- Email: [email protected]
The improved defects detection method of optical fiber winding
Optik 125 (2014) 675–678
Contents lists available at ScienceDirect
Optik journal homepage: www.elsevier.de/ijleo
The improved defects detection method of optical fiber winding Chenxia Guo ∗ , Ruifeng Yang School of Information and Communication Engineering, North University of China, Taiyuan city 030051, Shanxi province, China
a r t i c l e
i n f o
Article history: Received 6 March 2013 Accepted 5 July 2013
Keywords: Mathematical morphology Target detection algorithm Defects detection Optical fiber coil winding
a b s t r a c t The paper discusses mainly how to check the defects such as fiber-stacking fiber-cut and fiber-uneven with an improved method in the process of fiber optic gyroscope coil winding. In the paper, it is aimed at the gray level image of optic fiber coil winding to get binary image using mathematical morphology and to get optic fiber position image using the improved moving target detection algorithm, on the base of the optic fiber position image, to figure out the relative position of adjacent optic fiber. Through the value of the relative position of adjacent optic fiber, the status of optic fiber winding can be estimated. Experimental results show that the entire image processing and defect detection method can effectively distinguish the defects in the process of optical fiber coil winding. © 2013 Elsevier GmbH. All rights reserved.
1. Instruction Among the many factors contributing to FOG performance, the quality of the sensing coil is one of the most important [1]. Fiber placement is known to be important factor in winding a hign performance. The Sagnac effect relies on the fact that the shift in the phase of the light beams is due solely to the rotation of the coil. Fiber sensing coil must undergo a rigorous process of winding, that is, these fibers on the same layer or on the different layers should closely packed around, not to appear fiber-stacking (in Fig. 3), fibercut (in Fig. 4) and fiber-uneven (in Fig. 2), winding optic fiber like that can reduce the optical loss and depolarization caused by the environmental change [2]. One of the greatest challenges is to be able to wind an entire coil with few or no errors. Therefore, the winding process of the fiber sensing coil needs to be real-time detection, in order to trace and exclude the various defects that may occur in the process of optic fiber winding. Machine vision technology is applied to monitor the process of optic fiber winding in the paper. The images collected by machine vision are analyzed and processed by some algorithms to determine the status of the fiber winding. The machine vision technology is the use of optical devices for non-contact perception, automatic acquisition and interpretation of the image of a real scene, to get the information and control the machine [3]. The pretreated images are processed by using target detection algorithm to identify the detected target in the process of fiber winding. The position and status of the detected optical fiber is always changing, so the
∗ Corresponding author. Tel.: +86 13513634936; fax: +86 03513925235. E-mail addresses: [email protected] (C. Guo), [email protected] (R. Yang). 0030-4026/$ – see front matter © 2013 Elsevier GmbH. All rights reserved. http://dx.doi.org/10.1016/j.ijleo.2013.07.054
moving target detection method is applied to processing and detection of fiber winding images. In the applications environment of defect detection in fiber winding, image processing and target detection should be fast, accurate and robust. This paper applies mathematical morphology directly on the gray level image of the collected images by machine vision, which can greatly reduce image preprocessing time. On the base of traditional moving target detection algorithm, an improved algorithm is proposed, which is suitable for detecting the defect of fiber winding. After the images are processed by the above process, the images show the position of every fiber. On the base of the processed images, the fiber pixel location in difference images can be calculated by combining the pixel size of the chosen CCD industrial cameras with related position of fiber in qualified fiber coil to determine whether there are defects such as fiber-stacking, fiber-cut and fiber-uneven in the process of fiber winding. The clarity of the fiber winding images shows that the methods used in the paper are effective and correct. 2. The application and the basic idea of mathematical morphology Mathematical morphology is one kind of new method used in the field of image processing and pattern recognition [4]. The basic idea of using mathematical morphology for performing image processing is that there is a certain specific shape of the structural elements such as rectangle, circle, or diamond of a certain size detection target image, by checking the validity of the structural elements of caving and fill method in the target area of the image to obtain information about the image morphology, and thus achieve the purpose of analysis and recognition image. The morphology operation is a collection of object shape and the interaction between the structural elements, which is not sensitive
676
C. Guo, R. Yang / Optik 125 (2014) 675–678
optical flow field is proposed in reference [6], but due to the complexity of the calculation formula, under the conditions without special hardware support, it is difficult to achieve real-time requirements, so its poor application and poor real-time are not suitable in defect detection of fiber winding. There is a small amount of calculation in the inter-frame difference method, easy to implement, which use two or three adjacent frames of video sequence to get time difference threshold to extract the target. So it has strong adaptability in dynamic environment [7]. However, this method cannot completely extract the relevant characteristic pixel point, and the feature of depending on the selection time interval also becomes the drawback of this method. The method is characterized by the following formula:
Fig. 1. Regular status.
Fig. 2. Fiber-uneven.
Dk (x, y) = |Ik+1 (x, y) − Ik (x, y)| Fig. 3. Fiber-stacking.
Fig. 4. Fiber-cut.
to the edge direction, but which can detect the true edge of the image and the noise of the image can be largely suppressed [5]. So, the morphology operation has unique advantages in the description of the characteristics of the object in the image. Therefore, the mathematical morphology for edge detection can effectively filter out the noise and retain the original details in the image with better edge detection effect. This paper applies opening and closing operation of mathematical morphological to extract the outer contour of optical fiber winding image. The closing operation generally will narrow the gap linking and fill the hole smaller than the structural elements. The opening operation can remove the object area that cannot contain the structure elements and can remove the small protruding portions. The method is characterized by the following formula, where formula (1) is opening operation expression and formula (2) is closing operation expression. A ◦ B = (AB) ⊕ B
(1)
A • B = (A ⊕ B)B
(2)
The experimental results show that the application of mathematical morphology for processing Fig. 8, that is gray level image of original image of optic fiber winding to get binary image of optical fiber winding, that is Fig. 9. As can be seen from Fig. 9, the image fully meets the needs of the subsequent analysis. 3. Improved moving target detection algorithm for optical fiber winding 3.1. Traditional moving target detection algorithms Moving target detection algorithms are diverse. Every algorithm is different in its specific application environment. In a static context, traditional moving target detection algorithms are mainly optical flow field method, inter-frame difference method and background difference method. These moving target detection algorithms depend on the range of applications, each with its own advantages and disadvantages. Optical flow field method is represented by a two-dimensional image of the velocity field in the three-dimensional motion of the object point, which uses the optical flow characteristics, that is moving target changing over time to create the optical flow constraint equation for target detection, the calculation method of
(3)
where Dk (x, y), Ik+1 (x, y), and Ik (x, y) are respectively difference image, the k + 1 original image, and the k original image. The defect detection of fiber winding needs to detect the exact location of each optical fiber, so the inter-frame difference method is not suitable for the application environment. Like inter-frame difference method, the background difference method also extracts the target region using different image. But its difference to the inter-frame difference method is that the background difference method is not the difference image of the current frame image with the adjacent frame image, but is the difference image of the current frame image with a continuously updated background image [8]. The method is characterized by the following formula: Dx (x, y) = |Ik+1 (x, y) − Bk (x, y)|
(4)
where Dx (x, y), Ik (x, y), and Bk (x, y) are respectively the difference image, the current frame image, and the background image. In the defect detection of fiber winding, each frame image is different and the difference image of the adjacent frame image is not the detected target. So using background difference method for defect detection of fiber winding will cause larger detection error, it is not suitable to the detection environment of fiber winding defect. In view of the above analysis, the traditional target detection algorithms exist their shortcomings in the defect detection of optical fiber winding. The improved algorithm was forwarded for the particularity of the detected targets in the paper. 3.2. Improved moving target detection algorithm based on inter-frame difference method For fiber winding process, the interval of adjacent frames is very short; there is a strong correlation and continuity on adjacent frame image. Using the inter-frame difference method alone to detect target will be missed target. Due to the limits of fiber winding speed, there are more than two images on the interval of two fibers appearing on the same end respectively. For background difference method alone there are similar problems to inter-frame difference method, and the background difference method needs to constantly update the background image that causes the heavy workload. Therefore, in fiber winding, in order to obtain fast detection speed, according to the speed of optical fiber winding and the FPS of chosen CCD industrial cameras, the images of the input images collected by machine vision are sampled to meet accurately detected defect of fiber winding. Based on the basic principle of the inter-frame difference method, the images formed in the process of fiber winding will be re-combinated to form a new sequence of images, which are suitable for the application of defect detection and moving object detection method. There are lots of video images in the process of fiber winding, but in the judgment of the winding defect, need to select an image with a new and entire fiber on the end, as Fig. 5.
C. Guo, R. Yang / Optik 125 (2014) 675–678
677
three new image sequence with the inter-frame difference method for processing, as shown in formula (7) and formula (8). DK (x, y) = |EK (x, y) − EK−1 (x, y)|
(7)
DK+1 (x, y) = |EK+1 (x, y) − EK (x, y)|
(8)
where DK (x, y) and DK+1 (x, y) respectively are the difference images of the new image sequence K frame and the new image sequence K − 1 frame and the difference image of the new image sequence K + 1 frame and the new image sequence K frame. The flow chart of image preprocessing and target identification in fiber winding is shown in Fig. 5. 4. Defect detection method of fiber winding
Fig. 5. The flow chart of image preprocessing and target identification.
According to the speed of optical fiber winding, the paper selects appropriate time intervals such as 12 fps of the chosen industrial cameras to acquire images, from the acquired images to sample the images with the entire fiber information as the analysis of the reference images, these original images referred as I (x, y) are recorded as a collection of the following sequences: I(x, y) =
I1 (x, y), I2 (x, y), ·I3 (x, y), . . ., Ii (x, y)i = 1∼n
After the above processing, the output images display the pixel position of the single optical fiber at different times, combining pixel dimensions of chosen CCD industrial cameras with related position between fiber in qualified fiber coil (in Fig. 1) to determine fiber defect (in Figs. 2–4). The method is shown in Fig. 6. A defect is considered any place in the coil where there is a deviation from the pattern, which can degrade the performance of the coil produced. So we need to detect these defects during fiber winding. There are three threshold values, respectively referred to as T1, T2, and T3, which are set to determine category of fiber winding defect. The three threshold values are set according to pixel dimensions of chosen CCD industrial cameras with related position between fibers in qualified fiber coil. When the absolute difference of the minimum row of fiber pixels coordinating in the two difference images is greater than the threshold value T1, the judgment result is uneven in fiber winding; when the absolute difference of the minimum column of fiber pixels coordinating in the two difference images is greater than the threshold value T2, the judgment result is fiber-stacking in fiber winding; when the absolute difference of the minimum column and row of fiber pixel coordinating
(5)
From the above sequence, sampling i = 10, i = 20,. . . i.e. 10 integer multiples of the original images (this value is selected according to the speed of fiber winding and the FPS of chosen CCD industrial cameras) conduct a new sequence of image frames F (x, y), as the new sequence shown in the following sequences. F(x, y) =
F1 (x, y), F2 (x, y), ·F3 (x, y), . . ., Fi (x, y)j = 10i, ·i = 1∼n
(6)
In order to improve the quality of the detected image and the accuracy of detection, a new sequence of image frames need to be preprocessed for eliminating the irrelevant information included in the images and restoring useful authentic information, and enhancing the detected performance of relevant information and greatly simplifying data. The conventional methods of processing image include these steps such as graying, denoising, enhanced, edge sharpening, and binarization, then to extract specific information or certain features contained in the image for further processing using detecting algorithm, this paper uses mathematical morphology to greatly reduce the time of image pre-processing, and acquire clear target images. Preprocess the new image sequence, and applicate the improved moving target detection method to detect the position of fiber in the image. Here, the new image sequences are preprocessed as consecutive frames, to be written as: EK−1 , EK , and EK+1 , to the above
Fig. 6. Software method of the defect detection.
678
C. Guo, R. Yang / Optik 125 (2014) 675–678
between the two difference images is all close to T3, it is judged that a fiber breaks. When one of the above defects appears, the detection system alarms and returns to main program to wait for the further command. In Fig. 6, the input images are the two difference images processed by improved moving target detection algorithm. Fig. 12. The first difference image.
5. Implementation and results The proposed algorithm and detection method succeeds in processing the images of fiber winding and detecting the defects of fiber winding. In the paper, some original images of fiber-uneven in fiber winding and processed images are shown to verify the feasibility and accuracy of the proposed algorithm and detection method. The experimental platform is on Windows XP, Intel DualCore (2.2 GHz), and 2GB RAM. Fig. 7 is the original image which shows tightly coiled fiber and a single optical fiber on the next lay. The gray level image (in Fig. 8) is processed using the method of the mathematical morphology to get the binary image (in Fig. 9). The method of processing the images including two and three fibers on tightly coiled fiber layer is similar to the above method of processing Fig. 7 to get two binary images, as shown in Figs. 10 and 11. For the three binary images processed by using improved moving target detection algorithm to
Fig. 7. The original image.
Fig. 13. The second difference image.
get two difference images, as shown in Figs. 12 and 13. Fig. 12 is the difference image of Figs. 9 and 10. As can be seen from Fig. 10 and software calculation result, the fiber of Fig. 12 is the proper place on the fiber winding. Fig. 13 is the difference image of Figs. 10 and 11. As can be seen from Fig. 11 and software calculation result, the fiber of Fig. 13 is spaced-out from previous turn on the fiber winding. According to the clarity of the images that can be seen, using the mathematical morphology to preprocess images can clearly indicate the location of the single fiber. The application of improved moving target detection algorithm on the binary images can completely differentiate between current fiber and previous fiber. The results satisfy the needs of subsequent calculation. As can be seen from Fig. 11, the new fiber turn is spaced out close to one fiber from previous turn that is fiber-uneven. The result observed from Fig. 11 is consistent to the result calculated by the method as shown in Fig. 6. 6. Conclusion
Fig. 8. The gray level image.
Fig. 9. The first binary image.
Fig. 10. The second binary image.
Fig. 11. The third binary image.
Fiber sensing coil winding quality is an important factor in the accuracy of FOG. For real-time and accuracy requirements of the defect detection, this paper proposes mathematical morphology and improved moving target detection method to process images and combines software algorithm with pixel dimensions of chosen CCD industrial cameras and related position of qualified fiber turn to detect defect of fiber winding. According to the processed images and software calculation result, it is shown that the proposed image processing algorithm and defect detection method can fast and accurately distinguish fiber. Experimental results verify the feasibility of the algorithm and the method of defect detection in fiber winding. References [1] M.L. Stephen, Design and Development of an Automated Fiber Optic Gyroscope Coil Winding Machine, Massachusetts Institute of Technology, Cambridge, 1997, pp. 20–24, B.S. Mechanical Engineering. [2] Z. Ding, Study of Complete Quality Measurement for FOG Fiber Coils, TianJin University, 2010, pp. 3–6. [3] W.E. Snyder, H. Qi, Machine Vision, China Machine Press, 2005, pp. 1–2. [4] H. Nan, Edge detection using morphological algorithm, Comput. Simul. 29 (3) (2012) 288–291. [5] A.A. Purwita, K. Adityowibowo, et al., Automated microaneurysm detection using mathematical morphology, in: International Conference on Instrumentation, Communication, Information Technology and Biomedical Engineering, Bandung, Indonesia, 8–9 November, 2011. [6] I. dar, K. Janusz, Occlusion-aware optical flow estimation, IEEE Trans. Image Process. 17 (8) (2008) 1443–1451. [7] A. Ton, H. Fujiyoshi, R. Patil, Moving target classification and tracking from realtime video, in: Proceedings IEEE Workshop on Application of Computer Vision, Princeton, NJ, 8–14, 1998. [8] Taohui, D. Harwood, L. Davis, W4: Real-time surveillance of people and their activities, IEEE Trans. Pattern Anal. Mach. Intell. 22 (8) (2000) 809–830.
Contents lists available at ScienceDirect
Optik journal homepage: www.elsevier.de/ijleo
The improved defects detection method of optical fiber winding Chenxia Guo ∗ , Ruifeng Yang School of Information and Communication Engineering, North University of China, Taiyuan city 030051, Shanxi province, China
a r t i c l e
i n f o
Article history: Received 6 March 2013 Accepted 5 July 2013
Keywords: Mathematical morphology Target detection algorithm Defects detection Optical fiber coil winding
a b s t r a c t The paper discusses mainly how to check the defects such as fiber-stacking fiber-cut and fiber-uneven with an improved method in the process of fiber optic gyroscope coil winding. In the paper, it is aimed at the gray level image of optic fiber coil winding to get binary image using mathematical morphology and to get optic fiber position image using the improved moving target detection algorithm, on the base of the optic fiber position image, to figure out the relative position of adjacent optic fiber. Through the value of the relative position of adjacent optic fiber, the status of optic fiber winding can be estimated. Experimental results show that the entire image processing and defect detection method can effectively distinguish the defects in the process of optical fiber coil winding. © 2013 Elsevier GmbH. All rights reserved.
1. Instruction Among the many factors contributing to FOG performance, the quality of the sensing coil is one of the most important [1]. Fiber placement is known to be important factor in winding a hign performance. The Sagnac effect relies on the fact that the shift in the phase of the light beams is due solely to the rotation of the coil. Fiber sensing coil must undergo a rigorous process of winding, that is, these fibers on the same layer or on the different layers should closely packed around, not to appear fiber-stacking (in Fig. 3), fibercut (in Fig. 4) and fiber-uneven (in Fig. 2), winding optic fiber like that can reduce the optical loss and depolarization caused by the environmental change [2]. One of the greatest challenges is to be able to wind an entire coil with few or no errors. Therefore, the winding process of the fiber sensing coil needs to be real-time detection, in order to trace and exclude the various defects that may occur in the process of optic fiber winding. Machine vision technology is applied to monitor the process of optic fiber winding in the paper. The images collected by machine vision are analyzed and processed by some algorithms to determine the status of the fiber winding. The machine vision technology is the use of optical devices for non-contact perception, automatic acquisition and interpretation of the image of a real scene, to get the information and control the machine [3]. The pretreated images are processed by using target detection algorithm to identify the detected target in the process of fiber winding. The position and status of the detected optical fiber is always changing, so the
∗ Corresponding author. Tel.: +86 13513634936; fax: +86 03513925235. E-mail addresses: [email protected] (C. Guo), [email protected] (R. Yang). 0030-4026/$ – see front matter © 2013 Elsevier GmbH. All rights reserved. http://dx.doi.org/10.1016/j.ijleo.2013.07.054
moving target detection method is applied to processing and detection of fiber winding images. In the applications environment of defect detection in fiber winding, image processing and target detection should be fast, accurate and robust. This paper applies mathematical morphology directly on the gray level image of the collected images by machine vision, which can greatly reduce image preprocessing time. On the base of traditional moving target detection algorithm, an improved algorithm is proposed, which is suitable for detecting the defect of fiber winding. After the images are processed by the above process, the images show the position of every fiber. On the base of the processed images, the fiber pixel location in difference images can be calculated by combining the pixel size of the chosen CCD industrial cameras with related position of fiber in qualified fiber coil to determine whether there are defects such as fiber-stacking, fiber-cut and fiber-uneven in the process of fiber winding. The clarity of the fiber winding images shows that the methods used in the paper are effective and correct. 2. The application and the basic idea of mathematical morphology Mathematical morphology is one kind of new method used in the field of image processing and pattern recognition [4]. The basic idea of using mathematical morphology for performing image processing is that there is a certain specific shape of the structural elements such as rectangle, circle, or diamond of a certain size detection target image, by checking the validity of the structural elements of caving and fill method in the target area of the image to obtain information about the image morphology, and thus achieve the purpose of analysis and recognition image. The morphology operation is a collection of object shape and the interaction between the structural elements, which is not sensitive
676
C. Guo, R. Yang / Optik 125 (2014) 675–678
optical flow field is proposed in reference [6], but due to the complexity of the calculation formula, under the conditions without special hardware support, it is difficult to achieve real-time requirements, so its poor application and poor real-time are not suitable in defect detection of fiber winding. There is a small amount of calculation in the inter-frame difference method, easy to implement, which use two or three adjacent frames of video sequence to get time difference threshold to extract the target. So it has strong adaptability in dynamic environment [7]. However, this method cannot completely extract the relevant characteristic pixel point, and the feature of depending on the selection time interval also becomes the drawback of this method. The method is characterized by the following formula:
Fig. 1. Regular status.
Fig. 2. Fiber-uneven.
Dk (x, y) = |Ik+1 (x, y) − Ik (x, y)| Fig. 3. Fiber-stacking.
Fig. 4. Fiber-cut.
to the edge direction, but which can detect the true edge of the image and the noise of the image can be largely suppressed [5]. So, the morphology operation has unique advantages in the description of the characteristics of the object in the image. Therefore, the mathematical morphology for edge detection can effectively filter out the noise and retain the original details in the image with better edge detection effect. This paper applies opening and closing operation of mathematical morphological to extract the outer contour of optical fiber winding image. The closing operation generally will narrow the gap linking and fill the hole smaller than the structural elements. The opening operation can remove the object area that cannot contain the structure elements and can remove the small protruding portions. The method is characterized by the following formula, where formula (1) is opening operation expression and formula (2) is closing operation expression. A ◦ B = (AB) ⊕ B
(1)
A • B = (A ⊕ B)B
(2)
The experimental results show that the application of mathematical morphology for processing Fig. 8, that is gray level image of original image of optic fiber winding to get binary image of optical fiber winding, that is Fig. 9. As can be seen from Fig. 9, the image fully meets the needs of the subsequent analysis. 3. Improved moving target detection algorithm for optical fiber winding 3.1. Traditional moving target detection algorithms Moving target detection algorithms are diverse. Every algorithm is different in its specific application environment. In a static context, traditional moving target detection algorithms are mainly optical flow field method, inter-frame difference method and background difference method. These moving target detection algorithms depend on the range of applications, each with its own advantages and disadvantages. Optical flow field method is represented by a two-dimensional image of the velocity field in the three-dimensional motion of the object point, which uses the optical flow characteristics, that is moving target changing over time to create the optical flow constraint equation for target detection, the calculation method of
(3)
where Dk (x, y), Ik+1 (x, y), and Ik (x, y) are respectively difference image, the k + 1 original image, and the k original image. The defect detection of fiber winding needs to detect the exact location of each optical fiber, so the inter-frame difference method is not suitable for the application environment. Like inter-frame difference method, the background difference method also extracts the target region using different image. But its difference to the inter-frame difference method is that the background difference method is not the difference image of the current frame image with the adjacent frame image, but is the difference image of the current frame image with a continuously updated background image [8]. The method is characterized by the following formula: Dx (x, y) = |Ik+1 (x, y) − Bk (x, y)|
(4)
where Dx (x, y), Ik (x, y), and Bk (x, y) are respectively the difference image, the current frame image, and the background image. In the defect detection of fiber winding, each frame image is different and the difference image of the adjacent frame image is not the detected target. So using background difference method for defect detection of fiber winding will cause larger detection error, it is not suitable to the detection environment of fiber winding defect. In view of the above analysis, the traditional target detection algorithms exist their shortcomings in the defect detection of optical fiber winding. The improved algorithm was forwarded for the particularity of the detected targets in the paper. 3.2. Improved moving target detection algorithm based on inter-frame difference method For fiber winding process, the interval of adjacent frames is very short; there is a strong correlation and continuity on adjacent frame image. Using the inter-frame difference method alone to detect target will be missed target. Due to the limits of fiber winding speed, there are more than two images on the interval of two fibers appearing on the same end respectively. For background difference method alone there are similar problems to inter-frame difference method, and the background difference method needs to constantly update the background image that causes the heavy workload. Therefore, in fiber winding, in order to obtain fast detection speed, according to the speed of optical fiber winding and the FPS of chosen CCD industrial cameras, the images of the input images collected by machine vision are sampled to meet accurately detected defect of fiber winding. Based on the basic principle of the inter-frame difference method, the images formed in the process of fiber winding will be re-combinated to form a new sequence of images, which are suitable for the application of defect detection and moving object detection method. There are lots of video images in the process of fiber winding, but in the judgment of the winding defect, need to select an image with a new and entire fiber on the end, as Fig. 5.
C. Guo, R. Yang / Optik 125 (2014) 675–678
677
three new image sequence with the inter-frame difference method for processing, as shown in formula (7) and formula (8). DK (x, y) = |EK (x, y) − EK−1 (x, y)|
(7)
DK+1 (x, y) = |EK+1 (x, y) − EK (x, y)|
(8)
where DK (x, y) and DK+1 (x, y) respectively are the difference images of the new image sequence K frame and the new image sequence K − 1 frame and the difference image of the new image sequence K + 1 frame and the new image sequence K frame. The flow chart of image preprocessing and target identification in fiber winding is shown in Fig. 5. 4. Defect detection method of fiber winding
Fig. 5. The flow chart of image preprocessing and target identification.
According to the speed of optical fiber winding, the paper selects appropriate time intervals such as 12 fps of the chosen industrial cameras to acquire images, from the acquired images to sample the images with the entire fiber information as the analysis of the reference images, these original images referred as I (x, y) are recorded as a collection of the following sequences: I(x, y) =
I1 (x, y), I2 (x, y), ·I3 (x, y), . . ., Ii (x, y)i = 1∼n
After the above processing, the output images display the pixel position of the single optical fiber at different times, combining pixel dimensions of chosen CCD industrial cameras with related position between fiber in qualified fiber coil (in Fig. 1) to determine fiber defect (in Figs. 2–4). The method is shown in Fig. 6. A defect is considered any place in the coil where there is a deviation from the pattern, which can degrade the performance of the coil produced. So we need to detect these defects during fiber winding. There are three threshold values, respectively referred to as T1, T2, and T3, which are set to determine category of fiber winding defect. The three threshold values are set according to pixel dimensions of chosen CCD industrial cameras with related position between fibers in qualified fiber coil. When the absolute difference of the minimum row of fiber pixels coordinating in the two difference images is greater than the threshold value T1, the judgment result is uneven in fiber winding; when the absolute difference of the minimum column of fiber pixels coordinating in the two difference images is greater than the threshold value T2, the judgment result is fiber-stacking in fiber winding; when the absolute difference of the minimum column and row of fiber pixel coordinating
(5)
From the above sequence, sampling i = 10, i = 20,. . . i.e. 10 integer multiples of the original images (this value is selected according to the speed of fiber winding and the FPS of chosen CCD industrial cameras) conduct a new sequence of image frames F (x, y), as the new sequence shown in the following sequences. F(x, y) =
F1 (x, y), F2 (x, y), ·F3 (x, y), . . ., Fi (x, y)j = 10i, ·i = 1∼n
(6)
In order to improve the quality of the detected image and the accuracy of detection, a new sequence of image frames need to be preprocessed for eliminating the irrelevant information included in the images and restoring useful authentic information, and enhancing the detected performance of relevant information and greatly simplifying data. The conventional methods of processing image include these steps such as graying, denoising, enhanced, edge sharpening, and binarization, then to extract specific information or certain features contained in the image for further processing using detecting algorithm, this paper uses mathematical morphology to greatly reduce the time of image pre-processing, and acquire clear target images. Preprocess the new image sequence, and applicate the improved moving target detection method to detect the position of fiber in the image. Here, the new image sequences are preprocessed as consecutive frames, to be written as: EK−1 , EK , and EK+1 , to the above
Fig. 6. Software method of the defect detection.
678
C. Guo, R. Yang / Optik 125 (2014) 675–678
between the two difference images is all close to T3, it is judged that a fiber breaks. When one of the above defects appears, the detection system alarms and returns to main program to wait for the further command. In Fig. 6, the input images are the two difference images processed by improved moving target detection algorithm. Fig. 12. The first difference image.
5. Implementation and results The proposed algorithm and detection method succeeds in processing the images of fiber winding and detecting the defects of fiber winding. In the paper, some original images of fiber-uneven in fiber winding and processed images are shown to verify the feasibility and accuracy of the proposed algorithm and detection method. The experimental platform is on Windows XP, Intel DualCore (2.2 GHz), and 2GB RAM. Fig. 7 is the original image which shows tightly coiled fiber and a single optical fiber on the next lay. The gray level image (in Fig. 8) is processed using the method of the mathematical morphology to get the binary image (in Fig. 9). The method of processing the images including two and three fibers on tightly coiled fiber layer is similar to the above method of processing Fig. 7 to get two binary images, as shown in Figs. 10 and 11. For the three binary images processed by using improved moving target detection algorithm to
Fig. 7. The original image.
Fig. 13. The second difference image.
get two difference images, as shown in Figs. 12 and 13. Fig. 12 is the difference image of Figs. 9 and 10. As can be seen from Fig. 10 and software calculation result, the fiber of Fig. 12 is the proper place on the fiber winding. Fig. 13 is the difference image of Figs. 10 and 11. As can be seen from Fig. 11 and software calculation result, the fiber of Fig. 13 is spaced-out from previous turn on the fiber winding. According to the clarity of the images that can be seen, using the mathematical morphology to preprocess images can clearly indicate the location of the single fiber. The application of improved moving target detection algorithm on the binary images can completely differentiate between current fiber and previous fiber. The results satisfy the needs of subsequent calculation. As can be seen from Fig. 11, the new fiber turn is spaced out close to one fiber from previous turn that is fiber-uneven. The result observed from Fig. 11 is consistent to the result calculated by the method as shown in Fig. 6. 6. Conclusion
Fig. 8. The gray level image.
Fig. 9. The first binary image.
Fig. 10. The second binary image.
Fig. 11. The third binary image.
Fiber sensing coil winding quality is an important factor in the accuracy of FOG. For real-time and accuracy requirements of the defect detection, this paper proposes mathematical morphology and improved moving target detection method to process images and combines software algorithm with pixel dimensions of chosen CCD industrial cameras and related position of qualified fiber turn to detect defect of fiber winding. According to the processed images and software calculation result, it is shown that the proposed image processing algorithm and defect detection method can fast and accurately distinguish fiber. Experimental results verify the feasibility of the algorithm and the method of defect detection in fiber winding. References [1] M.L. Stephen, Design and Development of an Automated Fiber Optic Gyroscope Coil Winding Machine, Massachusetts Institute of Technology, Cambridge, 1997, pp. 20–24, B.S. Mechanical Engineering. [2] Z. Ding, Study of Complete Quality Measurement for FOG Fiber Coils, TianJin University, 2010, pp. 3–6. [3] W.E. Snyder, H. Qi, Machine Vision, China Machine Press, 2005, pp. 1–2. [4] H. Nan, Edge detection using morphological algorithm, Comput. Simul. 29 (3) (2012) 288–291. [5] A.A. Purwita, K. Adityowibowo, et al., Automated microaneurysm detection using mathematical morphology, in: International Conference on Instrumentation, Communication, Information Technology and Biomedical Engineering, Bandung, Indonesia, 8–9 November, 2011. [6] I. dar, K. Janusz, Occlusion-aware optical flow estimation, IEEE Trans. Image Process. 17 (8) (2008) 1443–1451. [7] A. Ton, H. Fujiyoshi, R. Patil, Moving target classification and tracking from realtime video, in: Proceedings IEEE Workshop on Application of Computer Vision, Princeton, NJ, 8–14, 1998. [8] Taohui, D. Harwood, L. Davis, W4: Real-time surveillance of people and their activities, IEEE Trans. Pattern Anal. Mach. Intell. 22 (8) (2000) 809–830.