KR101991186B1 - Bar detection method and apparatus - Google Patents

Abstract

A reinforcing bar detection method and an apparatus thereof are disclosed. The reinforcing bar detection method comprises the steps of: (a) receiving a reinforcing bar image; (b) generating a binary image of a reinforcing bar region by applying the reinforcing bar image to a learned random forest model; (c) generating a super pixel from the reinforcing bar image, and generating a reinforcing bar cross-sectional image by classifying the super pixel into the reinforcing bar region and a background region by using the binary image; and (d) detecting an individual reinforcing bar in the reinforcing bar cross-sectional image to detect the quantity of the reinforcing bar.

Description

Technical Field [0001] The present invention relates to a bar detection method and apparatus,

The present invention relates to a method and apparatus for detecting a reinforcing bar.

In order to know the quantity of rebar that is brought in at the construction site, it must be measured by human eyes. Reinforcing bars are usually packed in a bundle of 100 ~ 200 pieces, and multiple bundles are carried in the vehicle. It is difficult to ensure accurate accuracy due to its simple and highly concentrated nature.

 Therefore, in order to accurately and quickly measure the amount of rebar that is brought into the construction site, it is necessary to develop a system capable of automatically analyzing the images taken of the reinforced bar taken in the field. Conventionally, since the environment for photographing a reinforced bar image is constantly maintained by installing a dark film and an auxiliary lighting device in a reinforcing bar manufacturing factory, it is difficult to apply it to a construction site where the characteristics of the image are remarkably changed by photographing position, time, .

The present invention provides a reinforcing bar detecting method and apparatus capable of efficiently detecting the amount of reinforcing bars at a construction site.

In addition, the present invention provides a reinforcing bar detecting method and apparatus capable of real-time confirmation of the amount of reinforcing bars in various construction sites.

According to an aspect of the present invention, there is provided a method of detecting a reinforcing bar capable of efficiently detecting the amount of reinforcing bars at a construction site.

According to an embodiment of the present invention, there is provided a method of manufacturing a reinforcing bar, comprising: (a) receiving a reinforcing bar image; (b) generating a binary image for a reinforcing bar region by applying the reinforcing bar image to a learned random forest model; (c) generating a superpixel from the reinforcing bar image, and classifying the superpixel into a reinforcing bar area and a background area using the binary image to generate a reinforcing bar sectional image; And (d) detecting each of the individual reinforcing bars in the cross-sectional image of the reinforcing bars to detect the amount of reinforcing bars.

In the step (b), the rebar image is applied to the learned random forest model so that a pixel value of a pixel corresponding to a reinforcing bar area is designated as a first value, and a pixel value of a pixel corresponding to the background is designated as a second value So that a binary image can be generated.

The step (c) may include counting the number of pixels of the reinforcing bar area corresponding to the reinforcing bar area among the pixels included in the super pixel based on the binary image; And classifying the super pixel as a reinforcing area if the number of the counted number of reinforcing bar area pixels is equal to or greater than a reference number of pixels included in the super pixel, and classifying the super pixel as a background area if the number of reinforcing area pixels is less than the reference number Thereby generating a reinforcing section image.

In the step (d), the central points of the reinforcing bars may be respectively detected from the sectional images of the reinforcing bars, and the individual reinforcing bars may be detected to detect the amount of reinforcing bars.

The step (d) may include inverting the cross-sectional image of the reinforcing bar; Detecting a boundary of each reinforcing bar in the inverted reinforcing bar image; Accumulating the voting value in the voting area based on the boundaries of the respective bars; And detecting the position of the individual reinforcing bars and the amount of reinforcing bars by detecting a pixel having the highest value in the voting area as a center point of each reinforcing bar.

The voting area may include at least one of a starting point, a voting direction, a cone template length, and an angle for a cone template width, and the starting point may be set as a boundary of the reinforcing bar.

The voting value may be calculated by accumulating values reflecting the Gaussian distribution weighting along the voting direction about the starting point.

According to an aspect of the present invention, there is provided a reinforcing bar detecting apparatus capable of efficiently detecting a rebar quantity at a construction site.

According to an embodiment of the present invention, there is provided an apparatus comprising: a memory for storing at least one instruction; And a processor for executing the instructions stored in the memory, the instructions executed by the processor comprising the steps of: (a) receiving a rebar image; (b) generating a binary image for a reinforcing bar region by applying the reinforcing bar image to a learned random forest model; (c) generating a superpixel from the reinforcing bar image, and classifying the superpixel into a reinforcing bar area and a background area using the binary image to generate a reinforcing bar sectional image; And (d) detecting each of the individual reinforcing bars in the cross-sectional image of the reinforcing bars to detect the amount of reinforcing bars.

By providing the method and apparatus for detecting a reinforcing bar according to an embodiment of the present invention, it is possible to efficiently detect the amount of reinforcing bars at a construction site.

In addition, the present invention can confirm the quantity of reinforcing bars at various construction sites in real time.

Accordingly, the present invention has an advantage that it is easy to grasp the rebar inventory.

1 is a flowchart illustrating a method of detecting a reinforcing bar according to an embodiment of the present invention.
FIG. 2 is a view showing a superpixel and a binary-shape reinforced cross-sectional image according to an embodiment of the present invention; FIG.
3 is a diagram illustrating an algorithm for extracting a reinforcing bar area from a reinforcing bar image according to an embodiment of the present invention.
4 is a view illustrating a result of detecting a center point of a steel bar according to an embodiment of the present invention.
5 is a diagram illustrating a bar center point detection algorithm according to an embodiment of the present invention.
6 is a block diagram schematically illustrating the internal structure of a reinforcing bar detecting apparatus according to an embodiment of the present invention.

As used herein, the singular forms "a", "an" and "the" include plural referents unless the context clearly dictates otherwise. In this specification, the terms "comprising ", or" comprising "and the like should not be construed as necessarily including the various elements or steps described in the specification, Or may be further comprised of additional components or steps. Also, the terms "part," " module, "and the like described in the specification mean units for processing at least one function or operation, which may be implemented in hardware or software or a combination of hardware and software .

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

FIG. 1 is a flowchart showing a method of detecting a reinforcing bar according to an embodiment of the present invention. FIG. 2 is a view showing a superpixel and a binarized cross-sectional image according to an embodiment of the present invention. FIG. 4 is a view illustrating a result of detecting a center point of a steel bar according to an embodiment of the present invention. FIG. 5 is a view illustrating an example of a method of extracting a steel bar area according to an embodiment of the present invention FIG. 3 is a view showing a bar center point detection algorithm according to an embodiment of the present invention.

In step 110, the reinforcing bar detecting apparatus 100 acquires the reinforcing bar image. Here, the reinforcing bar image is taken at the construction site and may include various objects as a background.

In step 115, the reinforcing bar detecting apparatus 100 generates a super pixel from the reinforcing bar image. A superpixel represents a set (pixel group) of pixels having the same information. A superpixel represents a set of pixels with the same information.

For example, the reinforcing bar detecting apparatus 100 may generate a super pixel based on SLIC considering at least one of color and depth in a reinforcing bar image. The method of generating super pixels based on the SLIC itself will be obvious to those skilled in the art, so a detailed description thereof will be omitted.

In step 120, the steel wire detecting apparatus 100 generates a feature image and a feature vector of the reinforcing bar image based on the learned random forest, and generates a binary image of the reinforcing bar area based on the feature image and the feature vector.

A method of learning a random forest in order to facilitate understanding and explanation will be briefly described.

Random forests are used for regression analysis such as classification, linear regression, etc. as an ensemble learning algorithm. Random forest is a set of multiple classification trees. After generating several decision trees from given data, the decision trees synthesize the class labels inferred by each decision tree and output the most inferred class as a result.

Decision trees are tree-shaped decision models. The tree is extended from the root node to the child node in order to define the partition function to create one decision tree. In the Random Forest, the feature that best distinguishes the input data is not used in the partition function but is selected randomly instead. Assignment to partition function reduces learning time and avoids data overfitting problem. These partition functions are learned from the input data so that the class of the input data can be best separated through the characteristic to be examined by the user.

For random forest learning, it is necessary to have a data set that contains information about the rebar images taken at the construction site and which pixels correspond to the reinforced area. Therefore, feature information having the same size is extracted for the reinforcing bars of each data set for random forest learning.

According to an embodiment of the present invention, feature information can be extracted using various known methods such as a Sobel and Gaussian algorithm for extracting feature information from a reinforcing bar image. , But is not limited to a specific algorithm.

Depending on the method of extracting the feature information, the feature information may include various information. For example, it is assumed that a Sobel algorithm is used to extract feature information of a reinforcing bar image. The feature information may include edge information for the reinforcing bar area. As another example, suppose a Gaussian algorithm is used to extract feature information. The feature information may include texture information with reduced noise. As described above, the feature information on the reinforcing bar region of the reinforcing bar image may be different according to various algorithms used.

After generating a data set including feature information on the reinforcing bar image and the reinforcing bar area, the reinforcing bar detecting apparatus 100 generates a feature vector by randomly selecting pixels in the data set. Here, the feature vector is formed of a vector using feature information values of positions corresponding to one pixel. The individual values constituting the vector will be referred to as characteristics. When a feature vector is generated, a decision tree is generated to classify the feature vectors, and a random forest is learned. When the random forest learning ends, the generated tree information is stored in the random forest model. The stored random forest model may contain information necessary for feature vector generation and partition functions necessary for generating a decision tree.

Thus, the reinforcing bar detecting apparatus 100 generates a random forest based on the stored random forest model. The reinforcing bar detecting apparatus 100 generates feature information and a feature vector for all the pixels of the reinforcing bar image based on the generated random forest, classifies the reinforcing bar area, and sets the pixel value of the pixel corresponding to the reinforcing bar as a first pixel value (For example, white), and a pixel value of a pixel corresponding to the background is set to a second pixel value (for example, black) to generate a binary image.

In step 125, the reinforcing bar detecting apparatus 100 generates a binary image in which pixels corresponding to the cross-section of the reinforcing steel are separated (hereinafter, referred to as reinforcing cross-sectional image for convenience) using the binary image.

For example, the reinforcing bar detecting apparatus 100 counts the number of pixels included in the reinforcing bar area or the background area for each super pixel included in the super pixel list. Next, when the number of pixels corresponding to the reinforcing bar area is greater than or equal to half of each super pixel, the reinforcing bar detecting device 100 can classify the corresponding super pixel as the reinforcing bar area. However, if the number of pixels corresponding to the background area is more than half of the pixels included in each super pixel, the reinforcing bar detecting apparatus 100 can classify the corresponding super pixel as the background area.

Therefore, the reinforcing bar detecting apparatus 100 can output a binary image (i.e., a reinforced cross-sectional image) that is a result of classifying each super pixel into a reinforcing bar area and a background area based on the binary image.

The algorithm for this is shown in FIG.

In addition, the results of the binary image (i.e., the reinforced cross-sectional image) obtained by classifying each super-pixel into the reinforcing area and the background area are as shown in FIG.

In step 130, the reinforcing bar detecting apparatus 100 detects the amount of reinforcing bars by detecting individual reinforcing bars in the reinforcing frame sectional images.

This will be described in more detail.

The reinforcing bar detecting apparatus 100 can detect the position and the quantity of the individual reinforcing bars by detecting the center points of the reinforcing bars in each of the reinforcing frame sectional images. For this purpose, the reinforcing bar detecting apparatus 100 uses a voting algorithm.

As described above, the reinforced cross-sectional image according to one embodiment of the present invention is a binary image, and the value of each pixel is set to black or white. Therefore, it is possible to perform high-speed computation without detecting the gradients of all the pixels as in the prior art in order to detect the boundaries of the reinforcing bars.

That is, the reinforcing bar detecting apparatus 100 according to an embodiment of the present invention can detect the boundaries of each reinforcing bar simply by reversing each pixel value of the reinforcing frame sectional image, which is a binary image.

)으로 정의할 수 있다. 이어, 철근 검출 장치(100)는 보팅 후보 픽셀 집합에 포함된 각 픽셀에 대해 초기 보팅 방향을 계산할 수 있다. 이때, 초기 보팅 방향은 소벨 오퍼레이터의 근사 미분 계수를 보통 후보 픽셀 집합에 포함된 픽셀에만 적용해 계산될 수 있다.For example, the reinforcing bar detecting apparatus 100 may refer to a pixel position where a value of each pixel is greater than zero in an inverted reinforced cross-section image,

). Next, the reinforcing bar detecting apparatus 100 can calculate the initial voting direction for each pixel included in the voting candidate pixel set. At this time, the initial voting direction can be calculated by applying the approximate differential coefficient of the Sobel operator to only the pixels included in the ordinary candidate pixel set.

), 콘 템플릿의 너비를 위한 각(angle,

)으로 정의된다. Next, the reinforcing bar detecting apparatus 100 accumulates the voting values in the voting area for each of the pixels included in each voting candidate pixel set. Here, the voting area includes a start point, a voting direction, a length of a cone template (

), The angle for the width of the cone template (angle,

).

That is, the reinforcing bar detecting apparatus 100 may calculate the voting value by performing voting on the voting area of each pixel included in the voting candidate pixel set. The reinforcing bar detecting apparatus 100 may calculate and accumulate the voting value while updating the voting direction in the direction of the center point in the voting area by setting each pixel included in the voting candidate pixel set to the voting start position.

At this time, the voting value can be calculated by reflecting the two-dimensional Gaussian distribution weighting at all positions in the cone template region.

This can be expressed by the following equation (1).

는 가우시안 가중치를 나타내며,

는 표준편차를 나타낸다. Here, x and y represent each pixel included in the voting candidate pixel set. Also, u and v represent included pixels in the voting area. C is a constant,

Represents a Gaussian weight,

Represents the standard deviation.

The reinforcing bar detecting apparatus 100 updates the voting direction so as to indicate a point where the cumulative voting value in the voting area is the maximum when all the starting points (i.e., each pixel included in the voting candidate pixel set is once completed) 100) performs a voting process of accumulating values reflecting the two-dimensional Gaussian distribution weights at all positions in the cone template region after locating the cone template (cone shape) region from the reinforcing bar boundary to the center point direction to detect the center point of the reinforcing bar do.

보다 큰 위치(픽셀)들을 철근의 중심점으로 반환할 수 있다. 이를 통해 철근 단면 영상에서 철근 중심점을 검출함으로써 철근 위치 및 철근 수량을 검출할 수 있다. 도 4에는 철근 검출 장치에 의해 검출된 각 철근 중심점 위치를 나타낸 일 예가 도시되어 있다. The voting image (V) accumulates the voting value and has the same dimension as the input image (the reinforcing bar image). When the voting process is completed, the reinforcing bar detecting apparatus 100 determines whether the voting value (V)

Larger locations (pixels) can be returned as the center point of the rebar. It is possible to detect the position of reinforcing bars and the amount of reinforcing bars by detecting the center point of the reinforcing bars in the reinforcing frame images. FIG. 4 shows an example of the position of each center point of the reinforcing bars detected by the reinforcing bar detecting device.

FIG. 5 shows an algorithm for detecting the position and amount of individual reinforcing bars by detecting the center point of the reinforcing bars based on the reinforcing frame sectional image, which is a binarized image.

FIG. 6 is a block diagram schematically showing the internal structure of a steel bar detecting apparatus according to an embodiment of the present invention.

Referring to FIG. 6, a steel reinforcing bar 100 according to an embodiment of the present invention includes a memory 610 and a processor 615.

The memory 610 stores at least one instruction.

The processor 615 is a means for detecting individual reinforcing bars in the reinforcing bars image and grasping the quantity thereof by executing the instructions stored in the memory 610 in cooperation with the memory 610. [

The method of grasping the position of the individual reinforcing bars and the amount of the reinforcing bars is the same as that already described with reference to FIG. 1, so that a duplicate description will be omitted.

In addition, embodiments of the present invention may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions recorded on the medium may be those specially designed and constructed for the present invention or may be available to those skilled in the art of computer software. Examples of computer-readable media include magnetic media such as hard disks, floppy disks and magnetic tape; optical media such as CD-ROMs and DVDs; magnetic media such as floppy disks; Examples of program instructions, such as magneto-optical and ROM, RAM, flash memory and the like, can be executed by a computer using an interpreter or the like, as well as machine code, Includes a high-level language code. The hardware devices described above may be configured to operate as one or more software modules to perform operations of one embodiment of the present invention, and vice versa.

As described above, the present invention has been described with reference to particular embodiments, such as specific elements, and limited embodiments and drawings. However, it should be understood that the present invention is not limited to the above- Various modifications and variations may be made thereto by those skilled in the art to which the present invention pertains. Accordingly, the spirit of the present invention should not be construed as being limited to the embodiments described, and all of the equivalents or equivalents of the claims, as well as the following claims, belong to the scope of the present invention .

100: Rebar detection device
610: Memory
615: Processor

Claims (9)

(a) receiving a reinforcing bar image;
(b) generating a binary image for a reinforcing bar region by applying the reinforcing bar image to a learned random forest model;
(c) generating a superpixel from the reinforcing bar image, and classifying the superpixel into a reinforcing bar area and a background area using the binary image to generate a reinforcing bar sectional image; And
(d) detecting the number of reinforcing bars by detecting the respective central points of the reinforcing bars from the sectional images of the reinforcing bars to detect the individual reinforcing bars, respectively.
The method according to claim 1,
The step (b)
The rebar image is applied to the learned random forest model so that a pixel value of a pixel corresponding to a reinforcing bar area is designated as a first value and a pixel value of a pixel corresponding to the background is designated as a second value to generate a binary image Wherein the reinforcing bars are formed by a plurality of reinforcing bars.
The method according to claim 1,
The step (c)
Counting the number of pixels of the reinforcing bar area corresponding to the reinforcing bar area among the pixels included in the super pixel based on the binary image; And
Classifying the super pixels into a reinforcing area if the number of pixels of the counted reinforcing bar area is greater than or equal to a reference number among the pixels included in the super pixels and classifying the super pixels into a background area if the number of reinforcing bar pixels is less than the reference number And generating a reinforcing cross-sectional image.
delete The method according to claim 1,
The step (d)
Inverting the reinforced cross-sectional image;
Detecting a boundary of each reinforcing bar in the inverted reinforcing bar image;
Accumulating the voting value in the voting area based on the boundaries of the respective bars; And
Detecting a position of each reinforcing bar and a number of reinforcing bars by detecting a pixel having the highest value in the voting area as a center point of each reinforcing bar.
6. The method of claim 5,
Wherein the voting area includes at least one of a starting point, a voting direction, a cone template length, and an angle for a cone template width,
Wherein the starting point is set as a boundary of the reinforcing bars.
The method according to claim 6,
Wherein the voting value is calculated by accumulating a value reflecting the Gaussian distribution weighting along the voting direction about the starting point.
A computer-readable recording medium having recorded thereon a program code for performing the method according to any one of claims 1 to 3 and 5 to 7.
A memory for storing at least one instruction; And
A processor for executing instructions stored in the memory,
The instructions executed by the processor,
(a) receiving a reinforcing bar image;
(b) generating a binary image for a reinforcing bar region by applying the reinforcing bar image to a learned random forest model;
(c) generating a superpixel from the reinforcing bar image, and classifying the superpixel into a reinforcing bar area and a background area using the binary image to generate a reinforcing bar sectional image; And
(d) detecting the number of reinforcing bars by detecting the respective central points of the reinforcing bars from the sectional images of the reinforcing bars to detect the individual reinforcing bars, respectively.

Images