KR102295619B1 - Object Identification Accelerate method by Pre-Shape Discrimination in Sonar Images - Google Patents

Abstract

The method for accelerating target identification through pre-shape discrimination according to the present invention comprises: a target detection process of performing shape detection, feature point extraction, and feature quantity detection for target detection with respect to a sonar image; and a target classification process of classifying a target based on the shape detection, the feature point extraction, and the feature amount detection. On the other hand, the target classification process, the target shape and feature amount collecting process of collecting the shape and feature information from the target shape and feature DB unit; and a similarity determination process in which the similarity determination unit compares the shape information with the target table to select a shape priority, and performs a feature amount similarity comparison according to the shape priority, wherein in the similarity determination process, the similarity is a reference value If this is the case, target matching may be performed, and detection and classification of various targets in the sonar image may be performed at high speed, thereby performing target recognition according to target matching.

Description

Object Identification Accelerate method by Pre-Shape Discrimination in Sonar Images

The present invention relates to a method and apparatus for speeding up target identification through pre-shape discrimination in sonar images. More particularly, it relates to a method for improving the speed of a target classification unit by using shape and feature information for target recognition of a sonar image, and an apparatus for performing the same.

Since the underwater image is contaminated with image information due to natural and artificial phenomena, the target detection unit and the classification unit are performed after performing the pre-processing unit to recognize the target. The shape detection unit, which is a sub-step of the target detection unit, detects the shape information of the image to be recognized by searching for the key point candidates before performing the key point extraction unit. The key point extraction unit performs key point extraction for each candidate group. Based on the extracted feature points, the feature-quantity detector is performed to detect the feature, and the target classifier determines the similarity between the feature-quantity DB and the detected feature-quantity to perform matching.

On the other hand, the present invention relates to a high-speed method for improving the target identification speed of an acquired sonar image. In this regard, FIG. 1 is a flowchart illustrating a process for target recognition using a sonar image. Referring to Figure 1, it is composed of a pre-processing unit (Pre-processing), target detection unit (Object Detection), and target classification unit (Object Classification).

As shown in FIG. 1, representative techniques for performing target recognition using feature points include SIFT, SURF, BRIEF, and BRISK. The detailed representation of the target classification unit of FIG. 1 is as shown in FIG. 2 .

Before performing target recognition, as a preparatory step, build a DB of targets to be searched for. The target feature DB part consists of three steps (target table configuration part, feature data collection part, data classification part), and in the case of the target table configuration part in the first stage, Construct a table of targets. The feature data collection unit of the second step detects the feature and collects feature data to be converted into DB, and Fig. 3 shows the feature detection of SIFT with excellent performance among representative techniques for performing target recognition using feature points. An example of the method is shown.

16 영역에서 수식 1을 이용하여 방위각과 강도를 획득하고 방향 히스토그램(Orientation Histogram)을 구성한다. SIFT's feature quantity detection method is 16 where the feature point is located.

In 16 areas, using Equation 1, the azimuth and intensity are obtained and an orientation histogram is constructed.

[Equation 1]

와

는 특징점에 대한 강도와 방위각을 나타내며 이를 이용하여 각 특징점들에 대한 특징량을 벡터합으로 검출한다. 검출된 특징량에 대하여 유사도 측정 기법(유클리디안거리(Euclidean Distance), 코사인유사도(Cosine Similarity), 마할라노비스거리(Mahalanobis Distance) 등)을 적용하고 가장 높은 유사도를 가진 특징량을 하나 선별하여 데이터로 수집한다. 다음은 유사도 판별 기법의 예시로 유클리디안거리 수식을 나타낸다.here

Wow

denotes the intensity and azimuth for the feature point, and using this, the feature amount for each feature point is detected as a vector sum. A similarity measurement technique (Euclidean Distance, Cosine Similarity, Mahalanobis Distance, etc.) is applied to the detected feature quantity, and one feature with the highest similarity is selected. collect as data. The following shows the Euclidean distance equation as an example of a similarity determination technique.

[Equation 2]

,

의 범위를 가지며,

,

,

,

는 각각 영상의 크기와 픽셀이다. 세 번째 단계는 선별된 각 데이터를 구성된 테이블에 분류하여 DB 구축을 완성하는 단계이며 도 4는 특징량 DB 구축의 예시를 나타낸다.here

,

has a range of

,

,

,

are the size and pixels of the image, respectively. The third step is a step of classifying each selected data into a configured table to complete DB construction, and FIG. 4 shows an example of feature DB construction.

In relation to the target recognition technique, FIG. 2 shows the detailed configuration of the target classification unit related to the present invention. In this regard, when the similarity determination unit is performed, the feature quantity information obtained from the target detection unit is sequentially compared with each table in the DB in the feature quantity similarity sequential comparison unit of FIG. 2 and a full search is performed on the entire table. . In this method, a global search between the target table and the feature information is performed to perform a sequential comparison of similarity to determine the maximum similarity target. As this occurs, there is a problem in that time is wasted.

Accordingly, it is an object of the present invention to provide a method and apparatus for speeding up target identification through pre-shape discrimination in order to solve the above problems.

In addition. It is an object of the present invention to provide a method for improving the speed of a target classification unit by using shape and feature information for target recognition of a sonar image, and an apparatus for performing the same.

The method for speeding up target identification through pre-shape discrimination according to the present invention according to the present invention for solving the above problems is a target detection process that performs shape detection, feature point extraction, and feature amount detection for target detection with respect to a sonar image ; and a target classification process of classifying a target based on the shape detection, the feature point extraction, and the feature amount detection. On the other hand, the target classification process, the target shape and feature amount collecting process of collecting the shape and feature information from the target shape and feature DB unit; and a similarity determination process in which the similarity determination unit compares the shape information with the target table to select a shape priority, and performs a feature amount similarity comparison according to the shape priority, wherein in the similarity determination process, the similarity is a reference value If this is the case, target matching may be performed, and detection and classification of various targets in the sonar image may be performed at high speed, thereby performing target recognition according to target matching.

In an embodiment, in the target detection process, the shape detection unit detects shape information from the sonar image, and in the similarity determination process, the detected shape information and the target table are compared to select the shape priority. have.

In an embodiment, the target shape and feature amount DB unit may build a database DB based on the detected target shape and the feature amount information.

In an embodiment, the similarity determination unit may perform the similarity determination process based on the detected shape information and the feature amount information.

In an embodiment, the target table configuration unit of the target shape and feature DB unit may configure a target table for a target selected for target recognition.

In an embodiment, the shape and feature data collection unit of the target shape and feature DB unit may collect the shape information and the feature information detected from the target for DB construction.

In an embodiment, the data classification unit of the target shape and feature amount DB unit may classify the collected shape information and feature amount information to match the type of the target table.

In an embodiment, the shape similarity sequential comparison unit of the similarity determining unit may perform sequential comparison between the constructed DB and the shape information obtained from the shape detection unit.

In an embodiment, the priority-based feature amount similarity comparison unit of the similarity determining unit may preferentially compare the order determined by the shape priority selection unit. In this case, the priority-based feature amount similarity comparison unit may preferentially perform comparison based on target outer edge shape information among the target shapes. In the similarity determination process, if the similarity according to the target outer edge shape information is equal to or greater than a reference value, the target matching may be performed.

Accordingly, according to at least one embodiment of the present invention, there is an advantage in that detection and classification of various targets in the sonar image can be performed at high speed to perform target recognition according to target matching.

In addition, according to at least one embodiment of the present invention, there is an advantage that it can be applied in various ways in improving the target classification speed not only in the sonar image but also in various images.

1 is a flowchart illustrating a process for target recognition (Object Recognition) using a sonar image.
2 shows a detailed configuration of a target classification unit related to the present invention.
3 is a conceptual diagram illustrating a feature amount detection process of the SIFT technique in feature point-based target recognition according to the present invention.
4 shows an example of DB construction for the extracted feature quantity according to the present invention.
5 is a flowchart of a method for speeding up target identification through pre-shape discrimination according to the present invention.
6 shows a detailed configuration of a target recognition apparatus according to the present invention.
7 shows a detailed configuration of a target classification unit according to the present invention.
8 shows an example of the target shape and feature DB construction according to the present invention.
9 shows an original image for evaluating the speed of the target classifier according to the present invention.
10 shows a shape detection binary image and a feature detection image according to the present invention.
11 is a table comparing the experimental results according to the present invention and the existing method in relation to the target classification method.
12 is a graph comparing the experimental results according to the present invention with the existing method in relation to the target classification method.

The features and effects of the present invention described above will become more apparent through the following detailed description in relation to the accompanying drawings, and accordingly, those of ordinary skill in the art to which the present invention pertains can easily implement the technical idea of the present invention. will be able

Since the present invention can have various changes and can have various embodiments, specific embodiments are illustrated in the drawings and described in detail in the detailed description. However, this is not intended to limit the present invention to a specific embodiment, it should be understood to include all modifications, equivalents and substitutes included in the spirit and scope of the present invention.

In describing each figure, like reference numerals are used for like elements.

Terms such as first, second, etc. may be used to describe various elements, but the elements should not be limited by the terms. The above terms are used only for the purpose of distinguishing one component from another.

For example, without departing from the scope of the present invention, a first component may be referred to as a second component, and similarly, a second component may also be referred to as a first component. and/or includes a combination of a plurality of related listed items or any of a plurality of related listed items.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

Terms such as those defined in a commonly used dictionary should be interpreted as having a meaning consistent with the meaning in the context of the related art, and should not be interpreted in an ideal or excessively formal meaning unless explicitly defined in the present application. shouldn't

The suffix module, block, and part for the components used in the following description are given or mixed in consideration of only the ease of writing the specification, and do not have distinct meanings or roles by themselves.

Hereinafter, a preferred embodiment of the present invention will be described with reference to the accompanying drawings so that a person skilled in the art can easily implement it. In the following description of embodiments of the present invention, if it is determined that a detailed description of a related known function or a known configuration may unnecessarily obscure the gist of the present invention, the detailed description thereof will be omitted.

Hereinafter, a method and apparatus for speeding up target identification through pre-shape discrimination in a sonar image according to the present invention, and more specifically, a method of improving the target classification unit speed by using shape and feature information for target recognition of a sonar image and to an apparatus for performing the same.

In this regard, FIG. 3 is a conceptual diagram illustrating a feature amount detection process of the SIFT technique in feature point-based target recognition according to the present invention. In addition, FIG. 4 shows an example of DB construction for the extracted feature quantity according to the present invention.

Meanwhile, FIG. 5 shows a flowchart of a method for speeding up target identification through pre-shape discrimination according to the present invention. Referring to FIG. 5 , the method for accelerating target identification through pre-shape discrimination includes a target detection process ( S100 ), a target classification process ( S200 ), and a target matching process ( S300 ). In addition, the target classification process ( S200 ) includes a process of collecting a target shape and feature amount ( S210 ) and a process of determining the similarity ( S220 ).

Meanwhile, FIG. 6 shows a detailed configuration of a target recognition apparatus according to the present invention. In this regard, in the present invention, as shown in FIG. 6 , shape information obtained through pre-shape discrimination is utilized when the target classifier performs similarity comparison. In addition, we present a high-speed target classification method that performs target identification when the similarity is greater than or equal to the threshold through the Early Decision Scheme, which performs comparative search with priority on targets judged to be similar in shape.

Accordingly, with reference to FIGS. 5 and 6 , a method for accelerating target identification through pre-shape discrimination using the target recognition apparatus according to the present invention will be described in detail as follows.

In the target detection process (S100), shape detection, feature point extraction, and feature quantity detection are performed for target detection on the sonar image. In addition, in the target classification process ( S200 ), the target is classified based on the shape detection, the feature point extraction, and the feature amount detection.

Specifically, in the target shape and feature data collection process ( S210 ), shape and feature data are collected through the target shape and feature data DB unit.

Also, in the similarity determination process ( S220 ), the shape information and the target table are compared through the similarity determination unit to select a shape priority, and a feature amount similarity comparison is performed according to the shape priority. Accordingly, in the similarity determination process ( S220 ), if the degree of similarity is equal to or greater than a reference value, target matching ( S300 ) may be performed.

Meanwhile, the method for speeding up target identification through pre-shape discrimination according to the present invention will be described in terms of the detailed configuration of the target recognition apparatus as follows.

In this regard, in the target detection process (S100), the shape detection unit detects shape information from the sonar image, and in the similarity determination process (S220), the shape priority is determined by comparing the detected shape information with the target table. can choose

In this case, the target shape and feature amount DB unit may build a database DB based on the detected target shape and the feature amount information. Also, the similarity determination unit may perform the similarity determination process ( S220 ) based on the detected shape information and the feature amount information.

Meanwhile, the target table configuration unit of the target shape and feature DB unit may configure a target table for a target selected for target recognition. In addition, the shape and feature data collection unit of the target shape and feature DB unit may collect the shape information and the feature information detected from the target for DB construction.

Meanwhile, the shape and feature data collection unit of the target shape and feature DB unit may collect the shape information and the feature information detected from the target for DB construction. In addition, the data classification unit of the target shape and feature amount DB unit may classify the collected shape information and feature amount information to match the type of the target table.

Meanwhile, the shape similarity sequential comparison unit of the similarity determining unit may perform sequential comparison between the constructed DB and the shape information obtained from the shape detecting unit. Also, the shape priority selecting unit of the similarity determining unit may select a priority based on the similarity derived from the shape similarity sequential comparison unit.

Specifically, the rank-based feature quantity similarity priority comparison unit of the similarity determining unit may preferentially compare the order determined by the shape priority selection unit. In an example, the priority-based feature amount similarity comparison unit may preferentially perform comparison based on target outer edge shape information among the target shapes. In this case, the priority-based feature amount similarity comparison unit may preferentially perform comparison based on target outer edge shape information rather than using specific coordinate values among the target shapes.

In another embodiment, when it is difficult to detect the target outer edge shape information, as shown in FIG. 7 , information about the surrounding object other than the outer edge shape may be used. Also, based on the motion information of the corresponding object, an inference operation on the shape of the outer edge may be performed. In addition, the comparison may be performed based on information on other features until the shape of the outer rim can be grasped.

Accordingly, in the similarity determination process, if the degree of similarity based on the target outer edge shape information is equal to or greater than the reference value, the process may be prematurely terminated and the target matching may be performed.

In this regard, FIG. 7 shows a detailed configuration of a target classification unit according to the present invention. Referring to FIG. 7 , the target shape and feature DB unit performs the target table configuration unit to configure a table for each target, and applies shape information that has not been utilized before to perform the shape and feature data collection unit. Shape information with the highest shape similarity by applying cross-correlation techniques (SCC (Simple Cross-Correlation), NCC (Normalized Cross-Correlation), ZNCC (Zero Mean Normalized Cross Correlation, etc.) Select one and collect it as data The following is an example of the shape similarity discrimination technique, and the formula of ZNCC is shown.

[Equation 3]

은 픽셀의 개수,

와

는 각각 비교를 수행할 표적의 형상이다. 그리고

,

는 각 표적형상의 픽셀평균

,

는 표적형상의 표준편차이다. 이후 선별된 형상정보는 데이터 분류부를 통해 테이블별 데이터분류를 수행하고 분류결과로 DB를 구축한다. 이와 관련하여, 도 8은 본 발명에 따른 표적형상 및 특징량 DB구축의 예시를 나타낸다.in Equation 3

is the number of pixels,

Wow

is the shape of the target to be compared, respectively. and

,

is the pixel average of each target shape

,

is the standard deviation of the target shape. After that, the selected shape information is classified by table through the data classification unit, and a DB is built with the classification results. In this regard, Figure 8 shows an example of the target shape and feature DB construction according to the present invention.

Prior to performing the similarity determination unit, shape information is obtained from the target detection unit, and a priority-based feature quantity similarity priority comparison unit based on the determined priority is performed by applying a similarity determination technique. Unlike the existing sequential comparison, if the similarity of the target to be compared is determined from the table that is judged to be similar, and the similarity is greater than or equal to the standard value, target matching is carried out with an early decision. If the criterion is not reached when the similarity reference value is compared, the procedure of performing the similarity comparison on the next ranking table as shown in FIG. 7 is repeatedly performed until the reference value target is derived. At this time, if the similarity criterion is not reached during the procedure and a global search is performed on the entire table, it is classified as a target that cannot be matched.

In the above, a method and apparatus for accelerating target identification through pre-shape discrimination in a sonar image according to the present invention, and more specifically, a method of improving the target classification unit speed by using shape and feature information for target recognition of a sonar image and a device for performing this has been described. Experiments and results for verifying the technical effects of the present invention are as follows.

In the present invention, the target classifier speed of the devised technique was evaluated through the priority-based early decision technique. In this regard, FIG. 9 shows an original image for evaluating the speed of the target classifier according to the present invention. As for the image, four types of images acquired from the sonar were made into a comprehensive image as shown in FIG. 9 and the experiment was conducted, and the experiment was performed based on the resolution: 1920x1080 image.

Meanwhile, FIG. 10 shows a shape detection binary image and a feature amount detection image according to the present invention. As shown in FIG. 10 , shape information (left) of the target can be obtained from the shape detection unit as a binary image, and the characteristic quantity (right) can be derived by performing the characteristic quantity detection unit.

Meanwhile, FIG. 11 is a table comparing the experimental results according to the present invention with the existing method in relation to the target classification method. As shown in FIG. 11, the experimental results showed that the speed of the target classification unit of the devised method was reduced compared to the existing techniques (SIFT, SURF, BRISK, BRIEF) for each technique: 172ms, 160ms, 172ms, and 167ms. expressed. That is, FIG. 12 is a graph comparing the experimental results according to the present invention with the existing method in relation to the target classification method.

According to the experimental results, if the present invention is applied to the target classification unit for target recognition, it is expected that various contributions will be possible in improving the target classification speed not only in sonar images but also in various images in the future.

Accordingly, according to at least one embodiment of the present invention, there is an advantage in that detection and classification of various targets in the sonar image can be performed at high speed to perform target recognition according to target matching.

In addition, according to at least one embodiment of the present invention, there is an advantage that it can be applied in various ways in improving the target classification speed not only in the sonar image but also in various images.

According to the software implementation, not only the procedures and functions described in this specification but also the design and parameter optimization for each component may be implemented as a separate software module. The software code may be implemented as a software application written in a suitable programming language. The software code may be stored in a memory and executed by a controller or a processor.

Claims (10)

a target detection process of performing shape detection, feature point extraction, and feature quantity detection for target detection with respect to the sonar image; and
a target classification process of classifying a target based on the shape detection, the feature point extraction, and the feature amount detection,
The target classification process is
a target shape and feature quantity collection process of collecting shape and feature quantity information from the target shape and feature quantity DB unit; and
a similarity determination process in which the similarity determination unit compares the shape information with the target table to select a shape priority, and performs a feature amount similarity comparison according to the shape priority;
In the similarity determination process, if the degree of similarity is greater than or equal to a reference value, target matching is performed;
In the target detection process, the shape detection unit detects shape information from the sonar image,
In the similarity determination process, the shape priority is selected by comparing the detected shape information with the target table,
The target table configuration part of the target shape and feature DB part configures a target table for a target selected for target recognition,
The order-based feature quantity similarity priority comparison unit of the similarity determination unit preferentially compares the order determined by the shape priority selection unit,
The rank-based feature amount similarity priority comparison unit does not use the coordinate value of the target, but information on the outer edge of the target having the highest shape similarity based on the average and standard deviation of each pixel including the outer edge of the target of the target shape Comparison is performed preferentially by
In the similarity determination process, if the degree of similarity by the target outer edge shape information is greater than or equal to a reference value, the target identification is performed using the coordinate value of the target. delete According to claim 1,
The target shape and feature amount DB unit builds a database (DB) based on the detected target shape and the feature amount information, a method for accelerating target identification through pre-shape discrimination. delete According to claim 1,
Target identification speed method through pre-shape discrimination, characterized in that the target table configuration part of the target shape and feature DB part configures a target table for a target selected for target recognition. 6. The method of claim 5,
The shape and feature data collection unit of the target shape and feature data DB unit collects the shape information and the feature data detected from the target for DB construction, characterized in that the target identification speed up method through pre-shape discrimination. 7. The method of claim 6,
The data classifying unit of the target shape and feature amount DB unit classifies the collected shape information and feature amount information to match the type of the target table. 4. The method of claim 3,
The shape similarity sequential comparison unit of the similarity determining unit, characterized in that the sequential comparison with respect to the shape information obtained from the built-up DB and the shape detection unit, target identification speeding method through pre-shape discrimination. 9. The method of claim 8,
The shape priority selection unit of the similarity determination unit selects a priority based on the similarity derived from the shape similarity sequential comparison unit, the method for speeding up target identification through pre-shape determination. delete

Images