JP3499807B2 - Target classification method and apparatus for radar - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a classifying method and apparatus for classifying a target in a radar using an ISAR (Inverse Synthetic Aperture Radar) image created from radar echoes. The present invention relates to a method and a device for classifying by expressing and analyzing in a vector space.

[0002]

2. Description of the Related Art Generally, in identifying a target from an ISAR image generated by processing a radar echo reflected from the target in a radar, the target is projected in various different directions, and further sufficient for identification. It is necessary to consider the problem that information is contained in only a limited number of frames. Therefore, conventionally, in order to solve this problem, a method of selecting a frame from a time-series ISAR image that captures a moment that is easy to identify and comparing it with a target model prepared in advance, or a method for identifying the frame It was general to apply the method of performing the image processing of. However,
Since such processing often requires manual work, the load on the operator becomes a bottleneck, and it has been difficult to realize automatic class identification at high speed.

On the other hand, Japanese Patent Laid-Open No. 11-281731 (hereinafter referred to as known document 1) discloses a method of automatically selecting an optimum observation time. However, in this method, a threshold is set so that the spread in the range direction is matched in the radar video whose amplitude is detected, and only the radar video exceeding the threshold is extracted and converted into ISAR image information. Need to go through. Therefore, it cannot be said that only the ISAR image selectively obtained by this method is sufficient for the purpose of classifying the target.

I proposed by MUSUMAN et al.
EEE Transactions on Aerospaceand Electronic Systems
In the automatic recognition of ISAR ship images published in Vol.32, No.4 (hereinafter referred to as known document 2), the target line of the target ship is obtained using Hough transformation, and then the target is based on this. Techniques for detecting other characteristics of the above have been proposed, and it is possible to automatically extract some parameters that are used as identification materials. However, this method does not always obtain the center line of the target ship so that it accurately passes through the bow and stern, and it may not always be possible to perform stable classification.

[0005]

As described above, in the prior art, when a radar target is classified using an ISAR image, the target is generally projected in various directions according to the principle of image generation. With regard to this, a frame that captures an easily identifiable moment is selected from the ISAR image sequence, and image processing for identification is performed on the frame. However, since the image processing involves manual work, the load on the operator is increased. Was a bottleneck, and it was difficult to realize automatic class identification at high speed. Further, the above-mentioned known document 1,
The method described in No. 2 also has a problem that the target classification cannot be performed easily and stably.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a class identifying method and apparatus capable of quickly and easily and stably classifying a target while reducing a load on an operator in a radar.

[0007]

In order to solve the above problems, according to the present invention, a radar echo is processed to generate a first time series image, and an unknown target feature is extracted from the first time series image. Are expressed as at least one first feature vector, and known target features included in the second time-series image prepared in advance as a dictionary are expressed as at least one second feature vector. Feature vector and second
Based on the matching with the feature vector of, the unknown target class is identified.

Further, the class identifying apparatus according to the present invention includes an image generating means for processing radar echoes to generate a first time series image, and a target for extracting an unknown target feature from the first time series image. Feature extraction means and a second prepared in advance as a dictionary
Target feature registration means for registering a known target feature included in the time-series image as at least one second feature vector, and at least one first feature expressing the feature extracted by the target feature extraction means as a vector. It is characterized in that it comprises an identifying means for identifying a class of an unknown target on the basis of matching between the vector and the second feature vector.

The target feature obtained by distorting variously is expressed as a feature vector from the time-series image created by processing the radar echo as described above, and the feature vector is analyzed to identify the target type. Can be performed easily and stably.

Further, the present invention causes a computer to function as means corresponding to the present invention (or causes the computer to realize the function corresponding to the present invention).
It also holds as a computer-readable recording medium recording a program.

[0011]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings, taking a case where a target is a ship as an example. FIG. 1 is a block diagram showing the overall configuration of a class identification device according to an embodiment of the present invention. This classifier is I
The SAR image generation unit 11, the target feature extraction unit 12, the dictionary creation unit 13, and the target identification unit 14 are included.

The ISAR image generator 11 receives the radar echo as an input and processes the radar echo to generate an ISAR image which is a time series image (first time series image). The target feature extraction unit 12 uses the ISAR generated by the ISAR image generation unit 11.
An unknown target silhouette or the like obtained by processing the image is extracted as a feature, and the extraction result is output to the target identifying unit 13. The target identifying unit 14 includes various known target features registered in advance in the dictionary creating unit 13, and the target feature extracting unit 1.
By comparing the characteristics of the unknown target extracted in 2 in the vector space, the target classification is performed. Hereinafter, details of each unit in FIG. 1 will be described in order.

(A) Target Feature Extraction Unit 12 FIG. 2 is a block diagram showing a detailed configuration of the target feature extraction unit 12. As described above, the target feature extraction unit 12 includes the target feature point detection unit 21, the ISAR image normalization unit 22, and the ISA.
The R image superimposing unit 23 is included. Target feature point detector 2
In 1, the target feature point is detected from the ISAR image generated by the ISAR image generator 11 of FIG. ISAR
The image normalization unit 22 performs normalization by correcting the distortion for each target frame in the ISAR image based on the target feature point position detected by the target feature point extraction unit 21. The images normalized by the ISAR image normalization unit 22 are grouped into a plurality of pieces of information such as contours by the ISAR image superposition unit 12, and the output thereof is input to the target identification unit 14 of FIG. 1 for use in class identification. To be done.

(A-1) Target Feature Point Detecting Section 21 FIG. 3 shows a detailed configuration example of the target feature point detecting section 21 in FIG. The target feature point detection unit 21 calculates the luminance distribution along the distance direction along the axis representing the distance direction in the first time-series image, and the distance-direction luminance distribution generation unit 31 according to the generated luminance distribution. The target area setting unit 32 that measures the area where the target exists and the target edge detection unit 33 that detects the specific end of the target as a feature point based on the area where the specified target exists. It

More specifically, first, in the distance direction luminance distribution generation unit 31, the ISAR image generation unit 11 of FIG.
The distribution of the luminance value along the axis in the range (distance) direction of the ISAR image, which is the first time-series image generated in (1), is generated. Normally, the horizontal axis of the ISAR image corresponds to the spread in the target range direction, and the vertical axis corresponds to the Doppler frequency of the reflected wave. Therefore, the distance-direction luminance distribution generation unit 31 observes (detects) the luminance value in the vertical direction at each horizontal coordinate with reference to the horizontal axis, for example, and selects the maximum value to determine the luminance in the distance direction. Generates distribution values.

That is, the brightness value at the horizontal axis coordinate x = x _k and the vertical axis coordinate y = y is I (x _k , y), and the brightness distribution value at the horizontal axis coordinate x = x _k is H (x _k ). Then, H (x _k ) that is the value of the luminance distribution in the distance direction is obtained by the following equation.

[0017]

[Equation 1]

As H (x _k ), an average value of vertical brightness values may be used instead of the maximum value.

Next, in the target area specifying unit 32, according to the value of the distance direction luminance distribution generated by the distance direction luminance distribution generating unit 31, the target in the ISAR image generated by the ISAR image generating unit 11 of FIG. The existing area (hereinafter referred to as the target area) is specified. Since the brightness value of the ISAR image is high in the area where the target is present, the value of the distance direction brightness distribution is ideally high in the area where the target is present. Therefore, considering that the target is located near the center of the ISAR image, correspondingly, the distance-direction luminance distribution also has a high value in the portion corresponding to the center of the horizontal axis and a low value in both end portions. It will be. By utilizing the characteristics of the luminance distribution in the distance direction, by detecting points at which the values sharply increase from both ends of the luminance distribution in the distance direction toward the center, respectively, the range sandwiched between the coordinates of these two points can be obtained. It is possible to specify the range of the spread of the target in the distance direction on the horizontal axis, that is, the target region.

Here, it is applied in the target area specifying section 32,
An example of a method of detecting a point where the luminance distribution value in the distance direction changes abruptly will be shown below in the case of searching from the left end of the horizontal axis toward the center. Here, the horizontal axis direction (range direction)
At each point of, the average of the values at a certain interval on the left side of that point is compared with the value at that point, and when the difference exceeds a preset threshold, the value of the distance direction luminance distribution rapidly increases. Consider it done. When this condition is expressed by an equation, it becomes as follows.

[0021]

[Equation 2]

Here, n represents the number of pixels included in the constant interval considered, and Th represents a threshold value. In the process of sequentially searching, the point that first satisfies such a condition is set as R ₀ . Next, conversely to the above, in the case of sequentially searching from the right end of the horizontal axis toward the center, consider this symmetric condition.

[0023]

[Equation 3]

[0024] Similarly, if the point that first satisfies such a condition is R ₁ , then R ₀ <x <R in the horizontal direction.
_The range corresponding to ₁ will be included in the target area.

In the target edge detecting section 33, the target area specifying section 3
The specific end of the target is detected as a feature point based on the target area obtained in 2. Characteristic points in an ISAR image tend to appear in a portion having a relatively acute-angled feature in the target due to its nature. When the target is a ship, in addition to the bow and stern, the bridge, chimney, antenna, and the tip of the turret are the parts with the above-mentioned acute angle. Here, the target edge detection unit 33 detects the bow and stern as the feature points that can be used by the ISAR image normalization unit 22 of FIG.

According to the target area specifying method applied by the target area specifying unit 32 described above, the points R ₀ and R _{1 at} which the value of the brightness distribution in the distance direction changes rapidly are spread in the distance direction of the target. Corresponds to both ends of. Also, in the example where the target is a ship,
The bow and stern define the distance spread. Therefore, one of the bow and the stern is located near R ₀ , and the other is located near R ₁ .

On the other hand, since the bow and stern are acute-angled parts, the luminance value of the corresponding pixel in the ISAR image is high, and due to the structure of the ship, it is almost always the local maximum value. is there. Therefore, when viewed in the horizontal axis direction, R
By detecting points where the luminance has a local maximum value in the vicinity of ₀ and R ₁ , it is possible to set characteristic points representing the bow and stern. The characteristic points representing the bow and stern detected by the target end detection unit 33 in this manner are set as P and Q.

(A-2) ISAR image normalization unit 22 FIG. 4 shows a detailed configuration example of the ISAR image normalization unit 22 of FIG. The ISAR image normalization unit 22 includes a main axis setting unit 41 that sets a straight line that penetrates the center portion of the target as a standard for normalization based on the target feature points detected by the target feature point detection unit 21, and an IASR. The phase determination unit 42 that determines the phase that indicates whether the target shape appears inverted on both sides of the main axis in the image, and the image of the frame that is determined to be in the inverted phase by the phase determination unit 42 with respect to the main axis And a normalization processing unit 43 that performs normalization so that both sides are inverted.

More specifically, in the ISAR image normalizing section 22, for the ISAR image generated by the ISAR image generating section 11 of FIG. 1, based on the target feature points detected by the target feature point detecting section 21, Normalization is performed so that the target shape, which appears distorted in various different directions in each frame due to changes in pitch, roll, and yaw, is included in a spatially common framework.

The spindle setting unit 41 sets a straight line (spindle) passing through the center portion of the target as a standard for normalization. It is conceivable to use the long axis for the two-dimensional luminance distribution in the ISAR image for this straight line, but its position with respect to the ship is due to the influence of the reflection from the bridge, etc., which changes with time as the attitude of the ship changes. , Not always constant. Therefore, the positions P and Q of the bow and stern detected by the target end detection unit 33 in FIG. 3 are effectively used, and these two points P,
A straight line passing through Q is defined as the main axis. A straight line passing through the bow and stern penetrates the hull and can be stably obtained regardless of the attitude of the ship. The main axis passing through these points P and Q is y = ax
Described as + b. However, x and y are variables in the horizontal and vertical axis directions, and a and b are constants representing the inclination and position of the straight line.

The phase determination unit 42 determines whether the target posture, particularly whether the direction of change in pitch angle is positive or negative. More specifically, the target shape appears upside down in the image depending on whether the pitch angle at the time each frame of the ISAR image is generated changes in the positive or negative direction. The phase determination unit 42 determines whether or not is occurring.

This determination can be made by examining the relative deviation of the brightness values on both sides (upper and lower) of the spindle with the spindle set by the spindle setting section 41 as a reference. For example,
Using the average value of the luminance values, (y> ax +) above the main axis
b) and the area omega _u consisting of lower (y <When the number of pixels included in ax + b) and a region omega _l is n _u, n _l respectively, the luminance values I of the respective regions omega _u, omega _l The average values I _u and I _l follow the following equation.

[0033]

[Equation 4]

Then, these values are compared, and if I _u > I _l , it is determined that the target in the image is in the positive phase, and if I _u <I _l , it is in the negative phase, that is, the inverted phase. To do. on the other hand,
Determination of phase becomes subtle when the average brightness of the upper and lower region of the main shaft is substantially equal Iu~I _l. for that reason,
For example, when both the ratio Iu / I _l is close to 1 (1-δ <Iu
When / I _l <1 + δ), the phase cannot be determined.

The normalization processing section 43 performs the normalization processing on the ISAR image as follows based on the spindle setting result by the spindle setting section 41 and the phase determination result by the phase determination section 42. That is, the entire ISAR image is normalized by such a conversion that the main axis set by the main axis setting unit 41 is in the horizontal direction. At this time, with respect to the frame determined to be in the inverted phase by the phase determination unit 42, both sides of the main axis, that is, the upper and lower sides are inverted and output with respect to the normalized image. As a result, the target shapes that appear distorted in various different directions due to changes in pitch, roll, and yaw in each frame are aligned with the shapes when viewed from the direction along the horizontal plane and orthogonal to the main axis. become.

(A-3) ISAR image superimposing section 23 FIG. 5 shows a detailed configuration example of the ISAR image superimposing section 23 in FIG. The ISAR image superimposing unit 23 superimposes the images normalized by the ISAR image normalizing unit 22 in FIG. 2 (images in which the targets in each frame are aligned in a situation where the targets are viewed from the same direction), and useful information for class identification is displayed. A cycle setting unit 51 for setting a time interval at which the target movement amount is equal to or less than a predetermined level in a time interval of one cycle or more of the target motion, and the set time interval. As a unit, within a corresponding time, the ISAR image normalization unit 22
It is composed of a binarization and superposition unit 52 which takes the logical sum of the binarization results of the images of each frame normalized by

The cycle setting section 51 determines a time interval, that is, a cycle in which the outputs from the respective frames normalized by the ISAR image normalizing section 22 are superimposed. Targets navigating over the sea repeat periodic motions due to shaking except for the yaw (snake angle). On the other hand, if the target is continuously observed for a certain period of time or more, as the target moves, the position in the image also gradually shifts.

Therefore, as a time interval (time range) for superimposing the frame outputs, a range that sufficiently includes one cycle of the target motion and a position shift is not significant (amount of movement of the target position). Within a predetermined level), for example, a period of 1 cycle or more and 2 cycles or less is set.
This superimposes the output over multiple frames from different positions, essentially as viewed from the same direction.

The binarizing and superimposing section 52 superimposes the output in the time interval (cycle) set by the cycle setting section 51. However, while the body of the ship always has a high brightness value, the part corresponding to the tip of the bridge has a brightness value only when the rate of change in pitch angle is large. If the normalized image is superimposed as it is, an image in which only the body portion is emphasized is obtained. That is,
The output from the bridge that shows the characteristics of the target is relatively weak. Therefore, the binarizing and superimposing unit 52
Then, with respect to each frame image, so-called binarization processing is performed in which a luminance value exceeding a predetermined threshold value is converted into, for example, 1 and a luminance value not exceeding the threshold value is converted into 0.

Then, the binarizing and superimposing section 52 further outputs
By taking the logical sum of the binarized frame images, the target features appearing in each frame output in one cycle set by the cycle setting unit 51 are superimposed without loss. Specifically, in the case where an ISAR image of 3 frames per second is generated, the target motion cycle is 8 seconds, and the cycle setting unit 51 sets a cycle that matches the target motion cycle. As an example, the binarized images for a total of 24 frames output in the 8 seconds are superimposed.
However, frames for which phase determination cannot be performed by the phase determination unit 42 are not subject to superimposition. As a result of the superposition in the binarization superposition unit, a binarized image of the target silhouette is obtained, and this is used as the output of the target feature extraction unit 12.

(B) Dictionary creation unit 13 The dictionary creation unit 13 of FIG. 1 is a target IS whose type is known.
AR images (second time-series images) are collected and subjected to the same processing as that of the target feature extraction unit 12 to be described later, thereby creating data regarding known target features and using this as a dictionary. Register as. Although the target data of the kind that can be a candidate is collected, collecting various kinds of data for each candidate is expected to enhance the versatility of the dictionary.

(C) Target Discriminating Unit 14 FIG. 6 is a block diagram showing the detailed structure of the target discriminating unit 14 in FIG. As shown in FIG. 6, the target identifying unit 14 includes a contour vector generating unit 61, a dictionary matching unit 62, and an identification determining unit 63.

The contour vector generation unit 61 generates a target contour vector from the binarized image of the target silhouette obtained by the target feature extraction unit 1 of FIG. Dictionary matching unit 6
2 compares the target contour vector generated by the contour vector generation unit 61 with a plurality of contour vectors corresponding to known targets registered in advance in the dictionary creation unit 13 of FIG. The identification determination unit 63 determines the target feature type identification based on the result of the comparison in the dictionary matching unit 62.

The details of each part in FIG. 6 will be described below in order. The contour vector generation unit 61 generates a target contour vector as a first feature vector from the binarized image of the target silhouette obtained by the target feature extraction unit 12 of FIG. Here, in a ship, the shape of the bridge, turret, etc. located above the main axis connecting the bow and stern characterizes the target, and in the binarized image of the silhouette, each horizontal coordinate on the main axis On the other hand, the upper part of the main axis is searched in the vertical direction, and the vertical coordinate values in which pixels are inverted from 1 to 0 at the boundary of the silhouette are arranged in order, whereby the contour of the ship is expressed as a vector and used as the first feature vector.

That is, the contour which is the first feature vector
If the vector is s, horizontal axis coordinate x = x_k, Vertical shaft seat
Mark y = y_kThe pixel value of the binary image of the silhouette in
B (x _k, Y_k), S is each horizontal axis coordinate x = x_k
The value s (x_k) As an element
Be done. s (x_k) = {Y_k| B (x_k, Y_k) = 1, b (xk, y_k
+1) = 0} According to this, the area where the target exists at both ends of the ISAR image.
Outside the area, s (x_k) = 0, the contour vector s
By definition, these areas should be removed. here
So, the length of the contour vector s is
Since it appears differently depending on the direction of travel in
Normalize to length, eg l = 100. On the other hand, contour contour
The absolute value of the element value of Tor s is the target angular velocity.
The norm of s is, for example, |
Normalization is performed so that s | = 1. By these,
Regarding the data obtained in various situations, only the target shape is
It is possible to make a comparison under the conditions that appear as a characteristic.

The dictionary matching unit 62 includes a contour vector generation unit 6
The unknown target contour vector (first feature vector) generated in 1 is compared with the known target contour vector (second feature vector) created in advance in the dictionary creating unit 13 of FIG. To do. Since the vectors are compared with each other, basically, the similarity between the vectors can be calculated from the angle difference between the vectors, and the smaller the angle difference is, the higher the similarity is. Therefore, if the angle is θ, cos θ is taken, and the degree of similarity is numerically expressed in the range of 0 to 1.

Here, the dictionary creating unit 13 has a target feature point detecting unit, an ISAR image normalizing unit and an ISAR image superimposing unit, like the target feature extracting unit 12 shown in detail in FIG.
Further, it has a contour vector generation unit and a memory unit. Then, the feature point of the known target is detected from the ISAR image having the known target in the same manner as the target feature point detection unit 12 of FIG.
Based on this feature point, the ISAR image is normalized in the same manner as the ISAR image normalization unit 22 in FIG. 2, and the normalized image is superimposed in the same manner as in the ISAR image superimposing unit 23 in FIG. Is generated as a second feature vector and registered in the memory unit.

Normally, the dictionary creating unit 13 prepares a plurality of contour vectors of each target as the second feature vector. Therefore, the actual comparison is performed by the contour vector generating unit 61 as the first feature vector for an unknown target. The distance between the generated contour vector and the subspace formed by a plurality of contour vectors corresponding to a known target is measured, and for this, a subspace method or the like as a pattern recognition method may be used. . In addition, the mutual subspace method may be used for more stable determination. In the mutual subspace method, a plurality of contour vectors are acquired for an unknown target, and then the distance between the subspace and the target subspace prepared on the dictionary side is obtained. As a standard of this distance, for example, the minimum canonical angle between eigenvectors straddling each subspace can be used.

The discrimination determining unit 63 discriminates the target type based on the degree of similarity digitized by comparing the feature vectors of the known target and the unknown target in the dictionary matching unit 62. In the determination of class identification, basically, the target with the highest degree of similarity is selected from the dictionary and used as the identification result. However, the judgment result can be supplemented by performing some exception processing. For example, when the degree of similarity of the selected target is lower than a predetermined constant value, it is determined that there is no corresponding target in the dictionary. Further, when the similarity of the selected target and the similarities of the other candidates are very close to each other, it is necessary to pay attention to the reliability of the determination.

The present invention is not limited to the contents described in the above embodiment. For example, when the target traveling direction is obtained as an input in addition to the ISAR image, I
The target length can be calculated from the target area included in the SAR image, and the accuracy of the target determination can be expected to improve by reflecting the information on the target length in the creation of the contour vector.

In the target feature extraction unit 12, the output format is the target binarized silhouette image, but a method of directly inputting a grayscale image obtained by omitting the binarization process in the middle is also considered. To be In this case, in the target identifying unit 14, instead of vectorizing the contour of the target as described above, if the grayscale image including the target is expressed as a vector having the number of pixels as the length and the brightness value of each pixel as an element, Similar classification processing can be applied to the following.

[0052]

According to the present invention, an ISAR (Inverse Synthetic Aperture Radar) image, which is a time-series image created after processing a radar echo from a target in a radar, is variously distorted with noise. By performing stable analysis of the characteristics of the target acquired in step S in the vector space, it becomes possible to facilitate high-speed target class identification in the radar.

[Brief description of drawings]

FIG. 1 is a block diagram showing an overall configuration of a class identification device according to an embodiment of the present invention.

FIG. 2 is a block diagram showing a detailed configuration of a target feature extraction unit 12 of FIG.

3 is a block diagram showing a detailed configuration of a target feature point detection unit 21 of FIG.

4 is a block diagram showing a detailed configuration of an ISAR image normalization unit 22 of FIG.

5 is a block diagram showing a detailed configuration of an ISAR image superimposing unit 23 in FIG.

6 is a block diagram showing a detailed configuration of a target identifying section 14 in FIG.

[Explanation of symbols]

11 ... ISAR image generator 12 ... Target feature extraction unit 13 ... Dictionary creation unit 14 ... Goal identification section 21 ... Target feature point detection unit 22 ... ISAR image normalization unit 23 ... ISAR image superimposing section 31 ... Distance direction luminance distribution generation unit 32 ... Target area specifying unit 33 ... Target end detection unit 41 ... Spindle setting section 42 ... Phase determination unit 43 ... Normalization processing unit 51 ... Cycle setting section 52 ... Binary superposition unit 61 ... Contour vector generation unit 62 ... Dictionary matching unit 63 ... Identification determination unit

─────────────────────────────────────────────────── ─── Continuation of the front page (72) Kenichi Maeda 1 Komukai Toshiba Town, Komukai-shi, Kawasaki-shi, Kanagawa, Toshiba Research & Development Center Co., Ltd. (56) Reference JP-A-9-33649 (JP, A) Open 2000-3436 (JP, A) JP-A-8-145826 (JP, A) JP-A-9-138277 (JP, A) JP-A-11-265452 (JP, A) JP-A-7-311867 (JP , A) JP 7-73327 (JP, A) JP 2000-275338 (JP, A) JP 9-43348 (JP, A) JP 11-237476 (JP, A) (58) Survey Areas (Int.Cl. ⁷ , DB name) G01S ⁷ /00-7/42 G01S 13/00-13/95 G01T 1/00-1/20 G01T 3/00-5/50 G01T 7/00-7 / 60 G01T 9/00-9/40

Claims (6)

(57) [Claims] 1. An image generation step of processing a radar echo to generate a first time-series image; extracting an unknown target feature from the first time-series image;
A first feature vector generating step of generating a first feature vector expressing the feature as a vector; a known target feature included in the second time series image; and a second feature expressing the feature as a vector. A second feature vector generation step that generates a feature vector and prepares it as a dictionary in advance, and an identification that performs the type identification of the unknown target based on the collation of the first feature vector and the second feature vector. The first and second feature vector generation steps include: (a) detecting the unknown or known target feature point from the first or second time series image; and (b) detecting A straight line passing through the central portion of the unknown or known target based on the determined feature points is set as a main axis serving as a reference for normalization, and (c) the unknown or the unknown in the first or second time series image. It is determined whether or not a known target shape appears inverted on both sides of the main axis, and (d) an image of a frame where the unknown or known target shape appears inverted on both sides of the main axis. With respect to, the normalization is performed on the image of each frame so that both sides are inverted with respect to the main axis by a conversion in which the main axis is in the horizontal direction, and (e) the normalized image of each frame is binarized. By generating a binary image of the silhouette of the unknown or known target by superimposing the contour vector of the unknown or known target from the binary image.
A method for classifying targets in a radar, which is characterized in that it is generated as a feature vector. 2. The identifying step includes matching the two vectors by obtaining the similarity between the first and second feature vectors, and performing the class identification of the unknown target based on the similarity. The method for identifying a target type in a radar according to claim 1, wherein: 3. The identifying step includes determining the similarity using a subspace method or a mutual subspace method based on a subspace defined for each of the plurality of first and second feature vectors. The method for identifying a target type in a radar according to claim 2, wherein 4. Image generation means for processing a radar echo to generate a first time-series image, and extracting an unknown target feature from the first time-series image,
First feature vector generation means for generating a first feature vector expressing the feature as a vector, and a known target feature included in the second time-series image, and a second feature vector expressing the feature. Second feature vector generation means that generates a feature vector and prepares it in advance as a dictionary, and identification for performing the type identification of the unknown target based on the collation of the first feature vector and the second feature vector. The first and second feature vector generation means, the feature point detection means for detecting the unknown or known target feature point from the first or second time-series image; Spindle setting means for setting a straight line that penetrates the central portion of the unknown or known target as a spindle serving as a reference for normalization based on the characteristic points detected by the point detecting means, and the first or second time Determination means for determining whether or not the shape of the unknown or known target in the column image appears on both sides of the main axis, and the shape of the unknown or known target appears on both sides of the main axis in reverse The image of the frame determined to be normalization processing means for normalizing the image of each frame by a conversion in which the main axis is in the horizontal direction so that both sides are inverted with respect to the main axis, By binarizing and superimposing the images of the respective frames, two of the unknown or known target silhouettes can be obtained.
A radar characterized by comprising: superimposing means for obtaining a binarized image; and contour vector generating means for producing the unknown or known target contour vector from the binarized image as the first or second feature vector. Target classifier in. 5. The identifying means collates both vectors by obtaining a similarity between the first and second feature vectors, and classifies the unknown target based on the similarity. 5. The target class identification device in a radar according to claim 4, wherein. 6. The identification means comprises a plurality of the first and second plurality.
6. The target class identification device in a radar according to claim 5, wherein the similarity is obtained by using a subspace method or a mutual subspace method based on a subspace defined for each of the feature vectors of.