Tunnel water inrush measurement method based on multi-view fusion and optical flow field calculation

Abstract

A tunnel water inrush measurement method based on multi-view fusion and optical flow field calculation relates to the technical field of image recognition, and comprises the following steps: shooting a tunnel face water inrush video, carrying out image frame extraction on the obtained video, preprocessing the frame-extracted image, establishing a tunnel face image data set, labeling the preprocessed image, and carrying out data enhancement. Training a U-Net semantic segmentation model to identify a water inrush area image in the image data set, calculating an adjacent frame pixel displacement field in the water inrush area image in the image by adopting an RAFT optical flow model, extracting a multi-view effective flow velocity vector field, fusing the multi-view effective flow velocity vector field, and calculating the tunnel face water inflow; according to the method, pixel-level segmentation is carried out on the water inrush area by adopting U-Net through multi-view imaging, speed measurement is carried out through mask-constrained RAFT optical flow, speed vector fusion is carried out on a reference plane in a unified mode, influences of insufficient illumination, light reflection, local shielding and the like in a tunnel can be overcome, and the accuracy and stability of a result are improved.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, specifically to a method for measuring tunnel water inrush based on multi-view fusion and optical flow field calculation. Background Technology

[0002] During tunnel construction, sudden water inrushes at the tunnel face have a significant impact on engineering safety and structural stability. Traditional water inrush monitoring mainly relies on manual observation and measurement, which is inefficient, has limited accuracy, and poses safety hazards due to personnel operating close to the tunnel face. In recent years, some studies have attempted to achieve non-contact flow rate calculation using methods such as video processing, optical flow analysis, or particle image velocimetry, but these methods still have limitations: First, single-view information is insufficient to fully reflect the spatial characteristics of the water inrushes; second, traditional optical flow algorithms have poor robustness in low-texture and complex backgrounds, resulting in large errors in velocity estimation; and third, most methods remain at the level of flow field visualization and fail to be effectively combined with the actual water inrush measurement at the tunnel face.

[0003] Therefore, there is an urgent need for a comprehensive method that combines multi-view observation, automatic segmentation and identification, and high-precision flow rate calculation to achieve non-contact, automated, and refined calculation of water inflow at the tunnel face, providing technical support for tunnel construction and operation safety monitoring. Summary of the Invention

[0004] In view of this, the present invention proposes a tunnel water inrush measurement method based on multi-view fusion and optical flow field calculation. It uses a deep learning segmentation model to automatically extract the water inrush area and combines an advanced optical flow field calculation model to accurately invert the water flow velocity, thereby realizing non-contact, high-precision, and automated measurement of water inrush volume at the tunnel face, improving measurement efficiency and safety, and solving the problems mentioned in the background technology.

[0005] To solve at least one of the above-mentioned technical problems, the present invention provides a method for measuring tunnel water inrush based on multi-view fusion and optical flow field calculation, comprising the following steps: Step S1: Capture multi-view videos of the tunnel face; Step S2: Extract frames from the obtained video, preprocess the extracted frames, and establish a tunnel face image dataset. Step S3: Label the images of gushing water in the dataset, and train the U-Net semantic segmentation model to recognize the gushing water region images in the image dataset; Step S4: Calculate the displacement field of adjacent frame pixels in the image of the gushing water area using the RAFT optical flow model, and extract the effective flow velocity vector field from multiple perspectives; Step S5: Integrate the effective velocity vector field from multiple perspectives to calculate the water inflow at the working face.

[0006] The technical effects achieved by this invention are: This invention employs U-Net for pixel-level segmentation of the water inrush area through multi-view imaging and uses RAFT optical flow constrained by a mask to measure velocity, unifying the velocity to a reference plane. It then performs weighted fusion based on segmentation confidence and the angle between the line of sight and the tunnel face. Combined with calibration parameters and pixel-physical scale, it achieves objective, continuous, and non-contact measurement of water inrush volume at the tunnel face. This invention can overcome the influence of complex working conditions such as insufficient lighting, reflection, and local obstruction in the tunnel, thus improving the accuracy and stability of the results. Attached Figure Description

[0007] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0008] Figure 1 This is an overall flowchart of the present invention; Figure 2 This is a diagram of the experimental setup for this example; Figure 3 The probability of sudden water inrush obtained by inputting the U-Net model in this embodiment of the invention. Figure 2 A schematic diagram of the image after value-based processing. Detailed Implementation

[0009] The present invention will be further described in detail below with reference to the embodiments and accompanying drawings.

[0010] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention.

[0011] See Figure 1 A method for measuring tunnel water inrush based on multi-view fusion and optical flow field calculation includes the following steps: Step S1: Capture multi-angle video of water inrush at the tunnel face; The multi-view water inrush video is captured by a camera simultaneously from three perspectives. Artificial lighting can be used to increase the light intensity inside the tunnel during shooting. The camera is positioned at the following locations: directly opposite the water outlet at the tunnel face (facing the normal direction), at a 45° angle between the shooting axis and the main axis of the water outlet at the tunnel face, and at a 90° angle between the shooting axis and the main axis of the water outlet at the tunnel face (perpendicular to the normal direction). The distance between the camera and the tunnel face is 10m. The camera is set to autofocus mode and the frame rate is set to 30fps to capture a 20-second video of the water inrush at the tunnel face. The video type can be conventional image video or infrared imaging video. If it is conventional image video, its resolution should be no less than 1024×1024 and the frame rate should be no less than 30fps, preferably 50fps or higher.

[0012] By acquiring video of water inrush at the tunnel face from three perspectives, the shortcomings of insufficient information obtained from a single perspective are overcome, which can more comprehensively reflect the spatial characteristics of water inrush and improve the acquisition efficiency.

[0013] Step S2: Extract frames from the obtained video, preprocess the extracted frames, and establish a tunnel face image dataset. The tunnel face infrared thermal imaging video obtained in step S1 was read using the VideoCapture() function from the OpenCV library in Python. The frame rate of the infrared thermal imaging video was read using the get(cv2.CAP_PROP_FPS) function. 30 frames per second were extracted from the water inrush video at the tunnel face using the OpenCV read() function and saved to a local folder to obtain the raw tunnel face infrared image dataset. The GaussianBlur() function from OpenCV was used for noise reduction and smoothing of the tunnel face infrared images, with the Gaussian kernel size set to 3. Then, the Laplacian function from OpenCV was used to sharpen the images, enhancing the high-frequency components to make the images clearer. The Laplacian operator size was set to 3, the scaling factor to 1, the offset value to 1, and the boundary type to cv2.BORDER_DEFAULT. This yielded the preprocessed tunnel face image dataset.

[0014] By employing frame extraction technology, the dataset size is significantly expanded, avoiding the labor involved in manual photography or screenshotting. Gaussian blurring is used to reduce image noise caused by tunnel dust, vibration, and humidity, while the Laplacian algorithm restores image clarity and sharpens the image after Gaussian blurring. Step S2 effectively improves the visibility of water flow textures and boundaries, reduces subsequent segmentation and optical flow errors, compensates for the lack of detail in low-light environments in traditional methods, and thus improves measurement stability.

[0015] Step S3: Label the images of gushing water in the dataset, and train the U-Net semantic segmentation model to recognize the gushing water region images in the image dataset; Step S31: Label the images in the tunnel face water inrush image dataset, obtain the label files of the water inrush area, and convert the labels into semantic segmentation masks that correspond one-to-one with the original images in batches; First, using the LabelMe annotation tool, manually annotate the preprocessed image with polygons and export the annotation file. Then, fill the annotated polygons with foreground=1 and background=0 to create a single-channel semantic segmentation mask of the same size as the original image. The file name is the same as the original image. Figure 1 Save each item in a corresponding manner.

[0016] Step S32: Perform synchronous data augmentation on the image and mask to obtain the "image-mask" dataset; The Albumentations image enhancement library is used to simultaneously enhance the "image-mask" to improve the model's robustness to ambient light, slight blur, reflections and other ambient light disturbances. The enhancement features include RandomGridShuffle, random horizontal or vertical flipping, and random cropping and scaling. Enhanced samples are stored with the same structure as the original samples.

[0017] Synchronous data augmentation can effectively improve the model's generalization ability. For different scenarios, the accuracy can be restored by fine-tuning with a small number of samples of about 50 to 200.

[0018] Step S33: Establish a U-Net semantic segmentation model, wherein the model input is a normalized and size-aligned image of water inrush at the tunnel face, and the model output is a single-channel water inrush probability map of the same size as the input. The model is constructed using the U-Net semantic segmentation network. Each level of the encoder contains 3×3 convolution + BN + ReLU× and 2×2 max pooling, with a bottleneck of 1024 channels. The decoder uses upsampled transposed convolution and skip connections with the corresponding encoder features. The final 1×1 convolution outputs a single-channel probability map. The preferred input size is 512×512, and the output is the same size as the input.

[0019] Step S34: Read the data from the “Image-Mask” dataset as the training and validation sets to train the U-Net semantic segmentation model. After training, export the model weight file as the prediction channel. For model training and validation, the data was divided into training, validation, and test sets in a 7:2:1 ratio using the PyTorch deep learning framework, with stratified sampling covering different water inrush patterns. The DataLoader (batch size=8, shuffle=True) was used to load the samples. The model employed the BCEWithLogitsLoss loss function and the Adam optimizer (lr=1×10⁻⁶). - ³, weight_decay = 1 × 10-5 ), and combined with ReduceLROnPlateau(patience=5, factor=0.1, min_lr=1×10 -6 Adaptive learning rate; after 50 training epochs or until the validation set metrics no longer improve, save the optimal weight file U-Net.pth.

[0020] in This represents the model's raw output (logits) for pixel x. The actual mask label is 1 for water inrush and 0 for background, and N is the number of pixels involved in the calculation.

[0021] Step S35: Input the images from the tunnel face water inrush image dataset obtained in Step S2 that were not used for training into the trained U-Net model to obtain a pixel-level water inrush probability map. Threshold the segmentation probability map and filter it by morphology and connected components to obtain a water inrush binary mask M(x). The specific results are as follows: Figure 3 As shown.

[0022] Input the image from step S2 into the U-Net model trained in step S34 to obtain the pixel-level inrush water probability map P. (x) , for P (x) Thresholding (threshold τ=0.5) is performed to obtain an initial binary mask. Then, morphological operations (structuring element 3×3) and connected component filtering are performed to output the final gushing water binary mask and its contour and area properties, which serve as the constraint region for the subsequent optical flow calculation in step S4.

[0023] Step S4: Calculate the displacement field of adjacent frame pixels in the image of the gushing water area using the RAFT optical flow model, and extract the effective flow velocity vector field from multiple perspectives; The video sequence from step S2 is read using OpenCV, and adjacent frame pairs are constructed according to frame rate. Unify the image to a longer side of 512 pixels, and pad with zeros to multiples of 8; normalize to It is then converted into a PyTorch tensor (torch.from_numpy().permute(2,0,1).unsqueeze(0)) according to the RAFT input channel order (RGB) and sent to the GPU.

[0024] The RAFT optical flow model implemented in PyTorch is used for forward inference at mixed precision, and the pixel displacement field is obtained with iters=12 iterations: In the formula, This represents the dense optical flow vector output by RAFT; This represents the horizontal displacement vector, with rightward displacement being positive. This represents the vertical displacement vector, with downward being positive.

[0025] Read the binary mask of the inrush water in step S3 Optical flow is preserved only within the mask; a value of 1 is used for pixels representing gushing water, and a value of 0 is used for pixels representing the background. In the formula, This represents the optical flow vector within the mask constraint; Calculate amplitude and direction: In the formula, This represents the magnitude of the optical flow, i.e., the magnitude of the vector. The direction angle of the light flow is the angle of the light flow relative to the positive direction (rightward). Indicates the two-parameter arctangent; To suppress jitter noise, an amplitude threshold is set. Only retain The effective optical flow point. Then, based on the video frame rate (fps) and pixel-physical scale. (cm / px), converting displacement into pixel velocity field: In the formula, This represents the velocity vector of the image plane.

[0026] The superscript (i) indicates the i-th viewpoint. The output includes a vector field, amplitude map, and a set of effective point coordinates. Preferably, after step S4, the output may further include result statistics and output: statistically analyzing the average, median, and 95th percentile velocities and the proportion of effective optical flow points within the gushing water mask to generate a velocity heatmap and a vector overlay map; synchronously recording metadata such as frame rate, pixel-physical scale, and threshold, and archiving them together with the velocity field and mask for use in step S5 for fusion and cross-sectional area integration.

[0027] Step S4 constrains the RAFT optical flow calculation within the surge water mask obtained in S3, performs pixel estimation within the mask and converts it into physical velocity, effectively shielding against false motion caused by background reflections and camera shake.

[0028] Step S4 also requires robust filtering and smoothing of outlier vectors. Outlier vectors refer to outliers in optical flow caused by specular reflection, occlusion, strong edges, or tracking mismatch. Examples include excessively high optical flow velocity, unstable direction, and sudden large changes from the previous frame. The specific method for processing outlier vectors is to first remove extreme outliers in amplitude, and then apply... Median filtering fills in bad vectors with "multiple values" from their neighborhood.

[0029] Step S5: Integrate the effective velocity vector field from multiple perspectives to calculate the water inflow at the working face.

[0030] The specific steps of step S5 are as follows: Step S51: Based on the semantic segmentation mask of the gushing water obtained in step S31 and the multi-view effective velocity vector field obtained in step S4, the multi-view effective velocity vector field is mapped to the same reference plane according to the camera calibration parameters to obtain the velocity field and mask of each view on the reference plane. The reference plane is fused from the results of the three cameras (including coordinate alignment and unit unification) into the same geometric plane. The specific steps are as follows: Using the homography matrix H i The results from each viewpoint are projected onto a unified reference plane, as shown in the following equation: Depend on Obtain the displacement of the reference plane , and according to With reference plane pixel — physical scale The conversion speed is shown in the following formula: Indicates the first Each perspective is on the "reference plane" and at a point. The velocity vector at that point, in this invention, is uniformly defined as the working face; Step S52: Calculate the multi-view normalization scaling factor for each pixel, as shown in the following formula: In the formula, i, j, and k represent the i-th, j-th, and k-th viewpoints, respectively; Represents pixels x The normalized scaling factor at the i-th viewpoint, for different viewpoints, is obtained by applying the formula above. and Replacing i with j and k in the text allows you to calculate the scaling factor under different viewpoints; Represents pixels x The unnormalized contribution (also known as support) to the fusion result from the i-th perspective is used to characterize the relative credibility of that perspective to the final fusion result. This represents the pixel value obtained in S3 when the i-th viewpoint is on the reference plane. x It is the probability confidence level of "sudden water inrush"; k iα represents the absolute value of the cosine of the angle between the line of sight i and the normal of the reference plane; α and β represent weighting coefficients, where α∈[1,2] and β∈[0.5,1.5]. Step S53: Calculate the fused velocity field by summing the pixels within the surge water region segmented by U-Net. The formula for calculating the fused velocity field is as follows: In the formula, The velocity field of the reference plane representing viewpoint i; Step S54: In the inrush water region segmented by U-Net (i.e., the inrush water binary mask) =1) Set a hydraulic control surface, calculate the cross-sectional normal velocity, and perform discrete integration according to the inrush type to obtain the instantaneous inrush volume. The hydraulic control surface is a geometric set flux integration section that is artificially set in the calculation to cut across the water flow. The calculation method for the cross-sectional normal velocity is shown in the following formula: In the formula, V n ( x ) represents the cross-sectional normal velocity of the hydraulic control surface; n represents the normal unit vector of the hydraulic control surface; Water inflow types include diffuse water and stream-like water. For film-like water flowing along the wall (diffuse water), the control surface is a line set L orthogonal to the main flow direction, and its pixel thickness... From the side view (90) It is estimated that the control surface for a stream-shaped gush (stream-shaped water) is a circular cross-section A perpendicular to the axis of the gush, and the diameter d is obtained by observation.

[0031] The instantaneous inflow rate of seepage along the wall is calculated using the following formula: In the formula, Q1 represents the instantaneous inflow of dispersed water; h (x) This indicates the thickness of the water film measured by the side-view camera; S ref Represents the pixel-physical scale, determined by the actual distance between two pixels in the image in reality; The calculation method for the instantaneous inflow of a stream of water is shown in the following formula: In the formula, Q2 represents the instantaneous inflow of the stream of water; The average normal velocity on the cross section is represented by A; A represents the circular cross-sectional area of ​​the water jet, which is calculated by estimating the diameter of the circular cross-section obtained from the side view. Example: This embodiment uses a tunnel face physical model system to verify the tunnel inrush water measurement method based on multi-view fusion and optical flow field calculation. The experimental setup includes... Figure 2 As shown, it includes: 1) Constant head water supply system, acrylic panel type: A constant head water tank is equipped with an overflow pipe to limit the water level, an inlet valve, and connecting pipes; 2) Physical Model of Seepage at the Working Face (Quartz Sand Tank, 3D Printed Working Face): A quartz sand tank is set in front of the working face model to simulate the seepage medium channel in front of the working face; filters are installed at both ends of the quartz sand tank to prevent sand loss and ensure uniform inflow and outflow. 3D Printed Working Face Model Unit: Three working face panels with different crack widths are 3D printed using resin material and sprayed with cement mortar. Crack outlets are pre-set on the panels. The crack widths are 0.1cm, 0.2cm, and 0.4cm respectively.

[0032] 3) Image acquisition system: Three CMOS industrial cameras are used, positioned at the frontal view (0°), oblique view (45°), and side view (90°), respectively, and fixed camera brackets and stable supplementary lighting are used.

[0033] 4) Water collection and flow measurement system: The actual flow value is measured using a cumulative flow meter.

[0034] The parameter range involved in this embodiment includes: Crack width at the tunnel face: w: 0.1 cm, 0.2 cm, 0.4 cm; Constant head head difference: H is available in lengths of 10cm, 30cm, and 50cm. Video capture parameters: Frame rate 30fps, resolution 1024*1024. Frame rate and resolution remain consistent across all operating conditions; capture duration 20s.

[0035] The implementation steps are as follows: (1) Supply water to the acrylic plate constant head water tank so that the water level reaches the overflow port height and overflows, thereby stabilizing the water level at the preset height; by changing the overflow height, the target head difference H is obtained.

[0036] (2) Lay end filter screen in sand tank, fill with quartz sand and compact it in the preset manner; then carry out saturation treatment (to reduce the interference of air bubbles and unstable seepage on the test and form a stable seepage channel).

[0037] (3) Select a test specimen with the corresponding crack width, seal and connect the specimen to the water supply pipeline and fix it in place to ensure that the crack outflow area is within the field of view of the camera device.

[0038] (4) After the seepage stabilizes, start multi-view video acquisition to obtain video sequences.

[0039] (5) Perform the image preprocessing and U-Net automatic segmentation of the gushing water region in steps S1 to S3 above. The preprocessing methods include Gaussian blur, Laplacian sharpening, U-Net automatic segmentation of the gushing water region and binarization.

[0040] (6) Then, following the method in steps S4 to S5, the effective velocity vector field from multiple perspectives is fused to calculate the water inflow at the working face.

[0041] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the embodiments of the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for measuring tunnel water inrush based on multi-view fusion and optical flow field calculation, characterized in that, Includes the following steps: Step S1: Capture multi-angle video of water inrush at the tunnel face; Step S2: Extract frames from the obtained video, preprocess the extracted frames, and establish a tunnel face image dataset. Step S3: Label and augment the preprocessed images, and train the U-Net semantic segmentation model to recognize the images of the gushing water area in the image dataset; Step S4: Use the RAFT optical flow model to calculate the displacement field of adjacent frame pixels in the image of the water inrush area, and extract the effective flow velocity vector field from multiple perspectives; Step S5: Integrate the effective velocity vector field from multiple perspectives to calculate the water inflow at the working face.

2. The tunnel water inrush measurement method based on multi-view fusion and optical flow field calculation according to claim 1, characterized in that: The multi-view surge water video is captured synchronously by a three-view camera. The video type is conventional image video and / or infrared imaging video. The resolution of the conventional image video is not less than 1024×1024 and the video frame rate is not less than 30fps.

3. The tunnel water inrush measurement method based on multi-view fusion and optical flow field calculation according to claim 2, characterized in that: The three-view cameras are positioned at the locations directly opposite the water outlet of the working face, at an angle of 45° between the shooting axis and the main axis of the water outlet of the working face, and at an angle of 90° between the shooting axis and the main axis of the water outlet of the working face. The distance between the camera and the working face is 10m.

4. The tunnel water inrush measurement method based on multi-view fusion and optical flow field calculation according to claim 1, characterized in that: The specific steps of frame extraction and preprocessing in step S2 are as follows: OpenCV is used to read the frame rate and total number of frames of the video, the video images are extracted frame by frame to form the original image sequence, and the original image sequence is subjected to Gaussian filtering for noise reduction, smoothing, sharpening, normalization and size alignment to obtain the tunnel face water inrush image dataset.

5. The tunnel water inrush measurement method based on multi-view fusion and optical flow field calculation according to claim 4, characterized in that: The specific steps of step S3 are as follows: Step S31: Label the images in the tunnel face water inrush image dataset, obtain the label files of the water inrush area, and convert the labels into semantic segmentation masks that correspond one-to-one with the original images in batches; Step S32: Perform synchronous data augmentation on the image and mask to obtain the "image-mask" dataset; Step S33: Establish a U-Net semantic segmentation model, wherein the model input is a normalized and size-aligned image of water inrush at the tunnel face, and the model output is a single-channel water inrush probability map of the same size as the input. Step S34: Read the data from the "image-mask" dataset as the training set and validation set to train the U-Net semantic segmentation model. After training, export the model weight file as the segmentation model for the tunnel water inrush scene. Step S35: Input the images in the image dataset obtained in step S2 that were not used for training into the trained U-Net model to obtain a pixel-level inrush water probability map. Threshold the segmentation probability map and obtain a binary mask for inrush water by morphological and connected component filtering.

6. The tunnel water inrush measurement method based on multi-view fusion and optical flow field calculation according to claim 5, characterized in that: In step S34, the ratio of training set, validation set, and test set during training is 7:2:

1.

7. The tunnel water inrush measurement method based on multi-view fusion and optical flow field calculation according to claim 5, characterized in that: The specific steps of step S5 are as follows: Step S51: Based on the semantic segmentation mask of the gushing water obtained in step S31 and the multi-view effective flow velocity vector field obtained in step S4, the multi-view effective flow velocity vector field is mapped to the same reference plane according to the camera calibration parameters to obtain the velocity field and mask of each view on the reference plane. Step S52: Calculate the multi-view normalization scaling factor for each pixel, as shown in the following formula: In the formula, i, j, and k represent the i-th, j-th, and k-th viewpoints, respectively; Represents pixels x The normalized scaling factor from the i-th perspective; Represents pixels x The unnormalized contribution of the fusion result from the i-th perspective; This represents the pixel value obtained in S3 when the i-th viewpoint is on the reference plane. x It is the probability confidence level of "sudden water inrush"; k i α represents the absolute value of the cosine of the angle between the line of sight i and the normal of the reference plane; α and β represent weighting coefficients, where α∈[1,2] and β∈[0.5,1.5]. Step S53: Calculate the fused velocity field by summing the pixels within the surge water region segmented by U-Net. The formula for calculating the fused velocity field is as follows: In the formula, The velocity field of the reference plane representing viewpoint i; Step S54: Set a hydraulic control surface in the inrush water region segmented by U-Net, calculate the cross-sectional normal velocity, and perform discrete integration according to the inrush water type to obtain the instantaneous inrush water volume; The calculation method for the cross-sectional normal velocity is shown in the following formula: In the formula, V n ( x ) represents the cross-sectional normal velocity of the hydraulic control surface; n represents the normal unit vector of the hydraulic control surface; The types of water inflow include diffuse water and stream-like water. The instantaneous inflow rate of diffuse water is calculated using the following formula: In the formula, Q1 represents the instantaneous inflow of dispersed water; h (x) This indicates the thickness of the water film measured by the side-view camera; S ref Represents pixel-physical scale; The calculation method for the instantaneous inflow of a stream of water is shown in the following formula: In the formula, Q2 represents the instantaneous inflow of the stream of water; The value of the normal velocity on the cross section is represented by A; A represents the circular cross-sectional area of ​​the water jet.

8. The tunnel water inrush measurement method based on multi-view fusion and optical flow field calculation according to claim 7, characterized in that: The reference plane mentioned in step S51 is the tunnel face where water inrush occurs.

9. The tunnel water inrush measurement method based on multi-view fusion and optical flow field calculation according to claim 7, characterized in that: Step S4 requires robust filtering and smoothing of outlier vectors. The specific method is as follows: first remove extreme outlier magnitudes, then apply... Median filtering.

Images