Sewage pipeline closed water test leakage point automatic identification method adopting machine vision

Abstract

The invention provides a method for automatically identifying leakage points in a closed water test of a sewage pipeline by adopting machine vision. The method comprises the following steps: acquiring multi-modal image information in the pipeline through various sensors; constructing a three-dimensional point cloud model of the inner wall of the pipeline according to the multi-modal image information; obtaining and separating a static background and a dynamic leakage area of the image; acquiring flow velocity field information of the dynamic leakage area through a deep learning identification network; and analyzing and acquiring leakage information by combining the data information set and the flow velocity field information of the three-dimensional point cloud model. According to the method, the static background and the dynamic leakage area are effectively identified and distinguished through fusion analysis of the multi-modal images, so that the influence of environments such as pipeline stains is effectively avoided, leakage information is obtained by calculating the flow velocity field of the dynamic leakage area, and the detection accuracy is improved. Finally, the actual leakage point position of the pipeline is obtained through the image coordinates, and the method has the advantages of being accurate, efficient, convenient and rapid.

Description

Technical Field

[0001] The present invention relates to the field of pipeline water tightness tests, and particularly relates to an automatic identification method for leakage points in sewage pipeline water tightness tests using machine vision. Background Art

[0002] The sewage pipeline water tightness test is a key link for evaluating the sealing performance of pipelines. According to the requirements of municipal construction regulations, the water tightness test must be carried out before backfilling. In the water tightness test, after blocking both ends of the built pipeline, water is injected into the pipeline through the wellhead. The water surface height at the wellhead is generally about 2 meters higher than the top of the pipeline. After injection, it is soaked for a period of time (usually 24h) to avoid the influence caused by water absorption of the pipeline, etc. After the soaking is over, the water tightness test is carried out, that is, the change in the water surface height at the wellhead is measured. The leakage volume and leakage rate can be calculated through the change in the water surface height within a period of time. If leakage occurs, the leakage point needs to be found and repaired.

[0003] For the identification of leakage points, traditional methods rely on manual visual inspection or sensor monitoring. Manual detection has low efficiency and strong subjectivity, and it is difficult to detect tiny leaks; existing sensor technologies (such as pressure sensors, humidity sensors) require complex equipment to be deployed, with high costs and are easily interfered by the environment. In recent years, due to its non-contact, high-precision and high-efficiency characteristics, machine vision technology has been widely used in the field of industrial inspection, but it has not been maturely applied to the leakage detection of pipeline water tightness tests. Existing vision methods mostly rely on single image features (such as color changes), and have insufficient anti-interference ability for complex backgrounds (such as stains and water stains on the inner wall of the pipeline), resulting in a high false detection rate. Summary of the Invention

[0004] In order to solve the problems existing in the background art, the present invention proposes an automatic identification method for leakage points in sewage pipeline water tightness tests using machine vision.

[0005] An automatic identification method for leakage points in sewage pipeline water tightness tests using machine vision includes the steps of:

[0006] S100. Obtain multi-modal image information inside the pipeline through a variety of sensors;

[0007] S200. Construct a three-dimensional point cloud model of the inner wall of the pipeline according to the multi-modal image information;

[0008] S300. Obtain and separate the static background and dynamic leakage area of the image;

[0009] S400. Obtain the velocity field information of the dynamic leakage area through a deep learning recognition network;

[0010] S500. Combine the data information set of the three-dimensional point cloud model and the velocity field information, and analyze and obtain leakage information.

[0011] Based on the above, the sensor includes multiple groups of visible light cameras, near-infrared cameras, and thermal imagers. The visible light cameras are used to collect visible light image information, the near-infrared cameras are used to collect near-infrared image information, and the thermal imagers are used to collect thermal infrared band image information.

[0012] Based on the above, in step S200, a three-dimensional point cloud model of the pipeline inner wall is constructed by a stereo matching algorithm based on structured light projection and binocular vision.

[0013] Based on the above, in step S300, an improved ViBe background modeling algorithm is used to separate the static background and the dynamic leakage area. The improved ViBe algorithm adopts a spatio-temporal joint update strategy. The update formula for the background model B of pixel point x is:

[0014] B t+1 (x) = αB t (x) + (1 - α)(I t (x) + βΔD(x))

[0015] where α = 0.95, β = 0.2, ΔD is the spatio-temporal domain difference amount, and I represents the light intensity value at pixel x at the current frame time t.

[0016] Based on the above, in step S400, a three-channel deep learning recognition network including spatial texture features, temporal motion features, and thermodynamic features is constructed. For spatial texture feature extraction, an improved EfficientNet-B4 network with a channel attention module is used. For temporal motion feature extraction, a 3D ResNeXt structure is used, and the input is a stack of 10 consecutive frames of images. For thermodynamic feature analysis, an infrared image gradient histogram descriptor is used.

[0017] Based on the above, the leakage information includes leakage volume information and leakage velocity information. The leakage volume Q is:

[0018]

[0019] where v i is the flow velocity of the i-th pixel area, A i is the leakage area of the i-th area, and Δt is the time interval.

[0020] Based on the above, the sensor further includes an inertial navigation system, which is used to position the detection device, establish a corresponding unified coordinate system with the pipeline after obtaining the image information, associate the image pixel coordinate area with the actual position of the pipeline through image recognition, and obtain the actual leakage point position information of the pipeline according to the detected leakage point pixel area.

[0021] Based on the above, step S600 is included, where the detected leakage information is analyzed and compared with the actually measured leakage information in the closed water test, and the identified and detected leakage information is evaluated and verified according to the comparison result.

[0022] The present invention has prominent substantive features and significant progress compared with the prior art. Specifically, through the fusion analysis of multi-modal images, the present invention can effectively identify and distinguish static backgrounds and dynamic leakage areas, thereby effectively avoiding the influence of environments such as pipeline stains. By calculating the velocity field of the dynamic leakage area, leakage information is obtained, and the accuracy of detection and identification can be verified by comparing the leakage information with the actually measured leakage information. Finally, the actual leakage point position of the pipeline is obtained through image coordinates, which has the advantages of accuracy, high efficiency, convenience and speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 is a flowchart of the present invention.

[0024] Figure 2 is a table of comparative experimental results of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0025] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0026] As Figure 1 shown, an automatic leakage point identification method for the water tightness test of sewage pipelines using machine vision includes the steps of: S100, obtaining multi-modal image information inside the pipeline through multiple sensors; S200, constructing a three-dimensional point cloud model of the inner wall of the pipeline according to the multi-modal image information; S300, obtaining and separating the static background and dynamic leakage areas of the image; S400, obtaining the velocity field information of the dynamic leakage area through a deep learning recognition network; S500, analyzing and obtaining leakage information by combining the data information set of the three-dimensional point cloud model and the velocity field information.

[0027] In use, if there is leakage during the closed - water test, a visual robot is put into the pipeline through the wellhead. The robot carrier is a six - wheel all - wheel drive chassis. The traveling speed of the robot can be set as needed, such as 0.25 m / s. It is equipped with a six - axis IMU inertial navigation system with a positioning accuracy of ±2 cm. A variety of sensors are set on the visual robot. The visual robot moves along the axial direction of the pipeline and acquires the internal image information of the pipeline during the movement. The inertial navigation system is used to position the detection equipment. After obtaining the image information, a unified coordinate system corresponding to the pipeline is established. Through image recognition, the image pixel coordinate area is associated with the actual position of the pipeline, that is, the actual position area of the pipeline can be obtained corresponding to the image pixels. It should be noted that the leakage point only needs to be determined within a certain range. In reality, after knowing the leakage point area, workers can enter the pipeline to confirm and repair it, without having to check a large area of the pipeline, which greatly improves the efficiency. In this embodiment, the sensors include six groups of visible - light cameras, two groups of near - infrared cameras, and one group of thermal imagers. The visible - light cameras use Sony IMX586 sensors with a pixel size of 1.6 μm. The six groups of visible - light cameras are arranged in a ring shape and are used to collect visible - light image information; the response band of the near - infrared cameras is 900 - 1700 nm, and the quantum efficiency > 65% @ 1300 nm. The near - infrared cameras are used to collect near - infrared image information; the thermal imager is a FLIR thermal imager with a resolution of 640×512 and a temperature resolution of 0.05 °C. The thermal imager is used to collect thermal - infrared band image information. After constructing a three - dimensional point - cloud model of the pipeline inner wall based on multi - modal image information, an improved background extraction algorithm, the ViBe algorithm, is used to obtain and separate the static background and dynamic leakage areas of the image. Then, through a deep - learning recognition network, the flow - field information of the dynamic leakage area is obtained. Finally, combined with the data set of the three - dimensional point - cloud model and the flow - field information, the leakage information is analyzed and obtained.

[0028] Specifically, after constructing a three - dimensional point - cloud model of the pipeline inner wall through an existing stereo matching algorithm based on structured - light projection and binocular vision, an improved ViBe background - modeling algorithm is used to separate the static background and dynamic leakage areas. The improved ViBe algorithm adopts a spatio - temporal joint update strategy. The update formula for the background model B of pixel point x is:

[0029] B t+1 (x) = αB t (x)+(1 - α)(I t (x)+βΔD(x))

[0030] where α = 0.95, β = 0.2, ΔD is the spatio - temporal domain difference amount, and I represents the light intensity value at pixel x at time t of the current frame. According to the update information of the background model of the pixel point over time, the static background and dynamic leakage areas can be judged and separated.

[0031] Construct a three-channel deep learning recognition network that includes spatial texture features, temporal motion features, and thermodynamic features. For the extraction of spatial texture features, an improved EfficientNet-B4 network with a channel attention module is used. For the extraction of temporal motion features, a 3D ResNeXt structure is adopted, and the input is a stack of 10 consecutive frames of images. For the analysis of thermodynamic features, an infrared image gradient histogram descriptor is used. Through the three-channel feature fusion learning recognition of the spatial texture features, temporal motion features, and thermodynamic features combined with image information, the flow velocity field information of the leakage point is obtained. By combining the three-dimensional point cloud data with the flow velocity field information, parameters such as the leakage area and flow velocity are output through the deep learning recognition network, and then the leakage volume Q can be calculated as follows:

[0032]

[0033] where v i is the flow velocity of the i-th pixel area, A i is the leakage area of the i-th area, and Δt is the time interval.

[0034] In practice, the detected leakage information is also analyzed and compared with the leakage information actually measured in the water tightness test. That is, the leakage flow velocity and leakage volume information identified and detected are compared and analyzed with the leakage velocity and leakage volume detected during the water tightness test, so as to judge whether the results of the identification and detection are accurate, and then evaluate and verify the results of the identification and detection. Taking a cast iron pipe with a pipe diameter of DN1000 and a detection section length of 50m as an example, with 3 actual leakage points set (the leakage velocities are 0.1L / min, 2L / min, and 10L / min respectively), compared with the traditional single-image feature detection method and the single acoustic wave detection method, the comparison results are as Figure 2 shown, and the accuracy and efficiency of the recognition method in this embodiment are greatly improved.

[0035] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, in any aspect, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present invention. Any reference signs in the claims should not be regarded as limiting the claimed rights.

Claims

1. An automatic leakage point identification method for closed water test of sewage pipelines using machine vision, characterized in that, Including the steps: S100. Obtain multi-modal image information inside the pipeline through multiple sensors; S200. Construct a three-dimensional point cloud model of the inner wall of the pipeline according to the multi-modal image information; S300. Obtain and separate the static background and the dynamic leakage area of the image; S400. Obtain the flow velocity field information of the dynamic leakage area through a deep learning recognition network; S500. Combine the data information set of the three-dimensional point cloud model and the flow velocity field information to analyze and obtain the leakage information.

2. The automatic leakage point identification method for the water tightness test of sewage pipelines using machine vision according to claim 1, characterized in that: The sensors include multiple groups of visible light cameras, near-infrared cameras, and thermal imagers. The visible light cameras are used to collect visible light image information, the near-infrared cameras are used to collect near-infrared image information, and the thermal imagers are used to collect thermal infrared band image information.

3. The automatic leakage point identification method for the water tightness test of sewage pipelines using machine vision according to claim 1, characterized in that: In step S200, a three-dimensional point cloud model of the inner wall of the pipeline is constructed through a stereo matching algorithm based on structured light projection and binocular vision.

4. The automatic leakage point identification method for the water pressure test of sewage pipelines using machine vision according to claim 1, wherein: In step S300, an improved ViBe background modeling algorithm is used to separate the static background and the dynamic leakage area. The improved ViBe algorithm adopts a spatio-temporal joint update strategy, and the background model B update formula for pixel point x is: B t+1 B(x) = αB t (x) + (1 - α)(I t (x) + βΔD(x)) where α = 0.95, β = 0.2, ΔD is the spatio-temporal domain difference amount, and I represents the light intensity value at pixel x at the current frame time t.

5. The automatic leakage point identification method for the water tightness test of sewage pipelines using machine vision according to claim 1, characterized in that: In step S400, a three-channel deep learning recognition network including spatial texture features, temporal motion features, and thermodynamic features is constructed. The spatial texture feature extraction uses an improved EfficientNet-B4 network with a channel attention module added, the temporal motion feature extraction uses a 3D ResNeXt structure, and the input is a stack of 10 consecutive frames of images. The thermodynamic feature analysis uses an infrared image gradient histogram descriptor.

6. The automatic leakage point identification method for the water tightness test of sewage pipelines using machine vision according to claim 1, characterized in that: The leakage information includes leakage volume information and leakage velocity information. The leakage volume Q is: Among them, v i is the flow velocity of the i-th pixel region, A i is the leakage area of the i-th region, and Δt is the time interval.

7. The automatic leakage point identification method for the water tightness test of sewage pipelines using machine vision according to claim 1, characterized in that: The sensors also include an inertial navigation system, which is used to position the detection device, establish a corresponding unified coordinate system with the pipeline after obtaining the image information, associate the image pixel coordinate area with the actual position of the pipeline through image recognition, and obtain the actual leakage point position information of the pipeline according to the detected leakage point pixel area.

8. The automatic leakage point identification method for the water-closed test of sewage pipelines using machine vision according to claim 1, characterized in that: Including step S600, analyze and compare the detected leakage information with the actually measured leakage information of the water tightness test, and evaluate and verify the detected leakage information according to the comparison result.

Images