Coal mine underground water flow trajectory capturing and flow velocity measuring method and system

Abstract

The invention discloses a coal mine underground water flow trajectory capturing and flow velocity measuring method and system, belongs to the field of fluid image processing, and aims to deal with sudden accidents possibly encountered in underground operation, enhance underground operation safety and reduce casualties. Physical constraint parabola fitting is combined to ensure that a track accords with a physical law, visual angle deviation is processed through angle correction and coordinate transformation, real-time stability is improved through Kalman filtering, and finally the water flow velocity is calculated based on a physical equation. The system comprises an image acquisition module and a water outlet detection module, and realizes non-contact, high-precision and real-time monitoring. According to the method, the problems of contact interference and high cost of a traditional method and low precision and insufficient physical constraint of existing vision measurement are solved, and the method is completely suitable for a specific scene of underground operation of a coal mine from the perspective of safety.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of fluid image processing, and specifically presents a coal mine underground water flow trajectory capture and flow velocity measurement method and system, which is particularly suitable for real-time monitoring of water flow velocity in the drainage scene of underground operation. BACKGROUND

[0002] Water flow velocity is a core parameter for evaluating water flow state, and its measurement accuracy directly affects engineering safety, environmental assessment and industrial production efficiency. Existing measurement technologies mainly include traditional physical sensor method and computer vision-based method, both of which have significant defects, as follows:

[0003] Traditional methods rely on physical devices such as flow meters and ultrasonic flow meters, which require direct contact with water flow or close deployment, and cannot adapt to the scene of coal mine underground operation, and have the following problems: traditional flow meters need to be immersed in water flow, and the rotation of impeller or the contact of electrodes will change the natural flow state of water flow, resulting in measurement deviation. For example, in small-diameter pipeline measurement, the water flow cross-section occupied by the flow meter is 5%-10%, which will make the measured flow velocity 8%-12% higher than the true value. In addition, contact-type devices have safety hazards in underground operation, which is not conducive to the development of explosion-proof work; the cost of a single ultrasonic flow meter device is generally more than 10,000 yuan, and the 5t probe needs to be calibrated regularly (at least twice a year), which will increase the hardware and maintenance costs exponentially for large-scale monitoring (such as river) that requires multiple devices. The measurement range and real-time performance are poor, the flow meter measurement range is usually limited to a single point, making it difficult to cover large areas of water, and the sampling frequency of ultrasonic flow meter is at most 1Hz, which cannot provide millisecond-level continuous monitoring data and cannot capture the instantaneous changes of water flow.

[0004] With the development of computer vision, image processing methods have been gradually applied to water flow monitoring, but existing technologies still have key bottlenecks. The flow velocity measurement method described in existing public literature 1 (A river surface water flow velocity measurement method based on machine vision, 2023) has poor environmental adaptability and is easily disturbed, relying too much on external non-water flow features for tracking, ignoring some features of water flow itself, and the measurement logic has limitations, relying too much on manual experience for image acquisition through a single camera. The flow velocity measurement method described in existing public literature 1 (River surface water flow velocity measurement method and system based on machine vision, 2025) has high hardware and operation complexity, requiring the deployment of two cameras and accurate calculation of the included angle, and has strict requirements for installation location. Moreover, its correction logic and scene adaptation are insufficient, and when correcting the flow velocity based on Navier-Stokes equation, the river water density, kinematic viscosity, pressure gradient and other parameters need to be input, which need to be measured by professional instruments, increasing the cost of equipment and measurement steps.

[0005] Therefore, there is a need for a water flow velocity measurement technology that adapts to the scene of coal mine underground operation. Summary of the Invention

[0006] This invention provides a method and system for capturing the trajectory and measuring the velocity of water flow in underground coal mines, thereby solving the problems in the prior art.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A method for capturing the trajectory and measuring the velocity of underground water flow in coal mines includes the following steps:

[0009] Step 1, Multimodal water flow detection: This step integrates three algorithms: motion detection, color detection, and edge detection. Motion detection uses a combination of the MOG2 background subtraction algorithm and frame difference method. Color detection is based on the HSV color space with dual thresholds and auxiliary color ranges. Edge detection integrates the Canny, Sobel, and Laplacian algorithms. The output is a set of water flow trajectory points, which is used as the input for the next step of fitting.

[0010] Step 2, Physically Constrained Parabolic Fitting: Based on the trajectory point set output in Step 1, and using monotonicity constraints and a multi-strategy fitting algorithm, a parabolic trajectory and coefficients conforming to the parabolic motion law of water flow are obtained.

[0011] Step 3, Angle Correction and Coordinate Transformation: The parabolic coefficients obtained in Step 2 are processed to offset the viewing angle θ and the pipe inclination angle α, and the corrected standard parabola is output.

[0012] Step 4, Kalman filter stabilization: Based on the parabolic coefficients corrected in Step 3, a state model and a noise model are constructed, and combined with physical constraints, real-time stable monitoring results are provided.

[0013] Step 5, Physical velocity calculation: The physical initial velocity is calculated based on the parabolic equation. When the water flows horizontally, the initial velocity is obtained based on the parabolic relationship without a first-order term. When the water flows at an angle, the horizontal and vertical components are solved simultaneously. The unit conversion is achieved by combining the water outlet diameter with pixel ratio calibration.

[0014] As a further aspect of this invention, in the multimodal water flow detection stage, motion, color, and edge information are extracted in parallel and then fused to generate a trajectory point set. For motion detection, MOG2 background modeling and frame difference methods are used in parallel to mutually verify dynamic pixels. A "dilation-erosion" morphological processing method is combined to suppress isolated noise, and pixels jointly determined by both methods are prioritized. Considering the imaging characteristic that water flow in the well scene mainly moves to the lower right, the pixel weight in this area is appropriately increased to ensure the continuity and integrity of the main channel. For color detection, a primary and secondary threshold are set in the HSV space to cover color gamut drift under varying illumination. Closure operations are first performed to fill water body holes, and then dilation is used to enhance connectivity, resulting in a continuous water body region. Edge detection fuses Canny, Sobel, and Laplacian gradient information, either bitwise or merged to preserve detailed boundaries as much as possible. The three results are converged into a reliable trajectory observation based on a consistency criterion, providing input for subsequent fitting.

[0015] As a further aspect of this invention, based on the prior knowledge of "the approximate parabolic motion of a water jet under gravity," physical constraint parabolic fitting is performed on the trajectory point set to avoid drift caused by noise and occlusion in pure numerical fitting. First, monotonicity constraint filtering is implemented, including spatial constraints that retain only pixels to the right of the outlet, vertical range constraints based on the outlet, and continuity checks to ensure that the vertical variation between adjacent points does not exceed a threshold. This eliminates outliers that do not conform to physical laws from the data level, ensuring that the fitted samples are consistent with the actual motion trend.

[0016] As a further aspect of this invention, a multi-strategy constraint and optimization mechanism is introduced to enhance physical consistency and geometric anchoring. When the pipe outlet is not tilted, a "vertex constraint" is applied, with the outlet as the vertex of the parabola. At the same time, an "outlet constraint" is applied, forcing the fitted curve to pass through the center of the outlet. The solution level employs least squares under equality constraints, weighted least squares with weighted key anchor points, and analytical solution that strictly satisfies the constraints using the Lagrange multiplier method. This comprehensively improves the fitting stability and accuracy of the outlet neighborhood and the global system, ensuring that the obtained coefficients both fit the observations and follow the dynamics.

[0017] As a further aspect of this invention, considering the imaging geometric errors caused by monocular deployment, a unified processing flow for angle correction and coordinate transformation is proposed. First, the fitted parabola in the initial pixel coordinate system is standardized by applying horizontal and vertical viewing angle parameters. Then, the coordinates are translated with the outlet as the origin, converting the curve equation in the image space into a standard form with clear physical meaning. Based on this, various physical coefficients and the corrected equations are derived, so that the velocity can be solved on the same coordinate and coefficient reference for observations under different installation postures.

[0018] As a further aspect of this invention, a Kalman filter model with parabolic coefficients as state variables is constructed. The approximate assumption of unit transition reflects the slowly changing characteristics of the coefficients. Combined with process and observation noise covariance modeling, state prediction and measurement updates are performed in each frame. A physical constraint such as "opening direction" is introduced for posterior verification, so that the filtered output maintains a physically consistent smooth trajectory parameter sequence under adverse conditions such as illumination jitter, splashing, and intermittent occlusion, providing a stable input for subsequent velocity calculation.

[0019] As a further aspect of this invention, the two working conditions of horizontal and inclined water discharge are handled separately based on the corrected standard parabolic analytical relationship. For horizontal water discharge, the initial horizontal velocity is directly calculated based on the trajectory form without a first-order term. For inclined water discharge with pipe pitch, the initial velocity is decomposed into horizontal and vertical components and synthesized into a total velocity. The gravitational acceleration constant is used for calculation. To achieve the mapping from pixels to physical units, a pixel ratio calibration based on the outlet diameter is adopted. This not only supports automatic conversion of the ratio of the measured diameter to the pixel diameter input by the user, but also provides recommended values ​​based on the YOLO detection frame width and common diameter database when no manual input is required, which are then confirmed by the user. This achieves a balance between general deployment and accuracy assurance.

[0020] As a further aspect of this invention, the system is configured with a modular structure around the aforementioned algorithm chain: the front end uses a regular RGB camera to capture images of the water flow in the well; water flow scene detection reduces background interference; water flow trajectory detection outputs a set of trajectory points; a curve fitting module generates parabolic coefficients; a filtering and stabilization module outputs time-stable parameters; an angle correction module unifies the geometric reference system; a velocity calculation module completes unit conversion and velocity calculation; and a result display module provides real-time curves and statistical analysis. It also includes extended functions for parameter configuration, historical data management, and pixel calibration, supporting settings for viewing angle parameters, gravitational acceleration, and outlet diameter; historical velocity and coefficient tracking; and both automatic and manual calibration paths. These modules work together to form a closed-loop process of "parameters—detection—fitting—correction—filtering—velocity calculation—display," satisfying the real-time performance, stability, and maintainability requirements of the well scene.

[0021] The technical effects and advantages of this invention, a method and system for capturing water flow trajectory and measuring flow velocity in underground coal mines, are as follows: This invention, through the fusion of multiple technologies and physical constraint design, possesses significant advantages over existing technologies. The multi-modal detection algorithm integrates motion, color, and edge information, and suppresses environmental interference through complementary methods. Experimental verification shows that, in scenarios with ±30% changes in illumination and background clutter, the accuracy of water flow trajectory point detection reaches over 92%, a 35% improvement compared to single motion detection algorithms. Monotonicity constraints are used to filter out anomalies, and multi-strategy fitting forces the trajectory to conform to parabolic motion laws. In inclined water outlet scenarios, the deviation between the fitted result and the actual water flow trajectory is ≤3%, far lower than the 15%-20% of unconstrained fitting. The system features a 100% deviation rate; horizontal or vertical viewing angle offset correction and coefficient correction algorithms that can handle viewing angle deviations within ±20°. When the horizontal viewing angle offset is 15°, the corrected velocity calculation error is ≤5%. The Kalman filter algorithm controls the fluctuation range of velocity monitoring results within ±2%, with a sampling frequency of up to 15Hz, meeting the real-time monitoring needs of scenarios such as water conservancy flood discharge and industrial pipelines. Measurement is achieved based on a common RGB camera, eliminating the need for contact with water flow and avoiding interference from traditional flow meters on water flow patterns. It is especially suitable for special scenarios such as sewage and corrosive fluids. The automatic pixel ratio calibration function adapts to different shooting distances and resolutions, eliminating the need for manual parameter adjustment, lowering the operational threshold, and enabling rapid deployment at different monitoring sites. Attached Figure Description

[0022] Figure 1 A schematic diagram of the overall architecture of a machine vision-based system for capturing the trajectory and measuring the velocity of downhole water flow.

[0023] Figure 2 Flowchart of multimodal water flow detection algorithm.

[0024] Figure 3 Flowchart of the physical constraint parabolic fitting algorithm.

[0025] Figure 4 Flowchart of angle correction and coordinate transformation algorithm.

[0026] Figure 5 Flowchart of the Kalman filter stabilization algorithm.

[0027] Figure 6 Flowchart of the physical velocity calculation algorithm.

[0028] Figure 7 A flowchart of a machine vision-based system for capturing downhole water flow trajectories and measuring flow velocity. Detailed Implementation

[0029] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying 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. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0030] Embodiment 1

[0031] In this embodiment, according to the overall system architecture diagram as shown in Figure 1 below, a method for capturing the water flow trajectory and measuring the flow velocity in a coal mine underground includes the following steps:

[0032] Step 1, a multi-modal water flow detection algorithm, as shown in Figure 2 below, by fusing the three detection algorithms of motion, color, and edge, the complementary advantages are utilized to improve the recognition accuracy of the water flow area, and the anti-interference ability is significantly enhanced.

[0033] 1.1 The motion detection module uses the background subtraction algorithm and the MOG2 background modeler to statistically construct the background model from historical frames. The parameter settings are as follows: the number of historical frames = 300 (to ensure the stability of the background model), the variance threshold = 25 (to distinguish foreground / background pixels). For the pixel point (x, y), its Gaussian mixture model probability density function is:

[0034]

[0035] where ω k,t is the weight of the k-th Gaussian component, μ k,t is the mean value, is the variance; when P(I(x, y, t)) < T (T = 25), it is determined as a foreground (water flow) pixel.

[0036] The motion detection module detects by the frame difference method, calculates the gray difference between adjacent frames, and enhances the recognition of dynamic water flow. The formula is as follows:

[0037] D(x, y, t) = |I(x, y, t) - I(x, y, t - 1)|

[0038] where I(x, y, t) is the gray value at (x, y) at time t; the threshold is set to 8, and when D(x, y, t) > 8, it is marked as a motion pixel; subsequent morphological processing (dilation first and then erosion) is performed to eliminate noise.

[0039] The motion detection module fuses motion pixels, preferentially selects the dynamic pixels jointly detected by MOG2 and the frame difference method, and at the same time doubles the pixel weight in the main flow direction of the water flow to ensure that there is no missed detection of the water flow near the water outlet.

[0040] 1.2 The color detection module accurately identifies water flows of specific colors (such as industrial dyeing water flows, water flows marked in specific monitoring scenarios; in this experiment, the water flow is a dyed light red water flow) based on the HSV color space. HSV dual-threshold detection is used, as the HSV space separates color and brightness, providing strong resistance to light interference. The main red range is set as follows: H∈[0,10], S∈[30,255], V∈[100,255]; the auxiliary red range is set as follows: H∈[170,180], S∈[30,255], V∈[100,255]. The dual ranges cover the color gamut of red under varying lighting conditions, avoiding missed detections by a single threshold.

[0041] 1.3 The edge detection module integrates multiple algorithms to extract water flow edges and clarify trajectory contours. The multi-edge algorithm fusion includes: Canny algorithm (double threshold [20, 60], non-maximum suppression to preserve fine edges); Sobel algorithm (threshold = 30, calculating horizontal / vertical gradients); and Laplacian algorithm (threshold = 20, detecting isolated edges using second derivatives). Bitwise OR operations are used to merge the results, ensuring complete edge information and providing a basis for trajectory point extraction.

[0042] Step two, physical constraint parabolic fitting algorithm, such as Figure 3 As shown, based on the physical law that water flow "moves in a parabolic motion under the action of gravity", multiple constraints are introduced to avoid the fitting results from deviating from reality.

[0043] 2.1 Monotonic constraint filtering is used to remove abnormal trajectory points, retaining samples that conform to physical laws, including:

[0044] ① Right-side filtering: Only pixels to the right of the outlet are retained, i.e., for the trajectory point (x... i ,y i ) and the center of the outlet (x outlet ,y outlet ), satisfying x i >x outlet ;

[0045] ② Vertical range constraint: Vertical offset relative to the outlet ∈ [-30, 800] pixels (y i -y outlet ∈[-30,800]), remove distant noise points;

[0046] ③ Continuity check: Vertical variation between adjacent points ≤ 10 pixels (|y i+1 -y i |≤10), ensuring the trajectory is continuous and conforms to the parabolic trend.

[0047] 2.2 Multi-strategy constraint fitting ensures the parabola strictly conforms to the physical motion of water flow, including:

[0048] ① Vertex constraint fitting: When the pipe opening is not tilted (angle parameter = 0), with the outlet as the vertex of the parabola, the equation is:

[0049] yy outlet =a(xx) outlet ) 2

[0050] ② Outlet constraint fitting: Force the parabola to pass through the center of the outlet circle, with the following constraint conditions:

[0051]

[0052] (a, b, c are the general form of a parabola y = ax) 2 +bx+c coefficient).

[0053] ③ Constrained Least Squares Method: In the case of constraints Next, minimize the fitting error:

[0054]

[0055] Substitute c into E, take the partial derivatives with respect to a and b and set them to 0, then solve for the optimal coefficients.

[0056] ④ Weighted Least Squares Method: Outlet point weight = 1000, other points weight = 1, weighted error:

[0057]

[0058] Improve the fitting accuracy near the water outlet.

[0059] ⑤ Lagrange multiplier method: Construct a Lagrange function that strictly satisfies the following constraints:

[0060]

[0061] Take partial derivatives with respect to a, b, c, λ and set them to 0, then solve the system of equations to obtain the optimal coefficients.

[0062] Step 3, angle correction and coordinate transformation algorithm, such as Figure 4 As shown, processing viewpoint shift and nozzle tilt to restore the true trajectory shape includes:

[0063] 3.1 Viewpoint Correction Algorithm

[0064] For the initial pixel, fit a parabola: When performing perspective correction, the horizontal perspective x' = x p *cos(θ h ), vertical perspective y' = y p *cos(θ v ), where θ h For the horizontal viewpoint offset angle, θ vThe vertical viewing angle offset (in radians) is then simplified to obtain the standard form:

[0065] 3.2 Coordinate Transformation Algorithm

[0066] Translation of the coordinate origin: A coordinate transformation with the outlet (x0, y0) as the origin. The translation formula is x... ′ =x - x0, y ′ =y - y0, the equation of the parabola after translation is in Substituting (x0 = 0, y0 = 0) into the equation, we get...

[0067] Step four, Kalman filter stabilization algorithm, such as Figure 5 As shown, by suppressing noise interference, the stability of real-time monitoring results is ensured, including:

[0068] 4.1 State Model

[0069] ①State vector: X t =[a t ,b t ,c t ] T (parabolic coefficients at time t)

[0070] ②State transition matrix: F = I 3×3 (Identity matrix, assuming coefficients change slowly)

[0071] ③State prediction:

[0072] 4.2 Noise Model

[0073] ① Process noise covariance: Q = diag(1e-5, 1e-5, 1e-5) (small uncertainty in coefficient variation)

[0074] ② Observation matrix: H = I 3×3 (Directly observed coefficients)

[0075] ③ Observation noise covariance: R = diag(1e-3, 1e-3, 1e-3) (The fitting result contains a small amount of noise)

[0076] ④ Kalman gain: K t =P t|t-1 H T HP t|t-1 H T +R) -1

[0077] ⑤ Status Update: (Z t(current fitting coefficients)

[0078] ⑥ Physical constraint check: after forced filtering, a t >0 (The parabola opens downwards, consistent with gravity). Step 5, physical velocity calculation: Based on the physical equations of parabolic motion, combined with pixel-to-physical unit conversion, calculate the actual flow velocity, including:

[0079] 5.1 Calculation of horizontal water outflow velocity

[0080] When the water flows horizontally, it is only subject to gravity, and the trajectory equation has no coefficient for the first-order term:

[0081]

[0082] Compare the parabola y = ax 2 The initial horizontal velocity is obtained as follows:

[0083]

[0084] Where g = 980 cm / s 2 (Gravity acceleration). In actual calculations, the formula derived in section 5.2 below can be used for uniform calculation; simply set the coefficient b to 0.

[0085] 5.2 Calculation of inclined outlet velocity: When the pipe outlet is inclined, the initial velocity of the water flow contains horizontal and vertical components. The trajectory equation simplifies to y = ax when the outlet is the origin. 2 +bx, where Combining physical motion decomposition: v x =v0*cos(α), v y =v0*sin(α), from which the formula for the total velocity can be derived:

[0086]

[0087] Based on the relationship between the velocity direction and the coefficient b, the horizontal component... Vertical component, And the physical equations are known:

[0088]

[0089] Compare the relationships between the coefficients: The final velocity formula is obtained as follows:

[0090]

[0091] When applying this method, the initial velocity of each frame is calculated and averaged, where n represents the number of frames.

[0092]

[0093] 5.3 Unit conversion algorithm, which converts pixel units to physical units (centimeters) to ensure accurate and reliable speed results, including:

[0094] Pixel ratio calibration is used, with the outlet diameter as the reference. Let the actual outlet diameter be φ (unit: cm), and the number of pixels in the image detecting the outlet diameter be φ. p Then the pixel-to-centimeter conversion ratio For example, if the actual diameter of the water outlet is 10cm, but it is detected as 20 pixels in the image, then the conversion ratio is 0.5cm / pixel.

[0095] The parabolic pixel coefficients a obtained by image fitting p Convert to physical coefficient 'a', formula:

[0096]

[0097] If a p =0.004 pixels -1 If the conversion ratio is 0.5cm / pixel, then

[0098] When the user does not manually input the actual diameter of the outlet, the system estimates φ based on the width of the YOLO detection frame. p It combines a database of common outlet diameters for matching and recommendation, and automatically calculates the conversion ratio after user confirmation, reducing human error.

[0099] Example 2

[0100] To verify the effectiveness of this invention, water flow data with outlet diameters of 8mm, 12mm, and 16mm were manually measured sequentially in a simulated scenario. Experiments were conducted under various conditions, including no angular offset scenario, viewpoint offset scenario (horizontal / vertical offset of 15° to 45°), and scenarios with and without pipe pitch angle parameters (gravitational acceleration g was taken as 960cm / s²). 2 The following data compares traditional manual measurement data with the calculation results of this invention:

[0101] Table 1 shows data from a water flow monitoring experiment using an 8mm orifice pipe. The experiment verified that there was no horizontal or vertical viewing angle shift at a 15° upward angle from the pipe opening. The pixel conversion ratio was 1px = 0.0176cm. Based on the water flow velocity analysis report, 50 velocity samples were collected, and an average velocity analysis was performed on every 5 samples. The average physical equation obtained using this system is y = 0.077204x. 2 Substitute the coefficient -0.001881x into the formula. The average water flow velocity was 79.67 ± 0.04 cm / s. The final velocity range stabilized at 80–84 cm / s, with an error range of 1.09%–3.80% compared to the manually measured velocity.

[0102] Table 1. Manually measured flow velocity data for an 8mm orifice.

[0103]

[0104] Example 3:

[0105] Table 2 data comes from a water flow monitoring experiment using a 12mm orifice pipe. The experiment included a 15° leftward offset horizontal viewing angle and no pipe inlet pitch angle or vertical viewing angle offset for verification. The pixel conversion ratio was 1px =

[0106] 0.0420 cm. Based on the water flow velocity analysis report, 50 velocity samples were collected, and an average velocity analysis was performed based on every 5 samples. The average water flow velocity obtained using this system is 66.19 ± 0.06 cm / s, and the average physical equation is: y = 0.111855x 2 The final velocity range stabilizes at 66–68 cm / s. The corrected physical equation is y = 0.118187x. 2 Substitute into the formula The average water flow velocity was 64 cm / s, with an error of 1.78% compared to the manually measured velocity.

[0107] Table 2. Manually measured flow velocity data for a 12mm orifice.

[0108]

[0109] Example 4:

[0110] Table 3 shows data from a water flow monitoring experiment using a 16mm orifice pipe. The experiment verified that there was no horizontal or vertical viewing angle shift at a 15° upward angle from the pipe opening, and the pixel conversion ratio was 1px = 0.0306cm. Based on the water flow velocity analysis report, 50 velocity samples were collected, and average velocity analysis was performed based on every 5 samples. The average physical equation obtained using this system is y = 0.109690x. 2 Substitute the coefficients into the formula The average water flow velocity was 66.84 ± 0.11 cm / s. The final velocity range stabilized at 66–68 cm / s, with an error range of 0.645%–1.564% compared to the manually measured velocity.

[0111] Table 3. Manually measured flow velocity data for a 16mm orifice.

[0112]

[0113]

[0114] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0115] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method and system for capturing water flow trajectory and measuring flow velocity in underground coal mines, the system comprising modules for image acquisition, water flow scene detection, water flow trajectory detection, physical constraint fitting, filtering stabilization, angle correction, velocity calculation, user interaction, and result display, characterized in that, Includes the following steps: Step 1, Multimodal water flow detection: This step integrates three algorithms: motion detection, color detection, and edge detection. Motion detection uses a combination of the MOG2 background subtraction algorithm and frame difference method. Color detection is based on the HSV color space with dual thresholds and auxiliary color ranges. Edge detection integrates the Canny, Sobel, and Laplacian algorithms. The output is a set of water flow trajectory points, which is used as the input for the next step of fitting. Step 2, Physically Constrained Parabolic Fitting: Based on the trajectory point set output in Step 1, and using monotonicity constraints and a multi-strategy fitting algorithm, a parabolic trajectory and coefficients conforming to the parabolic motion law of water flow are obtained. Step 3, Angle Correction and Coordinate Transformation: The parabolic coefficients obtained in Step 2 are processed to offset the viewing angle θ and the pipe inclination angle α, and the corrected standard parabola is output. Step 4, Kalman filter stabilization: Based on the parabolic coefficients corrected in Step 3, a state model and a noise model are constructed, and combined with physical constraints, real-time stable monitoring results are provided. Step 5, Physical velocity calculation: The physical initial velocity is calculated based on the parabolic equation. When the water flows horizontally, the initial velocity is obtained based on the parabolic relationship without a first-order term. When the water flows at an angle, the horizontal and vertical components are solved simultaneously. The unit conversion is achieved by combining the water outlet diameter with pixel ratio calibration.

2. The method for capturing the trajectory and measuring the velocity of underground water flow in a coal mine according to claim 1, characterized in that, The motion detection includes background subtraction and frame difference methods to detect dynamic pixels. Subsequent morphological processing eliminates noise, and dynamic pixels detected by both algorithms are preferentially fused to increase the weight of pixels in the main water flow area in the lower right corner of the target detection region, thereby capturing more complete water flow motion features. The color detection includes HSV color space dual-threshold water flow detection. After detection, a closing operation is used to fill holes in the water flow area, and a dilation operation is used to enhance regional connectivity. The edge detection includes the fusion of Canny, Sobel, and Laplacian algorithms. The edge detection results of the three algorithms are combined using a bitwise OR operation to ensure complete water flow edge information.

3. The method for capturing the trajectory and measuring the velocity of underground water flow in a coal mine according to claim 1, characterized in that, The physical constraint parabolic fitting includes: monotonicity constraint, right-side filtering, vertical range constraint, and continuity check; the vertex constraint fitting includes a fitting algorithm with the outlet as the vertex of the parabola, and when all angle parameters are 0, the fitting equation is yy outlet =a(xx) outlet ) 2 The outlet constraint fitting includes a constraint algorithm where the parabola passes through the center of the marked outlet circle, and the constraint condition is...

4. The method for capturing the trajectory and measuring the velocity of underground water flow in a coal mine according to claim 1, characterized in that, The angle correction and coordinate transformation include: viewpoint correction, translation of the coordinate origin, and coefficient correction; the equation of the parabola after viewpoint correction is: Where θ h For the horizontal viewpoint offset angle, θ v This refers to the vertical viewing angle offset; a p b p is a coefficient.

5. The method for capturing the trajectory and measuring the velocity of underground water flow in a coal mine according to claim 1, characterized in that, The physical velocity calculation includes: horizontal water discharge, inclined water discharge, and unit conversion; in the horizontal water discharge calculation, only the initial horizontal velocity is considered, and the initial vertical velocity is not considered, v x =v0*cos(α)=v0,v y =v0*sin(α)=0; In the calculation of the inclined water outlet, the equation of the parabola with the vertical initial velocity as the origin is considered: a is the corrected parabolic physical coefficient, b is the corrected parabolic physical coefficient, φ is the actual diameter of the outlet, φ p This parameter represents the number of pixels representing the outlet diameter in the image. It supports automatic calibration based on the YOLO detection frame width and a database of common outlet diameters. It can also be manually calibrated according to the actual water flow velocity, and this parameter is used in subsequent flow velocity measurements.

6. The method and system for capturing the trajectory and measuring the velocity of underground water flow in a coal mine according to claim 1, characterized in that, The image acquisition module includes: acquiring water flow video images, supporting ordinary RGB cameras; the water flow scene detection module includes: detecting overall moving water flow based on a target recognition model to reduce background interference; the water flow trajectory detection module includes: implementing a multimodal water flow detection algorithm, outputting a set of water flow trajectory points through motion detection, color detection, and edge detection sub-modules; the curve fitting module includes: implementing a physically constrained parabolic fitting algorithm, including outputting parabolic coefficients through monotonicity constraint and multi-strategy fitting sub-modules; the filtering stabilization module includes: implementing a Kalman filter stabilization algorithm, constructing a state model and a noise model, performing state prediction, measurement update, and physical constraint checking, and outputting the stabilized parabolic coefficients; the velocity calculation module includes: implementing a physical velocity calculation algorithm, including a unit conversion sub-module and a velocity solution sub-module, used to output water flow velocity; the result display module includes: displaying real-time monitoring results and statistical analysis, supporting graphical interaction.

7. The method and system for capturing the trajectory and measuring the velocity of underground water flow in a coal mine according to claim 1, characterized in that, The system also includes parameter configuration, data management, and pixel calibration modules; The angle parameter configuration module includes: configuring horizontal viewing angle, vertical viewing angle, and pipe inlet pitch angle; the physical parameter configuration module includes: configuring gravitational acceleration and outlet diameter; the historical data management module includes: managing historical speed records and supporting data query and statistical analysis; the pixel calibration module includes: automatic pixel ratio calibration and manual calibration functions, with calibration results synchronized to the speed calculation module in real time.

8. The method and system for capturing the trajectory and measuring the velocity of underground water flow in a coal mine according to claim 1, characterized in that, The system supports the following practical functions: real-time monitoring, providing real-time water flow velocity monitoring with a delay of ≤100ms and a velocity display accuracy of 0.01m / s; multi-scene adaptation, supporting various scenarios such as horizontal water discharge, tilted water discharge, and viewpoint shift, and automatically switching between fitting and velocity calculation logic; data export, supporting the export of monitoring data, including original trajectory points, fitting coefficients, and velocity results; and visualization display, providing trajectory visualization, velocity curves, statistical analysis information, and supporting keyboard interactive parameter adjustment.

Images