| Sign up

PDF (1,003.3 KB)

Cite

Collect

Submit Manuscript

Show Outline

Figures (4)

Fig. 1

Fig. 2

Fig. 3

Fig. 4

Tables (1)

Table 1

Open Access

Part Deviation Correction Method Based on Geometric Feature Recognition

Guoqing Zhang^¹(), Hongbo Sun^¹

1School of Computer and Control Engineering, Yantai University, Yantai 264005, China

Show Author Information

Abstract

To realize the automatic loading process of parts, one of the core tasks is to identify the geometric contour of the part’s surface and the angular direction. Since the angular direction of each part is not the same when it arrives at the loading position, for example, there are two same types of parts with the same pattern, when they arrive at the loading position, the pattern on one part may be on the right side of the part surface, and the pattern on the other part may be on the left side of the part surface, the gripper of the mechanical arm needs to rotate above the parts in order to grab the parts during each loading process. If the rotation angle is wrong, there will be an impact between the gripper and the parts. Therefore, in order to solve the problem of different angles, this paper proposes a method of parts deviation correction based on geometric features. In this work, firstly, the acquired image is preprocessed, the image background is separated, and the geometric features of the parts are obtained. Then edge detection is used to obtain the set of edge pixels to obtain the contour in the image. Finally, the image moment and measurement model are used to output angular direction. Through 500 repeated detection experiments, the results show that this method can perform better angular direction correction. The maximum angular direction difference is 0.073°, which is 0.856° and 1.793° higher than the Least square method and Hough transform circle detection accuracy, respectively. The average detection time is 1.89 s and is 0.336 s and 1.39 s less than the Least square method and Hough transform circle detection, which meets the requirements of industrial applications.

Keywords

surface geometric contour image processing edge detection image moment measurement model

References

[1]

L. Pérez, Í. Rodríguez, N. Rodríguez, R. Usamentiaga, and D. García, Robot guidance using machine vision techniques in industrial environments: A comparative review, Sensors, vol. 16, no. 3, p. 335, 2016.

Crossref Google Scholar

[2]

A. W. Hashmi, H. S. Mali, A. Meena, I. A. Khilji, M. F. Hashmi, and S. N. B. M. Saffe, Machine vision for the measurement of machining parameters: A review, Materials Today: Proceedings, vol. 56, pp. 1939–1946, 2022.

Crossref Google Scholar

[3]

S. K. Yi, R. M. Haralick, and L. G. Shapiro, Optimal sensor and light source positioning for machine vision, Comput. Vis. Image Underst., vol. 61, no. 1, pp. 122–137, 1995.

Crossref Google Scholar

[4]

S. B. Dworkin and T. J. Nye, Image processing for machine vision measurement of hot formed parts, J. Mater. Process. Technol., vol. 174, nos. 1–3, pp. 1–6, 2006.

Crossref

[5]

Y. Zhao and B. Zuo, Analysis on application of machine vision in fabric defect detection, Computer Engineering and Applications, vol. 56, no. 2, pp. 11–17, 2020.

Google Scholar

[6]

J. Yu, X. Cheng, L. Lu, and B. Wu, A machine vision method for measurement of machining tool wear, Measurement, vol. 182, p. 109683, 2021.

Crossref Google Scholar

[7]

Q. Tan, Y. Kou, J. Miao, S. Liu, and B. Chai, A model of diameter measurement based on the machine vision, Symmetry, vol. 13, no. 2, p. 187, 2021.

Crossref Google Scholar

[8]

D. Shen, S. Zhang, W. Ming, W. He, G. Zhang, and Z. Xie, Development of a new machine vision algorithm to estimate potato’s shape and size based on support vector machine, J. Food Process. Eng., vol. 45, no. 3, p. e13974, 2022.

Crossref Google Scholar

[9]

T. Tao, H. He, X. Zheng, and J. Xu, Study on the detection technology of piston position and orientation based on digital image processing, Industrial Instrumentation & Automation, no. 5, pp. 3–8, 2018.

Google Scholar

[10]

C. Jie, Q. Yuan, and B. Li, Method for measuring rotation angle of faucet handle based on machine vision, IEEE Access, vol. 8, pp. 202412–202427, 2020.

Crossref Google Scholar

[11]

C. Wang, J. Li, W. Wang, and J. Zhang, Research on roundness detection and sorting of oil cooling pipe based on machine vision, Machine Tool & Hydrauics, vol. 50, no. 15, pp. 31–37, 2022.

Google Scholar

[12]

J. Zhou, Y. Liu, and L. Huo, Machine vision-based rotation angle measurement of bolt looseness, Machine Design and Research, vol. 37, no. 4, pp. 159–162,175, 2021.

Google Scholar

International Journal of Crowd Science

Volume 7 Issue 3,
September 2023

Pages 113-119

DOI: 10.26599/IJCS.2023.9100005

Cite this article:

Zhang G, Sun H. Part Deviation Correction Method Based on Geometric Feature Recognition. International Journal of Crowd Science, 2023, 7(3): 113-119. https://doi.org/10.26599/IJCS.2023.9100005

Return

Algorithm 1　Measurement method
Input:
I // I is the collected image by the camera
TG //TG is the template of the Gaussian
TO //TO is the template of the open operator
TS_x //Sobel’s template along the X direction
TS_y // Sobel’s template along the Y direction
Sigma // Sigma is the $σ$ of TG
Output: $θ$
/Gaussian Filter/
Step 1:
NI [IW, IH] ← 0
/IW (width of the image I), IH (height of the image I), and set a temporary image to black/
[IW, IH] ← size (I) // Get I dimensions
[TGW, TGH] ← size (TG)
/* TGW (width of the template TG), TGH (height of the template TG), and get TG dimensions*/
CG ← floor (TG/2)+1 // CG is a half of TG
sum ← 0 // we will calculate the normalization for i ← 1 to TGW do
for j ← 1 to TGH do
(TG_i, TG_j) ← exp (−((j−center) × (j−center)+(i−center) × (i−center)))/(2 × sigma)
sum ← $\sum$ TG
end
end
$T G \leftarrow T G / s u m$ //Normalization processing
Void Convolve (I, TG) //Convolution function
$f o r x \leftarrow C G t o I W d o$
//Address all columns except border
$f o r y \leftarrow C G t o I H d o$
//Address all rows except border
C $R \leftarrow 0$
$f o r d_{x} \leftarrow 1 t o T G W d o$
$f o r d_{y} \leftarrow 1 t o T G H d o$
$C R \leftarrow \sum I (y + d_{y} - c e n t e r, x + d_{x} - c e n t e r) \times T (d_{y}, d_{x})$
end
end
$N I (y, x) \leftarrow$ CR
end
end
$G I (y, x) \leftarrow N$ ormalise(NI)// GI is Gaussian fitter result
end Convolve
/The result of the convolution is to use an N×N* convolution template to traverse the entire image, multiply and then add the corresponding coordinates, and then permutate the values of the corresponding coordinates of the original image. This article contains many of convolution operations, so the following article directly calls the convolution function*/
/* Image binarization*/
Step 2:
$f o r e a c h ({G I}_{i}, {G I}_{j}) i n G I d o$
$i f (\exists N, 0 ⩽ N ⩽ 255, ({G I}_{i}, G I) > N)$ then
$({G I}_{i}, {G I}_{j}) \leftarrow 255$
//Segment the foreground of the GI
$e l s e$
$({G I}_{i}, {G I}_{j}) \leftarrow 255$
//Segment the background of the GI
end if
end
/* Open operation*/
Step 3:
$[G I W, G I H] \leftarrow s i z e (G I)$
/Get GI dimensions, and the GIW (width of the image GI) and GIH (height of the image GI) are equal IW and IH/
$[T O W, T O H] \leftarrow s i z e (T O)$
/* TOW (width of the template TO),TOH (height of the template TO), and get TO dimensions
$C O \leftarrow f l o o r (T O / 2) + 1$ // CO is a half of TO
$S O \leftarrow 0$ //we will judge the total
$N U M \leftarrow$ The total number of nonzero in TO
/The GI is first Erosion processing /
$f o r x \leftarrow C O t o G I W d o$
$f o r y \leftarrow C O t o G I H d o$
$f o r d_{y} \leftarrow 1 t o T O W d o$
$f o r d_{y} \leftarrow 1 t o T O W d o$
$S O \leftarrow \sum D D (y + d_{y} - C O, x + d_{x} - C O) \times T O (d_{y}, d_{x})$
$i f (S O < (255 \times N U M))$ then
$({G I}_{x}, {G I}_{y}) \leftarrow 0$
end $i f$
end
end
end
end
/The GI is then do Dilation operation /
$f o r x \leftarrow C O t o D D W d o$
$f o r y \leftarrow C O t o D D H d o$
$f o r d y \leftarrow 1 t o T O W d o$
$f o r d_{y} \leftarrow 1 t o T O W d o$
$S O \leftarrow \sum D D (y + d_{y} - C O, x + d_{x} - C O) \times T O (d_{y}, d_{x})$
$i f (S O ⩾ 255)$ then
$({G I}_{x}, {G I}_{y}) \leftarrow 255$
end $i f$
end
end
end
end
/* Obtain geometric contours */
Step 4:
${S I}_{x} \leftarrow C o n v o l v e (G I, {T S}_{x})$
/The edge image ${S I}_{x}$ (image after Sobel operation along X direction template) in the X-direction is generated by convoluting with the template ${T S}_{x}$ along the X-direction/
${S I}_{y} \leftarrow C o n v o l v e (G I, {T S}_{y})$
/The edge image ${S I}_{y}$ (image after Sobel operation along Y direction template) in the Y-direction is generated by convoluting with the template ${T S}_{y}$ along the Y-direction/
$S I \leftarrow {S I}_{x} + {S I}_{y}$
// The edge image $S I$ (image I after Sobel operation along X and Y directions template) of GI if formed by ${S I}_{x}$ and ${S I}_{y}$
$I C \leftarrow$ Image binarization SI/* The IC (contours of the image I) geometric contours contained in the image are obtained by using image binarization on SI*/
/* Obtain geometric contours centroid*/
Step 5:
${{C o n t o u r s}_{i}} \leftarrow I C$
/* Assign all the information of the i contour in IC, including coordinates and pixel values, to ${C o n t o u r s}_{i}$ */
for each Contours_i in IC do
$f o r e a c h (x, y)$ in ${C o n t o u r s}_{i}$ do
$m_{00} \leftarrow \sum {C o n t o u r s}_{i} (x, y)$
$m_{10} \leftarrow \sum x {C o n t o u r s}_{i} (x, y)$
$m_{10} \leftarrow \sum y {C o n t o u r s}_{i} (x, y)$
$C_{x_{i}} \leftarrow m_{10} / m_{00}$
// The i contours x centroid coordinate
$C_{y_{i}} \leftarrow m_{01} / m_{00}$
// The i contours y centroid coordinate
end
end
/* Measurement*/
Step 6:
$(l_{x}, l_{y}) \leftarrow {C_{x_{i}}, C_{y_{i}}}$
/* The $l_{x}, l_{y}$ (centroid coordinate of the little contours) of the black circle hole is filtered according to the size of the contour*/
$(b_{x}, b_{y}) \leftarrow {C_{x_{i}}, C_{y_{i}}}$
/* The $b_{x}, b_{y}$ (centroid coordinate of the big contours) of the resembling the double-headed arrow contour is filtered according to the size of the contour*/
$K \leftarrow b_{x} - l_{x}$
$W \leftarrow b_{y} - l_{y}$
/* The image coordinate system is converted to a planar rectangular coordinate system*/
$i f (K \leftarrow 0 a n d W > 0)$ then
$θ \leftarrow$ 90°
$i f (K \leftarrow 0 a n d W < 0)$ then
$θ \leftarrow$ −90°
$i f (K > 0 a n d W \leftarrow 0)$ then
$θ \leftarrow$ 0°
$i f (K < 0 a n d W \leftarrow 0)$ then
$θ \leftarrow$ −180°
$i f (K > 0 a n d W > 0)$ then
$θ \leftarrow a r c t a n \frac{\| K \|}{\| W \|}$
$i f (K < 0 a n d W < 0)$ then
$θ \leftarrow - a r c t a n \frac{\| K \|}{\| W \|}$
$i f (K < 0 a n d W < 0)$ then
$θ \leftarrow (- 90^{\circ} + a r c t a n \frac{\| K \|}{\| W \|})$
$i f (K > 0 a n d W < 0)$ then
$θ \leftarrow 90^{\circ} + a r c t a n \frac{\| K \|}{\| W \|}$
Out $θ$

Table 1Comparison of several methods.

Method	Range (°)	Mean measurement time (s)
Hough transformation	$θ =$ 1.866	3.28
Least square method	$θ =$ 0.929	2.226
Purposed method	$θ =$ 0.073	1.89