River channel water flow velocity measurement method based on video identification and CFD simulation

Abstract

The invention is suitable for the technical field of hydrographic survey, and provides a river and channel water flow velocity measurement method based on video recognition and CFD simulation, and the method comprises the steps: S1, completing data collection and constructing a data set, S2, building a three-dimensional water flow model of a target river or channel, S3, generating a vertical flow velocity distribution model, S4, carrying out the image preprocessing of an obtained video stream, and S5, carrying out the calculation of a vertical flow velocity distribution model. S5, performing target detection on the preprocessed image and calculating an area of the flow velocity, S6, performing optical flow calculation on the extracted specific area and calculating the surface velocity, S7, establishing a distribution relationship between the surface velocity and the vertical velocity, and S8, calculating the real flow velocity of the river channel according to the surface velocity and the vertical velocity distribution. According to the method, the relation between the cross section vertical flow velocity obtained through simulation and the surface flow velocity obtained through calculation of the optical flow method is established, the real flow velocity is finally obtained, multi-source fusion of data is achieved, and the calculation accuracy and the flow velocity measurement precision are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of hydrological measurement, and more particularly to a method for measuring water flow velocity in a river channel based on video recognition and CFD simulation. Background Art

[0002] Accurate measurement of river cross-section flow is an important part of optimizing water resources allocation and scientific management. It is also a technical problem that urgently needs to be solved for real-time and accurate measurement of water intake in irrigation areas and water diversion projects.

[0003] The existing estimation and measurement of river and channel flow velocity has always been a hot issue in the water conservancy industry. Traditional water flow velocity measurement methods are mainly divided into two categories. The first category is the traditional contact flow meter measurement method, and the second category is measurement through non-contact acoustic Doppler effect, optics, and radar.

[0004] However, existing traditional measurement methods have many disadvantages such as high cost, complex installation, and limited measurement range. With the rise of visual recognition technology, more and more water flow velocity measurement methods based on visual recognition have attracted attention. However, since visual recognition methods analyze the surface motion characteristics of water flow, they can only estimate the surface velocity and cannot reflect the actual flow velocity distribution of the water flow cross section. Especially under complex hydraulic conditions, the error is large. Summary of the Invention

[0005] In view of the shortcomings of the existing technology, the purpose of the present invention is to provide a river channel water flow velocity measurement method based on video recognition and CFD simulation, which effectively improves the accuracy of water flow velocity measurement and overcomes the limitation of traditional visual velocity measurement methods that cannot reflect the true flow velocity of the water flow cross section.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for measuring water flow velocity in a river channel based on video recognition and CFD simulation, the steps comprising: S1, completing data acquisition and constructing a data set, including obtaining river and channel geography, hydrological information, and a video stream of the river channel;

[0008] S2. Based on the acquired hydrological and geographic information, a three-dimensional flow model of the target river or channel is established using the CFD simulation software OpenFOAM to generate the vertical flow velocity distribution, including determining the computational domain and meshing the computational domain.

[0009] Setting the fluid mechanics equation, the present invention uses the following equation to describe the behavior of water flow:

[0010] (1) Navier-Stokes equations: describe the conservation of momentum of water flow and calculate the influence of velocity distribution, pressure distribution and turbulence behavior on water flow.

[0011]

[0012] Where ρ is the fluid density, u is the flow velocity, p is the pressure, τ is the viscous stress tensor, and f is the external force (such as gravity);

[0013] (2) Mass conservation equation: used to ensure the conservation of mass of water flow

[0014]

[0015] (3) Turbulence model: The turbulence model k-ε is used to simulate the turbulent characteristics of the flow, and the turbulent kinetic energy (k) and turbulent dissipation rate (ε) are calculated to accurately describe the impact of turbulence on the water flow;

[0016] Set the boundary conditions and initial conditions for CFD simulation;

[0017] The finite volume method (FVM) is used to discretize the governing equations, the SIMPLE algorithm, a pressure-velocity coupling solver, is used to handle the coupling relationship between velocity and pressure, and the turbulence model k-ε is used to simulate turbulence in the flow.

[0018] S3, generating a vertical velocity distribution model based on the three-dimensional velocity field of the water flow in the target area obtained through CFD simulation, and then outputting the vertical velocity distribution data as prior information;

[0019] S4, performing image preprocessing on the acquired video stream;

[0020] S5. Perform target detection on the preprocessed image based on YOLO v8, extract the target area, and define the area where the flow velocity needs to be calculated using a bounding box;

[0021] S6. Perform optical flow calculation on the extracted specific area and calculate the surface velocity, including:

[0022] Use the Shi-Tomasi algorithm to detect stable corner points in the target area and set specific parameters (maximum number of corner points N, quality level q and minimum distance d);

[0023] The Pyramidal Lucas-Kanade algorithm is used to calculate the motion vector of each corner point in consecutive frames, and Δx and Δy are solved by minimizing the sum of squared errors in a local window.

[0024] Use forward-backward validation or error threshold methods to eliminate outliers and ensure stable tracking results;

[0025] Calculate the average displacement of all valid corner points and convert the pixel displacement into actual flow velocity. The flow velocity u is obtained by calibration by obtaining the conversion coefficient k and the inter-frame time Δt. s ;

[0026] S7, establishing the relationship between surface velocity and vertical velocity distribution, correlating the velocity vertical distribution u(z) generated by CFD simulation with the water surface velocity us obtained by visual recognition, and establishing a mathematical model;

[0027] S8. Calculate the true flow velocity of the river channel based on the surface velocity and vertical velocity distribution, including surface velocity, vertical velocity and cross-sectional velocity.

[0028] A river channel water flow velocity measurement system based on video recognition and CFD simulation includes a data acquisition module, a CFD simulation module, a visual recognition module, a model building module and a flow velocity calculation module.

[0029] The advantages of the present invention are:

[0030] First, the present invention integrates three types of data: geographic hydrological information, visual sensing, and CFD simulation. It introduces geographic parameters such as the cross-sectional shape, slope, and roughness of the river channel to provide spatial constraints for the model. Based on these parameters, CFD simulation is run to generate the vertical flow velocity of the cross section as prior information. Then, a camera is used to capture the water surface video stream, and the surface flow velocity is extracted based on the optical flow method. The simulated vertical flow velocity of the cross section is then linked to the surface flow velocity calculated by the optical flow method to finally obtain the true flow velocity. This realizes multi-source fusion of data and significantly improves the accuracy of calculation and the precision of flow velocity measurement.

[0031] Secondly, the present invention can realize fast surface flow velocity analysis by integrating the lightweight YOLOv8s target detection algorithm and pyramid-style Lucas-Kanade optical flow tracking. Combined with the pre-stored CFD parameter table, the system can output the vertical flow velocity distribution, cross-sectional average flow velocity and surface flow velocity in real time.

[0032] Third, the present invention makes flow rate detection more applicable by combining visual recognition with CFD simulation. CFD simulation enables flow rate detection to adapt to different river channel shapes, while video recognition demonstrates its superiority in scenarios that are difficult to detect with traditional methods, such as scenarios where traditional sensors fail (narrow / shallow rivers). BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a flow chart of a river channel water flow velocity measurement method based on video recognition and CFD simulation according to the present invention;

[0034] Figure 2 This is a block diagram of the river channel water flow velocity measurement system based on video recognition and CFD simulation of the present invention. DETAILED DESCRIPTION

[0035] See also Figure 1 , the present invention provides the following technical solutions:

[0036] Specifically, it refers to a method for measuring water flow velocity in river channels based on video recognition and CFD simulation. The specific steps are as follows:

[0037] S1. Complete data collection and build a dataset, including obtaining river and channel geography, hydrological information, and video streams of river channels.

[0038] S1-1. Obtain hydrological and geographic information as input parameters for CFD simulation to establish a water flow model. This information includes: ① hydrological information: flow rate (Q), water depth (h); ② geographic information: river channel geometry (cross-sectional shape, slope, width), riverbed roughness.

[0039] S1-2. Obtain the video stream of river and channel water flow captured by the camera, with a frame rate ≥ 30fps and a resolution > 1080p.

[0040] S2. Based on the acquired hydrological and geographic information, a three-dimensional water flow model of the target river or channel is established using the CFD simulation software OpenFOAM.

[0041] S2-1. Determine the computational domain. The computational domain should cover the geometric shape of the river channel to ensure accurate simulation of water flow behavior. Divide the computational domain into grids and use unstructured technology to accurately simulate complex river channel shapes. Especially in areas with large changes in water flow (such as high flow velocity or sharp turns), the grid should be locally refined.

[0042] S2-2. Set the fluid mechanics equation. The present invention uses the following equation to describe the behavior of water flow:

[0043] Navier-Stokes equations: Describes the conservation of momentum in water flow and calculates the effects of velocity distribution, pressure distribution, and turbulent behavior on water flow;

[0044]

[0045] Where ρ is the fluid density, u is the flow velocity, p is the pressure, τ is the viscous stress tensor, and f is the external force (such as gravity);

[0046] Mass conservation equation: used to ensure the conservation of mass in water flow

[0047]

[0048] Turbulence model: The turbulence model k-ε is used to simulate the turbulent characteristics in the flow, and the turbulent kinetic energy (k) and turbulent dissipation rate (ε) are calculated to accurately describe the impact of turbulence on the water flow.

[0049] S2-3. Set the boundary conditions and initial conditions for the CFD simulation. Set the velocity or flow boundary at the inlet of the simulation model, set the outlet to a zero pressure condition (i.e., a free outflow boundary), set a no-slip boundary condition at the bottom of the river channel and canal, and use the Darcy-Weisbach friction coefficient model to describe the effect of riverbed roughness on flow. Set a no-slip condition for the side walls of the river channel, and also introduce a roughness model to simulate the interaction between water flow and the side walls; set free surface boundary conditions, and use the Volume of Fluid (VOF) method to track the water surface morphology and position.

[0050] S2-4. The finite volume method (FVM) is used to discretize the control equations. The finite volume method is suitable for fluid flow simulation under complex boundary conditions. The pressure-velocity coupling solver SIMPLE (Semi-implicit Method for Pressure-Linked Equations) algorithm is used to deal with the coupling relationship between flow velocity and pressure. The turbulence model k-ε is used to simulate turbulence in the flow.

[0051] S3. The three-dimensional velocity field of the water flow in the target area obtained by CFD simulation includes the velocity components of the water flow in each grid unit. In order to generate the vertical velocity distribution, it is first necessary to extract the vertical velocity components (w) of the water flow at different water depths from the simulation results, and then record the vertical velocity data of different water depths on multiple cross sections, and construct the vertical velocity profile of each cross section. The vertical velocity profile data of multiple cross sections are mapped to the entire water area to generate a vertical velocity distribution model, and then the vertical velocity distribution data is output as prior information.

[0052] S4. Perform image preprocessing on the acquired video stream.

[0053] S4-1. Perform grayscale conversion on the acquired video stream in OpenCV to convert the color image into a grayscale image.

[0054] S4-2. Perform image enhancement and noise reduction processing on the grayscale image.

[0055] S5. Perform target detection on the preprocessed image based on YOLO v8, extract the target area, and define the area where the flow velocity needs to be calculated using a bounding box.

[0056] S5-1. Perform object detection on the image based on YOLO v8, extracting ripples and vortices. Specifically, the YOLO v8s model is used, including Backbone, Neck, Head, and Post-processing.

[0057] S5-2. Build the YOLOv8s model and perform targeted training to improve its ability to extract water surface features such as ripples and vortices. For classification, use the BCEWithLogitsLoss loss function. For regression, use the CIoU Loss combined with the Distribution Focal Loss (DFL) loss function to train the model. Additionally, to improve the model's ability to extract small objects like ripples and vortices, increase the bounding box loss weight appropriately.

[0058] S6. Perform optical flow analysis on the extracted specific area;

[0059] S6-1. Use the Shi-Tomasi algorithm to detect stable corner points in the target area and set specific parameters (maximum number of corner points N, quality level q and minimum distance d);

[0060] S6-2. Use the Pyramidal Lucas-Kanade algorithm to calculate the motion vector of each corner point in consecutive frames, and solve Δx and Δy by minimizing the sum of squared errors in a local window. The local minimization formula of the Lucas-Kanade optical flow is as follows:

[0061] In the optical flow calculation, it is assumed that the grayscale of the image remains unchanged in the local window, that is, for the pixel located at (x, y), its grayscale value at time t and t+Δt satisfies:

[0062] I(x,y,t)≈I(x+Δx,y+Δy,t+Δt)

[0063] In order to solve the pixel displacement (Δx, Δy), the sum of squared errors is minimized within the local window w. The formula is expressed as:

[0064] E(Δx, Δy)=∑ (x,y)∈w [I(x+Δx,y+Δy,t+Δt)-I(x,y,t)] 2 ;

[0065] S6-3. Use forward-backward verification or error threshold method to eliminate outliers to ensure stable tracking results;

[0066] S6-4. Calculate the average displacement of all valid corner points and convert the pixel displacement into the actual flow velocity. The flow velocity u is obtained by calibrating the conversion coefficient k and the inter-frame time Δt. s ;

[0067] S6-4-1. The calculation formula for characteristic point displacement is as follows:

[0068] For each feature point i tracked in consecutive frames, its displacement vector is (Δx i , Δy i), the calculation formula for the displacement amplitude of a single feature point is:

[0069]

[0070] In the ROI area, for the M valid feature points retained, the average displacement Δd is calculated as follows:

[0071]

[0072] S6-4-2. Calculate the actual flow velocity on the river surface. To convert the average displacement Δd (in pixels) obtained by the optical flow method into actual physical displacement, it is necessary to introduce the conversion relationship k between pixels and actual distance (unit: meter / pixel, obtained by camera calibration), and combine it with the inter-frame time interval Δt (unit: second, determined by the video frame rate, ), which is calculated as follows:

[0073]

[0074] S7, establishing the relationship between surface velocity and vertical velocity distribution, correlating the velocity vertical distribution u(z) generated by CFD simulation with the water surface velocity us obtained by visual recognition, and establishing a mathematical model;

[0075] S7-1. Arrange the CFD simulation results into a segmented working condition data table, each segment contains the surface flow rate and the corresponding stratified flow velocity u(z i );

[0076] S7-2, vertical stratification of adjacent CFD conditions i , establish the surface velocity The laminar flow velocity u(z i )’s linear relationship:

[0077]

[0078] Among them, a i is the slope of the i-th layer, which indicates the change in the flow velocity of this layer when the surface flow velocity increases by 1 m / s; b i is the intercept of the i-th layer, reflecting the baseline value of the bottom layer flow velocity.

[0079] S8. Calculate the true flow velocity of the river channel based on the surface velocity and vertical velocity distribution, including surface velocity, vertical velocity and cross-sectional velocity.

[0080] S8-1, the surface flow velocity u obtained by video recognition calculation s Substitute the above model, interpolate and calculate the velocity of each layer to obtain the velocity distribution of the cross section;

[0081] S8-2, performing a weighted average of the surface flow velocities obtained by video recognition and CFD simulation to obtain the true surface flow velocity;

[0082] S8-3, the calculation formula of the average flow velocity of the cross section is as follows:

[0083]

[0084] in, is the average flow velocity of the cross section, and H is the water depth.

[0085] See also Figure 2 According to the above scheme, the present invention provides a new technical solution: a river channel water flow velocity measurement system based on video recognition and CFD simulation, including a data acquisition module, a CFD simulation module, a visual recognition module, a model building module and a flow velocity calculation module.

[0086] The data acquisition module collects river and channel hydrological and geographic information through measurement equipment or geographic information system (GIS), obtains high-resolution river and channel water surface video streams, and performs image preprocessing to ensure that the video streams are suitable for subsequent video recognition;

[0087] The CFD simulation module uses the CFD simulation software OpenFOAM to build a three-dimensional water flow model of the target river or channel and generate the vertical flow velocity distribution, which is stored as prior information;

[0088] The visual recognition module performs target detection on the preprocessed image based on YOLO v8, extracts the target area, uses the optical flow algorithm to extract the motion characteristics of the water surface, and calculates the surface velocity;

[0089] The model building module establishes the relationship between surface velocity and vertical velocity distribution, and associates the vertical velocity distribution u(z) generated by CFD simulation with the surface velocity us obtained by visual recognition;

[0090] The velocity calculation module calculates the actual flow velocity of the river channel based on the surface velocity and vertical velocity distribution, including surface velocity, vertical velocity and cross-sectional velocity.

Claims

1. A method for measuring water flow velocity in a river channel based on video recognition and CFD simulation, characterized by: The steps include, S1, completing data collection and constructing a data set, including obtaining river and channel geography, hydrological information and video streams of river channels; S2. Based on the acquired hydrological and geographic information, a three-dimensional flow model of the target river or channel is established using the CFD simulation software OpenFOAM to generate the vertical flow velocity distribution, including determining the computational domain and meshing the computational domain. Setting the fluid mechanics equation, the present invention uses the following equation to describe the behavior of water flow: (1) Navier-Stokes equations: describe the conservation of momentum of water flow and calculate the influence of velocity distribution, pressure distribution and turbulence behavior on water flow. Where ρ is the fluid density, u is the flow velocity, p is the pressure, τ is the viscous stress tensor, and f is the external force (such as gravity); (2) Mass conservation equation: used to ensure the conservation of mass of water flow (3) Turbulence model: The turbulence model k-ε is used to simulate the turbulent characteristics of the flow, and the turbulent kinetic energy (k) and turbulent dissipation rate (ε) are calculated to accurately describe the impact of turbulence on the water flow; Set the boundary conditions and initial conditions for CFD simulation; The finite volume method (FVM) is used to discretize the governing equations, the SIMPLE algorithm, a pressure-velocity coupling solver, is used to handle the coupling relationship between velocity and pressure, and the turbulence model k-ε is used to simulate turbulence in the flow. S3, generating a vertical velocity distribution model based on the three-dimensional velocity field of the water flow in the target area obtained through CFD simulation, and then outputting the vertical velocity distribution data as prior information; S4, performing image preprocessing on the acquired video stream; S5. Perform target detection on the preprocessed image based on YOLO v8, extract the target area, and define the area where the flow velocity needs to be calculated using a bounding box; S6. Perform optical flow calculation on the extracted specific area and calculate the surface velocity, including: Use the Shi-Tomasi algorithm to detect stable corner points in the target area and set specific parameters (maximum number of corner points N, quality level q and minimum distance d); The Pyramidal Lucas-Kanade algorithm is used to calculate the motion vector of each corner point in consecutive frames, and Δx and Δy are solved by minimizing the sum of squared errors in a local window. Use forward-backward validation or error threshold methods to eliminate outliers and ensure stable tracking results; Calculate the average displacement of all valid corner points and convert the pixel displacement into actual flow velocity. The flow velocity u is obtained by calibration by obtaining the conversion coefficient k and the inter-frame time Δt. s ; S7, establishing the relationship between surface velocity and vertical velocity distribution, correlating the velocity vertical distribution u(z) generated by CFD simulation with the water surface velocity us obtained by visual recognition, and establishing a mathematical model; S8. Calculate the true flow velocity of the river channel based on the surface velocity and vertical velocity distribution, including surface velocity, vertical velocity and cross-sectional velocity.

2. The method for measuring water flow velocity in a river channel based on video recognition and CFD simulation according to claim 1, characterized in that: The hydrological information in S1 includes flow (Q) and water depth (h); the geographic information includes river channel geometry (cross-sectional shape, slope, width) and riverbed roughness; the video stream obtained during data collection should meet the requirements of a frame rate of 30fps and a resolution >1080p.

3. The method for measuring water flow velocity in a river channel based on video recognition and CFD simulation according to claim 1, characterized in that: The three-dimensional velocity field of the water flow in S3 includes the velocity components of the water flow in each grid unit. In order to generate the vertical velocity distribution, it is first necessary to extract the vertical velocity components (w) of the water flow at different water depths from the simulation results, and then record the vertical velocity data of different water depths on multiple cross sections, and construct the vertical velocity profile of each cross section. The vertical velocity profile data of multiple cross sections are mapped to the entire water area to generate a vertical velocity distribution model, and then the vertical velocity distribution data is output as prior information.

4. The method for measuring water flow velocity in a river channel based on video recognition and CFD simulation according to claim 1, characterized in that: The image preprocessing in S4 includes performing grayscale conversion on the acquired video stream in OpenCV, converting the color image into a grayscale image, and performing image enhancement and noise reduction on the grayscale image.

5. The method for measuring water flow velocity in a river channel based on video recognition and CFD simulation according to claim 1, characterized in that: In S5, target detection is performed on the image based on YOLO v8, and ripples and vortices are used as target extraction objects. Specifically, the YOLO v8s model is selected, including: Backbone, Neck, Head, and Post-processing; The YOLOv8s model was built and trained to improve its ability to extract water surface features such as ripples and vortices. The classification loss used the BCEWithLogitsLoss loss function, and the regression part used the CIoU Loss combined with the Distribution FocalLoss (DFL) loss function to train the model. In addition, to improve the model's ability to extract small objects such as ripples and vortices, the bounding box loss weight was appropriately increased.

6. The method for measuring water flow velocity in a river channel based on video recognition and CFD simulation according to claim 1, characterized in that: In S6, the average displacement of all valid corner points is calculated, and the pixel displacement is converted into the actual flow velocity. The flow velocity u is obtained by calibration by obtaining the conversion coefficient k and the inter-frame time Δt. s The specific implementation steps are as follows: Calculate the displacement of the feature point. The calculation formula of the feature point displacement is as follows: For each feature point i tracked in consecutive frames, its displacement vector is (Δx i , Δy i ), the calculation formula for the displacement amplitude of a single feature point is: In the ROI area, for the M valid feature points retained, the average displacement Δd is calculated as follows: To calculate the actual flow velocity on the river surface, the average displacement Δd (in pixels) obtained by the optical flow method is converted into the actual physical displacement. It is necessary to introduce the conversion relationship k between pixels and actual distance (unit: meter / pixel, obtained by camera calibration), and combine the inter-frame time interval Δt (unit: second, determined by the video frame rate, ), which is calculated as follows:

7. The method for measuring water flow velocity in a river channel based on video recognition and CFD simulation according to claim 1, characterized in that: The specific steps of establishing the mathematical model in S7 are: arranging the CFD simulation results into segmented working condition data tables, each of which contains surface flow velocity and the corresponding stratified flow velocity u(z i ); Vertically layer adjacent CFD cases z i , establish the surface velocity The laminar flow velocity u(z i )’s linear relationship: Among them, a i is the slope of the i-th layer, which indicates the change in the flow velocity of this layer when the surface flow velocity increases by 1 m / s; b i is the intercept of the i-th layer, reflecting the baseline value of the bottom layer flow velocity.

8. The method for measuring water flow velocity in a river channel based on video recognition and CFD simulation according to claim 1, characterized in that: The specific step in S8 is to calculate the surface flow velocity u obtained by video recognition. s Substitute the above model, interpolate and calculate the velocity of each layer to obtain the velocity distribution of the cross section; The surface velocity obtained by video recognition and CFD simulation is weighted averaged to obtain the true surface velocity; The formula for calculating the average flow velocity of the cross section is as follows: in, is the average flow velocity of the cross section, and H is the water depth.

9. A river channel water flow velocity measurement system based on video recognition and CFD simulation, based on the river channel water flow velocity measurement method based on video recognition and CFD simulation according to any one of claims 1 to 8, characterized in that: It includes data acquisition module, CFD simulation module, visual recognition module, model building module and flow rate calculation module.

Images