JP2983019B1 - Target detection method and target detection device - Google Patents

Abstract

A method and apparatus for satisfactorily detecting the shape of a target from an image in which the contrast between the background and the luminance value of pixels constituting the target are not good, and for quickly detecting the target in the image. SOLUTION: A pixel constituting a target candidate is detected from an input original image to generate a binarized image, and a target candidate configuration is generated using the input original image and the binarized image generated above. The presence / absence of the target candidate is determined from the statistical value of the pixel and the statistical value of the background constituent pixels, and based on the result of the presence / absence determination, if the target candidate is present, the target candidate present area is set in the vicinity thereof. Enlarge and output the target candidate area data, take the logical product of the binarized image generated above and the target candidate area data, label the image obtained by the logical product, and calculate the center of gravity of the labeled target candidate Then, the target is determined from all the target candidates for which the center of gravity has been calculated, and the target is detected from the input image.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and an apparatus for detecting a target existing in an input image, and more particularly to a method for detecting a target such as a ship from an image acquired by a synthetic aperture radar mounted on a satellite. The present invention relates to a target detection method and a target detection device.

[0002]

2. Description of the Related Art There are various methods for detecting a target in the related art. Here, two target detection methods will be described. FIG. 29 is a diagram showing an overall configuration of a target detection device 1 that executes a conventional target detection method 1. In FIG. 29, a target detection device 1 includes a target constituent pixel detection unit 10a for detecting target constituent pixels from an input original image, and a target presence / absence determination unit 30 for performing target presence / absence determination from the original image.
a, which calculates the logical product of the output of the target constituent pixel detection unit 10a and the output of the target presence / absence determination unit 30a
It is composed of a D unit 50a.

FIG. 30 is a diagram showing a processing flowchart of the target detecting device 1 which executes the conventional target detecting method 1 shown in FIG. In FIG. 30, the processing of target detection method 1 includes target constituent pixel detection processing S10a for detecting target constituent pixels from an input image (indicated by an original image O), target presence / absence determination processing S30a for performing target presence / absence determination from an input image, and target processing. Output of constituent pixel detection processing S10a (binary image, indicated by B)
AND processing S50a for calculating the logical product of the output (target area data, indicated by A) of the target presence / absence determination processing S30a
Consists of The output of the AND processing S50a is also a binarized image (indicated by B '), like the output of the target constituent pixel detection processing S10a.

FIG. 31 shows a detailed process of the target constituent pixel detection process S10a in FIG. The target constituent pixel detection unit 10a in FIG. 29 performs the original image reading process S101 in FIG.
In a, the original image O shown in FIG. 32 is read,
In the target constituent pixel determination region / background region generation processing S102a, each small region including the target constituent pixel determination region O1 and the background region O2 surrounding the target constituent pixel determination region O1 in the original image O is generated.
In the background area statistical calculation processing S103a, the background area O
2 (here, for example, the average value μ and the standard deviation σ) are calculated, and the binarization determination value setting processing S1 is performed.
At 04a, a binarization determination value D is set.

In the background area statistical calculation processing S103a, for example, an average value μ _B and a standard deviation σ _B of the background area are calculated.
In the binarization determination value setting processing S104a, a binarization determination value D given by the following equation is calculated from these values. D = μ _B + x × σ _B Here, x: constant In the target constituent pixel determination processing S105a, the luminance value X (i, j) of each pixel p (i, j) in the target constituent pixel determination area O1 shown in FIG. j) is greater than or equal to a binarization determination value D. When X (i, j) ≧ D, the pixel is determined to be a pixel constituting a target, and X (i, j) ) <D, the pixel is determined to be a pixel that does not constitute the target. Here, i is the x coordinate of the pixel in the original image O, and y is the y coordinate of the pixel in the original image O. In step S106a, it is determined whether the region is the final region. If not, the target component pixel determination region O1 'which is the next region and the background region O2 surrounding the target region as shown in FIG. 'Repeated. Then, the processing is sequentially executed, and if the area is the final area, the target constituent pixel detection processing S10a ends.

FIG. 33 shows a detailed process of the target presence / absence determination process S30a of FIG. In FIG. 33, in an original image reading process S301a, an original image O is read as shown in FIG. In the target determination area / background area generation processing S302a, a target determination area Q1 and a surrounding background area Q2 in the original image O are generated. In the target determination area / background area statistical value calculation processing S303a, the average value μ _T and variance σ _T ² of the target determination area Q1 and the average value μ _B and variance σ of the background area Q2 are obtained.
To calculate the _B ^2. In the target determination value setting process S305a, for example, the contrast value C is calculated based on the following equation.
Is calculated. In _{C = {(μ T -μ B} ) 2 + σ T 2} / σ B 2 target determining process S305a, by comparing the above contrast value C and the predetermined threshold value Tc, to detect the presence of target. That is, it is determined that C ≧ Tc: the target exists, and C <Tc: the target does not exist. In step S306a, it is determined whether or not the area is the last area.
As shown in the figure, a target determination area Q1 ′ which is the next area
And the background area Q2 'surrounding the area. Then, this process is sequentially performed, and if it is the last area, the target presence / absence determination processing S30a ends.

FIG. 35 is a diagram showing the AND processing S50a. In the AND processing S50a of FIG. 35, the target constituent pixel binary image B shown in FIG. 35A obtained in the target constituent pixel detection processing S10a and the target presence / absence determination processing S30a
From the target area data A shown in FIG.
All the pixels that are the target constituent pixels and are determined to be the target existence area in the target area data are determined to be the pixels that constitute the target. From the result of the determination, a target constituent pixel binary image B ′ as shown in FIG. 35 (c) is obtained.

FIG. 36 shows K. Eldhuset "Automatic ship an
d ship wake detection system for Spaceborne SAR ima
ges in Coastral Regions "(IEEE Transactions on Geo
science and Remote Sensing, Vol. 34, No. 4, July 1996)
3 is a block diagram showing a configuration of a target detection method 2 described in FIG. FIG. 37 is a flowchart showing a process of the conventional target detection method 2.

FIG. 36 is a block diagram showing the overall configuration of a conventional target detection device 2. As shown in FIG. In FIG. 36, the target detection device 2 includes a target candidate constituent pixel detection unit 10a, a labeling unit 60a, and a target candidate centroid calculation unit 70.
a, a target determination unit 80a.

FIG. 37 is a flowchart showing the processing of the conventional target detecting device 2. In FIG. 37, the target detection device detects pixels constituting the target candidate in the input original image in the target candidate constituent pixel detection processing S10a, and detects the pixels constituting the detected target candidate in the labeling processing S60a. Are subjected to labeling, and the target candidates are recognized as target candidates. In a target candidate center of gravity calculation process S70a, the center of gravity of each labeled target candidate is calculated.
In the target determination processing S80a, a target determination area is set for each target candidate for which the center of gravity has been calculated, and it is determined whether or not the target is a target, and the target is detected.

A target candidate constituent pixel detecting unit 1 shown in FIG.
0a, as in FIG. 31, the original image reading process S101a and subsequent processes (S102a to S10a) are performed.
6a) is performed, and a binarized image B as shown in FIG. 38 is output. In FIG. 38, a pixel P1 detected as a target candidate constituent pixel in the binary image B has, for example, “1”.
Is displayed. As shown in FIG. 39, the labeling unit 60a performs a labeling process S60a for detecting a connected component of each pixel and attaching a label for recognizing each target candidate, and outputs a labeling image L. You. After the labeling process S60a, each target candidate is recognized as "100", "101", "109" in the labeling image L. Thereafter, the target candidate center of gravity calculation unit 70a calculates the center of gravity of the target candidate in order to set the target determination area of each target candidate. A target determination area is provided around the calculated position of the center of gravity, and a contrast value between the target candidate and the background in the determination area is calculated. When the contrast value is equal to or greater than a predetermined threshold, the target for determining the target candidate as the target is determined. Determination process S8
0a is performed and the target is detected.

[0012]

Of the two processing methods described above, the target detection method 1 is the target constituent pixel detection processing S10a when it is determined that the target exists in the target determination area A1 as shown in FIG. Among the pixels detected as pixels constituting the target in the background U2 other than the target U1
Is detected as a target when erroneously is detected as a target constituent pixel. This is because, for example, as shown in FIG. 42, when the contrast between the pixel constituting the target and the background is not good, that is, the background pixel having a luminance distribution like W1 and the target constituent pixel having a luminance distribution like W2 When the frequency distributions of the above overlap, when the threshold value of the determination is set to T1, the background pixel is erroneously detected as the target constituent pixel, and the image is simply divided to set the target presence / absence determination area. This is because the presence or absence of the target is determined without recognition. Further, as shown in FIG. 41A, when the target extends over a plurality of target determination areas A1, A3, and A4, for example, in the target presence / absence determination processing S30a, for example, A1 and A4 are targeted. If it is determined that the region does not exist, the target has a shape like the target U1 ′ in FIG. 41B by the AND processing S50a.
There is a disadvantage that the shape of the target is not well detected.

In the target detection method 2, a target determination area is set around the center of gravity of each target candidate in order to solve the problem of the target detection method 1, and the target determination is performed using pixels constituting each target candidate and background pixels. Is what you do. However, as described above, when the frequency distribution of the background pixel having the luminance distribution like W1 as shown in FIG. 42 and the frequency distribution of the pixels constituting the target having the luminance distribution like W2 overlap, the target detection method 2 also As in the case of the target detection method 1, the background pixel is erroneously detected as the target constituent pixel. Therefore, in this case, FIG.
As shown in (1), when all the target candidates U1 and U2 are labeled with respect to the result of the target constituent image detection processing S10a and the target determination processing is performed, the processing amount becomes enormous, and there is a disadvantage that the processing time is required.

SUMMARY OF THE INVENTION The present invention has been made in view of the above problems, and when a target is detected from an image, the shape of the target is preferably detected particularly from an image in which the contrast between the pixels constituting the target and the background is not good. And a target detection method and a target detection device capable of detecting a target at high speed.

[0015]

In order to solve the above-mentioned problems, a target detecting method according to a first aspect of the present invention detects a pixel constituting a candidate for a target from an input original image, and performs a method for detecting a target.
Generate a binarized image, determine the presence or absence of the target candidate from the statistical value of the target candidate constituent pixels and the statistical value of the background constituent pixels using the input original image and the binarized image generated above, If the result of the presence / absence determination indicates that the target candidate exists, if the target candidate exists, the target candidate existing area is enlarged to the periphery thereof, and the target candidate area data is output. The logical product of the region data is taken, the image obtained by the logical product is labeled, each target candidate is recognized, the center of gravity of the labeled target candidate is calculated, and the center of gravity is calculated for all the target candidates recognized above. Is set, a target candidate statistic and a background statistic are calculated in each target determination area, and the target is determined.

In the target detecting method according to a second aspect of the present invention, a small target candidate removing process is added between the labeling process and the target candidate center-of-gravity calculating process, so that the number of pixels of the target candidate is smaller than a predetermined number. Is configured to be removed.

In the target detection method according to the third invention of the present invention, dilation processing is added between the labeling processing and the target candidate center-of-gravity calculation processing, and pixels surrounded by the same label number and having no label number are added. It is configured to have the same label number.

In the target detection method according to a fourth aspect of the present invention, a small target candidate removal process and an expansion process are added between the labeling process and the target candidate centroid calculation process, and the number of pixels of the target candidate is less than or equal to a predetermined number. The small target candidate is removed, and the pixels surrounded by the same label number and having no label number are configured to have the same label number.

A target detecting apparatus according to a fifth aspect of the present invention includes target candidate constituent pixel detecting means for detecting a pixel constituting a target candidate from an input original image and generating a binary image. Target candidate presence / absence determining means for determining the presence / absence of a target candidate from the statistical values of the target candidate constituent pixels and the statistical values of the background constituent pixels using the original image and the binarized image generated above, In response to the result, if the target candidate exists, if the target candidate exists, the target candidate existing area is expanded to its surroundings, and target candidate existing area expanding means for outputting target candidate area data; AND processing means for taking a logical product of the image and the target candidate area data, labeling means for labeling the image obtained by the logical product to recognize each target candidate, and a center of gravity for all the target candidates recognized above Sets a target determination region centered calculates a target candidate statistics and background statistic in each target determination region, configured with a target determining means for determining a target.

A target detecting apparatus according to a sixth aspect of the present invention comprises a labeling means and a target candidate center of gravity calculating means.
It is configured to add a small target candidate removing unit that removes small target candidates whose number of pixels of the target candidate is equal to or less than a predetermined number.

According to a seventh aspect of the present invention, there is provided a target detecting apparatus, comprising:
It is configured to add expansion processing means that makes pixels surrounded by the same label number and not given a label number the same label number.

According to an eighth aspect of the present invention, there is provided a target detecting apparatus, comprising: a labeling means and a target candidate center-of-gravity calculating means;
Small target candidate removing means for removing small target candidates in which the number of pixels of the target candidate is equal to or smaller than a predetermined number, and expansion processing means for surrounding pixels having the same label number and having no label number with the same label number. It is configured to add.

[0023]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiment 1 FIG. 1 is a block diagram showing the overall configuration of the target detection device according to the first embodiment of the present invention. In FIG. 1, a target detection device according to a first embodiment of the present invention includes a target candidate constituent pixel detection unit 10, a storage unit 20, a target candidate presence / absence determination unit 30, a target candidate existence region enlargement unit 40, and an AND processing unit 5.
0, a labeling unit 60, a target candidate center of gravity calculation unit 70, and a target determination unit 80.

FIG. 2 is a flowchart showing the processing of the target detecting device according to the first embodiment of the present invention. In FIG. 2, the target detecting device detects pixels constituting the target candidate in the input original image in a target candidate constituent pixel detecting process S10 of the input original image (O), and executes a target candidate presence / absence determining process. In S30, the input original image is divided into small regions, and the presence / absence of a target candidate is determined. In the target candidate presence region enlarging process S40, if the target candidate is present, the target candidate presence region is set to the target candidate presence region. The target candidate area data is enlarged and output to the periphery, and in the AND processing S50, the logical product of the binarized image and the target candidate area data is obtained.
In the labeling process S60, the pixels constituting the detected target candidates are labeled, and in the target candidate center-of-gravity calculation process S70, the center of gravity of each labeled target candidate is calculated. In the target determination process S80, the target A determination area is set, and it is determined whether the target is a target or not and the target is detected.

Hereinafter, the configurations shown in FIG. 1 and the processing performed by each of them will be described in detail. Target Candidate Constituent Pixel Detection Unit 10: FIG. 3 is a flowchart illustrating a target candidate constituent pixel detection process S10 according to the first embodiment of the present invention. The target candidate constituent pixel detection unit 10 is a unit that performs a target candidate constituent pixel detection process. As shown in FIG. 3, an original image reading process S101, a target candidate constituent pixel determination region / background region generation process S102, a background region Statistical value calculation processing S103, 2
It is a unit that performs a binarization determination value setting process S104, a target candidate constituent pixel determination process S105, and a repetition process S106. The data processed in the target candidate constituent pixel detection unit 10 is recorded in the shared storage unit 20.

Target candidate constituent pixel detection process S10:
The input original image O is read by an original image reading process S101. In the target candidate constituent pixel determination area / background area generation processing S102, the read original image O is subjected to the target candidate constituent pixel determination area O1 as shown in FIG.
And a background area O2 surrounding it. In the background region statistical value calculation processing S103, statistical values (average value μ _B , standard deviation σ _B, etc.) are calculated from all the pixels in the generated background region O2. Binarization determination value setting processing S10
4, the binarization determination value D of the background area is calculated by the following equation.
Is calculated. D = μ _B + K × σ _{B In the} target candidate constituent pixel determination processing S105, the luminance value X (i, j) of the pixel p (i, j) in the target candidate constituent pixel determination area O1.
And the binarization determination value D obtained in the background area O2 are sequentially compared to determine a target candidate constituent pixel by the following logic, for example.

That is, when X (i, j) ≧ μ _B + K × σ _B , it is determined that the pixel is a target candidate constituent pixel. When X (i, j) <μ _B + K × σ _B , the target candidate constituent pixel is determined. It is determined that the pixel is not a pixel. here,
i is the x coordinate in the original image O, y is the y coordinate in the original image O, K
Is a constant.

In the intensity image of the synthetic aperture radar, for example, the following equation can be used as the binarization determination value D. D = μ _B + K × √N where N is the number of looks.

Similarly to the above, in the target candidate constituent pixel determination processing S105, the pixel p (i, j) in the target candidate constituent pixel determination area O1 and the binarization determination value D obtained in the background area O2 are determined. It is sequentially compared by the following logic to determine whether or not the pixel is a target candidate constituent pixel.

That is, when X (i, j) ′ ≧ μ _B + K × √N, it is determined that the pixel is a target candidate constituent pixel. When X (i, j) ′ <μ _B + K × √N, the target pixel is determined. It is determined that the pixel is not a candidate constituent pixel. here,
X (i, j) '= X (i, j) + X (i + 1, j) + X (i, j + 1) + X (i + 1,
j + 1) −4 μ _B , X (i, j) is the luminance value of p (i, j).

Thereafter, target candidate constituent pixel detection processing S10
Generates a small area such that the next judgment areas O1 ′ and O2 ′ of the judgment areas O1 and O2 do not overlap the original image O, and sets a target candidate for all pixels in O1 ′. It is sequentially determined whether or not the pixel is a constituent pixel. In this way, the present process is performed on all the pixels in the original image O, and “1” is assigned to the pixels constituting the target candidate, for example, “0” is assigned to the other pixels.
And a target candidate constituent pixel binary image B as shown in FIG.
Is stored in the storage unit 20.

Target candidate presence / absence determination unit 30: FIG.
FIG. 4 is a diagram illustrating a flowchart of a target candidate presence / absence determination process S30 according to the first embodiment of the present invention. Next, as shown in FIG. 5, the target candidate presence / absence determination unit 30 performs an original image reading process S301, a target candidate constituent pixel binary image reading process S302, an original image and a binary image small area dividing process S
303, a target candidate / background statistical value calculation process S304, a target candidate determination process S305, and a repetition process S306. Note that the image reading order of step S301 and step S302 may be reversed.

Target candidate presence / absence determination processing S30: In the original image reading processing S301, the input original image O is read, and the target candidate constituent pixel binary image reading processing S3 is performed.
At 02, the binarized image B stored in the storage unit 20 in the target candidate constituent pixel detection processing S10 is read. In the original image and binarized image small area division processing S303, the original image and the binarized image are divided into small areas.
FIG. 6 shows the binarized image B as small areas A1, A2, A3,.
The figure shows a state of division into n.

In the target candidate / background statistical value calculation processing S304, the target candidate constituent pixel binary image reading processing S3
02, the binarized image B read from the storage unit 20
From FIG. 6, for example, the position of the target candidate constituent pixel U <b> 1 which is the portion “1” in FIG. 6 in the divided small area A <b> 1 is confirmed. Further, in the small area A1 of the original image O (FIG. 7) corresponding to the small area A1 divided by the target candidate constituent pixel binary image B (FIG. 6), the target candidate constituent pixel 2
From the luminance values of the target candidate constituent pixels (M1, M2,..., M8) on the original image O corresponding to the position of U1 on the binarized image B, the statistical values (eg, Mmax, Msum) of the target candidate constituent pixels and the background From the luminance values of the pixels (H1, H2,..., H17), statistical values (eg, Hmax, Hmin) of the background constituent pixels are obtained.

In the target candidate determination processing S305, the statistical values of the target candidate constituent pixels and the background constituent pixels obtained as described above are substituted into the following equation, and are compared with a predetermined threshold value T1. It is determined whether or not the target candidate exists in the small area thus set. Mmax / Hmax ≧ T1: Target candidate exists. Mmax / Hmax <T1: No target candidate exists. Or Mmax / (Hmax−Hmin) ≧ T1: Target candidate exists. Mmax / (Hmax−Hmin) <T1: No target candidate exists. Or Msum / Hmax ≧ T1: target candidate exists. Msum / Hmax <T1: No target candidate exists. Here, Mmax: the maximum luminance value of the target candidate constituent pixels, Msum: the total luminance value of the target candidate constituent pixels, Hmax: the maximum luminance value of the background constituent pixels, Hmin: the luminance value of the background constituent pixels Minimum value, T1: threshold value. According to the determination obtained in this way, the region where the target candidate exists is located in the region exceeding the threshold T1.
The area which does not exceed the threshold T1 is set to 1 and the area where the target candidate does not exist is written to the storage unit 20 as the target candidate area data as the area "0".
5 sequentially determines whether or not the target candidate exists in all of the divided areas A1, A2, A3,..., An. The determination result is stored in the storage unit 2 as described above.
0 is written as target candidate area data.

Target candidate existing area expanding unit 40: FIG.
FIG. 9 is a diagram showing a flowchart of a target candidate existing region expanding process S40 according to the first embodiment of the present invention. As shown in FIG. 8, the target candidate existing area expanding unit 40 includes a target candidate area data reading process S401, a target candidate existing determining process S402, an 8-neighbor expanding process S403, and a repetitive process S
This unit performs 404. The target candidate existence area expanding unit 40 is generated by the target candidate presence / absence unit 30;
The target candidate area data (A) stored in the storage unit 20 is read, and in the case of a target candidate area, a target candidate area in the storage unit 20 in which eight neighboring areas adjacent to the target candidate area are set as target areas. Rewrite the data (A).

Target candidate existence area expansion processing S40: FIG.
FIG. 8 is a diagram for explaining a target candidate existence region expanding process in the target detection method according to the first embodiment of the present invention. In target candidate area data reading processing S401 in FIG. 8, binary candidate target area data is read from the storage unit 20. In the target candidate presence determination processing S402, a region where the target candidate exists is extracted from the target candidate region data. An 8-neighbor enlargement process S403 is performed on the extracted target candidate existence region. As shown in FIG. 9, all the pixels constituting the target candidate are not always in the same area, and may extend over a plurality of areas. At this time, for example, assuming that the region A5 is determined to be a region where the target candidate exists, and the other regions A1, A2 and A4 are determined to be regions where the target candidate does not exist, the shape of the target cannot be detected well. Will be. In order to avoid this, in an area where the size of the target is considered in advance,
It is determined that the target exists near the target candidate. Thereby, the target shape can be detected well.

Next, the 8-neighbor enlargement process S403 will be described in detail. In the 8-neighbor enlargement process S403, FIG.
When it is determined that the target candidate exists in the area A1 in 9,
Eight neighbors adjacent to the area A1 (here, three neighbors A
2, A4, A5) is used as the target candidate existence area, and the enlargement processing is performed. Similarly, when it is determined that the target candidate exists in the area A2, eight neighbors adjacent to the area A2 (here, 5
The enlargement process is performed using the two neighbors A1, A3, A4, A5, and A6) as target candidate existence areas. Further, for example, when it is determined that the target candidate exists in the area A5, the area A5
To eight neighbors (here, eight neighbors A1, A2,
A3, A4, A6, A7, A8, A9) are used as the target candidate existence areas, and the enlargement processing is performed. When this enlargement processing is performed, the result rewrites the target candidate area data stored in the storage unit 20. Thus, the areas (A1, A2, A3,..., An) on all the target candidate area data
If it is determined that there is a target candidate for, the 8-neighbor enlargement process S403 is performed, and the target candidate area data is rewritten.

AND processing unit 50: FIG. 10 is a diagram showing a flowchart of the AND processing according to the first embodiment of the present invention. In FIG. 10, the AND processing unit 50
The target candidate constituent pixel binary image reading processing S50
1. This unit performs target candidate area data reading processing S502, logical product operation processing S503, and repetition processing S504. The AND processing unit 50 performs an AND operation on the binarized image B and the target candidate area data A.
1 ", that is, pixels that are target candidate constituent pixels and belong to the target candidate existence area are determined as target candidate constituent pixels, and the target candidate constituent pixel binary image B in the recording unit 20 is rewritten, as shown in FIG. Thus, the new binarized image B 'is stored in the storage unit 20. Step S
The image reading order of step 501 and step S502 may be reversed.

AND Process S50: FIG. 11 is a diagram for explaining the AND process according to the first embodiment of the present invention. In the target candidate constituent pixel binarized image reading processing S501, a target candidate constituent pixel detection processing S1 as shown in FIG.
The binary image B obtained at 0 is read. FIG.
In (a), pixels determined to be target candidate constituent pixels are surrounded by a thick line frame. On the other hand, in the target candidate area data read processing S502, the target candidate area data A obtained in the target candidate presence / absence determination processing S30 and the target candidate existence area enlargement processing S40 as shown in FIG.
Is read. In FIG. 11 (b), the regions where it is determined that the target candidates are present are similarly indicated by bold-line frames (A1, A1).
2, A3, A4, A5, A6, A7, A8, A9). In a logical product operation process S503, a logical product operation of the binary image B and the target candidate area data A is performed. The logical product operation is repeated by the repetition process S504 until the process is completed for all the pixels. as a result,
A target candidate constituent pixel binary image B ′ as shown in FIG. 11C is obtained. This result is stored in the storage unit 20 as a new binary image B ′.

Labeling Unit 60: FIG. 12 is a diagram showing a flowchart of the labeling process according to the first embodiment of the present invention. The labeling unit 60 is a unit that performs a target candidate constituent pixel binary image reading process S601 and a labeling execution process S602. The labeling unit 60 reads the binarized image B ′ of the target candidate constituent pixels, which is a result obtained by the AND processing unit, detects a connected portion of the target candidate constituent pixels, and determines which target candidate each pixel belongs to. judge. Thereby, the connected component is recognized as each target candidate. Processing result is labeling image L
Is written to the storage unit 20.

Labeling process S60: FIG. 13 is a diagram for explaining the labeling process according to the first embodiment of the present invention.
In the target candidate constituent pixel binary image reading processing S601, the target candidate constituent pixel binary image B ′ stored in the storage unit 20 in the above-described AND processing S50 is read. In the labeling execution process S602, a labeling process is performed on the read target candidate constituent pixel binary image B ′. The labeling process is performed on the target candidate constituent pixels (represented by “1”) in the binary image B ′ as shown in FIG. 13A, and the connected pixels as shown in FIG. 13B. The same label, for example, 100, 101, 1
02, etc., to create a labeling image L. As a result, connected pixels can be recognized as the same target candidate.

Target candidate center of gravity calculation unit 70: FIG.
FIG. 7 is a diagram showing a flowchart of a target candidate barycenter calculation process according to the first embodiment of the present invention. The target candidate center of gravity calculation unit 70 is a unit that performs a labeling image reading process S701, a center of gravity calculation process S702, and a repetition process S703. The target candidate center-of-gravity calculation unit 70 calculates the center of gravity of each target candidate in order to set the target determination region of each target candidate using the labeling image L obtained by the labeling unit 60.

Target candidate center of gravity calculation process S70: FIG.
FIG. 5 is a diagram illustrating a target candidate center of gravity calculation process in the target detection method according to the first embodiment of the present invention. A range U1 surrounded by a thick line in FIG. 15 is a target candidate for which the center of gravity G is calculated. In a labeling image reading process S701, target candidates with the same label number are read.
In the center-of-gravity calculation processing S702, the positions X and Y of the centers of gravity in the x and y directions of the target candidates assigned the same label number
Are calculated by the following equations, and the calculated center of gravity G of each target candidate is written to the storage unit 20 as target candidate center of gravity data.

[0045]

(Equation 1)

Here, (xn, yn): the position of the pixel forming the target candidate X: the position of the center of gravity of the target candidate in the x direction Y: the position of the center of gravity of the target candidate in the y direction T: the pixel forming the target candidate The center-of-gravity calculation processing S7
02, the centroids of all the target candidates are calculated.

Target determination unit 80: FIG. 16 is a diagram showing a flowchart of the target determination processing according to the first embodiment of the present invention. The target determination unit 80 includes a labeling image reading process S801, an original image reading process S802, a target candidate barycenter data reading process S803, a target determination region setting process S804, and a target candidate / background statistical value calculation process S8.
05, a unit for performing the determination processing S806 and the repetition processing S807. The target determination unit 80 determines whether or not a target target candidate is a target based on the labeling image, the original image, and the target candidate barycenter data. Note that the image reading order of step S801 and step S802 may be reversed.

Target determination processing S80: The labeling image L output from the labeling unit 60 is read in the labeling image reading processing S801, and the original image O is read in the original image reading processing S802. On the other hand, in the target candidate center-of-gravity data reading processing S803, each target candidate center-of-gravity data which is the result calculated by the above-described target candidate center-of-gravity calculating unit 70 is sequentially read. In the target determination area setting processing S804, a target determination area is set, and a target is determined from the statistical value of the target candidate and the statistical value of the background to be determined in the area.

In the target candidate / background statistical value calculation processing S805, the target judgment of the original image O and the labeling image L centered on the center of gravity of the target candidate whose size is set in advance so as to include all the pixels constituting the target candidate In the region, the statistical value of each target candidate to be determined and the statistical value of the background are calculated. In the target determination processing S806, it is determined whether or not the target target candidate is a target by using these statistical values.

FIGS. 17 and 18 show Embodiment 1 of the present invention.
FIG. 8 is a diagram illustrating a target determination process S80 in the target detection method of FIG. Using these, the operation of the target determination process S80 will be described in further detail. In the labeling image L of FIG. 17 read by the labeling image reading process S801, target candidates (U1, U2, U
3, U4,..., Un) will be described. This judgment is made for all target candidates U
1, U2, U3, U4,..., Un
Here, only the target candidate U4 will be described as an example. In the target determination area setting processing S804, the target determination area A4 is set around the center of gravity G of the target candidate U4 using the target candidate gravity center data read in the target candidate gravity center data reading processing S803. In the target candidate / background statistical value calculation processing S805, the labeling image L
In the original image O corresponding to the target candidate U4 labeled “102” (M5, M6, M7, M8, M
9, M10, M11) and the background pixel (H
1, H2, H3,..., H17), the same processing as the target candidate / background statistical value calculation processing S304 in the target candidate presence / absence determination unit 30 described above is performed.

That is, in the target candidate / background statistical value calculation processing S805, the statistical values (eg, Mmax, Msum) and the background pixel of the target candidate are calculated from the luminance values of the target candidate constituent pixels on the original image O in the target determination area A4. Of the background (for example, Hmax, Hmin) are obtained from the luminance value of.

In the target determination process S805, the target candidate and the statistical value of the background obtained as described above are substituted into the following equation, and are compared with a predetermined threshold T2 to determine whether the target is a target. Is done. Mmax / Hmax ≧ T2: target. Mmax / Hmax <T2: Not target. Or Mmax / (Hmax−Hmin) ≧ T2: target. Mmax / (Hmax−Hmin) <T2: Not target. Or Msum / Hmax ≧ T2: target. Msum / Hmax <T2: Not target. Here, Mmax: the maximum luminance value of the target candidate pixel, Msum: the total luminance value of the target candidate pixel, Hmax: the maximum luminance value of the background pixel, Hmin: the luminance value of the background pixel Minimum value, T2: threshold value. According to the judgment obtained in this way, judgment processing S
At 806, the target candidate U4 exceeding the threshold T2 is determined to be a target, and the target candidate U4 not exceeding the threshold T2 is determined not to be a target. At this time, the pixel constituting the target of the area exceeding the threshold value T2 is set to "1" and all the other pixels are set to "0", and the result is written into a newly generated binary image B ''. Thereafter, the same process is performed on all the target candidates, the target determination process S80 is completed, and the detection of the target is completed.

Embodiment 2 FIG. 19 is a block diagram illustrating a configuration of the target detection device according to the second embodiment of the present invention.
The target detecting apparatus according to the second embodiment is configured by adding a small target candidate removing unit 90 to the configuration of the first embodiment shown in FIG. In the second embodiment, a small target candidate removal process S90 can be added after the labeling process S60 in FIG. 1 to remove small target candidates. The configuration other than the small target candidate removal unit 90 is the same as that of FIG.
Description is omitted.

FIG. 20 is a flowchart showing a process of the target detecting method according to the second embodiment of the present invention. The configuration other than the small target candidate removal processing S90 is the same as that in FIG.

Small target candidate removing unit 90: FIG.
FIG. 14 is a diagram illustrating a flowchart of a small target candidate removal process according to the second embodiment of the present invention. Small target candidate removal unit 90
Is a unit that performs a labeling image reading process S901, a target candidate constituent pixel counting process S902, a pixel number comparing process S903, a removing process S904, and a repeating process S905. The small target candidate removing unit 90 removes a target candidate in which the number of pixels of the target candidate with the same label on the labeling image L is smaller than a preset target candidate pixel number. Here, the number of target candidate pixels to be set can be set according to the pixel size and the target size to be targeted. For example, a threshold value of the number of target candidate pixels can be set so that a ship of 20 m or more is detected and a ship of less than 20 m is not detected.

Small Target Candidate Removal Processing S90: FIG. 22 is a diagram for explaining the small target candidate removal processing according to the second embodiment of the present invention. In the labeling image reading process S901, as shown in FIG.
Is read out. In the target candidate constituent pixel counting process S902, the target candidates (U1, U2, U3, U4,.
Un) is counted. Pixel number comparison processing S
At 903, target candidates (U1, U2, U3, U4,
, Un) is a predetermined number (for example, 2
Below) compare if: In the removal processing S904,
A target candidate having a predetermined number of pixels or less is removed. By repeating this process for the target candidates in the image in the repetition process S905, a labeling image L 'from which the small target candidates are removed as shown in FIG. 22B is generated.

As described above, in the second embodiment, by introducing the small target candidate removal processing S90, it is possible to remove a target candidate that is unlikely to become a target because the shape of the target candidate is small. become. Therefore, the subsequent target candidate centroid calculation processing S70 for the removed target candidate,
There is no need to perform the target determination processing S80, and the processing time can be reduced.

Embodiment 3 FIG. FIG. 23 is a block diagram showing a configuration of the target detection device according to the third embodiment of the present invention.
The target detection device of the third embodiment is configured by adding an expansion processing unit 100 to the configuration of the first embodiment shown in FIG. By adding the expansion processing unit 100, the target shape can be detected better. The configuration other than the expansion processing unit 100 is the same as that of FIG.

FIG. 24 is a flowchart showing a process of the target detecting method according to the third embodiment of the present invention. The configuration other than the expansion processing S100 is the same as that of FIG.

Expansion processing unit 100: FIG. 25 is a diagram showing a flowchart of expansion processing according to the third embodiment of the present invention. The expansion processing unit 100 performs labeling image reading processing S1001 and expansion processing area presence / absence determination processing S
1002, an expansion execution process S1003, a repetition process S1004 in the target candidate, and a repetition process S1005 for all the target candidates. The expansion processing unit 100 determines whether or not there is an area to be subjected to the expansion processing from the labeling image L, and performs the expansion processing on the area.

Expansion processing S100: FIG. 26 is a view for explaining expansion processing S100 according to the third embodiment of the present invention. In the third embodiment, a target shape can be detected more favorably, and an expansion process S100 is added between the labeling S60 and the target candidate center-of-gravity calculation process S70 in FIG. In the labeling image reading process S1001, as shown in FIG.
The labeling image L generated at 60 is read. The presence / absence of an area to be expanded is determined from the labeling image L, and the expansion processing is performed on the area. In the presence / absence determination processing S1002 of the expansion processing area, an unlabeled pixel area P2 is detected in the labeled target candidate U1. In the dilation execution processing S1003, the pixel area P2 surrounded by four pixels in the target candidate U1 in the read labeling image L is determined as a pixel belonging to the target candidate, and is labeled, thereby performing the target candidate U1.
Pixel. FIG. 26 (b) shows the labeling image L 'obtained after the expansion processing thus obtained.

As described above, in the third embodiment, by introducing the expansion processing S100, the pixel area P2 surrounded by four pixels in the target candidate U1 is determined as a pixel belonging to the target candidate. , The shape of the target candidate can be detected better. The repetition process S10
In step 04, the same processing is performed for all the same target candidate areas, and in the repetition processing S1005, for all labeled target candidates, the shapes of all target candidates can be detected satisfactorily.

Embodiment 4 FIG. 27 is a block diagram showing a configuration of the target detection device according to the fourth embodiment of the present invention.
The target detecting apparatus according to the fourth embodiment is configured by adding a small target candidate removing unit 90 and an expansion processing unit 100 to the configuration of the first embodiment shown in FIG. Configurations other than the small target candidate removing unit 90 and the expansion processing unit 100 are the same as those in FIG.

FIG. 28 is a flowchart showing a process of the target detecting method according to the fourth embodiment of the present invention. In the fourth embodiment, after the labeling processing S60 in FIG. 1, the small target candidate removing unit 90 and the expansion processing S10
0 can be added, thereby removing small target candidates and better detecting the shape of the target.

Small target candidate removing unit 90 and dilation processing unit 100: Small target candidate removing unit 90 and dilation processing unit 100 in the fourth embodiment are respectively small target candidate removing unit 90 in the second embodiment.
And since it is the same as that of the expansion processing unit 100 in Embodiment 3, detailed description is omitted.

Small target candidate removal processing S90 and expansion processing S100: Small target candidate removal processing S90 and expansion processing S
100 denotes a small target candidate removal process S90 in the second embodiment and an expansion process S1 in the third embodiment, respectively.
Since it is the same as 00, detailed description is omitted.

In the fourth embodiment, the merits of the second and third embodiments can be simultaneously obtained by introducing both the small target candidate removal processing S90 and the expansion processing S100. That is, the small target candidates are removed, the target determination processing is reduced and the speed is increased, and the shape of the target candidate image can be detected more favorably.

[0068]

As described above, the first and fifth embodiments of the present invention are described.
According to the present invention, the present invention detects a pixel constituting a target candidate from an input original image to generate a binarized image,
Using the input original image and the binarized image generated above, the presence / absence of the target candidate is determined from the statistical values of the target candidate constituent pixels and the statistical values of the background constituent pixels. If the target candidate area is present, the target candidate existing area is enlarged to its surroundings, the target candidate area data is output, and the logical product of the binarized image generated above and the target candidate area data is obtained. Labeling the image obtained by the logical product, recognizing each target candidate, calculating the center of gravity of the labeled target candidate, and determining the center of gravity with respect to the center of gravity for all the target candidates recognized above. And set
Since the target candidate statistic and the background statistic are calculated and the target is determined in each target determination area, the pixels included in the target candidate existence area and detected as the target candidate constituent pixels are referred to as subsequent target pixels. As a target candidate constituent pixel for processing,
In an image in which the contrast between the background and the pixels constituting the target is not good because the target candidates can be limited in the image and labeling of unnecessary target candidates, target candidate center-of-gravity calculation processing, and target determination processing need not be performed, the conventional target configuration is used. Compared with the detection method 2, the processing amount can be reduced and the processing time can be reduced.

According to the second and sixth aspects of the present invention, according to the present invention, a small target candidate removing process is added between the labeling process and the target candidate centroid calculating process, and the number of pixels is equal to or less than a predetermined number. Is configured to remove small candidate candidates for
If the size of the target, in particular, the minimum size of the target is known in advance, by removing the target candidates with the number of pixels smaller than the size, it is possible to obtain a result at higher speed.

According to the third and seventh aspects of the present invention, the present invention adds an expansion process between the labeling process and the target candidate barycenter calculation process, and is surrounded by the same label number. Are configured to have the same label number as the pixels without the mark, so that among the pixels constituting the target, the internal pixels that are not detected as the target candidate constituent pixels are newly set as the target candidate constituent pixels. Can be better detected.

Further, according to the fourth and eighth aspects of the present invention, the present invention adds a small target candidate removal process and an expansion process between the labeling process and the target candidate center of gravity calculation process to reduce the number of pixels. Since it is configured such that pixels smaller than a predetermined number of small target candidates are removed and pixels that are surrounded by the same label number and have no label number are given the same label number, the size of the target, especially the minimum size of the target It is possible to obtain a result at high speed by removing target candidates with a number of pixels less than the size if the value is known in advance, and re-examine internal pixels that are not detected as target candidate constituent pixels. By using the target candidate constituent pixels, the target shape can be detected more favorably.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a target detection device according to a first embodiment of the present invention.

FIG. 2 is a flowchart illustrating a process of a target detection method according to the first embodiment of the present invention.

FIG. 3 is a flowchart illustrating a target candidate constituent pixel detection process according to the first embodiment of the present invention;

FIG. 4 is a diagram illustrating target candidate constituent pixel detection processing in the target detection method according to the first embodiment of the present invention.

FIG. 5 is a diagram illustrating a flowchart of a target candidate presence / absence determination process according to the first embodiment of the present invention.

FIG. 6 is a diagram illustrating a target candidate presence / absence determination process according to the first embodiment of the present invention.

FIG. 7 is a diagram illustrating a target candidate presence / absence determination process according to the first embodiment of the present invention.

FIG. 8 is a diagram illustrating a flowchart of a target candidate existing region expanding process according to the first embodiment of the present invention.

FIG. 9 is a diagram for explaining a target candidate existing region enlarging process in the target detecting method according to the first embodiment of the present invention.

FIG. 10 is a diagram showing a flowchart of an AND process according to the first embodiment of the present invention.

FIG. 11 is a diagram illustrating an AND process according to the first embodiment of the present invention.

FIG. 12 is a diagram illustrating a flowchart of a labeling process according to the first embodiment of the present invention.

FIG. 13 is a diagram illustrating a labeling process according to the first embodiment of the present invention.

FIG. 14 is a diagram illustrating a flowchart of a target candidate barycenter calculation process according to the first embodiment of the present invention.

FIG. 15 is a diagram illustrating a target candidate center-of-gravity calculation process in the target detection method according to the first embodiment of the present invention.

FIG. 16 is a diagram illustrating a flowchart of a target determination process according to the first embodiment of the present invention.

FIG. 17 is a diagram illustrating a target determination process in the target detection method according to the first embodiment of the present invention.

FIG. 18 is a diagram illustrating a target determination process in the target detection method according to the first embodiment of the present invention.

FIG. 19 is a block diagram illustrating a configuration of a target detection device according to a second embodiment of the present invention.

FIG. 20 is a flowchart illustrating a process of a target detection method according to the second embodiment of the present invention.

FIG. 21 is a diagram illustrating a flowchart of a small target candidate removal process according to the second embodiment of the present invention.

FIG. 22 is a diagram illustrating a small target candidate removal process in the target detection method according to the second embodiment of the present invention.

FIG. 23 is a block diagram illustrating a configuration of a target detection device according to a third embodiment of the present invention.

FIG. 24 is a diagram illustrating a flowchart showing processing of a target detection method according to the third embodiment of the present invention.

FIG. 25 is a diagram illustrating a flowchart of expansion processing of the target detection method according to the third embodiment of the present invention.

FIG. 26 is a diagram illustrating expansion processing of the target detection device according to the third embodiment of the present invention.

FIG. 27 is a block diagram illustrating a configuration of a target detection device according to a fourth embodiment of the present invention.

FIG. 28 is a diagram illustrating a flowchart illustrating processing of a target detection method according to the fourth embodiment of the present invention.

FIG. 29 is a block diagram showing a configuration of a conventional target detection device 1.

FIG. 30 is a flowchart showing a process of a conventional target detection method 1.

FIG. 31 is a view showing a flowchart of conventional target constituent pixel detection processing.

FIG. 32 is a diagram illustrating a conventional target constituent pixel detection process.

FIG. 33 is a diagram showing a flowchart of a conventional target presence / absence determination process.

FIG. 34 is a diagram illustrating a conventional target presence / absence determination process.

FIG. 35 is a diagram illustrating a conventional AND process.

FIG. 36 is a block diagram showing a configuration of a conventional target detection device 2.

FIG. 37 is a flowchart showing a process of a conventional target detection method 2.

FIG. 38 is a diagram illustrating a target candidate constituent pixel detection process of the conventional target detection device 2.

FIG. 39 is a diagram illustrating labeling in the conventional target detection method 2.

FIG. 40 is a diagram illustrating an example of erroneous detection of a target in the conventional target detection method 1.

FIG. 41 is a diagram illustrating an example where a target shape is not well detected in the conventional target detection method 1.

FIG. 42 is a diagram illustrating an example of a histogram of an image in which the contrast between the pixels constituting the target and the background is not good.

FIG. 43 is a diagram illustrating an example of an image after subject candidate constituent pixel detection processing has been performed on an image in which the contrast between the pixels constituting the target and the background is not good.

[Explanation of symbols]

10: target candidate constituent pixel detection unit, 20: storage unit, 30: target candidate presence / absence determination unit, 40: target candidate existence area enlargement unit, 50: AND processing unit
Reference numeral 60: labeling unit, 70: target candidate center of gravity calculation unit, 80: target determination unit, 90: small target candidate removal unit, 100: expansion processing unit

────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A-10-232206 (JP, A) JP-A-10-38523 (JP, A) JP-A-8-137931 (JP, A) JP-A-7- 177370 (JP, A) JP-A-7-198324 (JP, A) JP-A-7-79428 (JP, A) JP-A-7-105383 (JP, A) JP-A-6-201715 (JP, A) JP-A-5-68245 (JP, A) JP-A-1-24881 (JP, A) (58) Fields investigated (Int. Cl. ⁶ , DB name) G01S 7/00-7/42 G01S 13/00 -13/95 G06F 15/62 G06F 15/70

Claims (8)

(57) [Claims] 1. A binarized image is generated by detecting pixels constituting a target candidate from an input original image, and a target is generated by using the input original image and the binarized image generated above. The presence / absence of the target candidate is determined from the statistical value of the candidate constituent pixel and the statistical value of the background constituent pixel, and based on the result of the presence / absence determination, if the target candidate exists, the target candidate existing area is determined. The target candidate area data is output by enlarging the area, the logical product of the generated binarized image and the target candidate area data is taken, and the image obtained by the logical product is labeled to recognize each target candidate. Calculating the center of gravity of the labeled target candidate; setting a target determination area centered on the center of gravity for all of the target candidates recognized in the above; a target candidate statistic and a background in each target determination area Calculate statistics, A target detection method, comprising: determining a target. 2. The method according to claim 1, wherein a small target candidate removal process is added between the labeling process and the target candidate centroid calculation process, and a small target candidate whose number of pixels of the target candidate is equal to or less than a predetermined number is determined. A target detection method characterized by removing. 3. The method according to claim 1, wherein an expansion process is added between the labeling process and the target candidate center-of-gravity calculation process, so that pixels surrounded by the same label number and having no label number are the same. A target detection method characterized by using a label number. 4. The method according to claim 2, wherein an expansion process is added between the small target candidate removal process and the target candidate centroid calculation process, and a small target candidate whose number of pixels of the target candidate is equal to or less than a predetermined number is determined. A target detection method, wherein pixels are removed, and pixels surrounded by the same label number and having no label number are assigned the same label number. 5. A target candidate constituent pixel detecting means for detecting a pixel constituting a target candidate from an input original image to generate a binarized image, and further comprising: the input original image; Target candidate presence / absence determining means for determining the presence / absence of the target candidate from the statistical values of the target candidate constituent pixels and the statistical values of the background constituent pixels using the binarized image; A target candidate existing area expanding means for enlarging the target candidate existing area to the periphery thereof and outputting target candidate area data; and a logical product of the generated binarized image and the target candidate area data AND processing means for taking the AND, labeling an image obtained by the logical product and recognizing each target candidate, and a target determination area centered on the center of gravity for all the target candidates recognized above. Set, calculates a target candidate statistics and background statistic in each target determination region, the target detecting device, characterized in that it comprises a target determining means for determining a target. 6. The apparatus according to claim 5, further comprising a small target candidate removing unit between the labeling unit and the target candidate center of gravity calculating unit, the small target candidate removing unit for removing a small target candidate whose number of pixels of the target candidate is equal to or less than a predetermined number. A target detection device characterized by being added. 7. The expansion device according to claim 5, wherein pixels surrounded by the same label number and having no label number are assigned the same label number between the labeling means and the target candidate gravity center calculating means. A target detection device, characterized by adding processing means. 8. The apparatus according to claim 6, wherein pixels surrounded by the same label number and having no label number are assigned the same label number between the small target candidate removing means and the target candidate centroid calculating means. A target detection device characterized by adding expansion processing means.