Number counting device for bound bars
The use of Mask R-CNN for object detection in an imaging system allows for accurate and automatic counting of binding bars by identifying characteristic values, overcoming brightness and shape variations to ensure precise counting.
Patent Information
- Application Number
- JP2023085660
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-24
- Publication Date
- 2024-12-05
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Conventional methods for counting binding bars require visual inspection and are prone to errors due to variations in brightness and shape caused by factors like paint coating, surface cutting unevenness, or attached labels, making accurate identification of end faces difficult.
An imaging system using Mask R-CNN for object detection to identify end faces of bundled bars by calculating characteristic values such as mask region area, center-to-center distance, and minimum distance to the outer perimeter, excluding non-end face images based on tolerance values.
Accurately recognizes end faces of bars without visual inspection, enabling reliable automatic counting.
Smart Images

Figure 2024168743000001_ABST
Abstract
Description
[Technical field]
[0001] The present invention relates to a rod counting device that counts the number of rods (including wire rods) bundled together with a belt or wire. [Background technology]
[0002] Patent Document 1 shows an example of an apparatus for counting the number of bundled rods. In this apparatus, illumination light is applied to the end faces of multiple bundled rods, which are photographed by a camera, and the photographed image is binarized using a predetermined brightness threshold. Island regions (lands) in the binarized image with characteristic values such as the center of gravity coordinates, maximum radius, and peripheral length within a predetermined range are then judged to be normal lands, i.e., images of rod end faces, and the number of rods is counted. The above judgment is repeated while appropriately changing the threshold value to count the number of rods, and the end face images of the counted rods are deleted from the binarized image, and the remaining end face images are visually confirmed and counted. [Prior art documents] [Patent documents]
[0003] [Patent Document 1] Patent Publication 2012-173901 Summary of the Invention [Problem to be solved by the invention]
[0004] However, with the above-mentioned conventional devices, although the burden is reduced, visual counting is still required, and the brightness of the actual end faces of bar stock varies greatly depending on factors such as the inclination caused by paint coating or uneven surface cutting, and whether or not a label is attached, making it difficult to accurately identify the end faces of bar stock from the brightness threshold value alone.
[0005] Therefore, the present invention is intended to solve such problems and aims to provide a device for counting the number of bundled rods that can accurately recognize the end faces of the rods without the need for visual inspection and automatically perform reliable number measurement. [Means for solving the problem]
[0006] In order to achieve the above object, the present invention provides a device for counting the number of bundled rods, the device comprising: an imaging means (2) for obtaining an overall image (2a) of a plurality of bundled rods (4) photographed from the end face side; an object detection AI unit (31) for obtaining end face images (4a) of the rods (4) from the obtained overall image (2a); and a number confirmation unit (32) for detecting predetermined characteristic values for each obtained end face image (4a) and counting the number of end face images (4a) whose characteristic values are within allowable values as the number of the rods (4).
[0007] The object detection AI unit (31) may be configured with Mask R-CNN. In this case, the characteristic value may be at least one of the area of the mask region (44) in each of the end face images (4a) generated by Mask R-CNN, the distance between the centers of gravity of adjacent mask regions (44), and the minimum distance between the center of gravity and the periphery of the mask region (44).
[0008] According to the present invention, even if the brightness and shape of the bar end faces vary widely, the object detection AI unit can obtain end face images of each bar. Then, predetermined characteristic values are detected for each end face image, and only end face images in which these characteristic values are within the allowable values are regarded as actual end face images of the bar. Therefore, for example, end face images of labels attached to the end faces are excluded, and the number of bars can be accurately calculated only from the end face images of the bar.
[0009] The reference characters in parentheses indicate, for reference, the correspondence with specific means described in the embodiments described later. Effect of the Invention
[0010] As described above, the device for counting the number of bundled rods of the present invention can accurately recognize the end faces of the rods without the need for visual inspection, and can automatically and reliably count the number of rods. [Brief description of the drawings]
[0011] [Figure 1] FIG. 2 is a diagram showing the overall configuration of the number counting device. [Diagram 2] FIG. 2 is a perspective view of the end surface of bundled round steel bars. [Diagram 3] FIG. 1 is a diagram illustrating the conceptual configuration of Mask R-CNN. [Figure 4] This is a flowchart showing the learning process of Mask R-CNN. [Diagram 5] FIG. 13 is a diagram illustrating an example of training data. [Figure 6] 10 is a flowchart showing a processing procedure of a processing device of the bottle counting device. [Figure 7] FIG. 2 is a diagram showing an example of an end face image of a round steel bar. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0012] It should be noted that the embodiment described below is merely an example, and various design improvements made by those skilled in the art without departing from the gist of the present invention are also included in the scope of the present invention.
[0013] 1 shows the overall configuration of the bar counting device E, which includes an illumination device 1 that illuminates the end faces of multiple bundled bars (round steel bars in this embodiment), and a camera 2 that photographs the illuminated bars from the end face side to obtain an overall image, and overall image data 2a is sent from the camera 2 to a processing device 3. The processing device 3 has a built-in computer, and an object detection AI unit 31 and a number confirmation unit 32, which will be described in detail below, are realized by the computer.
[0014] The object detection AI unit 31 identifies the end face images of each bar from the overall image 2a, as described below, and the number confirmation unit 32 calculates the characteristic values of the identified end face images and calculates the number of end face images whose characteristic values are within the allowable values, which is used as the number of bundled bar materials.
[0015] FIG. 2 shows a perspective view of the end face of the round steel bar 4 as a bundled bar material. Table 1 shows an example of the specifications of the round steel bar 4. As is clear from Table 1, the product diameter, product temperature, product length, and number of products to be bundled of the round steel bar 4 vary, and the brightness and shape of the product end face vary widely depending on the heat treatment conditions, the presence or absence of a label, the chamfering or face cutting method, the paint coating color, etc. Table 2 shows an example of the type of heat treatment of the round steel bar 4, the presence or absence and number of labels attached, the type of chamfering or face cutting, the presence or absence of paint coating, and the paint coating color. In the example shown in FIG. 2, a label 41 is attached to the end face of all (all) of the round steel bars 4.
[0016] [Table 1]
[0017] [Table 2]
[0018] In the overall image 2a captured of the bundled end faces of such a round steel bar 4, the brightness varies greatly from end face to end face, or even from part to part within the same end face, making it difficult to accurately identify the end faces using the conventional method of binarizing the image captured by a camera at a predetermined brightness threshold.
[0019] Therefore, as described above, the number counting device E is provided with an object detection AI unit 31 to identify the end face image from the overall image 2a, and in this embodiment, the object detection AI unit 31 is a Mask R-CNN (Convolutional Neural Network).
[0020] The conceptual configuration of Mask R-CNN is shown in Figure 3. In Figure 3, 311 is a convolutional layer that inputs the entire image, 312 is a feature map output from the convolutional layer 311, 313 is a region proposal network (RPN) that sets object regions on the feature map, 314 is a feature map of a fixed size for each object region, 315 is a fully connected layer, and 316 is a mask generation layer. When the entire image 2a is input, Mask R-CNN generates an end-face image in which a bounding box, its class, and the mask region are displayed on the entire image 2a, as described below.
[0021] The learning process of Mask R-CNN is shown in Figure 4. First, the required amount of training data is created (step 101). An example of the training data is shown in Figure 5. The training data is obtained by enclosing the periphery of the approximately circular original end face images 42b of each round steel bar in the overall image 2a with a paragon bounding box 43 consisting of multiple points and straight lines connecting these, and the original end face images 42 within each paragon bounding box 43 are assigned the class "marubo".
[0022] In step 102 of Figure 4, transfer learning is performed using training data, i.e., the head part of the existing trained Mask R-CNN (see Figure 3) is modified and its parameters are updated. In the following step 103, the overall parameters including the backbone part and RPN are updated to prepare a new trained Mask R-CNN (object detection AI unit 31).
[0023] The processing procedure of the processing device 3 of the number-counting device E is explained with reference to the flowchart in Figure 6. In step 201, the end faces of multiple bundled round steel bars 4 (Figure 2) are photographed to obtain an overall image 2a. Next, product information such as the product dimensions of the round steel bars 4 and the number of bars to be bundled is obtained from the upstream process (step 202), and based on this, characteristic values such as the end face (cross-sectional) area of the round steel bars 4, the distance between the centers of gravity (centers) of adjacent round steel bars 4, and the minimum distance (radius) between the center of gravity and the outer periphery are calculated in step 203, and allowable values are set.
[0024] In step 204, the trained Mask R-CNN of the object detection AI unit 31 obtains end face images 4a of each round steel bar 4 from the overall image 2a. An example of the end face image 4a is shown in FIG. 7. In each end face image 4a, a substantially circular mask region 44 corresponding to the end face of the round steel bar 4 is partitioned by a rectangular bounding box 45, and the class "marubo" is displayed probabilistically in each bounding box 45. The trained Mask R-CNN reliably identifies the end face of the round steel bar 4, which has various changes in brightness and shape, as the mask region 44 (instance segmentation).
[0025] However, when a label 41 (Figure 2) is attached to the end face of a round steel bar 4, Mask R-CNN may erroneously recognize the attached label 41 as an end face image 4c having an approximately circular mask area 48 defined by a rectangular bounding box 47, as shown in Figure 7.
[0026] 6, an analysis is performed to determine whether or not characteristic values, such as the area of the mask region 44, 48, the distance between the centers of gravity of the adjacent mask regions 44, 48, and the minimum distance between the center of gravity and the outer periphery of the mask region 44, 48, for each end face image 4a, 4c are within the allowable values set in the above step 203. If the allowable values are exceeded, the end face image 4c to which the mask region 48 belongs is deemed not to be an end face image 4a corresponding to the end face of the round steel bar 4, and is excluded from the calculation of the number of end face images 4a. The number of remaining end face images 4a is then calculated and this is set as the number of bundled round steel bars 4 (step 206).
[0027] In step 207, the calculated number of round steel bars is compared with the number of bars to be bundled, and if they match, it is deemed normal, but if they do not match, it is deemed an abnormality and a notification is issued to that effect.
[0028] The characteristic value may be at least one of the areas of the mask regions 44, 48, the distance between the centers of gravity of the adjacent mask regions 44, 48, and the minimum distance between the centers of gravity and the outer periphery of the mask regions 44, 48. The bar material is not limited to round steel bars. [Explanation of symbols]
[0029] 2... camera (imaging means), 2a... overall image, 3... processing device, 31... AI unit for object detection, 32... number confirmation unit, 4... round steel bar (bar material), 4a... end face image.
Claims
1. A device for counting the number of bundled rods comprising: an imaging means for obtaining an overall image of multiple bundled rods photographed from the end face side; an object detection AI unit for obtaining images of each end face of the rods from the obtained overall image; and a number confirmation unit for detecting predetermined characteristic values for each obtained end face image and counting the number of end face images whose characteristic values are within acceptable values as the number of rods.
2. The device for counting the number of bundled rods as described in claim 1, wherein the object detection AI unit is composed of Mask R-CNN.
3. The device for counting the number of bundled rods described in claim 2, wherein the characteristic value is at least one of the area of the mask region in each end face image generated by the Mask R-CNN, the distance between the centers of gravity of adjacent mask regions, and the minimum distance between the center of gravity and the outer periphery of a mask region.
Citation Information
Patent Citations
Method and device for counting number of steel material in bound steel material bundle
JP2012173901A
