 can compare the feature extraction image and the template image with the same magnification.

The comparator 104 calculates the correlation between the two images having the same magnification. If the correlation is higher than a predetermined threshold, the comparator 104 determines that the subject of the feature extraction image and the subject of the template image are the same and outputs the recognition result. I do.

Here, the operation of the object recognition device will be described more specifically with reference to FIG. A feature extraction image is obtained from the input image, and this feature extraction image is 8 in 200 pixels.
If it is 00 mm, the image magnification is 800 mm /
200 Pixel = 4 mm / Pixel.

On the other hand, if the preset template image is 2 mm / Pixel, a good recognition characteristic cannot be obtained as it is.
The magnification is corrected for the Pixel. That is, 2 mm / Pix
el is scaled to 4 mm / Pixel.

Then, by comparing the feature extraction image with the same image magnification and the template image, a good recognition characteristic can be obtained.

FIG. 3 is a graph showing recognition characteristics. As before, the upper part shows the recognition rate and the lower part shows the erroneous recognition rate.
From these graphs, since the magnification of the template image is changed so that the magnification becomes the same even if the magnification of the input image changes, it is possible to always compare images with the same magnification,
It turns out that it is always possible to obtain a high recognition rate stably. Similarly, a low value of the erroneous recognition rate is always stably obtained.

In the conventional example, a plurality of template images having different image magnifications were compared with the input image at the same time, so that the erroneous recognition rate was greatly deteriorated.
Since the comparison operation is performed only on one template image having the same magnification, the erroneous recognition rate is not deteriorated and always becomes the best state.

Although FIG. 3 shows the condition in which there is a step in the variable power step of the variable power transformer 103, the characteristics are rippled in a ripple shape. Are better characteristics with less fluctuation.

(Second Embodiment) FIG. 4 shows a second embodiment of the present invention.
1 illustrates an object recognition device according to an embodiment. The same parts as those in the first embodiment are denoted by the same reference numerals and description thereof will be omitted.

In FIG. 4, reference numeral 205 denotes a memory for storing a plurality of template images having the same contents at different image magnifications; 206, a memory for storing scaling information as magnification information of each template image; A template selector (selecting means) for selecting a template image to be used for (1). Other configurations are the same as those of the first embodiment.

In the present embodiment, instead of adjusting the magnification of the template images using a magnification, template images having a plurality of image magnifications are prepared in advance, and an appropriate template image is selected from these template images. The first embodiment differs from the first embodiment in that the magnification of an input image (feature extraction image) is adjusted.

In the present embodiment, first, the magnification calculator 107 calculates the image magnification of the input image in the same manner as in the first embodiment, and inputs it to the template selector 208. The template selector 208 selects a template image having a magnification that substantially matches the magnification information from the magnification calculator 107 from a plurality of preset template image magnifications, and
Output to Hereinafter, as in the first embodiment, object recognition is performed by correlation calculation or the like. The specific operation of the subject recognition device of the present embodiment will be described with reference to FIG. The memory 206 has 2 mm / Pixel, 3 mm / Pixel and 4 mm / Pixel.
Three types of template images having the same content and having an image magnification of mm / Pixel are stored. Since the feature extraction image is 4 mm / Pixel, a 4 mm / Pixel template image with a magnification corresponding to this is selected, and subject recognition is performed by correlation calculation or the like.

FIG. 6 is a graph showing recognition characteristics. First
As in the embodiment, the upper part shows the recognition rate, and the lower part shows the erroneous recognition rate. From these graphs, a template image having a magnification that matches the magnification of the input image is selected from the three template images of the image magnifications T1, T2, and T3 corresponding to the change in the magnification of the input image. A high recognition rate can be obtained. Similarly, a low value of the false recognition rate can be obtained almost stably.

In this embodiment, the case where the number of magnification types of the template image is set to three has been described. However, the more the number of the types, the more stable the characteristics with less fluctuation as the recognition characteristics. .

(Third Embodiment) FIG. 7 shows a third embodiment of the present invention.
1 illustrates an object recognition device according to an embodiment. The same parts as those in the above embodiments are denoted by the same reference numerals, and the description will not be repeated.

In FIG. 7, reference numeral 308 denotes a feature extraction circuit 10.
3 is a magnifier that takes the output image of the input 3 as an input and scales the input image based on magnification information of the magnification calculator 107. Others are the same as the first embodiment.

This embodiment differs from the first embodiment in that the magnification of the template image and the feature extraction image are adjusted not by scaling the template image but by scaling the feature extraction image.

In the subject recognizing device of the present embodiment, first, a feature extraction image for subject recognition is output from the feature extraction circuit 103. The magnification calculator 107 calculates the image magnification of the feature extraction image in the same manner as in the first embodiment, and changes the value for making the image magnification substantially equal to the image magnification of the template image stored in the memory 106. Input to the container 308.

The scaler 308 scales the input feature extracted image based on the calculation result from the scale calculator 107 and outputs the scaled image to the comparator 104. The template image is directly input to the comparator 104, where the subject is recognized by a correlation operation or the like.

FIG. 8 shows the subject recognition operation of the present embodiment.
This will be specifically described with reference to FIG. In this figure, the image magnification of the template image is 2 mm / Pixel, and the image magnification of the feature extraction image is 4 mm / Pixel. For this reason, the feature extraction image is scaled to 2 mm / Pixel, and after matching the magnification, subject recognition is performed by a correlation operation with the template image or the like.

The characteristics of the recognition rate and the erroneous recognition rate at this time are the same as those of the graph of FIG. 3 described in the first embodiment. For this reason, even if the subject recognition device of the present embodiment is used, both the recognition rate and the erroneous recognition rate can always obtain good characteristics stably.

[0045]

As described above, according to the present invention,
Only the feature-extracted image and the subject standard image of approximately the same magnification are always compared for subject recognition, so that the subject can be recognized stably and reliability without erroneous recognition is reduced without performing complicated arithmetic processing. And a subject recognition device with high image quality can be provided.

[Brief description of the drawings]

FIG. 1 is a block diagram of a subject recognition device according to a first embodiment of the present invention.

FIG. 2 is a diagram showing an operation outline of the object recognition device.

FIG. 3 is a graph showing an outline of recognition / erroneous recognition characteristics in the object recognition device.

FIG. 4 is a block diagram of a subject recognition device according to a second embodiment of the present invention.

FIG. 5 is a diagram showing an outline of the operation of a subject recognition device according to the second embodiment.

FIG. 6 is a graph showing an outline of recognition / erroneous recognition characteristics according to the second embodiment.

FIG. 7 is a block diagram of a subject recognition device according to a third embodiment of the present invention.

FIG. 8 is a diagram showing an outline of the operation of a subject recognition device according to the third embodiment.

FIG. 9 is a block diagram showing a conventional object recognition device.

FIG. 10 is a diagram showing an outline of operation of a conventional object recognition device.

FIG. 11 is a graph showing an outline of recognition / erroneous recognition characteristics in a conventional object recognition device.

FIG. 12 is a diagram showing an operation outline of another conventional object recognition device.

FIG. 13 is a graph showing an outline of recognition / erroneous recognition characteristics in a conventional object recognition device.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1, 101 Photometry circuit 102 Distance measurement circuit 2, 103 Feature extraction circuit 3, 104 Comparator 105, 205 Template information memory 106, 206 Scaling information memory 107 Magnification calculator 108, 308 Magnifier 208 Template selector

Claims (3)

[Claims] A feature extraction unit configured to extract a feature portion of a captured object scene to form a feature extraction image; a magnification calculation unit configured to calculate an image magnification of the feature extraction image formed by the feature extraction unit; Scaling means for scaling any one of the feature extracted image and the subject standard image so that the image magnification calculated by the magnification calculating means substantially matches the preset image magnification of the subject standard image. A subject recognition apparatus, comprising: performing subject recognition based on the feature extraction image and the subject standard image having substantially the same image magnification. 2. A feature extraction unit for extracting a feature portion of a captured object scene to form a feature extraction image; a magnification calculation unit for calculating an image magnification of the feature extraction image formed by the feature extraction unit; Image selection means for selecting a subject standard image that substantially matches the image magnification detected by the magnification detection means from among a plurality of subject standard images having different set image magnifications, wherein the image magnification is substantially matched. A subject recognition apparatus that performs subject recognition based on a feature extraction image and the subject standard image. 3. The subject recognition apparatus according to claim 1, wherein the magnification calculating means calculates an image magnification of the feature extraction image based on at least information on a field distance at the time of imaging.