Dynamic image capture and geometric parameter measurement of moving objects on conveyor belts-EEWORLD

Collect

【Abstract】This paper introduces the image restoration and geometric parameter measurement of moving objects on conveyor belts by image processing methods, and applies them to the online curvature and thickness measurement of badminton feathers in badminton production. The principle of linear motion image restoration, geometric parameter calculation method and improvement measures are described in detail, and finally the system solution and measurement results are given.
Keywords: video capture, image restoration, online measurement

1 Introduction
In the process of automated production, conveyor belts and logistics systems are important components of the production line. Usually, in order to achieve the classification and processing of objects or complete the quality control of products, we need to understand some details of the objects such as geometric parameters. Therefore, the measurement method based on image processing is increasingly used. There are two main measurement methods: static and dynamic. Static means that the target object is stationary relative to the camera. In this way, a clear image can be obtained, while in the dynamic mode, a clear image cannot be obtained.
In some cases, static images cannot be obtained, and the imaging of moving targets will cause the image quality to deteriorate. In order to avoid image degradation, or considering other factors such as mechanical transmission system, motion characteristics, etc., we generally use the method of image capture and processing in motion. Although the use of a smooth motion system can reduce vibration during the transmission process, improve the operating efficiency of the system, and simplify the design of the system, the image processing time will also be prolonged.
This paper applies the dynamic image measurement method to the measurement of parameters such as the curvature of the feathers in the badminton production process, introduces the composition and operation of the entire system, and provides the main results of the process.
2 Principle of motion image restoration
To understand the principle of motion image restoration, we should first understand the motion model of the image.
According to the Gonzalez horizontal image motion model〔1〕, let the original image be f(x, y), and the original image moves a distance a in the horizontal direction within the exposure time T, the moving speed is constant and the exposure is linear, then

This is a recursive relationship, which means that the restored image at the current position can be calculated from the restored image a away from the current position, and the derivative of the blurred image g(x) can always be obtained. As long as the image with a length of a is obtained, the entire image can be obtained according to the above recursive relationship.
Let m be the integer part of x/a, and the restored image can be approximated by the following formula〔2〕

The quality of image restoration depends on the selection of each parameter in the restoration relationship. A and γ affect the background and contrast of the restored image, while a plays a decisive role in the quality of the restored image. Usually, a search method is used to obtain a suitable value. In reference [2], an automatic search method is implemented under the mean square error criterion. In the measurement environment of this paper, since the motion speed is constant, parameter a is treated as a parameter once it is determined.
3 Calculation of geometric parameters based on image processing
3.1 Badminton feather parameters
An important indicator of badminton quality is its flight stability, that is, no swaying or change of line during flight. The aerodynamic mechanism of badminton flight is very complicated and will not be studied here. As long as the feathers of the same shape can be inserted into the same ball, the badminton produced under normal procedures will have good flight stability. The traditional measurement method is not only slow, but also the fixture affects the measurement results of soft materials. This paper attempts to use image capture equipment to dynamically capture and process the feathers on the conveyor belt, obtain the shape parameters of the feathers, and then divide them into grades according to the shape parameters to ensure that 16 feathers with the same shape are inserted into the same badminton. The
main geometric parameters of the feathers include the curvature and arch of the feather shaft and the thickness of the top of the feather shaft. The curvature here refers to the horizontal distance between the center line of the hair rod and its tangent at the top position, and the arch is the arch height of the hair piece when it is laid flat. Here we mainly introduce the measurement method of curvature. First, extract the edge of the hair rod and calculate the center position, then fit the center line of the hair rod, and finally calculate the curvature value. In this process, the thickness is also obtained at the same time.
3.2 Edge extraction of image and collection of edge data
The edge of the object reflects the discontinuity of local characteristics in the image. The ideal edge has step type, roof type and flange type. Due to the existence of image noise, the actual edge becomes very complicated.
Edge detection is usually implemented using differential classification operators. Such operators include Sobel operator, Kirsh operator and Laplacian operator, etc. The first two operators are gradient operators, and the latter is a second-order differential operator. The Sobel gradient operator selects the value with the largest differential in two directions as its gradient value. Obviously, when the differential values in the two directions are equal, the gradient error is the largest; while Kirsh calculates the differential in eight directions and takes the maximum value as the gradient value〔3〕. The calculation result of the operator method is used as the basis for edge discrimination; the Laplacian operator is a second-order operator that does not depend on the edge direction and has rotation invariance.
Due to the inherent characteristics of the differential classification operator, edge detection will be affected by noise. The use of filtering methods can effectively suppress the interference of noise, but at the same time it also produces a certain degree of passivation to the edge. This passivation will affect the extraction of the edge, so the key lies in the selection of the filtering scheme. The above differential classification operators all use filtering methods. The Sobel operator uses a three-point weighted average. When the edge is in the horizontal or vertical direction, the actual filtering is performed along the edge, so the passivation effect of the filter on the edge is the smallest; when the edge is in the 45° or 135° direction, the difference between the filter point and the edge in direction is the largest, and the passivation effect of the filter on the edge is also the largest at this time. Therefore, selecting a kernel that is more consistent with the edge can achieve satisfactory results in noise suppression and edge preservation.
3.3 Measurement of hair piece parameters
The hair bar shape obtained from image processing will have burrs due to image noise and errors in the measurement process. The hair bar center obtained through processing is often not smooth and has discontinuous points.
The tangent line of a point on the curve is approximated by the line connecting two nearby points. When the distance between two points is close, the connecting line can be regarded as a tangent line. This method is simple to implement, but the measurement result error is very large in a noisy environment. The method adopted in this paper is to perform curve fitting on the finite points obtained by sampling the hair bar center, and then obtain the tangent line of a certain point from the fitted curve, and calculate the curvature of the hair piece.
The curve fitting adopts the least squares method of polynomials. For each hair bar, select N sets of coordinates (Xi, Yi) along the Y direction, take X as the variable and Y as the independent variable, and select the equation of the curve as

For the above linear equations, the polynomial coefficients can be obtained.
The last issue to be considered is the perspective correction of the curvature by the camber. Each feather has a different degree of curvature. When measuring the curvature, the feather shaft is projected onto the horizontal plane for measurement. Therefore, the measured values of feathers with the same curvature under different viewing angles are different. The purpose of viewing angle correction is to restore the actual parameters.
4 Application of dynamic image capture and geometric parameter measurement in feather sorting
4.1 Composition of measurement system
The dynamic measurement system of feather geometric parameters includes the following parts: (1) Conveyor belt. Driven by a steady-speed motor, the feathers placed on it can enter the camera area smoothly at a constant speed. A vibration reduction mechanism is used to reduce the up and down vibration of the conveyor belt. (2) Camera and video capture device. The image capture part consists of a camera, a capture card and a computer. A 480-line CCD camera, a Seiko 16mm manual aperture lens and a LifeView video capture card are used. The video and capture speed is 15 to 30fps, and the designed conveyor belt movement speed is 2s per frame. (3) Image processing software and computer. The image processing software includes motion compensation filtering, image feature parameter extraction, feather shaft contour fitting and image parameter output. The computer uses a PⅢ microcomputer.
The image feature parameters are derived from the extraction of the edge contour and center position of the hair bar. The edge extraction is detected by the differential classification operator. Since the edge direction of the hair bar usually changes within a small range near the vertical direction, according to the actual placement of the hair piece, the edge direction of the hair bar is generally within 10° to the left or right of the vertical, so a 5×3 vertical strip kernel is used.
The fitting of the hair bar centerline adopts the method described above. 15 points are selected from the top to the root of the hair bar to fit the four coefficients of the third-order polynomial. In fact, the third-order polynomial is sufficient for the curvature calculation of the hair bar.
In real-time measurement, the processing speed is particularly important, which will affect the efficiency and benefit of the entire system. According to the above analysis, the restoration of the motion image only needs to be performed in a limited number of rows, which can reduce the time required for image restoration.
The main way to reduce the calculation time is to use a recursive relationship. From formula (4), it can be seen that the restored image is grouped by width a, and the points at the same position in each group are accumulated to realize the restoration operation. When programming, opening a unit for storage will save a lot of time.
In order to improve the processing speed, the image is transferred to the memory when the video image is captured, and the program accesses the stored image to complete the processing operation. This can avoid peripheral access and reduce the processing time within a loop. 4.2 Main results After the uniformly placed hair pieces on the conveyor belt are dynamically captured, a blurred image is obtained, as shown in Figure 2. The capture moment is triggered by the conveyor belt positioning device. The middle part is the restored image, where A is 50 and a is 15. The size of a can be obtained by experimental methods or by estimation〔4〕. The right side shows the curvature measurement results, which are actually used as a reference for system debugging, such as lens aperture setting, threshold setting during processing, etc. In practical applications, image restoration and parameter calculation are performed as a task, and the restored image does not appear on the screen as an intermediate result. In terms of image restoration, the image restoration method introduced above can obtain a relatively clear image, but there is also a large noise, which is obviously related to the differential operation. The parameter measurement of the hair rod is also affected by noise, but it will be improved to a certain extent under the filtering effect of the 5×3 vertical kernel. There is a lot of noise in the black background, which is mainly caused by the electronic noise of the CCD. Using an appropriate threshold value to suppress the noise under low brightness can achieve good results. The measurement results obtained in this way are closer to the results under static conditions and can meet the requirements of raw film parameter measurement and sorting.

References

1 Gonzalez R, Woods R. Digital Image Processing. Ad-dison-Wesley Publishing, 1992
2 Lu Jun, Shu Zhilong, Ruan Qiuqi. Image Restoration Based on Scale Rotation. Journal of Communications, No. 7, 2000
3 Castleman KR. Digital Image Processing. Beijing: Tsinghua University Press, 1998
4 Tekalp AM. Digital Video Processing. Beijing: Publishing House of Electronics Industry, 1998

Reference address：Dynamic image capture and geometric parameter measurement of moving objects on conveyor belts

Previous article：Research on a novel, fast and high-precision transformer DC resistance tester
Next article：Why develop test-friendly technology

Popular Resources
Popular amplifiers