Vision-based intelligent monitoring method and system for discharge flow rate of drainage outlet, and medium

Abstract

The present invention relates to the technical field of drainage monitoring, and in particular to a vision-based intelligent monitoring method and system for the discharge flow rate of a drainage outlet, and a medium. The method comprises: S1, acquiring an image of a pipe drainage outlet area in real time by means of a camera; S2, on the basis of a pre-constructed deep learning network model, identifying a pipe drainage outlet and water flow in the image, and generating a corresponding bounding box for each identified object; S3, on the basis of the bounding boxes of the pipe drainage outlet and the water flow, calculating the ratio rw of the water flow width to a pipe diameter; and S4, on the basis of a pre-constructed polynomial regression model used for representing a nonlinear relationship between the flow rate Q of the drainage outlet and the ratio rw, calculating the flow rate of the current pipe drainage outlet. According to the present invention, computer vision and deep learning technologies are combined, so that non-contact, accurate and automatic estimation of the flow rate of the pipe drainage outlet can be realized.

Description

Vision-based intelligent monitoring method, system and medium for drainage outlet discharge Technical Field

[0001] The present invention relates to the technical field of drainage monitoring, and more particularly to a vision-based intelligent monitoring method, system and medium for drainage outlet discharge volume. Background Art

[0002] With the rapid development of computer vision technology and the growing emphasis on environmental protection, the discharge of industrial and domestic wastewater has received unprecedented attention. To ensure that wastewater treatment meets standards, accurate monitoring and counting of discharge volume at outlets is crucial. While physical sensors such as flow meters have made some progress in this area, their performance can be affected by prolonged exposure to high temperatures, high pressures, or corrosive environments, leading to inaccurate data. Furthermore, these devices require regular maintenance and calibration, which not only increases operating costs but can also lead to data errors due to improper maintenance.

[0003] Modern computer vision monitoring technology can analyze image data in real time, automatically identify anomalies, and provide real-time warnings to environmental protection departments. Furthermore, the installation and maintenance costs of visual monitoring equipment are relatively low, saving businesses and environmental protection departments significant operating costs in the long term.

[0004] Therefore, how to apply visual monitoring technology to drain outlet monitoring and accurately estimate the drain outlet flow is an urgent problem that technical personnel in this field need to solve.

[0005] Summary of the Invention

[0006] In view of this, the present invention provides a vision-based intelligent monitoring method, system and medium for drain outlet discharge, which can realize contactless and accurate estimation of the flow rate of pipeline drain outlets.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] A vision-based intelligent monitoring method for drainage outlet discharge volume includes the following steps:

[0009] S1, collecting images of the pipe drain area in real time through a camera;

[0010] S2, based on the pre-built deep learning network model, identifies the pipe drains and water flows in the image and generates corresponding bounding boxes for each identified object;

[0011] S3. Calculate the ratio r of the water flow width to the pipe diameter based on the bounding box of the pipe drain outlet and the water flow. w ;

[0012] S4, based on the pre-built method for characterizing the outlet flow Q and ratio r w The flow rate of the current pipe drain outlet is calculated using a polynomial regression model based on the nonlinear relationship between them.

[0013] Furthermore, S2 also includes:

[0014] caching the pipe drain outlet bounding box identified by the deep learning network model and regularly updating the cache;

[0015] Before the deep learning network model performs pipe drain outlet detection, it first determines whether there is cached data of the pipe drain outlet boundary box. If so, it continues to determine whether the cached data needs to be updated. If the cached data does not exist or needs to be updated, the deep learning network model performs pipe drain outlet detection on the current image, generates a drain outlet detection result, and saves the cached data.

[0016] If the cached data does not need to be updated, the detection result of the drain outlet in the cached data will be used as the detection result at the current moment, and it will be determined whether there is a pipe drain outlet and its boundary box in the detection result of the cached data. If not, it means that the camera is abnormal. If so, the water flow in the current image is detected based on the deep learning network model, and a water flow detection result is generated. If there is water flow information, the flow of the current pipe drain outlet is calculated. Otherwise, it is determined that the current drain outlet is in a non-drainage state.

[0017] Furthermore, in S3, the ratio r of the pipe radius to the water flow width is w The calculation process includes:

[0018] Determine the upper left coordinate point (x1, y1) and the lower right coordinate point (x2, y2) of the pipe drain outlet boundary box;

[0019] Determine the upper left position coordinate point (x3, y3) and the lower right position coordinate point (x4, y4) of the water flow bounding box;

[0020] Calculate the horizontal pixel width of the water flow: x4-x3, calculate the horizontal pixel width of the pipe drain: x2-x1;

[0021] Calculate the water flow ratio r w =(x4-x3) / (x2-x1).

[0022] Furthermore, after determining the location coordinates of the pipe drain outlet, the following steps are also included:

[0023] The bounding box of the pipe drain outlet is expanded according to a percentage k of (x2-x1), where k<0.5.

[0024] Furthermore, in S4, the expression of the polynomial regression model is: Q = a × r 2 +b×h 2 +c×r×h+d×r+e×h+f

[0025] Where h = 1-cos(arcsin(r w )), represents the water height; Q represents the flow rate of the drain outlet at the current moment, a, b, c, d, e and f are fitting parameters, which serve as weight parameters of the polynomial regression model.

[0026] Furthermore, in S4, the process of constructing the polynomial regression model includes:

[0027] Collect the actual radius r of various physical pipes and record the actual drainage flow Q and water height h under different pipe radii;

[0028] Clean up abnormal data;

[0029] Add restriction data to the collected data to limit the result output of the model data r≤0 and h≤0;

[0030] The gradient descent method is used to continuously update the model's weight parameters a, b, c, d, e, and f.

[0031] Furthermore, S4 also includes:

[0032] Using a sliding window approach, the average value of the flow data is calculated as the final flow output result;

[0033] If the valid data in the window is less than the set value, it is determined that the current state is non-drainage.

[0034] It can be seen from the above technical solutions that compared with the prior art, the present invention has the following beneficial effects:

[0035] This invention leverages advanced computer vision and deep learning technologies to quickly and accurately analyze outfall image data in real time, automatically identifying anomalies such as sudden increases in flow and color changes. This provides real-time warnings to environmental protection authorities without requiring human intervention, significantly improving efficiency. Compared to traditional physical sensors, this system is less invasive, does not cause secondary water pollution, is unaffected by corrosive substances in the water, and offers low maintenance costs.

[0036] This invention is not only applicable to industrial emissions but can also be widely used in various scenarios, such as urban drainage and farmland irrigation. Whether it is a large industrial outlet or a small farmland outlet, it can be effectively monitored. Through the camera's real-time video stream, the outlet's discharge status can be captured instantly, ensuring a rapid response to abnormal emissions and greatly improving the timeliness of environmental protection.

[0037] At the same time, the present invention uses sliding window technology for data outlet status feedback, which improves the stability of output data and can more accurately reflect the real-time status of the outlet, providing more stable and reliable data support for environmental monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0039] FIG1 is a flow chart of one embodiment of a method for intelligently monitoring the discharge volume of a drain outlet based on vision provided by the present invention;

[0040] FIG2 is a flow chart of another embodiment of the vision-based intelligent monitoring method for drainage outlet discharge provided by the present invention;

[0041] FIG3 is a schematic diagram of image information acquired by a camera and detection results of a deep learning network model provided by the present invention;

[0042] FIG4 is a flow chart of data processing using a sliding window method provided by the present invention;

[0043] FIG5 is a schematic diagram of the installation of the camera and edge computer provided by the present invention. DETAILED DESCRIPTION

[0044] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0045] As shown in FIG1 , an embodiment of the present invention discloses a vision-based intelligent monitoring method for drainage outlet discharge, comprising the following steps:

[0046] S1, collecting images of the pipe drain area in real time through a camera;

[0047] S2, based on the pre-built deep learning network model, identifies the pipe drains and water flows in the image and generates corresponding bounding boxes for each identified object;

[0048] S3. Calculate the ratio r of the water flow width to the pipe diameter based on the bounding box of the pipe drain outlet and the water flow. w ;

[0049] S4, based on the pre-built method for characterizing the outlet flow Q and ratio r w The flow rate of the current pipe drain outlet is calculated using a polynomial regression model based on the nonlinear relationship between them.

[0050] In a more advantageous embodiment, the present invention introduces cached drain outlet detection results, which can further reduce the resource consumption of repeated detection and can also regularly update the cached data of drain outlet detection results to ensure the accuracy and real-time performance of the detection results. Specifically, as shown in Figure 2, S2 also includes:

[0051] The monitoring data B of the pipe drain outlet bounding box identified by the deep learning network model d Cache and update the cache regularly;

[0052] Before the deep learning network model performs pipe drain outlet detection, it first determines whether there is a pipe drain outlet bounding box B d If the cached data exists, it is determined whether the cached data needs to be updated. If the cached data does not exist or needs to be updated, the pipeline drain outlet detection is performed on the current image through the deep learning network model, the drain outlet detection result is generated, and the cached data is saved. The cached data is the data of the area of ​​interest.

[0053] If the cached data does not need to be updated, the detection result of the drain outlet in the cached data is used as the detection result at the current moment, and a check is made to determine whether the pipe drain outlet and its bounding box exist in the detection result of the cached data. If not, it indicates that the camera has an abnormality (such as offset, obstruction, or failure). In this case, a prompt signal is issued to remind the user to adjust the camera posture so that it faces the drain outlet, or to remove the obstruction or perform maintenance to restore the camera to normal operation;

[0054] If there is a pipe outlet and a bounding box in the cached data, it is necessary to extract the image information of the current outlet and the surrounding area to reduce the impact of complex scenes on water flow detection. In order to more accurately detect the water flow data Bw, this embodiment uses a deep learning-based network model to analyze the water flow in the current image to generate an accurate water flow detection box Bw. If there is water flow bounding box information, continue to execute S3-S4 to calculate the water flow ratio r w And calculate the flow rate of the current pipe drain outlet, and finally output the calculation result after post-processing. Otherwise, it is determined that the current drain outlet is in a non-drainage state.

[0055] In order to accurately detect the outlet and water flow information, the present invention selects a deep learning network model as the core detection module. In the training phase of the model, a supervised learning method is adopted. To this end, in the specific experiment, a large number of outlet images need to be collected from different angles and scenes to construct a preliminary data set. After multiple iterative training, the model can recognize the outlet pipe and the current water flow, and generate a corresponding bounding box for each identified object, which is denoted as B d and B w By checking B d The existence of B can be used to determine whether the camera installation angle is appropriate; and by checking B w , you can determine whether the current outlet is draining.

[0056] After the camera collects the image of the current pipe drain area, it repairs the dark light and other conditions in the image, compresses it, and then passes it to the deep learning network model for drain pipe detection. First, it detects B d After that, the region of interest (IoU) is extracted from the image, and then the water flow B is detected. w After the model successfully detects the boundary boxes of the pipe and water flow, it can use the ratio of the horizontal pixels of the outlet to the horizontal pixels of the water flow to estimate the ratio of the current pipe diameter to the water flow, that is, the average horizontal pixel width of the water flow in the image divided by the average horizontal pixel width of the pipe in the image, recorded as r w , where the marked content and the detected image target are shown in Figure 3, where the red box B d The data obtained from the first detection of the left image is used. The second detection is performed by extracting the region of interest and cropping the entire image. The goal is to detect B w , and finally calculated

[0057] Specifically, in S3, the ratio r of the pipe radius to the water flow width is w The calculation process includes:

[0058] Determine the upper left coordinate point (x1, y1) and the lower right coordinate point (x2, y2) of the pipe drain outlet boundary box; expand the pipe drain outlet boundary box according to the percentage k of (x2-x1), where k<0.5. The expanded area here is the edge area of ​​the pipe drain outlet image information, because the directly detected pipe B d It may be relatively compact, containing only the information of the pipe outlet. In order to better obtain the water flow boundary box, it is necessary to expand the pixels around the pipe to obtain the area of ​​interest of the water flow. At this time, the expanded area is p long. d =k×(x2-x1) / 2, the new boundary data can be obtained as (x1-p d ,y 1- p d) and (x2+p d ,y2+p d Specifically, if the width of the original bounding box is W, the total width after expansion will be W + k * W. This approach can expand the monitoring range of the surrounding area while maintaining the original detection accuracy, and more comprehensively capture and analyze the water flow conditions at the pipe drain outlet.

[0059] According to the region of interest, the important outlet information area in the image is cut out, and the water flow data B is further detected in this area w , B w Like the pipeline data, it contains the upper left position coordinate point (x3, y3) and the lower right position coordinate point (x4, y4) of the water flow bounding box;

[0060] Calculate the horizontal pixel width of the water flow: x4-x3, calculate the horizontal pixel width of the pipe drain: x2-x1;

[0061] Calculate the water flow ratio r w =(x4-x3) / (x2-x1).

[0062] In order to calculate the discharge volume of the drain outlet more quickly, the present invention establishes a polynomial regression model. For most pipes placed approximately horizontally, the flow rate is mainly affected by the hydraulic radius (R). The flow rate (Q) can be expressed as (Q = V × S), where V is the average velocity of the cross section and S is the cross-sectional area of ​​the water flow. Assuming the water height in the pipe is (h), the pipe angle corresponding to the horizontal plane can be calculated Furthermore, the cross-sectional area S of the water flow can be expressed as S = (r 2 ×θ)-(rh)×r×sin(θ), hydraulic radius R=2r×θ.

[0063] In the experiment, it was inferred that there is a nonlinear relationship between the flow rate Q and the pipe radius r and the water height h. In order to better reflect the real-time performance and more efficiently fit the complex nonlinear function with a small amount of data, the present invention uses polynomial regression to establish the relationship between the data and adopts partial Taylor expansion, that is, polynomial function to approximate the smooth function, and its expression is: Q = a × r 2 +b×h 2 +c×r×h+d×r+e×h+f

[0064] Where Q represents the current outfall flow rate, h represents the water height, and a, b, c, d, e, and f are fitting parameters, serving as weights for the polynomial regression model. The pipe radius r is known. After the camera is installed and before official operation, the corresponding pipe radius must be measured. Otherwise, a default pipe radius of 1 meter is assumed.

[0065] The above method can quickly establish the numerical relationship between the water outlet height and flow rate. After successfully establishing the regression equation of drainage water flow height and flow rate, the water flow ratio r can be monitored in real time. w To calculate the water level h under the unit circle (assuming the drainage pipe section is a circle with a radius of 1) of the drainage pipe discharge, where h = 1-cos(arcsin(r w )), combined with the known physical drainage pipe radius r, the real-time flow rate Q can be further estimated.

[0066] Vision can only calculate the pixel ratio in the image coordinate system, that is, how many pixels are in the image, but mapping to the real world coordinate system requires corresponding conversion. That is, according to the water flow ratio r w We can only calculate the drainage volume of the unit circle width, and then multiply it by the real physical drainage pipe information to deduce the real target data.

[0067] Specifically, in S4, the construction process of the polynomial regression model includes:

[0068] Collect the actual radius r of various physical pipes and record the actual drainage flow Q and water height h under different pipe radii;

[0069] Clean up abnormal data, such as data with values ​​that are too large or caused by recording errors. These data are usually significantly inconsistent with the distribution of the overall data set and may be caused by equipment failure or measurement errors. For example, when measuring water flow velocity, if the recorded data deviates significantly from the normal range due to equipment setting errors or temporary external interference, these obviously abnormal data points need to be removed from the data set.

[0070] Add restriction data to the collected data to limit the output of the model data r≤0 and h≤0; mainly artificially introduce the situation of r=0, h=0 during the training process, and record Q=0 at this time. Then, after the overall fitting process, the model can normally feedback the physical meaning of no water flow when receiving these edge numbers;

[0071] Using a gradient descent method, the model's weight parameters, a, b, c, d, e, and f, are continuously updated to achieve an accurate estimate of displacement. During this process, the direction of the derivative provides the direction for the weight parameter update. Convergence is achieved when the error between the parameterized result and the true result approaches 0.

[0072] In order to ensure that the flow data output by the model is more stable and smooth, the present invention also introduces the sliding window technology for post-processing, as shown in Figure 4. Specifically:

[0073] A sliding window with a fixed size of w is used. In each window, the average value of the traffic data is calculated as the final traffic output result o. As new data is continuously acquired, the sliding window will continue to move forward. That is, each time new data enters the window, the earliest data will be moved out of the window, so that a certain amount of data is maintained in the window.

[0074] To ensure the accuracy of the data, we will simultaneously remove values ​​from the window that do not conform to the actual situation, such as negative numbers, extreme values, etc.

[0075] If the valid data in the window is less than the set value k (k≤w), it is determined to be in a non-drainage state, otherwise the smoothed result o is output.

[0076] This processing method effectively reduces the sharp fluctuations in flow data and also improves the judgment of the stability of the outlet status.

[0077] As shown in FIG5 , an embodiment of the present invention further provides a vision-based intelligent monitoring system for drainage outlet discharge, comprising: a camera and an edge computer;

[0078] The camera is used to collect image information of the pipe drain area in real time. It is installed facing the pipe drain so as to better capture the drainage situation.

[0079] The edge computer, equipped with a deep learning network model and a polynomial regression model, executes steps S2-S4 above after receiving image information captured by the camera. This system not only verifies in real time whether the camera is functioning properly but also estimates the current discharge volume. Once the image is processed and analyzed, the results are stored in a database. When other terminals need to query the discharge volume at the outlet, they can quickly retrieve the relevant data from the database.

[0080] In other embodiments, the present invention further discloses a computer storage medium on which a computer program is stored. When the computer program is executed by a processor, the above steps S2-S4 are implemented.

[0081] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0082] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A vision-based intelligent monitoring method for drainage outlet discharge, characterized in that: The following steps are involved: S1, collecting images of the pipe drain area in real time through a camera; S2, identifying pipe drains and water flows in the image based on a pre-built deep learning network model, and generating a corresponding bounding box for each identified object; S3. Calculate the ratio r of the water flow width to the pipe diameter based on the pipe drain outlet and the water flow boundary box w ; S4, based on the pre-built method for characterizing the outfall flow rate Q and ratio r w The polynomial regression model of the nonlinear relationship between the two is used to calculate the flow rate of the current pipeline drain outlet.

2. The vision-based intelligent monitoring method for drainage outlet discharge according to claim 1 is characterized in that: S2 also includes: caching the pipe drain outlet boundary box identified by the deep learning network model, and regularly updating the cache; Before the deep learning network model performs pipe drain outlet detection, it is first determined whether there is cached data of the pipe drain outlet boundary box, and if so, whether the cached data needs to be updated; if there is no cached data or the cached data needs to be updated, the deep learning network model performs pipe drain outlet detection on the current image, generates a drain outlet detection result, and saves the cached data; If the cached data does not need to be updated, the detection result of the drain outlet in the cached data is used as the detection result at the current moment, and it is determined whether there is a pipe drain outlet and its boundary box in the detection result of the cached data. If not, it indicates that the camera is abnormal. If so, the water flow in the current image is detected based on the deep learning network model, and a water flow detection result is generated. If the water flow information exists, the flow rate of the current pipe drain outlet is calculated; otherwise, the current drain outlet is determined to be in a non-draining state. If the water flow information exists, the water flow in the current image is detected based on the deep learning network model, and a water flow detection result is generated; if the water flow information exists, the flow rate of the current pipe drain outlet is calculated; otherwise, the current drain outlet is determined to be in a non-draining state.

3. The vision-based intelligent monitoring method for drainage outlet discharge according to claim 1 is characterized in that: In S3, the ratio r of the pipe radius to the water flow width w The calculation process includes: Determine the upper left position coordinate point (x1, y1) and the lower right position coordinate point (x2, y2) of the pipe drain outlet boundary box; Determine the upper left position coordinate point (x3, y3) and the lower right position coordinate point (x4, y4) of the water flow boundary box; Calculate the horizontal pixel width of the water flow: x4-x3, calculate the horizontal pixel width of the pipe drain: x2-x1; Calculate the water flow ratio r w =(x4-x3) / (x2-x1).

4. The vision-based intelligent monitoring method for drainage outlet discharge according to claim 3 is characterized in that: After determining the location coordinates of the pipe drain outlet, it also includes: The bounding box of the pipe drain outlet is expanded according to a percentage k of (x2-x1), where k<0.

5.

5. The vision-based intelligent monitoring method for drainage outlet discharge according to claim 1 is characterized in that: In S4, the expression of the polynomial regression model is: Q = a × r 2 +b×h 2 +c×r×h+d×r+e×h+f Where h = 1-cos(arcsin(r w )), represents the water height; Q represents the flow rate of the drain outlet at the current moment, and a, b, c, d, e and f are fitting parameters, which serve as weight parameters of the polynomial regression model.

6. The vision-based intelligent monitoring method for drainage outlet discharge according to claim 1 is characterized in that: In S4, the process of constructing the polynomial regression model includes: Collect the actual radius r of various physical pipes, and record the actual drainage flow Q and water height h under different pipe radii; Clean up abnormal data; Add restriction data to the collected data to limit the result output of the model data r≤0 and h≤0; The gradient descent method is used to continuously update the model's weight parameters a, b, c, d, e, and f.

7. The vision-based intelligent monitoring method for drainage outlet discharge according to claim 1 is characterized in that: S4 also includes: The sliding window method is used to calculate the average value of the flow data as the final flow output result; If the valid data in the window is less than the set value, it is determined that the current state is non-drainage.

8. The vision-based intelligent monitoring method for drainage outlet discharge according to claim 7 is characterized in that: Before calculating the average value of the flow data in the window, the values ​​that do not conform to the actual situation are eliminated.

9. A vision-based intelligent monitoring system for drainage outlet discharge, characterized in that: include: Cameras and edge computers; The camera is installed facing the pipe drain outlet and is used to collect image information of the pipe drain outlet area in real time; The edge computer is equipped with the deep learning network model and the polynomial regression model, and is used to execute steps S2-S4 in the monitoring method as described in any one of claims 1-8 after receiving image information captured by a camera.

10. A computer storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, steps S2-S4 in the monitoring method according to any one of claims 1-8 are implemented.

Images