CN109029861B - A pressure vessel air tightness detection method based on background modeling and centroid clustering - Google Patents

Abstract

The invention discloses a method for detecting the air _tightness of _a pressure vessel based on background modeling and centroid clustering. When P ₁ ‑P ₀ ≤Δ, Δ is the pressure difference threshold, which means no pressurization; when P ₁ ‑P ₀ ＞Δ, it means pressurization, and the air tightness test starts; Step 2: Take the air tightness test The n-frame images of the video after the test starts, n=t*f, where t is the test duration, f is the video frame rate; denoting the n-frame images as I ₁ , I ₂ ,..., I _n , for I _i , _i =1, 2, . , can accurately judge the air tightness of the pressure vessel and accurately locate the leaking area.

Description

A pressure vessel air tightness detection method based on background modeling and centroid clustering

technical field

The invention belongs to the field of safety supervision and inspection of special equipment, in particular to a method for detecting the air tightness of a pressure vessel in a water immersion method based on background modeling and a centroid clustering algorithm.

Background technique

A pressure vessel is a common device for storing gas or liquid and belongs to the category of special equipment. In addition to affecting the function of the product itself, the pressure vessel with hidden quality risks will also cause dangerous events such as fire and explosion in severe cases. Therefore, the air tightness test of the pressure vessel during the production process is an important link to ensure quality.

Most domestic pressure vessel manufacturers use the water immersion method for air tightness testing. The traditional method determines the air tightness of the pressure vessel by manually observing the air bubbles in the water. However, long-term uninterrupted testing can easily cause eye fatigue of the inspectors and affect the inspection accuracy. , resulting in the situation of "unchecked air leakage". At the same time, pressure vessel manufacturers usually adopt a piece-rate system, which is prone to the situation that workers subjectively underestimate or give up inspections, and there is a problem of "less inspection and leakage inspection".

Therefore, the use of computer vision technology to detect air bubbles in water, and at the same time, combined with the water immersion method to detect the air tightness of pressure vessels has high feasibility and practical significance. Johnsson F (Sweden, Chalmers University of Technology, 2004) and others used computer vision technology to study and analyze the size, velocity and void ratio of bubbles in a two-dimensional fluidized bed. When Busciglio A (Italy, University of Palermo, 2009) studied the behavior of bubbles in gas-solid two-phase fluids, combined with image processing technology, the size and velocity of bubbles were detected. O. Zielinski (Germany, University of Oldenburg, 2010) etc. applied the optical flow method to the detection of air bubbles in water, and analyzed the feasibility through experiments. Wang Hongyi (Tianjin University, 2010) and others studied the rising process of bubbles in the gas-liquid two-phase flow field, and used a high-speed camera to take continuous images of the bubbles generated by leaking points of different diameters during the rising process. Visual technology, the parameters of the bubble (including velocity, acceleration, radius, area, etc.) were measured. Shao Jianbin (Xi'an University of Technology, 2011) proposed a watershed algorithm based on morphological theory to extract bubbles from aerated water flow images, and proposed a bubble image preprocessing consisting of inversion, image grayscale adjustment and low-pass filtering. Strategy. Liu Wei (Northeast Electric Power University, 2013) based on digital image processing technology for image enhancement research on water bubble images. Through Matlab simulation experiment, the effect of several threshold segmentation is compared, which provides an effective algorithm for obtaining bubble shape features. Wu Chunlong (Zhejiang Sci-tech University, 2013) proposed a set of automatic air tightness detection system based on PLC control, and applied the Hom-Schunck image processing algorithm based on optical flow theory to the identification and tracking of bubbles in gas-liquid two-phase flow field , to realize the detection and identification of air bubbles at the leak point of pressure vessels. Gan Jianwei (Xihua University, 2015) proposed an image processing system for bubble edge detection based on FPGA. Taking FPGA as the core, the Sobel edge detection algorithm is used to obtain the bubble edge features, and the feasibility is verified by Matlab experiments.

The computer vision detection technology mentioned in the above literature has carried out preliminary research on water bubbles, but there are still many shortcomings:

1) The experimental conditions of most algorithms are idealized, and the interference of environmental factors is not considered, and impurities with shapes similar to bubbles cannot be well excluded;

2) Some algorithms have higher computational complexity and time complexity, such as the bubble recognition algorithm based on the optical flow method. In the case of fewer bubbles, slower speed rise and no other interference, the algorithm does have good feasibility. and reliability. However, when the calculation time of the detection algorithm is too long and there are many bubbles, it is impossible to meet the requirements of real-time detection.

3) The cameras in these methods take pictures of the pressure vessel through the glass on the side of the water tank of the detection pool. Only a single pressure vessel can be detected each time, focusing on theoretical research, recording the process of bubble generation, and analyzing the conditions of bubble generation. and development process. In the actual pressure vessel production environment, a batch of multiple pressure vessels is sent to the detection pool together, placed side by side, and tested by the water immersion method at the same time. Therefore, the detection algorithm of the bubbles shot from the side cannot be applied to the actual scene.

SUMMARY OF THE INVENTION

In order to realize objective and efficient air tightness detection of pressure vessels, based on the dynamic characteristics, continuity and quantity of bubbles emerging, in order to overcome the shortcomings of existing methods, the present invention proposes a pressure based on background modeling and centroid clustering. Container air tightness testing method.

A pressure vessel air tightness detection method based on background modeling and centroid clustering, characterized in that it includes the following steps:

Step 1: Set the original pressure value of the pressure vessel to P ₀ , and the pressure value after T time to P ₁ . When P ₁ -P ₀ ≤Δ, Δ is the pressure difference threshold, which means that it is not pressurized; when P ₁ -P ₀ When >Δ, it means that it is pressurizing, and the air tightness test starts;

Step 2: Take n frame images of the video after the air tightness test starts, n=t*f, where t is the test duration, f is the video frame rate; denote n frame images as I ₁ , I ₂ ,..., I _n , grayscale I _i , i=1,2,...,n to obtain a corresponding grayscale image G _i ;

Step 2.1: take _G1 for initialization, establish a background model, update the background model with G2... _Gn , and obtain foreground images _M2 ... _{Mn ;}

Step 2.2: Binarize M ₂ ... _Mn to obtain binary graphs B ₂ ... B _n ;

从而得到连通图像C_j,其中J_j＝{L_ju|u＝1,2,...,v_j}，j＝2,3,…,n，J_j表示连通图像C_j中连通域的集合，其中X为结构元素，L_ju为J_j中的第u个连通区域，v_j为J_j中的连通区域个数，运算符“Θ”为腐蚀操作，运算符为膨胀操作；Step 2.3: Perform morphological operations on images B ₂ ... B _n ,

Thus, a connected image C _j is obtained, where J _j ={L _ju |u=1,2,...,v _j }, j=2,3,...,n, J _j represents the connected domain in the connected image C _j set, where X is the structuring element, L _ju is the u-th connected region in J _j , v _j is the number of connected regions in J _j , the operator "Θ" is the erosion operation, the operator for the expansion operation;

Step 2.4: N _ju is the number of pixels in the connected region L _ju , all connected regions L _ju satisfying TL < N _ju <TU are retained, and those that are not satisfied are eliminated, and the connected image C _j is eliminated to obtain F _j , and F The connected domain set of _j ={L _jv |v=1,2,...,w _j }, where L _jv is the _vth connected region in D _j , and w _j is the number of connected regions in D _j number, TL is the lower bound of the number of pixels in the bubble area, TU is the upper bound of the number of pixels in the bubble area;

m₀₀表示连通区域L_jv的零阶矩，m₀₁、m₁₀表示连通区域L_jv的一阶矩；Step 2.5: Calculate the centroid coordinates of the connected region L _jv of the connected domain set D _j

m ₀₀ represents the zero-order moment of the connected region L _jv , m ₀₁ , m ₁₀ represent the first-order moment of the connected region L _jv ;

表示得到初始气泡区域的帧数：DBSCAN聚类半径设置为radius，邻域内最少数据点数设置为minPTs，得到的聚类区域B_k，k＝0,1,...,g，B_k即为候选漏气区域；Step 3: Use the DBSCAN clustering algorithm to analyze the point set Clustering, the point set P contains the coordinates of the centroid points of all connected regions in the connected domain set of 1 to m+1 frames,

Indicates the number of frames to obtain the initial bubble area: the DBSCAN clustering radius is set to radius, the minimum number of data points in the neighborhood is set to minPTs, and the obtained clustering area B _k , k=0,1,...,g, B _k is Candidate leak area;

包含m+2帧到n帧连通域集合的全部连通区域质心点坐标，N_k表示聚类区域B_k中包含点集Q中数据点的数量，若N_k＞T，表示对应的B_k为漏气区域，T为设定的阈值。Step 3.1: Re-judgment the collected candidate air leakage areas, point set

Contains the coordinates of the centroid points of all connected regions in the connected domain set of frames m+2 to n, N _k represents the number of data points in the point set Q included in the clustering region B _k , if N _k >T, it means that the corresponding B _k is Air leakage area, T is the set threshold.

The beneficial effects of the present invention are: based on background modeling and centroid clustering, combined with bubble size, quantity, and dynamic characteristics, the present invention can accurately determine the air tightness of the pressure vessel and accurately locate the air leakage area.

Description of drawings

Fig. 1 is the original video image of the pressurization start in the embodiment;

Fig. 2 is the image obtained after grayscale processing to I ₁₀₀ ;

Fig. 3 is the foreground image that Fig. 2 obtains through VIBE algorithm processing;

Fig. 4 is the binarized image of Fig. 3;

Figure 5 is an open operation template;

Fig. 6 is the connected image after morphological operation of Fig. 4;

Fig. 7 is the image after size screening of Fig. 6;

FIG. 8 is an image of the determined candidate air leakage area;

FIG. 9 is an image of the air leakage area determined after the judgment in FIG. 8 .

Detailed ways

The specific implementation of the method for detecting the airtightness of a pressure vessel based on background modeling and centroid clustering of the present invention will be described in detail below with reference to the examples.

The method for detecting air tightness of a pressure vessel based on background modeling and centroid clustering of the present invention specifically includes the following steps:

Step 1: Set the original pressure value of the pressure vessel to P ₀ , and the pressure value after T time to P ₁ . When P ₁ -P ₀ ≤Δ, Δ is the pressure difference threshold, which means that it is not pressurized; when P ₁ -P ₀ When >Δ, it means that it is pressurizing, and the air tightness test starts;

Step 2: Take n frame images of the video after the air tightness test starts, n=t*f, where t is the test duration, f is the video frame rate; denote n frame images as I ₁ , I ₂ ,..., I _n , grayscale I _i , _i =1, 2, . *220 size ROI, the video frame rate is 30, considering that the pressure holding time of the pressure vessel needs to be more than 1 minute, so t=60s, n=1800, the original video image is shown in Figure 1, for I ₁₀₀ after The grayscale image G ₁₀₀ obtained after grayscale processing is shown in Figure 2;

前景图像M₁₀₀如图3所示；Step 2.1: Step 2.2: take _G1 for initialization, establish a background model, update the background model with G2... _Gn , and obtain foreground images _M2 ... _{Mn ;} in this embodiment; use VIBE background modeling algorithm, its four parameters are N=20, min=20, R=20,

The foreground image M ₁₀₀ is shown in Figure 3;

Step 2.2: _Binarize M ₂ ... Mn to obtain a binary graph B ₂ ... B _n ; in this embodiment, set 127 as the threshold to perform binarization, and obtain a binary graph B ₂ As shown in Figure 4;

从而得到连通图像C_j,其中J_j＝{L_ju|u＝1,2,...,v_j}，j＝2,3,…,n，J_j表示连通图像C_j中连通域的集合，其中X为结构元素，L_ju为J_j中的第u个连通区域，v_j为J_j中的连通区域个数，运算符“Θ”为腐蚀操作，运算符

为膨胀操作；即先通过腐蚀操作去除小块噪点，再通过膨胀操作将气泡区域连接填充。在本实施例中结构元素模板X为如图5所示的3*3正方形，每个连通区域即代表一个疑似气泡，所得的连通图像C₁₀₀如图6所示；Step 2.3: Perform morphological operations on images B ₂ ... B _n ,

Thus, a connected image C _j is obtained, where J _j ={L _ju |u=1,2,...,v _j }, j=2,3,...,n, J _j represents the connected domain in the connected image C _j set, where X is the structuring element, L _ju is the u-th connected region in J _j , v _j is the number of connected regions in J _j , the operator "Θ" is the erosion operation, the operator

It is the expansion operation; that is, first remove small noises through the erosion operation, and then connect and fill the bubble area through the expansion operation. In this embodiment, the structural element template X is a 3*3 square as shown in FIG. 5 , each connected area represents a suspected bubble, and the obtained connected image C ₁₀₀ is shown in FIG. 6 ;

Step 2.4: N _ju is the number of pixels in the connected region L _ju , all connected regions L _ju satisfying TL < N _ju <TU are retained, and those that are not satisfied are eliminated, and the connected image C _j is eliminated to obtain F _j , and F The connected domain set of _j ={L _jv |v=1,2,...,w _j }, where L _jv is the _vth connected region in D _j , and w _j is the number of connected regions in D _j number, TL is the lower bound of the number of pixels in the bubble area, TU is the upper bound of the number of pixels in the bubble area; in this embodiment, the original image resolution is 640*480, according to the size of the bubble, TL=10, TU=100, use the cvfindcontours function , to obtain the contour of the connected domain. For the 01 binary image, the zero-order moment m ₀₀ of the contour is the area of the connected domain, namely N _ju . F ₁₀₀ is the size-filtered image of C ₁₀₀ , and F ₁₀₀ is shown in Figure 7;

m₀₀表示连通区域L_jv的零阶矩，m₀₁、m₁₀表示连通区域L_jv的一阶矩；Step 2.5: Calculate the centroid coordinates of the connected region L _jv of the connected domain set D _j

m ₀₀ represents the zero-order moment of the connected region L _jv , m ₀₁ , m ₁₀ represent the first-order moment of the connected region L _jv ;

聚类，点集P包含1到m+1帧连通域集合中的全部连通区域质心点坐标，

表示得到初始气泡区域的帧数：DBSCAN聚类半径设置为radius，邻域内最少数据点数设置为minPTs，得到的聚类区域B_k，k＝0,1,...,g,B_k即为候选漏气区域；在本实施例中，m＝300，结合图像大小和气泡冒出的频率设置参数radius＝50,minPTs＝300，即在300帧F_j中，若任意半径为50的区域有超过300个质心点，即认定是漏气区域同时标出漏气区域，找到的疑似漏气区域如图8所示；Step 3: Use the DBSCAN clustering algorithm to analyze the point set

Clustering, the point set P contains the coordinates of the centroid points of all connected regions in the connected domain set of 1 to m+1 frames,

Indicates the number of frames to obtain the initial bubble area: the DBSCAN clustering radius is set to radius, the minimum number of data points in the neighborhood is set to minPTs, and the obtained clustering area B _k , k=0,1,...,g,B _k is Candidate air leakage area; in this embodiment, m=300, the parameters radius=50, minPTs=300 are set in combination with the image size and the frequency of bubbles emerging, that is, in 300 frames _Fj , if any area with a radius of 50 has If there are more than 300 centroid points, it is determined to be an air leakage area and the air leakage area is marked. The found suspected air leakage area is shown in Figure 8;

包含m+2帧到n帧连通域集合的全部连通区域质心点坐标，N_k表示聚类区域B_k中包含点集Q中数据点的数量，若N_k＞T，表示对应的B_k为漏气区域，T为设定的阈值。在本实施例中，结合视频帧率，k＝2，T＝2*(n-m)＝2*(1800-300)＝3000，如图9所示为最终确定的漏气区域，该压力容器存在漏气情况；Step 3.1: Re-judgment the collected candidate air leakage areas, point set

Contains the coordinates of the centroid points of all connected regions in the connected domain set of frames m+2 to n, N _k represents the number of data points in the point set Q included in the clustering region B _k , if N _k >T, it means that the corresponding B _k is Air leakage area, T is the set threshold. In this embodiment, combined with the video frame rate, k=2, T=2*(nm)=2*(1800-300)=3000, as shown in FIG. 9 is the finally determined air leakage area, the pressure vessel exists air leakage;

The content described in the embodiments of the present specification is only an enumeration of the realization forms of the inventive concept, and the protection scope of the present invention should not be regarded as limited to the specific forms stated in the embodiments, and the protection scope of the present invention also extends to the field Equivalent technical means that can be conceived by a skilled person according to the inventive concept.

Claims (1)

1. a pressure vessel airtightness detection method based on background modeling and centroid clustering, is characterized in that, comprises the steps: Step 1: Set the original pressure value of the pressure vessel to P ₀ , and the pressure value after T time to P ₁ . When P ₁ -P ₀ ≤Δ, Δ is the pressure difference threshold, which means that it is not pressurized; when P ₁ -P ₀ When >Δ, it means that it is pressurizing, and the air tightness test starts; Step 2: Take n frame images of the video after the air tightness test starts, n=t*f, where t is the test duration, f is the video frame rate; denote n frame images as I ₁ , I ₂ ,..., I _n , grayscale I _i , i=1,2,...,n to obtain a corresponding grayscale image G _i ; Step 2.1: take _G1 for initialization, establish a background model, update the background model with G2... _Gn , and obtain foreground images _M2 ... _{Mn ;} Step 2.2: Binarize M ₂ ... _Mn to obtain binary graphs B ₂ ... B _n ; Step 2.3: Perform morphological operations on images B ₂ ... B _n ,

Thus, a connected image C _j is obtained, where J _j ={L _ju |u=1,2,...,v _j }, j=2,3,...,n, J _j represents the connected domain in the connected image C _j set, where X is the structuring element, L _ju is the u-th connected region in J _j , v _j is the number of connected regions in J _j , the operator "Θ" is the erosion operation, the operator for the expansion operation; Step 2.4: N _ju is the number of pixels in the connected region L _ju , all connected regions L _ju satisfying TL < N _ju <TU are retained, and those that are not satisfied are eliminated, and the connected image C _j is eliminated to obtain F _j , and F The connected domain set of _j ={L _jv |v=1,2,...,w _j }, where L _jv is the _vth connected region in D _j , and w _j is the number of connected regions in D _j number, TL is the lower bound of the number of pixels in the bubble area, TU is the upper bound of the number of pixels in the bubble area; Step 2.5: Calculate the centroid coordinates O _jv (x, y) of the connected region L _jv of the connected domain set D _j ,

m ₀₀ represents the zero-order moment of the connected region L _jv , m ₀₁ , m ₁₀ represent the first-order moment of the connected region L _jv ; Step 3: Use the DBSCAN clustering algorithm to analyze the point set

Clustering, the point set P contains the coordinates of the centroid points of all connected regions in the connected domain set of 1 to m+1 frames,

Indicates the number of frames to obtain the initial bubble area: the DBSCAN clustering radius is set to radius, the minimum number of data points in the neighborhood is set to minPTs, and the obtained clustering area B _k , k=0,1,...,g, B _k is Candidate leak area; Step 3.1: Re-judgment the collected candidate air leakage areas, point set

Contains the coordinates of the centroid points of all connected regions in the connected domain set of frames m+2 to n, N _k represents the number of data points in the point set Q included in the clustering region B _k , if N _k >T, it means that the corresponding B _k is Air leakage area, T is the set threshold.

Images