KR101071144B1 - Automatic grading method of lumber - Google Patents

Abstract

The present invention relates to an automatic classification method of lumber grade, and captures and combines images of lumber using an existing shape recognition system, and removes the background (conveyor belt) to detect knots, and then to the quality standards for each grade of lumber. Determine the grade accordingly.

According to the present invention, when determining the grade of lumber, it is possible to mechanically automate the grade determination in order to prevent an error that may occur due to human vision, and is being transferred using a low-cost area camera instead of an expensive line scan camera. Image processing of lumber is possible.

Lumber, grade, automatic, machine time

Description

AUTOMATIC GRADING METHOD OF LUMBER}

The present invention relates to an automated classification method of lumber grades that can improve the quality of products by automating the classification of lumber using machine vision.

The domestic timber industry, which relies on imports for most of its timber supply, needs to automate the wood processing process to increase timber yield and worker efficiency as raw material prices continue to rise.

Recognition of defects with the naked eye of a skilled worker, which is a method currently used, may change the decision criteria according to the psychological state of the worker or the degree of fatigue, resulting in a change in the consistency of work speed and quality.

Huber et al. (1985) measured the recognition rate of lumber surface defects by naked eyes for workers in a furniture factory, and measured 59 ~ 74% according to the skill of the workers. However, McMillin et al. (1984) reported that when the machine vision (machine vision) is used, the sound material discrimination rate is 99%, and the defect discrimination accuracy is 88%.

In the 1990's, line scan cameras (CCD line scan sensors) began to be used industrially, making it easier to input images during transport.

In addition, a dedicated processor for calculating image input and image processing algorithms has been developed for real-time image processing, which makes it possible to detect the shape and surface defects of lumber. Was used to.

The area camera is used to recognize the state of the object once stopped and to track the position of the object moving in the surveillance area, and the line scan camera is used for the inspection of the object having a small thickness displacement that is linearly moved at a constant speed.

Belt conveyors with more or less movement speed adjustment are inappropriate for input of distorted and uneven images when used with line scan cameras that require sophisticated movement speed control. Repeatedly causes a disadvantage in that the transfer efficiency is low.

In addition, there is a problem that a high level of equipment, such as a robot or a linear motor that can control the position of the camera when the object moves without moving the object is required.

Therefore, there is a need for a method capable of image processing of a lumber material being transferred using a low cost area camera that replaces an expensive line scan camera.

In order to solve the problems of the prior art, the present invention accurately processes the image of the lumber being transferred by using a low-cost area camera instead of the expensive line scan camera, and automatically detects the knot using the processed lumber image. The purpose is to provide an automated classification method for lumber grades for determining the grade of a grade.

In order to achieve the above object, the automatic classification method of the lumber grade of the present invention captures and combines the image of the lumber being transported with the area camera in real time with the area camera determined by considering the motion blur, and then detects the knot to automatically determine the grade of the lumber. You can judge.

The present invention relates to an automatic classification method of lumber grade, including capturing images of lumber being transferred in real time, and combining one or more images captured in real time, and including a background on the combined screen. Removing to display only the image of the lumber. The method may include detecting defects of knots in the image of the lumber, and determining the grade of the lumber based on the detected knot information based on the quality criteria for each grade of the lumber.

In the present invention, the transmission speed may be determined through the motion blur amount and the moving speed of the image.

In the present invention, combining the captured one or more images may combine the one or more images based on defects (feature points) present in the captured image.

In the present invention, a method for removing a background from the combined screen may include applying a red, green, and blue channel filter to the combined screen, and determining that a result value of the following equation according to the channel filter is 0 as a background. Can be removed.

Vr = ((R-G) + (R-G)), if then Vr <0 then Vr = 0

Vg = ((G-R) + (G-B)), if then Vg <0 then Vg = 0

Vb = ((B-R) + (B-G)), if then Vb <0 then Vb = 0

(Vr is the result of the red channel, Vg is the result of the green channel, Vb is the result of the blue channel, R is the pixel value of the red channel, G is the pixel value of the green channel, and B is the pixel value of the blue channel. )

Detecting the knot in the present invention includes the step of selecting as a threshold value through a threshold reduction method, and generating a binary image through the selected threshold value and labeling a region where a defect is detected, And detecting the knot according to information such as the number of pixels, the aspect ratio, the Y-axis length, and the circularity of the labeled region.

According to the present invention, as the grade of the lumber is determined using the machine vision irrespective of the decision criteria according to the psychological state and fatigue of the worker, there is an effect of improving the work efficiency by the consistent work speed and quality consistency.

In addition, using a low-cost area camera instead of an expensive line scan camera has the effect of reducing the process cost.

 Preferred embodiments of the present invention will be described with reference to the accompanying drawings. In adding reference numerals to components of the following drawings, it is determined that the same components have the same reference numerals as much as possible even if displayed on different drawings, and it is determined that they may unnecessarily obscure the subject matter of the present invention. Detailed descriptions of well-known functions and configurations will be omitted.

1 is a flowchart illustrating an automatic classification method of lumber grade according to an embodiment of the present invention.

Referring to Figure 1, the step of capturing the image (S100) is to capture the image of the lumber being conveyed at the feed rate in real time, the feed rate is in accordance with the amount and motion speed of the motion blur obtained using the test piece (lumber) Select.

The motion blur is caused by the relative movement between the camera and the test piece. When the motion of the test piece is faster than the operating speed of the camera shutter, it indicates that a blur or afterimage of an image is stored in the still image.

More specifically, if the object moves rapidly according to the characteristics of the camera, motion blur, which is an after-image of the object, is generated on the input image. A 25 * 30mm black rectangular pattern is printed on white paper and attached to the center of the conveyor belt. When the feed speed is changed, the size of the motion blur is also changed (see FIGS. 2A and 2B).

Therefore, when the moving speed is increased by adjusting the frequency of the inverter mounted on the conveyor, the motion blur generated in the square pattern is measured and the appropriate moving speed is selected based on the result.

Step S110 captures the image of the lumber being transported in step S100 in real time, and then combines the images to form a single lumber shape.

At this time, if the lumber is conveyed at a constant speed in the combination of the images, the image partially captured in step S100 may be collected at regular intervals, but an error occurs in combining the images by the irregular speed of the conveyor belt. .

Therefore, in the present invention, an image combining algorithm for determining the combined position of consecutive images based on the feature points (defects) of each partial image is applied to constitute an entire image.

The image combining algorithm composes the whole image through the process of entering the lumber, calculating the distance traveled by the lumber, extracting feature points, selecting preliminary corresponding regions, extracting corresponding points, combining images, passing through lumber, and ending the image combining.

The entry decision of the lumber is for determining that the lumber has entered the image input area, and when more than 1000 pixels whose red value is greater than or equal to the reference value among the pixels in the predetermined area are determined, the lumber is entered.

The hourly moving distance calculation of the lumber stores the position of the entry point of the lumber in the image, and then retrieves and stores the position of the lumber entry of the image. The image distance is calculated using the stored positions of the first image and the entrance of the second image, and the moving distance per hour is calculated using the image distance and the storage time in 0.01 seconds.

The feature extraction is performed using a feature extraction algorithm of Shi. Tomasi (1994) to detect feature points with strong corner components such as neck or knot in the current image and select feature regions based on the feature points.

The preliminary corresponding area selection is performed by calculating the distance of the lumber using the time difference of the lumber and the time difference between the current image and the next image. Select the preliminary response area for the next image.

The image combination extracts an area from FIG. 3B by an image distance of the feature point and the corresponding point of FIGS. 3A and 3B and combines the area into the storage area of the entire image.

The passage decision of the lumber is repeated from the feature point extraction to the image combining until the lumber passes through the surveillance region.

The image combining end is terminated when the lumber image is not recognized in the surveillance area.

In step S120, the background is removed, and the background representing the conveyor belt is removed from one image combined in step S110.

To remove the background, software filters of the red channel, the green channel, and the blue channel can be used. The point where each result value is 0 is determined as the background part in the following equation.

Vr = ((R-G) + (R-G)), if then Vr <0 then Vr = 0

Vg = ((G-R) + (G-B)), if then Vg <0 then Vg = 0

Vb = ((B-R) + (B-G)), if then Vb <0 then Vb = 0

(Vr is the result of the red channel, Vg is the result of the green channel, Vb is the result of the blue channel, R is the pixel value of the red channel, G is the pixel value of the green channel, and B is the pixel value of the blue channel. )

In more detail, when using the red channel filter, the green channel filter, and the blue channel filter to remove the background of the combined image, it can be seen that a mask image is generated as shown in FIGS. 4B to 4D.

FIG. 4A shows the combined image, FIG. 4B shows the background channel removed by the red channel filter, FIG. 4C the green panel filter, and FIG. 4D the blue channel filter.

In the present invention, the background is removed using the red channel filter with the cleanest background part removed.

In operation S130, the knot is detected through the process of threshold determination, binarization, labeling, and knot determination on the image of the lumber with the background removed in operation S120.

As a method of determining a threshold, a method of determining a threshold through histogram analysis, a method of using a pixel value distribution of an image, a method of using boundary line extraction, and a method of separating an entire region into partial regions and analyzing the threshold value may be used. As shown in FIGS. 5A to 5D, since blue and channel patterns are the most distinct among the channel images of the lumber, the threshold value can be determined using the blue channel.

In the present invention, the threshold reduction method is used as a method using the pixel value distribution of the image.

Looking at the threshold reduction method, in step 1, the average value of the blue channel frequency of the lumber image is selected as the primary threshold for knots. In step 2, a binary image is generated using a threshold value, and an open operation is performed on the black pixel group lower than the threshold value for noise removal and separation of adjacent knots. The operation is called open operation, and the algorithm for smoothing the outline of the pixel group or removing noise and separating adjacent groups is executed.

Step 3 specifies (labels) the number of independent black areas, such as knots, among the white and black areas of the binary image, and specifies the number of pixels (NoPix), vertical to aspect ratio (YXRatio), and Y-axis length (for each label area). It stores information such as YLength) and circularity (ELong).

In step 4, the label information is determined to be not knot, even if the number of pixels is 10 or less, the aspect ratio is 2.5 or more, the Y-axis length is 3 or less, and the circularity is 0.1 or less. Classify.

In step 5, if the number of mislabeled labels exceeds 500, it is determined as an inappropriate threshold value, and the value obtained by subtracting 3 from the threshold value is reassigned to the threshold value and the steps 2 to 5 are repeated. Because a high threshold value increases the frequency of detecting other characteristics such as grain, and eventually increases the number of mislabeled labels.

Therefore, in order to determine a threshold suitable for knot detection, the threshold value is gradually lowered until the number of misidentified labels becomes 500 or less, and when the final threshold value is determined, classification of the label detected by the knot is performed to obtain knot information.

Step S140 is a step of indicating the grade of the lumber according to the information (pixel number, aspect ratio, Y-axis length, circularity, etc.) detected in step S130 to the rating criteria of Table 1 to Table 3 below Applied to the grade classification program, the grade of lumber is automatically determined. (However, Table 1 shows one kind of structural member, Table 2 shows two kinds of structural member, and Table 3 shows three kinds of structural member.)

[Example]

1. As a conveying device, a belt Kenveyer with a belt length of 5.4m and a belt width of 60cm was manufactured, and an inverter capable of freely adjusting the conveying speed was mounted. In the illumination, two daylight fluorescent lamps (18W) were installed on the ceiling so that the color of the image would not be distorted. The image processing area camera was installed at a height of 80 cm from the conveyor belt, and the camera angle was kept perpendicular to the belt surface.

The camera (Qrio IPCam 300S) is a standard lens group with an angle of view of 58 °. The camera has a size of 545 * 405mm in the plane of 1m distance from the camera. The intensity value for and the color image is the intensity value for each channel of red, green, and blue).

2. The area camera was used to correct the distorted image for the images input to the lumber being transported.

3. The partial images were combined into one image for the corrected image (see Fig. 6A).

4. At this time, the combined image is rotated to be parallel to the conveying direction even when the lumber is inclined rather than parallel to the conveying direction, and the lumber image is removed by using a background removal filter to separate the lumber and the background from the combined image. Was extracted (see FIG. 6B).

5. The extracted image is converted into a binary image that is displayed only in black and white by applying a threshold reduction method, followed by knot detection (see FIG. 6C) to extract knot information from the detected knots. It was.

6. Using the extracted knot information, the grade of the lumber was determined using a program to which the rating criteria of Tables 1 to 3 were applied (see FIGS. 7A to 7B).

7A to 7B are screens showing lumber grades by an automatic classification method of lumber grades according to an embodiment of the present invention. Determine the grade of the final lumber by selecting.

Referring to Figure 7a shows the image generated by the image combining algorithm when transferred to a certain movement during the transmission of the lumber, Figure 7b shows the position of the lumber because the lumber is moved to the left and right when sent in the warp state of the lumber Also represents an image generated by the image combining algorithm even in the changed input image from time to time.

Those skilled in the art can change or modify the described embodiments without departing from the scope of the present invention, and such changes or modifications are within the scope of the present invention. In addition, the materials of each component described herein can be readily selected and coped by a variety of materials known to those skilled in the art. In addition, those skilled in the art may omit some of the components described herein without adding to the performance or add the components to improve performance. In addition, those skilled in the art may change the order of the method steps described herein depending on the process environment or equipment. Therefore, the scope of the present invention should be determined by the appended claims and equivalents thereof, not by the embodiments described.

1 is a flow chart illustrating an automatic classification method of lumber grade according to an embodiment of the present invention.

2a and 2b is a view showing the size of the motion blur according to the moving speed of the conveyor belt according to an embodiment of the present invention.

3A and 3B illustrate feature points and image distances for image combining according to an embodiment of the present invention.

4A to 4D are diagrams illustrating mask images passed through respective filters (red, blus, green) for background removal according to an embodiment of the present invention.

5A to 5D are views illustrating images of respective channels (red, blus, and green) for determining a threshold according to an embodiment of the present invention.

6a to 6c is a view showing an automatic classification method of lumber grade according to an embodiment of the present invention.

7A to 7B are screens showing a lumber lumber grade determination program by an automated classification method of lumber lumber grades according to an embodiment of the present invention.

Claims (5)

Capturing in real time an image of the lumber being conveyed at the feed speed; Combining one or more images captured in real time; Removing a background from the combined screen to display only the image of the lumber; Detecting knots in the image of the lumber; And Judging the grade of the lumber based on the detected knot information based on the quality criteria for each grade of the lumber. Combining the one or more images captured in real time, Combine the one or more images based on the defects present in the captured image, and determine by the predetermined image combining algorithm the entry of the lumber, the calculation of the distance traveled by the lumber per hour, the extraction of a predetermined feature point, and the selection of a predetermined preliminary corresponding region. Characterized in that it proceeds in the order of extraction of a predetermined corresponding point, image combining, pass judgment of lumber, and end of image combining, Method for removing a background from the combined screen, The red, green, and blue channel filters are applied to the combined screen, and the automatic classification method according to the grade of the lumber grade, characterized in that it is determined by the background value of the result of the following equation according to the channel filter is removed. Vr = ((R-G) + (R-G)), if then Vr <0 then Vr = 0 Vg = ((G-R) + (G-B)), if then Vg <0 then Vg = 0 Vb = ((B-R) + (B-G)), if then Vb <0 then Vb = 0 (Vr is the result of the red channel, Vg is the result of the green channel, Vb is the result of the blue channel, R is the pixel value of the red channel, G is the pixel value of the green channel, and B is the pixel value of the blue channel. ) The method of claim 1, wherein the transmission speed is, Automated classification method of lumber grade, characterized in that determined by the motion blur amount and the moving speed of the image. delete delete The method of claim 1, wherein detecting the ear canal comprises: Selecting a threshold through a threshold reduction method; Generating a binary image through a predetermined threshold and labeling a region where a defect is detected; And And detecting the knot according to the knot information of the number of pixels, the aspect ratio, the Y-axis length, and the circularity of the labeled region.

Images