JP3446177B2 - Target identification method and target identification device - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for identifying a target existing in an input image and its apparatus, and more particularly to transmitting an image acquired by a synthetic aperture radar mounted on a satellite to a ground station, The present invention relates to a target identification method and a target identification device which process the image to identify a target such as a ship at high speed and stably.

[0002]

2. Description of the Related Art There are various conventional methods for identifying a target. Here, one of the methods for identifying a target will be described. FIG. 46 is a diagram showing an overall configuration of a target identifying apparatus that executes a conventional target identifying method.

In FIG. 46, a target identifying device detects a pixel constituting a target (hereinafter, referred to as a target constituent pixel) from an inputted image to be identified and binarizes the target constituent pixel detecting unit 110. A template image center moving range setting unit 160 for setting the moving range of the template image for identification, and a correlation value of the similarity between the region where the template image is placed in the binarized target image and the template image. As a correlation value calculation unit 280, a correlation value calculated by the correlation value calculation unit 280, the best matching result determination unit 290 that detects the highest correlation value in the range in which the template image is moved, and all template images. In contrast to the above 16
All template image end determination unit 220 for determining whether or not the processing of 0 to 290 has been completed, extracting the template image with the highest correlation value among all template images,
The target name output unit 260 outputs the target name.

FIG. 47 is a process flowchart of the target identifying apparatus for implementing the conventional target identifying method shown in FIG. In FIG. 47, the conventional target identification method is a target component pixel detection process S110 of detecting a target constituent pixel from an input target image to be identified and generating a binarized image, and a template image for identifying a target. Template image center movement range setting processing S160, 2 for setting the movement range
Correlation value calculation processing S280 and correlation value calculation processing S28 for calculating the correlation value between the template image and the area where the current template image is placed in the binarized target image.
The best matching result determination processing S290 for detecting the highest correlation value in the range in which the template image is moved with respect to the correlation value calculated by 0, and the processing from S160 to S290 is completed for all template images. All template image end determination processing S22 for determining whether or not
0, the template image having the highest correlation value among all the template images is extracted, and the target name output process S260 is performed to output the target name.

FIG. 48 is a diagram showing an input image to be identified and a template image. FIG. 48A shows the inputted image to be identified, and FIG. 48B shows the template image. In the target constituent pixel detection process S110, in FIG. 48, each pixel forming the target is represented on the xy axis of the image to be identified, and the luminance of the pixel at the (x, y) coordinate is X (x, y). ) Is displayed. On the other hand, the template image is a binarized image of the pixels forming the target corresponding to the target to be identified. Template images are prepared for all targets that can be identified. The target is identified in principle by calculating the correlation value between the input image to be identified and all the prepared template images,
By comparison, a template image having a correlation value of 1 or a value closest to 1 is extracted, and the target in the image to be identified is determined to be the target existing in the template image.

Next, a detailed description will be given of the conventional target identifying device. In the target constituent pixel detection process S110, the target constituent pixels are detected from all the pixels in the image to be identified, the target constituent pixels are represented by "1", and the pixels estimated to be the background are represented by "0". Process.

The target constituent pixel detection process S110 compares whether or not the luminance value X (i, j) of each pixel in the image to be identified is greater than or equal to the binarization determination value D, and X (i, j) ) ≧ D 2, it is determined that the pixel is a target constituent pixel, and when X (i, j) <D 2, it is determined that the pixel is a pixel that does not constitute the target. Where i is the x coordinate of a pixel in the image to be identified,
j is the y coordinate of the pixel in the image to be identified. This is performed for all pixels in the image to be identified. The binarized image thus obtained is shown in FIG. FIG. 49
FIG. 49A shows a target constituent pixel binarized image, and FIG.
Indicates a template image (binarized image). FIG. 49
The "target" in the central portion of (a) is a portion determined to be a pixel constituting the target in the image to be identified, and is the "target area".
Is a portion to be overlapped with the template image (binarized image) shown in FIG. 49 (b), and the calculation of the correlation value between the overlapped “target area” and the template image (binarized image) can be calculated. Done.

Template image center moving range setting process S
At 160, as shown in FIG. 50, the target constituent pixel binarized image is searched for a position where the pixel constituting the target and the pixel constituting the target in the template image best overlap with each other. The template image is moved in the converted image, and the moving range in which the center of the template image moves at this time is set. For example, as shown in FIG. 50, the center position P (x, y) of the template image is ± α in the x-coordinate axis direction from the target center of gravity O (u, v) in the target constituent pixel binarized image, and the y-coordinate axis direction. Set to move within ± β. By this setting, the center P (x, y) of the template image can be moved within the range of P (u−α, v−β) to P ′ (u + α, v + β).

In the correlation value calculation processing S280, a predetermined template image is extracted from all the template images (binarized images) in order to identify the target existing in the image to be identified in the above moving range, and FIG. As shown, the correlation value C of the target area that is the current target in the target constituent pixel binarized image that is the result of the target constituent pixel detection processing S110 is calculated.

[0010]

[Equation 1] Where f (x, y) is the target component pixel binarized image, t (x, y) is the template image, and (u, v) is the position where the center of the template image is set in the target component pixel binarized image. , F _ave is the average value of all the pixels in the portion (target area) where the correlation value is calculated with the template image in the target constituent pixel binary image, t _ave is the average value of all the pixels inside the template image, and S is the target configuration. An area (target area) currently targeted in the pixel binarized image is shown.

In the best matching result determination processing S290, the template image is moved within the moving range u ± α and v ± β set by the template matching moving range setting processing, and the center position of the template image is shifted to obtain the correlation value in the entire moving range. If C is not calculated, the process returns to the template image center moving range setting process S160, and if the calculation is completed in the entire moving range, the maximum value of the correlation value between each target area within the moving range and the template image is detected. , Output as the best matching value.

All template image end determination processing S220
Then, for all template images, the above process S160-
It is determined whether S290 is completed. If not completed, the process returns to S160, and if completed, the next name output process S2 is executed.
The process of moving to 60 is performed.

In the target name output process S260, the template image having the largest value is extracted from the maximum correlation value of each template image which is the result of the above-described best matching result determination process S290, and the target image existing in the template image is extracted. The name is output as a target existing in the image to be identified.

[0014]

However, the above-mentioned method is the identification of the target by the correlation value between the binarized images and is an effective method when the target can be identified by the outer shape. 51 (a1) and 51 (a2) have the same outer shape but cannot be applied to two targets having different internal structures. That is, when the two target images are displayed in multiple values, the internal brightness values are different and the two target images are different as shown in FIGS. 51 (b1) and (b2). However, when the threshold level is set so that the outer shape is not lost in this multi-valued image, the two target outer shapes are the same as shown in FIGS. 51 (c1) and (c2). That is, the above-described conventional target identification device cannot identify the targets as separate targets.

Therefore, instead of using a binarized image in the process of calculating the correlation value between the target area of the image to be identified and the template image, both the brightness value or the absolute value of the brightness value is obtained for both the image to be identified and the template image. It is conceivable to perform the above-mentioned correlation value processing with the value converted to (for example, the backscattering coefficient in the synthetic aperture radar image) to identify the target.

At this time, as shown in FIGS. 52 (a1) and 52 (a2), if the same target shape locally changes for some reason, as shown in FIGS. 52 (a1) and 52 (a2). The brightness value of the portion where the form has changed changes. In this way, if the same target shape locally changes for some reason, the correlation between the changed target brightness value pattern and a similar template image becomes high, and an incorrect target identification result is provided. There is.

In order to solve this, there is a partial template matching method as shown in FIG. 53, in which a region in the image to be identified in which a target is present is divided and template matching is performed on the local region. FIG. 53 (a) is a diagram in which the target region in the image to be identified is divided into four local regions, and FIG. 53 (b) is a diagram in which the template image is divided into four local regions. In this case, there are many background portions in the local area of the target area and the local area of the template image in the image to be identified, as shown in FIG. 53A, and the background area is present between the target area and the local area of the template image. When the similarity is high, the correlation value in this area is high. Also,
In some cases, there is a background only part.

When the number of such regions increases, the target existing in the template image having the maximum sum of the correlation values of the respective local regions, the average value of the correlation values, or the correlation value higher than the threshold is identified. When the target exists in the target image, the identification result may cause an error.

Further, there is a drawback in that the correlation value must be calculated even when the number of pixels constituting the target is small and the number of backgrounds is large in the divided local area, resulting in a large amount of processing.

The present invention has been made in view of the above problems, and when the target which cannot be identified only by the outer shape is identified, the correlation of the local region is determined by the number of target constituent pixels of the divided local region. (EN) A target identification method and a target identification device capable of performing stable and high-speed target identification by significantly reducing the calculation amount by skipping value calculation.

[0021]

In order to solve the above-mentioned problems, a target identifying method according to the first invention of the present invention detects a pixel constituting a target from an inputted image to be identified and binarizes it. An image is generated, a labeling image is generated from the binarized image generated above, a target center of gravity is calculated from the labeling image, and a center moving range of the template image is set with respect to the input identification target image. , Determining the center of the target area on which the template image should be placed, extracting the target area, counting the number of target pixels in the local area of the labeling image, and using the count number and a predetermined threshold, the local area Correlation value is determined and the correlation value is calculated by extracting the target local area in the target area and the target local area in the template image corresponding to the local area,
Correlation value calculation determination is performed for all local regions of the target region, the sum of correlation values of all local regions determined to be correlation value calculation in the correlation value calculation determination process is calculated, and the moving range of the template image is calculated. The best matching result is determined from the sum of the correlation values calculated in step 1, and the target existing in the template image that has the best matching result among all the template images is the target existing in the identification target image. And outputs the name of the target.

In the target identifying method according to the second aspect of the present invention, after the target center of gravity calculation processing, the target long axis is calculated and the target direction is included in the range of the target direction ± predetermined angle calculated from the image to be identified. A target long axis calculation process for selecting only the template image to be performed is added, and the correlation value / correlation value sum calculation process is performed using only the template image selected in the target long axis calculation process.

In the target identifying method according to the third aspect of the present invention, the target length is calculated after the target center-of-gravity calculation process, and the target length is included in the target length ± predetermined length range calculated from the image to be identified. A target long axis calculation process for selecting only the template image to be performed is added, and the correlation value / correlation value sum calculation process is performed using only the template image selected in the target length calculation process.

The target identifying method according to the fourth aspect of the present invention further calculates a target length after the target center of gravity calculation processing,
A target long axis calculation process that selects only the template image included in the range of the target long axis direction ± predetermined angle in which the target direction is calculated from the identification target image, and calculates the target length,
The target length is calculated by adding the target length calculation process for selecting only the template image whose target length is calculated from the image to be identified ± the predetermined length range, and the target long axis calculation process is performed, and the target The correlation value / correlation value sum calculation process is performed using only the template image selected in the length calculation process.

The target identifying method according to the fifth aspect of the present invention is the target area in the target area corresponding to the position of the target constituent pixel in the target local area in the correlation value calculation processing of the first to fourth aspects of the present invention. The correlation value is calculated using only the area-corresponding position pixel and the target local area-corresponding position pixel in the template image.

[0026]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiment 1. 1 is a block diagram showing the overall configuration of a target identifying device according to a first embodiment of the present invention. Referring to FIG. 1, the target identifying apparatus according to the first embodiment of the present invention includes a target constituent pixel detection unit 110, a labeling unit 120, a target center of gravity calculation unit 130, a template image center moving range setting unit 160, a target area extraction unit 170, a local area. Area processing execution determination unit 180, target local area extraction unit 190 within target area,
Target local area extraction unit 200 in template image,
Correlation value / correlation value sum calculation unit 210, all template image end determination unit 220, all local region end determination unit 230, best matching result determination unit 24
0, an entire movement range end determination unit 250, a target name output unit 260, a storage unit 300, and a counter 400.

FIG. 2 is a flowchart showing the overall processing of the target identifying device according to the first embodiment of the present invention. In FIG. 2, the overall flow of the target identification processing is as follows: Target constituent pixel detection processing S110 for detecting the pixels forming the target in the input target image to be identified; Labeling processing S120 for performing labeling to generate a labeling image, target centroid calculation processing S130 for calculating the centroid of the labeled target, and template image center movement range for setting the movement range of the template image for the image to be identified Setting process S160, target region extraction process S17 for extracting a target region in the image to be identified to determine matching with the template image
0, a local area of a preset size is extracted on the labeling image, the target number of pixels in the local area is calculated, and the calculated number of pixels is greater than or equal to the preset number of pixels (threshold). Local region processing execution determination process S180 for determining to execute the following correlation value / correlation value sum total calculation process S210 only in a certain local region, target for extracting a local region corresponding to the local region determined above from the target region In-region target local region extraction process S190: One template image is read out, and in-template-image target local region extraction process S is performed to extract the target local region in the template image corresponding to the target region target local region.
200, correlation value / correlation value sum total calculation processing S210 for calculating the correlation value of the target local area in the target area and the target local area in the template image, and calculating the sum of the correlation values. All template image end processing S220 for determining whether or not the target local area extraction processing S200 in the template image and the correlation value / correlation value sum total calculation processing S210 have been executed, and all local areas in the target area currently targeted are processed. All local region end determination processing S230 for determining whether or not it has been performed, it is determined whether the degree of matching currently processed has reached the maximum value for each template image, and if it is the maximum value, it is updated to the maximum value. If not, the best matching result determination process S240 that retains the previous maximum value, the template image center moving range setting process S1 0 total travel end determination process determines whether all range set is processed in S25
0, the maximum value is detected from the degree of matching of all template images, the target existing in the template image having the maximum value is determined to be the target in the image to be identified, and the name of the target is determined. Output target name output process S
It is composed of 260. The processes performed by the apparatus configuration shown in FIG. 1 and each processing configuration shown in FIG. 2 will be sequentially described in detail below.

FIG. 3 is a flow chart for explaining the target constituent pixel detection processing S110 according to the first embodiment of the present invention. In FIG. 3, the target constituent pixel detection unit 110
Performs an identification target image reading process S111, a target constituent pixel extracting process S112, and a binarizing process S113. The target constituent pixel extraction process S112 may use a process of extracting pixels constituting a target by using a method disclosed in, for example, “Target Detection Method and Target Detection Device”, Japanese Patent Laid-Open No. 2000-180543.

FIG. 4 shows a target constituent pixel detection process S performed by the target constituent pixel detection unit 110 according to the first embodiment of the present invention.
It is a figure explaining 110. In the identification target image reading process S111, the input image is scanned to read each pixel of the identification target image as shown in FIG. Target constituent pixel extraction processing S112
The pixel portion that constitutes the target is extracted by, and binarization processing S1 is performed.
A binarization process is performed in 13 to obtain a binarized image in which, as shown in FIG. 4B, for example, the pixel forming the target is given "1" and the other pixels are given "0". The obtained target constituent pixel binarized image is stored in the storage unit 300.

FIG. 5 is a flowchart of the labeling process S120 performed by the labeling unit 120 according to the first embodiment of the present invention. The labeling process S120 includes a target component pixel binary image reading process S121 and a labeling execution process S122. Labeling process S12
0 reads the target constituent pixel binarized image which is the result obtained by the target constituent pixel detection unit 110, detects the connected portion of the target constituent pixels, and determines whether each pixel belongs to the target. Thereby, the connected component is recognized as the target. The processing result is stored as a labeling image in the storage unit 30.
Written to zero.

FIG. 6 is a diagram for explaining the labeling process S120 performed by the labeling unit 120 according to the first embodiment of the present invention. Target constituent pixel binary image reading process S
At 121, the target constituent pixel binarized image stored in the storage unit 300 in the above-described target constituent pixel detection processing S110 is read out at 121. In the labeling execution process S122, labeling is performed on the read target component pixel binarized image. In the labeling process S120, as shown in FIG. 6A, the target constituent pixels (represented by "1") in the binarized image of the target constituent pixels are connected as shown in FIG. 6B. The same label, for example, 100 is added to the label to create a labeling image. As a result, connected pixels (pixels given 100) can be recognized as the same target.

FIG. 7 is a flowchart of the target center-of-gravity calculation processing S130 performed by the target center-of-gravity calculation unit 130 according to the first embodiment of the present invention. The target centroid calculation process S130
Labeling image reading process S131, centroid calculation process S
It is composed of 132. The target centroid calculation process S130
By using the labeling image obtained by the labeling unit 120, the center of gravity of the target existing in the image to be identified is calculated. This center of gravity of the target is used in a later process to determine the range of movement of the template image on the image to be identified.

FIG. 8 is a diagram for explaining the target center of gravity calculation processing S130 performed by the target center of gravity calculating unit 130 according to the first embodiment of the present invention. The range surrounded by the thick line in FIG. 8 is the target for which the center of gravity is calculated. In the labeling image reading process S131, the labeling image is read from the storage unit 300. In the center-of-gravity calculation process S132, the positions X and Y of the target x-axis and y-axis centroids having the same label number are calculated by the following equations, and the calculated target centroid G is stored as target centroid data in the storage unit. Written to 300.

[0034]

[Equation 2] Here, (x _n , y _n ): the position of the pixel forming the target X: the position of the center of gravity in the x direction of the target Y: the position of the center of gravity of the target in the y direction T: the number of pixels forming the target

FIG. 9 is a flowchart of the template image center moving range setting process S160 performed by the template image center moving range setting unit 160 according to the first embodiment of the present invention. Template image center moving range setting process S
Reference numeral 160 includes a target centroid data reading process S161, a moving range calculation process S162 of the template image center, and a moving range counter initialization process S163. In the template image center moving range calculation processing S160,
The moving range of the center of the template image that is the basis for moving the template image in order to search for the place where the target in the image to be identified and the target in the template image best overlap with each other is calculated.

FIG. 10 is a diagram for explaining the template image center moving range setting processing S160. In FIG. 10, the whole is the image to be identified, the solid line is the target constituent pixel, the double line is the template image area, G is the center of gravity of the target calculated from the pixels forming the target in the image to be identified, and G ′ is the template. The center of the image and the alternate long and short dash line indicate the range in which the center of the template image is moved. The target center-of-gravity data reading process S161 reads the target center-of-gravity data stored in the storage unit 300, and in the moving range calculation process S162 of the template image center, it is assumed that the target in the template image well overlaps with the target in the image to be identified. The range (point a to point i) in which the center position G ′ of the template image can be moved is calculated so that the entire area to be covered can be covered. Next, in the moving range counter initialization processing S163, the moving range counter is initialized to “0” so as to correspond to the point “a”, and is written in the counter 400.

FIG. 11 is a flowchart of the target area extraction processing S170 performed by the target area extraction unit 170 according to the first embodiment of the present invention. The target area extraction processing S170 is
The target image reading processing S171, the moving range counter information reading processing S172, the target area center position calculation processing S173, the target area pixel extraction processing S174, and the local area counter initialization processing S175 are included.

FIG. 12 is a diagram for explaining the target area extraction processing S170. Identification target image reading process S171
Then, the image to be identified shown in FIG. 12 is read from the storage unit 300, and the moving range counter information reading process S is performed.
At 172, the value of the moving range counter stored in the counter 400 is read. Target region center position calculation process S17
3 uses the value of the moving range counter to calculate the position where the center of the template image should be placed on the image to be identified, and writes it in the storage unit 300 as the center position MC of the target area.

For example, first, as shown in FIG. 12, the center position MC of the target area is determined so that the center of the template image is located at point a on the image to be identified. Next, in the target area pixel extraction processing S174, the target area in which the template images overlap in the image to be identified is extracted and written in the storage unit 300. In the local area counter initialization processing S175, the local area counter used in the subsequent processing of the local area is initialized to "0".

FIG. 13 is a flowchart of the local area processing execution judgment processing S180 executed by the local area processing execution judgment unit 180 according to the first embodiment of the present invention. The local area processing execution determination processing S180 is a labeling image reading processing S181, a target area center position reading processing S182,
It is composed of a local area counter information reading processing S183, a target local area range calculation processing S184, a local area counter addition processing S185, a local area target pixel number calculation processing S186, and a processing execution determination processing S187.

FIG. 14 is a diagram for explaining the local area processing execution determination processing S180 according to the first embodiment of the present invention. Figure 1
4, the whole image is a labeling image, the dotted line is the target region, the double line is the local region, and the thick line is the target detected and labeled as the target constituent pixel. First, the labeling image previously stored in the storage unit 300 is read in the labeling image reading process S181, and the storage unit 3 is read in the target area center position reading process S182.
The target area center position stored in 00 is read out. next,
In the local area counter information reading process S183, the local area counter value stored in the counter 400 is read,
Target Local Area Range Calculation Processing S184 The position range of the target local area on the labeling image is calculated. In the target pixel number calculation processing S186 in the local area, the number of pixels forming the target in the position range of the target local area, that is, in the currently targeted local area is counted based on the label number. And then add up the numbers.

For example, in the local areas 0, 1 and 2 of FIG. 14, the target constituent pixel included in each local area is 1, and the number of target constituent pixels included in the local area 3 is 7.
In the process execution determination process S187, the number of target constituent pixels is compared with the number of preset threshold pixels, and if the target number of constituent pixels is greater than or equal to the threshold, the local region process execution determination process S180 is terminated. Then, the process proceeds to the target region in target region extraction process S190 which is the next process, and if it is less than the threshold, the local region counter addition process S185 increments the local region counter by 1, and the process returns to the target local region range calculation process S184. The position range of the next local region is calculated, and the target pixel number calculation process S in the local region is performed.
At 186, the target number of constituent pixels of the local area is calculated. In addition, in this description, the target constituent pixel number is calculated using the labeling image, but the target constituent pixel binarized image may be used.

FIG. 15 is a flow chart of the target area in-target area local area extraction processing S190 performed by the target area in-target area local area extraction unit 190 according to the first embodiment of the present invention. The target local area extraction processing S190 in the target area includes a target area read processing S191, a local area counter information read processing S192, and a target local area range calculation processing S19.
3 and target local area pixel extraction processing S194.

FIG. 16 is a diagram for explaining the target local area in-target area extraction processing S190. In FIG. 16, the whole is the target area, the thick line is the target within the target area, and the double line is the local area. In the target area read processing S191, the target area shown in FIG. 16 is read, and in the local area counter information read processing S192, the processing S175 or S185.
In step S193, the value of the local area counter stored in the counter 400 is read out, the target local area range calculation processing S193 determines the range in which the target local area is extracted, and the target local area pixel extraction processing S194 becomes the subsequent processing target. Pixels in the local area are extracted and written in the storage unit 300 as the target local area in the target area. Here, for example, when the value of the local area counter is 0, nine pixels of the local area 0 are extracted.

FIG. 17 is a flowchart of the template image target local area extraction process S200 performed by the template image target local area extracting unit 200 according to the first embodiment of the present invention. The target local area extracting process S200 in the template image is a template image reading process S2.
01, local area counter information read processing S202, target local area range calculation processing S203, and target local area pixel extraction processing S204.

FIG. 18 is a diagram for explaining the target local area extraction processing S200 in the template image. In FIG. 18, the whole indicates the template image, the thick line indicates the target in the template image, and the double line indicates the local area. In the template image reading process S201, the entire template image shown in FIG. 18 is read, in the local region counter information reading process S202, the value of the local region counter stored in the counter 400 is read, and in the target local region range calculation process S203, The range in which the target local area is extracted is determined, and in the target local area pixel extraction processing S204, the pixels of the local area to be processed thereafter are extracted and written in the storage unit 300 as the target local area in the template image.

FIG. 19 is a flowchart of the correlation value / correlation value sum total calculation process S210 performed by the correlation value / correlation value sum calculation unit 210 according to the first embodiment of the present invention. The correlation value / correlation value sum total calculation process S210 includes a target region in-target region local region reading process S211, a template image in-target region local region reading process S212, and a correlation value calculating process S21.
3 and the correlation value sum calculation processing S214.

FIG. 20 shows a correlation value / correlation value sum total calculation process S.
It is a figure explaining 210. 20A shows the target region as a whole, the thick line shows the target in the target region, the double line shows the local region in the target region, and the triple line shows the target local region in the target region. The whole of FIG. 20B shows a template image,
The thick line indicates the target in the template image, the double line indicates the local area in the template image, and the triple line indicates the target local area in the template image.

In the correlation value calculation processing S210, the target area in-target area local area reading processing S211 and the template image in-target area local area reading processing S212 are respectively performed on the current object in the triple-line target area shown in FIG. The local area that has become the target area and the local area that is currently the target in the template image of the triple line in FIG. The correlation value calculation process S213 calculates the correlation value of both the read target local regions. Here, the calculation of the correlation value is performed only for the local area determined to be processed in the local area processing execution determination processing S180,
Do not do anything else.

For example, in FIG. 20A, the target number of pixels in each of the local areas 0 and 3 is 4, and the local area 1
And 2 have the target pixel number of 1, respectively. Here, for example, if the predetermined threshold is set to 3, the correlation value is calculated in the local areas 0 and 3, and the correlation is not calculated in the local areas 1 and 2. As described above, the present invention is characterized in that the calculation of the correlation value can be omitted in the local region below the predetermined threshold. In the correlation value sum calculation processing S214, the correlation values calculated so as to obtain the sum of the correlation values of all the local areas in the target area are added for each template image, and the sum is written in the storage unit 300 as the sum of the correlation values. In this example, finally, the sum of the correlation values calculated in the local area 0 and the local area 3 becomes the sum of the correlation values.

FIG. 21 is a flowchart of the all template image end determination processing S220 performed by the all template image end determination unit 220 according to the first embodiment of the present invention. The all-template-image end determination unit 220 calculates the correlation value / correlation value sum total calculation process S21 in which the current target local area in the target area is the target local area of all template images.
It is determined whether 0 has been performed. Correlation value / correlation value sum total calculation processing S2 with target local areas of all template images
If 10 is completed, the process proceeds to a subsequent process, that is, a process of determining the end of all local regions S230, and if not, the process returns to the process S200 of extracting a target local region in the template image to process the next template image.

FIG. 22 is a flowchart of the total local area end determination processing S230 performed by the total local area end determination unit 230 according to the first embodiment of the present invention. The all-local area end determination unit 230 determines whether or not the processing of all the local areas in the target area that is currently the target has been completed. If the processing of all local areas has been completed, the processing proceeds to the best matching result determination processing S240, which is the subsequent processing, and if not, the local area counter addition processing of adding 1 to the local area counter so that the processing can move to the next local area. After performing S232,
The processing returns to the local area processing execution determination processing S180, and the next local area is processed.

FIG. 23 is a flowchart of the best matching result judgment processing S240 performed by the best matching result judgment unit 240 according to the first embodiment of the present invention. The best matching result determination process S240 is a maximum correlation value total reading process S241, a current correlation image total value reading process S242, a matching result comparison process S243, and a maximum correlation value total replacement process S244.
Composed of.

In the best matching result determination processing S240, the maximum correlation value sum total reading processing S241 reads the maximum correlation value sum total value for each template image from the storage unit 300, and the current correlation value sum total for each template image is read. Read processing S242 reads the current value of the correlation value sum of each template image, and the matching result comparison processing S243 compares the current correlation value sum of each template image with the maximum correlation value sum read on the storage unit 300. To do. Maximum correlation value sum replacement processing S24
In 4, if the current correlation value sum is larger, the maximum correlation value sum in the storage unit 300 is replaced with the current matching result sum of correlation values. In the above description, the sum of correlation values of all local regions is used in the matching result comparison process S243, but an average value of correlation values of all local regions may be used.

FIG. 24 is a flowchart of the total movement range end determination processing performed by the total movement range end determination unit 250 according to the first embodiment of the present invention. The entire moving range end determination process S250 is a moving range counter information reading process S25.
1 and template image center moving range setting processing S1
A moving range end determination process S252 for determining whether or not the process has been performed for all the moving ranges set at 60, and a moving range counter adding process S253.

FIG. 25 shows the entire movement range end determination processing S25.
It is a figure explaining 0. In FIG. 25A, the whole shows the image to be identified, the thick line shows the target in the image to be identified, and the double line shows the current target area. In FIG. 25 (b), the whole shows the image to be identified, the thick line shows the target in the image to be identified, and the double line shows the next target area.

In the entire movement range end determination processing S250, the movement range counter information reading processing S251 reads the value of the movement range counter from the counter 400, and is set in the template image center movement range setting processing S160 in the movement range end determination processing S252. It is determined whether or not the processes have been performed in all the template moving ranges, and if all the processes are completed, the target name output process S26, which is the subsequent process.
0, otherwise, in the moving range counter addition processing S253, 1 is added to the moving range counter in the counter 400 so that the center position of the target area becomes the next point, and the target area extraction processing S170 is executed. Returning to this, as shown in FIGS. 25A and 25B, the target area is moved to the next position.

FIG. 26 is a target name output process S260 performed by the target name output unit 260 according to the first embodiment of the present invention.
It is a flowchart of. Target name output process S260
Is a final best matching value initial value setting process S261,
Template image best matching result reading process S2
62, a comparison process S263 of the final best matching value and the best matching result of each read template image, a final best matching value replacement process S264, an all template image end determination process S265, and a target name output process S266.

In the target name output process S260, the final best matching value initial value setting process S261 first sets the final best matching value initial value, and in the template image best matching result read process S262, the template image best matching result is obtained. Read the value of. In the comparison process S263, the final best matching value is compared with the best matching value of the currently read template image. If the read value is larger, the final best matching value replacement process S264 updates the final best matching value to the read value. To do. All template image end determination processing S265
It is determined in steps S262 to S264 for all template images, and the target name existing in the template image determined to be the final best matching value in the target name output process S266 is output.

As described above, according to the first embodiment of the present invention, by skipping a local area having a lot of backgrounds, the calculation of the correlation value of the local area can be omitted, so that the target can be identified stably and at high speed. be able to.

Embodiment 2. In the first embodiment,
The example using the template prepared for only one direction of the target has been described. However, in reality, in order to accurately identify the target, it is necessary to prepare a template image of the target in all directions. It takes a huge amount of processing time to process all the template images prepared in this way. In the second embodiment, the target direction is included in the range of the above-described target long-axis direction ± predetermined angle from the calculation error of the target long-axis direction in the image to be identified and the assumed long-axis calculation error. By selecting only the template image to be processed, it is possible to provide the target identifying apparatus and method for reducing the number of template images to be subjected to the correlation value / sum of correlation value calculation processing, reducing the processing amount, and speeding up the processing.

FIG. 27 is a block diagram showing the structure of the target identifying apparatus according to the second embodiment of the present invention. The target identifying apparatus according to the second embodiment is similar to the target center-of-gravity calculating unit 130 of the configuration of the first embodiment shown in FIG.
40 is added. Since only the template images within the target long-axis direction ± predetermined angle range in the target image calculated by adding the target long-axis calculation unit 140 can be selected, the subsequent correlation value / correlation value sum calculation processing is performed. It is possible to significantly reduce the number of template images to be applied.

FIG. 28 is a flow chart showing the overall processing of the target identifying method according to the second embodiment of the present invention. The target major axis calculation process S140 calculates the target inertial moment from the labeling image and the target center of gravity data to calculate the inertial principal axis, and sets the inertial principal axis as the target major axis. The calculation of this long axis is a known technique, for example, Makoto Nagao, "Pattern Information Processing", The Institute of Electronics, Information and Communication Engineers, University Series, Corona,
Since it was published in July 1993, its detailed explanation is omitted. 27 and 28, the target long axis calculation unit 140 and the target long axis calculation process S140.
The configuration other than that is the same as that of FIG. 1 and FIG.

The target major axis calculation process S140 will be described in detail below. FIG. 29 shows target long axis calculation processing S14.
It is a figure which shows the flowchart of 0. The target long axis calculation process S140 includes a labeling image reading process S141, a target long axis calculation process S142, and a processing direction determination process S143.

FIG. 30 is a diagram for explaining the outline of the target long axis calculation processing S142. In the labeling image reading process S141 in FIG. 29, as shown in FIG. 30A, the labeling image generated in the labeling process S120 is read. In the labeling image, the label number of the target constituent pixel is "100" here. At this time, the target long-axis direction is calculated as shown in FIG. 30B by calculating the target long-axis direction using the conventional technique for the pixel to which the brightness value of “100” is given. .

Next, in the processing direction determination processing S143, the target long axis direction θ is calculated in the target long axis calculation processing S142, and the target direction in the template image is included in the range of the long axis direction θ ± φ degrees. Select only the template images that will be displayed. Here, φ is an angle of deviation from the long-axis direction θ that is predetermined in relation to the long-axis direction θ, and, for example, φ is set to 10% of θ or an absolute angle of 10 degrees. FIG. 31 is a diagram conceptually showing the selection of the template image described above. FIG. 31A shows a target in the image to be identified and the long-axis direction of the target, and FIG.
1) and (b2) show template images having targets in various directions. The target major axis direction in the template image of FIG. 31 (b1) is θ + γ, and here γ> φ, so the template image of FIG. 31 (b1) is not selected. Here, γ is the angular difference between the long-axis direction of the target in the image to be identified and the long-axis direction of the target in the template image.

On the other hand, the target major axis direction in the template image of FIG. 31 (b2) is θ + γ, and here, γ
Since <φ, the template image in FIG. 31 (b2) is selected. Thus, the template image in which the angular difference γ between the target direction in the template image and the calculated long-axis direction θ is γ> φ can be excluded. At this point, the front of the target cannot be determined, so the template image having the direction of the major axis of +180 degrees (or -180 degrees) is also selected.

FIG. 32 is a diagram showing all template images. There are all possible goals (Goal 1 to Goal n in the figure) in all template images, and for each goal there are m template images for every goal direction. In the case of the first embodiment, in the processes S200 to S220, the images to be identified and all the template images are processed. That is, it is necessary to perform the processes S200 to S220 for the target n × direction m template images.

On the other hand, in the second embodiment, for example, the target directions 2, 3, 3 are selected from all template images.
Since the processes S200 to S220 are performed only on 6 types of template images of 4 and r-1, r-2, r-3, it is sufficient to perform the process only on n × 6 types of template images. Here, “r−2” indicates the template image in the direction in which the target is rotated 180 degrees with respect to the direction 3.

As described above, the target center of gravity calculation process S in FIG.
By adding the target major axis calculation process S140 after 130, the amount of template images used to identify the target can be reduced, and the target identification process can be sped up compared to the first embodiment.

Third Embodiment As shown in the second embodiment, all template images include template images of all the targets (target 1 to target n in the figure) that may exist. At this time, there is a possibility that template images having significantly different target lengths exist. In this case, if the length of the target in the image to be identified is known in advance, only the template image in which the length of the target and the target close to the length exist are extracted in the same manner as in the second embodiment, and the process S200 The number of template images used in -220 can be reduced. In the third embodiment, only the template image in which the direction of the target is included in the above-described target length ± predetermined length range is selected from the calculation error of the target length and the assumed target length in the image to be identified. However, the number of template images to be used can be reduced, the amount of processing can be reduced, and a target identifying apparatus and method for speeding up processing are provided.

FIG. 33 is a block diagram showing the structure of the target identifying apparatus according to the third embodiment of the present invention. The target identifying apparatus according to the third embodiment includes a target length calculating unit 15 after the target center of gravity calculating unit 130 having the configuration of the first embodiment shown in FIG.
0 is added. By adding the target length calculation unit 150, it is possible to select only a template image in which the target is within the target length ± predetermined length range from the length of the target in the image to be identified. Therefore, the subsequent correlation value / correlation value sum calculation process It is possible to significantly reduce the number of template images to be applied.

FIG. 34 is a flowchart showing the overall processing of the target identifying method according to the third embodiment of the present invention. The target length calculation unit 150 calculates the target moment of inertia from the labeling image and the center of gravity data of the target to obtain the target length L. This calculation of the target length is a known technique. For example, the publication of a paper published in 1988 by Koichi Sasakawa et al. "Extraction of parallel shapes by extended spoke filter", IEICE Technical Report,
Since it is disclosed in PRU-88-10, detailed description will be omitted. 33 and 34, the configurations other than the target length calculation unit 150 and the target length calculation processing S150 are the same as those in FIGS. 1 and 2, respectively, and therefore description thereof will be omitted.

The target length calculation processing S150 will be specifically described below. FIG. 35 is a diagram showing a flowchart of the target length calculation processing S150. Target length calculation process S15
0 is composed of a labeling image reading process S151, a target length calculating process S152, and a process length determining process S153.

In the target length calculation processing S150, the labeling image reading processing S151 is performed as in the second embodiment.
At, the labeling image generated in the labeling process S120 is read. In the target length calculation process S152, the target length L of the target in the image to be identified is calculated from the read labeling image by a known method, and in the process length determination process S153, the target length in the template image is calculated. Selects only the template image included in the range of the target length L ± η. Here, η is a length predetermined in relation to the target length L in the image to be identified,
For example, η is defined as 10% of L, or 10 m in absolute length.

FIG. 36 is a diagram conceptually showing the selection of the template image according to the target length. FIG. 36A shows an image to be identified, and FIGS. 36B1 and 36B2 show target template images of various lengths. Fig. 36
The target length in the template image of (b1) is L-
Since λ, and here λ> η, FIG.
The template image of 1) is not selected. Where λ
Is the difference length between the target length in the image to be identified and the target length in the template image. On the other hand, FIG.
Since the target length L + λ in the template image of 2), and here λ <η, the template image of FIG. 36 (b2) is selected. In this way, template images in which the target length in the template image is not included in the range of L ± η can be excluded.

FIG. 37 is a diagram showing all template images. The template images for all targets 1 to n are included in all the template images. Furthermore, each target has m template images for each direction. Here, the target length L of target 1
1 is 150 m, target length L2 of target 2 is 100 m, target n
Assuming that the target length Ln of the target image is 145 m, the target length of the target image in the target length calculation process S152 is 1
50 m, and in the processing length determination processing S153, for example,
If a template image with a target length within a target length of 150 m ± 10 m is selected, a target 2 with a length of 100 m can be selected.
No template image of is selected, and template images of target 1 and target n are selected. That is, the processing may be performed on 2 × m different template images.

As described above, the target center of gravity calculation process S of FIG.
By adding the target length calculation process S150 after 130, the amount of template images used to identify the target can be reduced, and the target identification process in the first embodiment can be speeded up.

Fourth Embodiment In the second embodiment,
Regarding the method of selecting only the template image in the direction of the long axis of the target in the image to be identified ± the direction included in the predetermined angle range, in the third embodiment, the target length in the image to be identified ± the predetermined length The method of selecting only the template images included in the range has been described. In the fourth embodiment, by combining the above-described second and third embodiments, it is possible to further reduce the number of template images to be selected, thereby providing a target identifying apparatus and method capable of further shortening the processing time. .

FIG. 38 is a block diagram showing the structure of the target identifying apparatus according to the fourth embodiment of the present invention. In the fourth embodiment, the target center of gravity calculation unit 130 of FIG. 1 is followed by the target long axis calculation unit 140 and the target length calculation unit 150.
Is added. Either the target long axis calculation unit 140 or the target length calculation unit 150 may come first.

FIG. 39 is a flowchart showing the overall processing of the target detecting method according to the fourth embodiment of the present invention. FIG. 39
In the flowchart, the target long axis calculation process S140 and the target length calculation process S150 are added after the target center of gravity calculation process S130. As described above, either the target long axis calculation process S140 or the target length calculation process S150 may come first.

Target long axis calculation process S in the fourth embodiment
Since 140 and the target length calculation processing S150 are the same as the target long axis calculation unit 140 in the second embodiment and the target length calculation processing S150 in the third embodiment, respectively, detailed description thereof will be omitted.

In the fourth embodiment, by introducing both the target long axis calculation processing S140 and the target length calculation processing S150, both the merits of the second embodiment and the third embodiment can be obtained.

FIG. 40 is a diagram showing all template images. The template images for all targets 1 to n are included in all the template images. Furthermore, each target has m template images for each direction. Therefore, n as a whole
There are × m different template images. In this example, in the target long axis calculation process S140, only the template images included in the target long axis direction ± predetermined angle range in the target image to be identified can be selected. Only the shaded portion of the template image included in the target length ± predetermined length range in the image can be selected, and the target template image can be further reduced in steps S200 to S220. Therefore, the processing amount can be further reduced and the processing time can be further shortened.

Embodiment 5. In the first embodiment,
In the correlation value / correlation value sum calculation processing S210, the correlation value is calculated using all the pixels of the target local area in the target area and the target local area in the template image which are the target of the correlation value calculation. In that case, the background pixel in the local region is also used for calculating the correlation value, and the correlation value is not calculated only from the portion corresponding to the target constituent pixel.
In the fifth embodiment, the correlation value is calculated by extracting only the pixels corresponding to the positions of the pixels forming the target in the target local area in the target area and the target local area in the template image for which the correlation value is calculated. In this way, the number of pixels used for calculating the correlation value is reduced and the processing amount is reduced.

FIG. 41 is a block diagram showing the overall structure of the target identifying apparatus according to the fifth embodiment of the present invention. In FIG. 41, the local area processing execution determination unit 1 of the first embodiment
80, the correlation value / correlation value sum calculation unit 210 respectively determines the local area pixel processing execution determination / position recording unit 31.
0, which is replaced by the target constituent pixel position correlation value / correlation value sum calculation unit 320.

FIG. 42 is a flowchart showing the overall processing of the target identifying device according to the fifth embodiment of the present invention. Figure 4
2, the local area process execution determination process S180 and the correlation value / correlation value sum total calculation process S210 of the first embodiment are performed.
Are replaced with local area pixel processing execution determination / position recording processing S310 and target constituent pixel position correlation value / correlation value sum calculation processing S320, respectively.

FIG. 43 is a diagram for explaining the local area pixel processing execution determination / position recording processing S310. In FIG. 43, local area pixel processing execution determination / position recording processing S310
Reads the labeling image stored in the storage unit 300 in the labeling image reading process S181, and reads the labeling region center position reading process S182 in the storage unit 3
The value of the center position of the target area stored in 00 is read out. Next, in the local area counter information reading process S183, the value previously stored in the local area counter of the counter 400 is read out, and in the target local area position calculation processing S184, the current target local area in the labeling image is read. Calculate the range. Next, in the target constituent pixel position writing process S311, the target local region position calculation process S18.
The positions of the pixels that match the label numbers of the pixels forming the target in the range of the target local area calculated in step 4 are sequentially recorded and written in the storage unit 300.

Next, the target pixel number calculation process S18 in the local area is performed.
6, the number of pixels matching the label number is calculated, and when the target pixel number calculated in process S186 in the process execution determination process S187 is larger than the threshold, the process moves to the target local region in target region extraction unit 190 thereafter. , Otherwise, local area counter addition processing S1
At 85, the local area counter is incremented by 1.

FIG. 44 is a flowchart for explaining the target constituent pixel position correlation value / correlation value sum total calculation processing S320. In FIG. 44, first, in the target area target local area read processing S211, the target area target local area is read from the storage unit 300, and in the template image target local area read processing S212, the template image target local area is read. Next, in the target constituent pixel position reading process S321, the target constituent pixel position data is read from the storage unit 300, and in the target region target local area corresponding position local pixel corresponding position pixel extraction processing S322, the current target pixel target local area is read. The value of the pixel corresponding to the target constituent pixel position is written in the storage unit 300. Similarly, target local area corresponding position pixel extraction processing in the template image S323
Then, the value of the pixel corresponding to the current target constituent pixel position in the target local area in the template image read out above is written in the storage unit 300. All pixel end determination process S
At 324, it is determined whether all of the target constituent pixels have finished,
When the processing is completed, in step S325, the target area corresponding position pixel in the target area and the target area corresponding position pixel in the template image are read, and the process proceeds to correlation value calculation processing S213.
If all of the target constituent pixels have not ended in the all-pixel end determination processing S324, the target constituent pixel position reading processing S3
21, the next target constituent pixel position data is read out,
After that, in the correlation value sum total calculation processing S214, the correlation values calculated so as to obtain the sum total of the correlation values of all the local regions in the target region are added for each template image, and stored in the storage unit 300 as the sum of the correlation values. Write.

FIG. 45 is a diagram for explaining the details of the correlation processing in the target constituent pixel position correlation value calculation processing S320. The whole of FIG. 45 (a) shows the target area, the thick line is the target in the target area image, and the double line is the local area in the target area.
The triple line indicates the target local area within the target area that is currently the target of processing within the target area. The whole of FIG. 45 (b) shows the template image, the thick line shows the target in the template image,
The double line indicates the local area in the template image, and the triple line indicates the target local area in the template image which is currently the processing target in the template image. The hatching part is
A pixel corresponding to the position of the target constituent pixel in the local region of interest,
That is, the pixel for which the correlation value is calculated is shown. The hatched portion in FIG. 45 shows pixels used for calculating the correlation value in the correlation value calculation processing S213. That is, in the fifth embodiment, the correlation value is not calculated using all the pixels in the local area, but is calculated using only the portion surrounded by hatching.

As described above, in the fifth embodiment,
By calculating the correlation value using only the corresponding position pixel of the target local area in the target region and the corresponding position pixel of the target local area in the template image corresponding to the position of the target constituent pixel, the background pixel is used to calculate the correlation value. Since it does not need to be used, the processing amount can be reduced and the processing speed can be increased.

[0093]

As described above, the first and sixth aspects of the present invention are provided.
According to the present invention, since the present invention does not need to calculate the correlation value when the number of pixels forming the target in the local region is smaller than the threshold, the present invention is more efficient than the conventional partial pattern matching process. The amount of processing can be reduced,
Target identification processing can be performed stably and at high speed.

According to the second and seventh aspects of the present invention, the target major axis calculation process is added after the target centroid calculation process in the present invention, so that the direction of the target in the template image is the object to be identified. It is possible to select only the template images included in the target long-axis direction ± predetermined angle range calculated from the images, reduce the number of template images to be processed, and speed up the processing.

According to the third and eighth aspects of the present invention, since the target length calculation process is added after the target center of gravity calculation process in the present invention, the target length in the template image is the object to be identified. It is possible to select only the template images included in the target length ± predetermined length range calculated from the images, reduce the number of template images to be processed, and speed up the processing.

According to the fourth and ninth aspects of the present invention, since the target major axis calculation process and the target length calculation process are added after the target center of gravity calculation process in the present invention, the target in the template image Only the template image whose direction is included in the long-axis direction of the target calculated from the image-to-be-identified ± within a predetermined angle range, and whose target length is included in the target length calculated from the image-to-be-identified ± within a predetermined length range Can be selected, the number of template images to be processed can be further reduced, and the processing can be further speeded up.

According to the fifth and tenth aspects of the present invention, the present invention provides a target area target local area corresponding position pixel corresponding to a position of a target constituent pixel in the target local area and a template image target. Since the correlation value is calculated using only the local region corresponding position pixel, the number of pixels in the calculation of the correlation value can be reduced, and the processing speed can be increased.

[Brief description of drawings]

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

FIG. 2 is a flowchart showing processing of a target identifying method according to the first embodiment of the present invention.

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

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

FIG. 5 is a flowchart of a labeling process according to the first embodiment of the present invention.

FIG. 6 is a diagram illustrating a labeling process in the target identifying method according to the first embodiment of the present invention.

FIG. 7 is a flowchart of target centroid calculation processing according to the first embodiment of the present invention.

FIG. 8 is a diagram illustrating target centroid calculation processing in the target identification method according to the first embodiment of the present invention.

FIG. 9 is a flowchart of template image center moving range setting processing in the target identifying method according to the first embodiment of the present invention.

FIG. 10 is a diagram illustrating template image center moving range setting processing according to the first embodiment of the present invention.

FIG. 11 is a flowchart of target area extraction processing in the target identification method according to the first embodiment of the present invention.

FIG. 12 is a diagram illustrating target area extraction processing according to the first embodiment of the present invention.

FIG. 13 is a flowchart of local area processing execution determination processing in the target identifying method according to the first embodiment of the present invention.

FIG. 14 is a diagram illustrating local area processing execution determination processing according to the first embodiment of the present invention.

FIG. 15 is a flowchart of target local area within target area extraction processing in the target identification method according to the first embodiment of the present invention.

FIG. 16 is a diagram illustrating target local area extraction processing within a target area according to the first embodiment of the present invention.

FIG. 17 is a flowchart of target local area extraction processing in a template image in the target identification method according to the first embodiment of the present invention.

FIG. 18 is a diagram for explaining target local area extraction processing within a template image according to the first embodiment of the present invention.

FIG. 19 is a flowchart of a correlation value / correlation value sum total calculation process of the target identifying method according to the first embodiment of the present invention.

FIG. 20 is a diagram illustrating a correlation value / correlation value sum total calculation process according to the first embodiment of the present invention.

FIG. 21 is a flowchart of all template image end determination processing in the target identifying method according to the first embodiment of the present invention.

FIG. 22 is a flowchart of an all local area end determination process in the target identifying method according to the first embodiment of the present invention.

FIG. 23 is a flowchart of a best matching result determination process in the target identifying method according to the first embodiment of the present invention.

FIG. 24 is a flowchart of the entire movement range end determination processing in the target identifying method according to the first embodiment of the present invention.

[Fig. 25] Fig. 25 is a diagram for describing the end-of-all-movement-range determination processing according to the first embodiment of the present invention.

FIG. 26 is a flowchart of target name output processing in the target identifying method according to the first embodiment of the present invention.

FIG. 27 is a block diagram showing a configuration of a target identifying device according to a second embodiment of the present invention.

FIG. 28 is a flowchart showing a process of a target identifying method according to the second embodiment of the present invention.

FIG. 29 is a flowchart of target long axis calculation processing in the target identifying method according to the second embodiment of the present invention.

FIG. 30 is a diagram illustrating target long axis calculation processing according to the second embodiment of the present invention.

FIG. 31 is a diagram conceptually showing selection of a template image in the target long-axis direction according to the second embodiment of the present invention.

FIG. 32 is a diagram showing all template images according to the second embodiment of the present invention.

FIG. 33 is a block diagram showing a configuration of a target identifying device according to a third embodiment of the present invention.

FIG. 34 is a flowchart showing a process of a target identifying method according to the third embodiment of the present invention.

FIG. 35 is a flowchart of target length calculation processing in the target identifying method according to the third embodiment of the present invention.

FIG. 36 is a diagram conceptually showing selection of a template image based on a target length according to the third embodiment of the present invention.

FIG. 37 is a diagram showing all template images according to the third embodiment of the present invention.

FIG. 38 is a block diagram showing a configuration of a target identifying device according to a fourth embodiment of the present invention.

FIG. 39 is a flowchart showing the processing of the target identifying method according to the fourth embodiment of the present invention.

FIG. 40 is a diagram showing all template images according to the fourth embodiment of the present invention.

FIG. 41 is a block diagram showing a configuration of a target identifying device according to a fifth embodiment of the present invention.

FIG. 42 is a flowchart showing the processing of the target identifying method according to the fifth embodiment of the present invention.

FIG. 43 is a flowchart showing local area process execution determination / position recording process according to the fifth embodiment of the present invention.

FIG. 44 is a flowchart illustrating target component pixel position correlation value / correlation value sum total calculation processing according to the fifth embodiment of the present invention.

FIG. 45 is a diagram illustrating details of a correlation value calculation process in a target constituent pixel position correlation value / correlation value sum total calculation process according to the fifth embodiment of the present invention.

FIG. 46 is a block diagram showing a configuration of a conventional target identifying device.

FIG. 47 is a flowchart showing the processing of a conventional target identifying method.

FIG. 48 is a diagram showing an input image to be identified and a template image.

FIG. 49 is a diagram showing a target constituent pixel binarized image and a template image (binarized image) that have been binarized by the target constituent pixel detection processing.

FIG. 50 is a diagram illustrating template image center moving range setting processing in the conventional target identification method.

FIG. 51 is a diagram for explaining the reason why targets having the same outer shape but different internal structures cannot be processed by the conventional target identifying apparatus.

FIG. 52 is a diagram illustrating a case where the shape of the target in the image to be identified locally changes.

[Fig. 53] Fig. 53 is a diagram for describing partial template matching in which a region where a target exists in an image to be identified is divided and template matching is performed as a local region.

[Explanation of symbols]

110 ... Target constituent pixel detection unit, 120 ... Labeling unit, 130 ... Target centroid calculation unit, 140 ...
Target long axis calculation unit, 150 ... Target length calculation unit,
160 ... Template image center moving range setting unit,
170 ... Target area extraction unit, 180 ... Local area processing execution determination unit, 190 ... Target area in target area local area extraction unit, 200 ... Template image in target local area extraction unit, 210 ... Correlation value / correlation value sum calculation unit, 220 … All template image end judgment unit, 2
30 ... All local region end determination unit, 240 ... Best matching result determination unit, 250 ... Full movement range end determination unit, 260 ... Target name output unit, 280 ... Correlation value calculation unit, 290 ... Best matching result determination unit, 300 ... Storage unit 310 ... Local area process execution determination / position recording unit 320 ... Target constituent pixel position correlation value / correlation value sum calculation unit 400 ... Counter

Continuation of front page (51) Int.Cl. ⁷ identification code FI G06T 1/00 285 G06T 1/00 285 7/00 300 7/00 300D (56) Reference JP-A-11-306353 (JP, A) Kaihei 5-250475 (JP, A) JP 2000-180543 (JP, A) PRMU96-18 Recognition by structural description of two-dimensional figures, IEICE Technical Report, Japan, May 17, 1996, Vol. 96 No. 41, pp. 45-50 (58) Fields investigated (Int.Cl. ⁷ , DB name) G06T ⁷ /00-7/60 G01S 13/90 G06T 1/00

Claims (10)

(57) [Claims] 1. A binarized image is generated by detecting pixels constituting a target from an inputted image to be identified, a labeling image is generated from the binarized image generated in the above, and a labeling image is generated from the labeling image. The center of gravity of the target is calculated, the center moving range of the template image is set for the input identification target image, the center of the target area where the template image is to be placed is determined, the target area is extracted, and the target area is set. Corresponding to
The area of the labeling image is divided into local areas and
The correlation value is calculated for each of the divided local regions, and the correlation value is calculated.
The target existing in the template image based on
Eye markers presumed to be targets existing in the image to be identified
Alternatively, the target number of pixels in each of the local regions is counted, and it is determined whether the count is less than or equal to a predetermined threshold ,
If the count is less than the predetermined threshold
Does not calculate the correlation value of the local area,
If the number is equal to or larger than a predetermined threshold, it extracts the target area within a subject local region and the template image in the target local region corresponding to the local area to calculate the correlation values, obtained in the preceding Symbol correlation value calculation origin management calculating a sum of the correlation values of all the local regions, and determines the best matching result from the sum of the correlation values calculated by the moving range of the template image to obtain more best matching result in the entire template image A target identifying method characterized by estimating a target existing in a template image as a target existing in an image to be identified and outputting the name of the target. 2. The method according to claim 1, wherein a target long axis is calculated after the target centroid calculation process, and the target direction in the template image is the direction of the target long axis calculated from the image to be identified ±. A target long axis calculation process for selecting only a template image included in a range of a predetermined angle is added, and the correlation value and correlation value sum calculation process is performed using only the template image selected in the target long axis calculation process. Target identification method characterized by. 3. The method according to claim 1, wherein after the target center of gravity calculation processing, a target length is calculated and the target length in the template image is a target length ± calculated from the image to be identified.
A target long axis calculation process for selecting only template images included in a predetermined length range is added, and the correlation value and correlation value sum calculation process is performed using only the template image selected in the target length calculation process. Target identification method characterized by. 4. The method according to claim 2, further comprising calculating a target length before or after the target long axis calculation process,
The target length is calculated by adding the target length calculation process for selecting only the template image whose target length is calculated from the image to be identified ± the predetermined length range, and the target long axis calculation process is performed, and the target A target identifying method, characterized in that the correlation value and correlation value sum calculation processing is performed using only the template image selected in the length calculation processing. 5. The method according to claim 1, wherein the target local area corresponding position pixel in the target area and the target local area corresponding position in the template image corresponding to the position of the target constituent pixel in the target local area. Target identification method characterized by calculating a correlation value using only pixels 6. A means for generating a binarized image by detecting pixels constituting a target from the inputted image to be identified, and a means for generating a labeling image from the binarized image generated above. Means for calculating the center of gravity of the target from the labeling image, means for setting the center movement range of the template image with respect to the input identification target image, and determining the center of the target area in which the template image should be placed, Means for extracting a region and a Rabeli corresponding to the target region
The region of the image is divided into local regions, and
A means for calculating the correlation value for each local area, and a means for calculating the correlation value based on the correlation value.
The target existing in the template image is identified based on
And means for presuming that the target exists in the target image.
The target identification device counts the target number of pixels in each of the local regions, and determines whether the count number is less than or equal to a predetermined threshold.
Means and the count number is less than a predetermined threshold
In this case, the correlation value for the local area is not calculated and the
When the number of unds is greater than or equal to a predetermined threshold, a phase for calculating the correlation value by extracting the target local area in the target area and the target local area in the template image corresponding to the local area
And related value calculating means, and means for calculating the sum of the correlation values of all the local regions obtained by the correlation value calculation detemir stage, the best matching results from the sum of the correlation values calculated by the moving range of the template image Determining means, and means for estimating a target existing in the template image having the best matching result among all the template images as a target existing in the image to be identified, and outputting the name of the target A target identifying device comprising: 7. The apparatus according to claim 6, wherein after the target center of gravity calculation means, a long axis of the target is calculated, and the direction of the target in the template image is the target direction ± predetermined angle calculated from the image to be identified. Target long axis calculating means for selecting only template images included in the range, and performing only the template image selected by the target long axis calculating means to perform the sum calculation means of the correlation value and the correlation value. Characteristic target identification device. 8. The apparatus according to claim 6, wherein after the target center of gravity calculation means, a target length is calculated, and the target length in the template image is a target length ± calculated from the image to be identified.
A target long axis calculation means for selecting only template images included in a predetermined length range is added, and the correlation value and the sum of correlation values are calculated using only the template image selected by the target length calculation means. A target identification device characterized by the above. 9. The apparatus according to claim 7, wherein a target length is further calculated before or after the target long axis calculation means,
A target length calculation unit for selecting only a template image whose target length is calculated from the image to be identified ± target length range is added, and the target long axis calculation unit selects the target image. A target identifying device characterized in that the correlation value and the sum of correlation values are calculated using only the template image selected by the length calculation means. 10. The apparatus according to claim 6, wherein the target local area corresponding position pixel in the target area and the target local area corresponding position in the template image corresponding to the position of the target constituent pixel in the target local area. A target identifying device characterized in that a correlation value is calculated using only pixels.