CN109029861B - Pressure container air tightness detection method based on background modeling and centroid clustering - Google Patents

Abstract

The invention discloses a background modeling and quality based methodThe method for detecting the air tightness of the pressure container by heart clustering comprises the following steps: step 1, setting the original air pressure value of the pressure container as P₀After T time, the air pressure value is P₁When P is₁‑P₀When the pressure is less than or equal to delta, the delta is a pressure difference threshold value, and the pressure is not pressurized; when P is present₁‑P₀When the pressure is larger than delta, the pressure is applied, and the air tightness test is started; step 2: taking n frames of images of a video after the start of the air tightness test, wherein n is t f, t is the test duration, and f is the video frame rate; recording n frame images as I₁,I₂,...,I_nTo 1, pair_iAnd (3) graying the image with i being 1,2, …, and n to obtain a corresponding grayscale image G_i(ii) a The invention has the beneficial effects that: the method is based on background modeling and centroid clustering, and can accurately judge the air tightness of the pressure container and accurately position an air leakage area by combining the size, the quantity and the dynamic characteristics of the bubbles.

Description

Pressure container air tightness detection method based on background modeling and centroid clustering

Technical Field

The invention belongs to the field of safety monitoring and detection of special equipment, and particularly relates to a method for detecting the air tightness of a pressure container in a water immersion method based on background modeling and a centroid clustering algorithm.

Background

The pressure container is a common device for storing gas or liquid, and belongs to the field of special devices. The pressure container with the quality hidden trouble has the advantages that the functions of the product are influenced, and dangerous events such as fire, explosion and the like can be caused in serious cases, so that the air tightness test of the pressure container in the production process is an important link for ensuring the quality.

Most pressure vessel manufacturers in China use a water immersion method for air tightness detection, the air tightness of the pressure vessel is determined by manually observing the condition of bubbles in water in the traditional method, but long-time uninterrupted detection easily causes eye fatigue of detection personnel to influence detection precision, so that the condition of 'air leakage is not detected' is caused, meanwhile, the pressure vessel manufacturers usually adopt a piece counting system, the condition that workers subjectively detect less or give up detection is easily caused, and the problem of 'detecting less and detecting missing' is caused.

Therefore, the computer vision technology is used for detecting the bubbles in the water, and meanwhile, the feasibility and the practical significance are high when the water immersion method is combined for detecting the air tightness of the pressure container. Johnsson F (Chalmers university of Technology, 2004), et al, conducted research analysis of the size, velocity, and void fraction of bubbles in a two-dimensional fluidized bed using computer vision techniques. In the study of the behavior of bubbles in a gas-solid two-phase fluid, the size and the velocity of the bubbles are detected by using an image processing technology in combination with Busciglico A (Italy, University of Palermo, 2009) and the like. Zielinski (University of Oldenburg, 2010, germany) and others applied the optical flow method to the detection of bubbles in water and analyzed the feasibility by experiments. The method is characterized in that the rising process of bubbles in a gas-liquid two-phase flow field is researched by Wanghong-Ying (Tianjin university, 2010) and the like, a high-speed camera is used for shooting continuous images of bubbles generated by air leakage points with different diameters in the rising process, and parameters (including speed, acceleration, radius, area and the like) of the bubbles are measured by a computer vision technology. Shore construction and technology (university of western ann's science 2011) provide a watershed algorithm based on morphological theory, perform bubble extraction on a water-doped flow image, and provide a bubble image preprocessing strategy consisting of reverse color, image gray level adjustment and low-pass filtering. Liuwei (northeast electric university, 2013) conducted image enhancement studies on images of bubbles in water based on digital image processing techniques. By Matlab simulation experiments, the effects of several kinds of threshold segmentation are compared, and an effective algorithm is provided for obtaining the shape characteristics of the bubbles. Wuchunlong (Zhejiang university of rational engineers, 2013) provides a set of automatic air tightness detection system based on PLC control, and applies a Hom-Schunck image processing algorithm based on an optical flow theory to the identification and tracking of bubbles in a gas-liquid two-phase flow field, so that the detection and identification of bubbles at an air leakage point of a pressure container are realized. Ganjiawei (seihua university, 2015) proposes an FPGA-based bubble edge detection image processing system. The FPGA is used as a core, the Sobel edge detection algorithm is adopted to obtain the bubble edge characteristics, and the feasibility is verified through the Matlab experiment.

The computer vision detection technology mentioned in the above documents, which has been the preliminary research on bubbles in water, still has many disadvantages:

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

2) some algorithms are high in computational complexity and time complexity, such as an air bubble identification algorithm based on an optical flow method, and the algorithm has good feasibility and reliability under the conditions of fewer air bubbles, slow speed rise and no other interference. However, under the condition that the calculation time of the detection algorithm is too long and the number of bubbles is too large, the requirement of real-time detection cannot be met.

3) The cameras in the methods shoot the pressure container through the glass on the side surface of the water tank of the detection pool, only a single pressure container can be detected each time, the theoretical research is emphasized, the bubble generation process is recorded, and the bubble generation condition and the bubble development process are analyzed. In the actual production environment of the pressure containers, a plurality of pressure containers in a batch are sent into the detection pool together and are placed side by side, and the detection is carried out by the soaking method. Therefore, the detection algorithm for capturing bubbles from the side cannot be applied to an actual scene.

Disclosure of Invention

In order to realize objective and efficient pressure container air tightness detection, the dynamic characteristic, continuity and quantity of bubble emergence are taken as the basis, and in order to overcome the defects of the existing method, the invention provides a pressure container air tightness detection method based on background modeling and centroid clustering.

A pressure container airtightness detection method based on background modeling and centroid clustering is characterized by comprising the following steps:

step 1, arranging a pressure containerOriginal air pressure value is P₀After T time, the air pressure value is P₁When P is₁-P₀When the pressure is less than or equal to delta, the delta is a pressure difference threshold value, and the pressure is not pressurized; when P is present₁-P₀When the pressure is larger than delta, the pressure is applied, and the air tightness test is started;

step 2: taking n frames of images of a video after the start of the air tightness test, wherein n is t f, t is the test duration, and f is the video frame rate; recording n frame images as I₁,I₂,...,I_nTo 1, pair_iAnd (3) graying the image with i being 1,2, …, and n to obtain a corresponding grayscale image G_i；

Step 2.1: get G₁Initializing, establishing background model, and using G₂...G_nUpdating the background model to obtain a foreground image M₂...M_n；

Step 2.2: to M₂...M_nCarrying out binarization to obtain a binary image B₂...B_n；

Step 2.3: for image B₂…B_nThe morphological operation is carried out on the obtained object,

thereby obtaining a connected image C_jWherein J_j＝{L_ju|u＝1,2,...,v_j}，j＝2,3,…,n，J_jRepresenting connected images C_jA collection of mesomeric domains, wherein X is a structural element, L_juIs J_jThe u-th communication area of (1), v_jIs J_jThe number of connected regions in (1), operator 'theta' as corrosion operation, operatorIs an expansion operation;

step 2.4: n is a radical of_juIs a connected region L_juThe number of the pixel points is that all the pixels satisfy TL < N_juConnected domain L of < TU_juRetained, unsatisfied culling, connected image C_jRemoving to obtain F_jAnd F_jConnected domain set D_j＝{L_jv|v＝1,2,...,w_jIn which L is_jvIs D_jThe v-th communicating region of (1), w_jIs D_jTL is the lower bound of the number of the pixel points of the bubble area, and TU is the upper bound of the number of the pixel points of the bubble area;

step 2.5: computing a set of connected domains D_jIs connected to the region L_jvCoordinates of center of mass

m₀₀Indicates a connected region L_jvZero order moment of (m)₀₁、m₁₀Indicates a connected region L_jvThe first moment of (d);

and step 3: adopting DBSCAN clustering algorithm to collect point pairsClustering, wherein the point set P comprises the coordinates of the centroid points of all connected regions in the 1 to m +1 frame connected region set,

number of frames representing the area where the initial bubble was obtained: the clustering radius of the DBSCAN 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_kNamely the candidate air leakage area;

step 3.1: judging the collected candidate air leakage area again, and collecting the points

Coordinates of centroid points of all connected regions containing m +2 frames to N frames of connected region set, N_kRepresents a clustering region B_kThe number of data points in the point set Q, if N_k> T, denotes the corresponding B_kT is a set threshold value for the air leakage area.

The invention has the beneficial effects that: the method is based on background modeling and centroid clustering, and can accurately judge the air tightness of the pressure container and accurately position an air leakage area by combining the size, the quantity and the dynamic characteristics of the bubbles.

Drawings

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

FIG. 2 is a pair I₁₀₀Obtaining an image after gray processing;

FIG. 3 is a foreground image obtained from the processing of the VIBE algorithm of FIG. 2;

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

FIG. 5 is an open operation template;

FIG. 6 is a connected image of FIG. 4 after morphological operations;

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

FIG. 8 is an image of a determined leak candidate area;

fig. 9 is an image of the air-leakage area determined after the determination of fig. 8.

Detailed Description

The following describes in detail a specific embodiment of the present invention based on background modeling and centroid clustering for airtight detection of pressure vessel with reference to the following examples.

The invention discloses a pressure container airtightness detection method based on background modeling and centroid clustering, which specifically comprises the following steps:

step 1, setting the original air pressure value of the pressure container as P₀After T time, the air pressure value is P₁When P is₁-P₀When the pressure is less than or equal to delta, the delta is a pressure difference threshold value, and the pressure is not pressurized; when P is present₁-P₀When the pressure is larger than delta, the pressure is applied, and the air tightness test is started;

step 2: taking n frames of images of a video after the start of the air tightness test, wherein n is t f, t is the test duration, and f is the video frame rate; recording n frame images as I₁,I₂,...,I_nTo 1, pair_iAnd (3) graying the image with i being 1,2, …, and n to obtain a corresponding grayscale image G_i(ii) a In this embodiment, the original video image has a resolution of 640 × 480, an ROI size of 600 × 220 is defined, the video frame rate is 30, and considering that the pressure holding time period after pressurizing the pressure vessel needs to be 1 minute or more, t is 60s, n is 1800, and the original video image is shown in fig. 1, for I₁₀₀Gray scale image G obtained after gray scale processing₁₀₀As shown in fig. 2;

step 2.1: step 2.2: get G₁Initializing, establishing background model, and using G₂...G_nUpdating the background model to obtain a foreground image M₂...M_n(ii) a In the present embodiment; using the VIBE background modeling algorithm, the four parameters are N-20, min-20, R-20,

foreground image M₁₀₀As shown in fig. 3;

step 2.2: to M₂...M_nCarrying out binarization to obtain a binary image B₂...B_n(ii) a In this embodiment, 127 is set as a threshold value, and binarization is performed, so that a resulting binary image B is obtained₂As shown in fig. 4;

step 2.3: for image B₂…B_nThe morphological operation is carried out on the obtained object,

thereby obtaining a connected image C_jWherein J_j＝{L_ju|u＝1,2,...,v_j}，j＝2,3,…,n，J_jRepresenting connected images C_jA collection of mesomeric domains, wherein X is a structural element, L_juIs J_jThe u-th communication area of (1), v_jIs J_jThe number of connected regions in (1), operator 'theta' as corrosion operation, operator

Is an expansion operation; namely, small noise points are removed through corrosion operation, and then the bubble area is filled in a connecting mode through expansion operation. In this embodiment, the structural element template X is a 3 × 3 square as shown in fig. 5, each connected region represents a suspected bubble, and the connected image C is obtained₁₀₀As shown in fig. 6;

step 2.4: n is a radical of_juIs a connected region L_juThe number of the pixel points is that all the pixels satisfy TL < N_juConnected domain L of < TU_juPick of retained and unsatisfiedRemove, connect image C_jRemoving to obtain F_jAnd F_jConnected domain set D_j＝{L_jv|v＝1,2,...,w_jIn which L is_jvIs D_jThe v-th communicating region of (1), w_jIs D_jTL is the lower bound of the number of the pixel points of the bubble area, and TU is the upper bound of the number of the pixel points of the bubble area; in this embodiment, the original image resolution is 640 × 480, TL is 10 and TU is 100 according to the size of the bubble, a connected domain profile is obtained using cvfindcontours function, and for a 01 binary image, the zeroth moment m of the profile is₀₀I.e. the area of the connected domain, i.e. N_ju。F₁₀₀Is C₁₀₀Size-screened image, F₁₀₀As shown in fig. 7;

step 2.5: computing a set of connected domains D_jIs connected to the region L_jvCoordinates of center of mass

m₀₀Indicates a connected region L_jvZero order moment of (m)₀₁、m₁₀Indicates a connected region L_jvThe first moment of (d);

and step 3: adopting DBSCAN clustering algorithm to collect point pairs

Clustering, wherein the point set P comprises the coordinates of the centroid points of all connected regions in the 1 to m +1 frame connected region set,

number of frames representing the area where the initial bubble was obtained: the clustering radius of the DBSCAN 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_kNamely the candidate air leakage area; in the present embodiment, m is 300, the combined image size and bubble emergence frequency setting parameters radius is 50, minPTs is 300, i.e., 300 frames F_jIn the above, if there is any region with a radius of 50If the number of the centroid points exceeds 300, the area is determined to be an air leakage area, and the air leakage area is marked at the same time, and the found suspected air leakage area is shown in fig. 8;

step 3.1: judging the collected candidate air leakage area again, and collecting the points

Coordinates of centroid points of all connected regions containing m +2 frames to N frames of connected region set, N_kRepresents a clustering region B_kThe number of data points in the point set Q, if N_k> T, denotes the corresponding B_kT is a set threshold value for the air leakage area. In the present embodiment, in combination with the video frame rate, k is 2, T is 2 (n-m) is 2 (1800-;

the embodiments described in this specification are merely illustrative of implementations of the inventive concept and the scope of the present invention should not be considered limited to the specific forms set forth in the embodiments but rather by the equivalents thereof as may occur to those skilled in the art upon consideration of the present inventive concept.

Claims (1)

1. A pressure container airtightness detection method based on background modeling and centroid clustering is characterized by comprising the following steps:

step 1, setting the original air pressure value of the pressure container as P₀After T time, the air pressure value is P₁When P is₁-P₀When the pressure is less than or equal to delta, the delta is a pressure difference threshold value, and the pressure is not pressurized; when P is present₁-P₀When the pressure is larger than delta, the pressure is applied, and the air tightness test is started;

step 2: taking n frames of images of a video after the start of the air tightness test, wherein n is t f, t is the test duration, and f is the video frame rate; recording n frame images as I₁,I₂,...,I_nTo 1, pair_iAnd (3) graying the image with i being 1,2, …, and n to obtain a corresponding grayscale image G_i；

Step 2.1: get G₁Initializing, establishing background model, and using G₂...G_nUpdating the background model to obtain a foreground image M₂...M_n；

Step 2.2: to M₂...M_nCarrying out binarization to obtain a binary image B₂...B_n；

Step 2.3: for image B₂…B_nThe morphological operation is carried out on the obtained object,

thereby obtaining a connected image C_jWherein J_j＝{L_ju|u＝1,2,...,v_j}，j＝2,3,…,n，J_jRepresenting connected images C_jA collection of mesomeric domains, wherein X is a structural element, L_juIs J_jThe u-th communication area of (1), v_jIs J_jThe number of connected regions in (1), operator 'theta' as corrosion operation, operatorIs an expansion operation;

step 2.4: n is a radical of_juIs a connected region L_juThe number of the pixel points is that all the pixels satisfy TL < N_juConnected domain L of < TU_juRetained, unsatisfied culling, connected image C_jRemoving to obtain F_jAnd F_jConnected domain set D_j＝{L_jv|v＝1,2,...,w_jIn which L is_jvIs D_jThe v-th communicating region of (1), w_jIs D_jTL is the lower bound of the number of the pixel points of the bubble area, and TU is the upper bound of the number of the pixel points of the bubble area;

step 2.5: computing a set of connected domains D_jIs connected to the region L_jvCentroid coordinate O of_jv(x,y)，

m₀₀Indicates a connected region L_jvZero order moment of (m)₀₁、m₁₀Indicates a connected region L_jvThe first moment of (d);

and step 3: clustering with DBSCANAlgorithm point-to-point set

Clustering, wherein the point set P comprises the coordinates of the centroid points of all connected regions in the 1 to m +1 frame connected region set,

number of frames representing the area where the initial bubble was obtained: the clustering radius of the DBSCAN 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_kNamely the candidate air leakage area;

step 3.1: judging the collected candidate air leakage area again, and collecting the points

Coordinates of centroid points of all connected regions containing m +2 frames to N frames of connected region set, N_kRepresents a clustering region B_kThe number of data points in the point set Q, if N_k> T, denotes the corresponding B_kT is a set threshold value for the air leakage area.

Images