Radar flow measurement method based on video assistance

Abstract

The invention relates to a radar flow measurement method based on video assistance. The method comprises the following steps: S1) measuring the vertical distance between the water level radar device and the liquid level by using the water level radar device, and sending the vertical distance value to a visual system with a video acquisition device; s2) acquiring a first water surface flow velocity by using a visual system, and transmitting the first water surface flow velocity to a flow velocity radar device; and S3) a speed measurement interval is set based on the first water surface flow velocity, the flow velocity radar device calculates a Doppler frequency shift interval corresponding to the speed measurement interval, a frequency reflecting the water flow velocity is found in the Doppler frequency shift interval, and the final water surface flow velocity is calculated according to the frequency.

Description

Technical Field

[0001] This invention belongs to the technical field of water flow velocity measurement equipment, specifically relating to a video-assisted radar flow measurement method. Background Technology

[0002] Water surface velocity measurement is a core component of hydrological and water conservancy monitoring, and the accuracy of its data directly impacts the effectiveness of flow measurement, water resource allocation, flood warning, and watershed management. Currently, radar flow measurement technology, with its outstanding advantages such as non-contact measurement, convenient installation and maintenance, and strong resistance to harsh environments, has been widely applied in the field of hydrological and water conservancy monitoring. The core principle of this technology is based on the radar Doppler effect: when the microwave signal emitted by the radar illuminates the moving water surface (including ripples), the reflected wave is received by the radar. The received radar wave signal has a frequency shift relative to the emitted wave signal, and the water surface velocity can be calculated by calculating this frequency shift.

[0003] However, existing radar current velocity measurement technology faces the following technical bottlenecks that urgently need to be addressed:

[0004] 1) Technical bottleneck of insufficient measurement accuracy caused by "signal-to-noise ratio" under low flow velocity conditions: Radar velocity measurement relies on the effective reflection of radar waves by the "roughness" (ripples, small waves) of the water surface. When the water surface velocity is very low, such as < 0.3 m / s, the water surface tends to be calm, and the water surface ripples are characterized by "small amplitude and few number". At this time, the reflected signal is extremely weak, making it difficult for the radar to identify the correct frequency component of the reflected signal. Therefore, false alarms or inaccurate measurements often occur in low flow velocity scenarios. For example, a flow velocity of 0 m / s may be falsely reported as 0.15 m / s.

[0005] 2) Significant technical bottleneck due to wind disturbance (wind-generated wave) error: When the wind speed on the water surface is ≥3m / s, the wind force will cause a large number of irregular ripples or waves on the water surface. The direction of motion of such ripples is often inconsistent with the direction of the main water flow. The radar cannot distinguish the motion of "wind-generated waves" from the motion of "water flow driven waves", resulting in the radar receiving a reflected wave containing both "wind disturbance components" and "water flow components". It will mix the "wind disturbance components" and "water flow components" for processing. When the wind speed is high and the water flow is slow, the "surface velocity" measured by the radar actually includes the velocity component of the wind-driven water wave motion, resulting in a measured value that is much higher than the actual velocity of the water body. For example, when the water flow is approximately still (flow velocity ≈ 0), a water flow velocity of 0.2~0.5m / s will appear, producing a serious positive error.

[0006] 3) Interference from complex surface motion patterns: Natural water bodies often exhibit complex flow regimes with multiple motion components coexisting, such as main currents, backflows, eddies, and turbulence. A radar beam illuminates a region (typically several meters to tens of meters in diameter), and its echo is the combined result of all motion components within that region. It cannot effectively identify and separate the "main current" component that represents the overall transport of the water body. Under complex flow regimes, the measured values ​​may differ significantly from the actual main current velocity, with deviations reaching over 0.3 m / s. Summary of the Invention

[0007] To address the above problems, this invention provides a video-assisted radar flow measurement method, comprising:

[0008] S1) Use a water level radar device to measure the vertical distance between itself and the water surface, and send the vertical distance value to a vision system with a video acquisition device;

[0009] S2) Using the vision system to acquire a first water surface flow velocity, and transmitting the first water surface flow velocity to the flow velocity radar device, wherein the step of acquiring the first water surface flow velocity includes:

[0010] S2-1) Collect video data of the water surface, convert it into single-channel image data in grayscale form, and use the mask method to extract the single-channel image data of the region of interest;

[0011] S2-2) Perform smoothing filtering on the single-channel image data of the region of interest to remove abnormal noise;

[0012] S2-3) Adaptive local Gaussian thresholding is performed on the single-channel image data after filtering to remove abnormal noise to obtain a binarized image;

[0013] S2-4) Feature points are selected from the binarized image using the ORB feature point detection method;

[0014] S2-5) The sparse optical flow of the feature points is calculated using the pyramid-based Lucas-Kanade method to obtain the pixel displacement of the feature points;

[0015] S2-6) Calculate the first water surface velocity using pixel displacement, pixel-to-actual-length conversion coefficient, and time interval;

[0016] S3) Based on the first water surface velocity, a velocity measurement interval is set, and the Doppler frequency shift interval corresponding to the velocity measurement interval is calculated using a flow velocity radar device. The frequency reflecting the water flow velocity is found in the Doppler frequency shift interval, and the final water surface velocity is calculated based on the frequency.

[0017] Beneficial effects of the present invention

[0018] The video-assisted radar flow measurement method provided by this invention first obtains high-confidence surface velocity using a visual flow measurement method, namely optical flow. During the visual flow measurement process, by extracting precise feature points such as floating objects and bubbles, high-confidence surface velocity can still be obtained even in low-velocity conditions with weak surface ripples, greatly improving the measurement accuracy in low-velocity situations. By tracking only the feature points of floating objects moving synchronously with the water flow direction, wind interference errors caused by wind ripples can be greatly reduced or eliminated, improving or overcoming the problem of false velocity due to wind interference faced by traditional radar flow measurement technology. Then, by reasonably setting a velocity measurement interval, the target interval is further locked and transformed into a Doppler frequency shift interval, which can further filter out frequency interference components and effectively avoid measurement deviations caused by multiple motion components in complex water surface motion patterns, greatly improving the accuracy and reliability of measuring water flow velocity in complex hydrological environments. Attached Figure Description

[0019] Figure 1 This is a flowchart of the video-assisted radar flow measurement method of the present invention;

[0020] Figure 2 This is a flowchart of the method for obtaining water surface flow velocity using a vision system, as used in this invention. Detailed Implementation

[0021] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0022] like Figure 1 As shown, this invention provides a video-assisted radar flow measurement method. The method involves constructing a water level radar device, a flow velocity radar device, and a vision system. The video acquisition devices in the water level radar device and the vision system are set to have equal vertical distances to the water surface, allowing the water level radar device to measure the vertical distance from the video acquisition device to the water surface. The water level radar device, flow velocity radar device, and vision system can use existing known devices. Many models of radar water level gauges (i.e., water level radar devices) and radar flow velocity meters (i.e., flow velocity radar devices) are available on the market, and not listed here. The vision system includes: a high-definition industrial camera or webcam with a resolution ≥1920×1080 and a frame rate ≥25fps as the video acquisition device, memory, storage, and data processing devices. This vision system can be a computer device connected to the high-definition industrial camera or webcam via wired or wireless means. It can be purchased commercially or built independently.

[0023] The video-assisted radar flow measurement method includes:

[0024] Step S1) Use a water level radar device to measure the vertical distance between itself and the water surface, and send the vertical distance value to a vision system with a video acquisition device.

[0025] The water level radar device emits radar waves to the water surface, calculates the vertical distance between itself and the water surface by calculating the round-trip time of the beam, and transmits the vertical distance value to the vision system in real time.

[0026] The video acquisition device can be a high-definition industrial camera or webcam as described above, preferably a video acquisition device with a 360° omnidirectional rotating lens.

[0027] Step S2) Use the vision system to obtain the first water surface velocity and transmit the first water surface velocity to the flow velocity radar device.

[0028] The first water surface velocity can be obtained by optical flow method. The software program for executing the optical flow method to obtain the first water surface velocity is pre-stored in the vision system. When the software program is run, the first water surface velocity is obtained by combining the data obtained through the hardware device.

[0029] The optical flow method used to obtain water surface velocity can employ existing known methods. However, existing publicly available optical flow methods for obtaining water surface velocity suffer from insufficient accuracy. Current optical flow velocimetry relies on the fundamental assumption of "constant brightness," meaning that the brightness or color intensity of a point on the same object remains constant across two consecutive image frames. "Constant brightness" implies a uniform spatial distribution of light, meaning that there are no drastic, localized changes in light intensity throughout the observed scene. Under this assumption, the movement of pixels in the image (i.e., optical flow) is entirely caused by the movement of objects. However, in practical applications, brightness is not constant. Complex outdoor lighting conditions, such as changes in ambient light due to weather, and shadows cast by man-made structures or natural objects, can all interfere with the imaging results.

[0030] Therefore, this invention proposes an improved optical flow method to obtain water surface velocity, thereby improving the accuracy of the obtained water surface velocity.

[0031] The method for obtaining water surface flow velocity proposed in this invention includes:

[0032] Step S2-1) Collect video data of the water surface, convert it into single-channel image data in grayscale form, and use the mask method to extract single-channel image data of the region of interest.

[0033] Video data of the water surface is acquired using a camera or webcam, with a resolution ≥1920×1080 and a frame rate ≥25fps. During the conversion of the acquired video data into a single-channel grayscale image, the R, G, and B channels of the video data are weighted and fused. The color (RGB ratio) of a moving object remains essentially constant under uniform lighting conditions, but the R, G, and B values ​​change individually with the object's motion. Optical flow algorithms require tracking a single, stable signal. Calculating the R, G, and B channels independently simultaneously yields three potentially inconsistent motion vectors, leading to contradictions and ambiguities. The grayscale conversion process is essentially a "weighted fusion" process; based on the sensitivity of the human eye, it fuses the information from the R, G, and B channels into a single value that best represents the perceived brightness, making it easier for optical flow methods to track object motion.

[0034] Performing masking operations on a single-channel image can filter out irrelevant background interference and extract the region of interest. The mask here uses a binary image of the same size as the original video. Masking operations are performed on the single-channel image, and each masking operation can process data in groups of one frame of video data. White areas with a value of 255 correspond to the region of interest, and black areas with a value of 0 correspond to the background area.

[0035] Converting video data into single-channel image data can reduce the computational load of subsequent processing. Furthermore, with the overall light intensity remaining constant, the grayscale values ​​obtained through weighted fusion remain relatively stable, allowing the algorithm to focus on the geometric motion itself. This avoids the overall RGB value ratio changes caused by changes in the color temperature of the light source under illumination. Such changes are not caused by motion, but by strong or localized changes in brightness, such as the brightness change caused by strong sunlight being suddenly blocked by clouds.

[0036] Since optical flow methods essentially calculate the motion vector of pixels across the entire image, while radar flow measurement methods only focus on water flow-related targets, random noise optical flow can be generated by static but textured background images such as grass and rocks on the bank, image noise, and slight changes in illumination. This reduces the accuracy of optical flow methods in measuring water velocity. Performing mask operations on single-channel images can effectively filter out such interference, ignore background areas, and focus only on brightness changes within the region of interest.

[0037] Step S2-2) Perform smoothing filtering on the single-channel image data of the region of interest to remove abnormal noise.

[0038] Gaussian blur filtering can be used for smoothing. This involves convolving single-channel image data using discrete odd-numbered Gaussian kernels. The kernel size can be 3x3, 5x5, or 7x7, and the Gaussian function is defined as follows:

[0039]

[0040] Where (x,y) are the coordinates relative to the kernel center, and σ is the standard deviation. The larger the σ is, the blurrier the image.

[0041] Step S2-3) Perform adaptive local Gaussian thresholding on the single-channel image data after filtering to remove abnormal noise, and obtain a binarized image.

[0042] The method for adaptive local Gaussian thresholding includes: firstly, determining the threshold for each pixel using preset initial parameters, which include the neighborhood window size (blockSize) and the threshold bias value (represented by a constant C).

[0043] The threshold T(x,y) of a pixel is the Gaussian weighted average of the pixels in the neighborhood window minus the threshold bias value.

[0044] Then, the gray value I(x,y) of the current pixel is compared with the threshold T(x,y) of the pixel. If the gray value I(x,y) of the pixel is greater than or equal to the threshold T(x,y) of the pixel, the pixel is set to 255 (white) in the grayscale image; otherwise, it is set to 0 (black), thereby converting the grayscale image into a binary image.

[0045] Gaussian weighting means that pixels closer to the center pixel within the neighborhood window contribute more to the threshold.

[0046] The threshold bias value controls the sensitivity of the binarized image by adjusting the threshold offset of each pixel.

[0047] The advantage of using the adaptive local Gaussian thresholding method is that even if the overall image illumination is uneven, such as the difference in brightness caused by water reflections, the threshold for each pixel is calculated based on the grayscale characteristics of its local neighborhood. The brightness change in the local area is relatively gradual, which can accurately distinguish the foreground (floating objects and other velocity-related targets) from the background. Using the Gaussian weighted average of pixels within the neighborhood window can better preserve local features and reduce the impact of noise on the threshold calculation. The slow change in background brightness caused by water reflections and ripples makes the calculated threshold close to the background grayscale value, thus binarizing the background into black or white and highlighting floating objects as contrasting colors (white and black). The resulting binary image with uniform background color and clear target outlines can satisfy the "constant brightness" assumption of the optical flow method and avoid the interference of water reflection fluctuations on the measured flow velocity.

[0048] Step S2-4) Select feature points by using the ORB feature point detection method on the binarized image.

[0049] The process of selecting feature points using the ORB feature point detection method includes: using the FAST algorithm to detect corner points in the image as keypoints; constructing an image pyramid (multi-scale scaling of the image) and detecting the keypoint at each layer of the pyramid to achieve scale invariance; calculating the principal direction of the keypoint and rotating the surrounding pixels to the principal direction to make the keypoint rotation invariant; generating a binary string as a descriptor by comparing the brightness of randomly selected pixel pairs around the keypoint; generating a BRIEFF descriptor for each keypoint; and finally performing feature matching, using Hamming distance to measure the similarity between two BRIEFF descriptors, and selecting the closest point pair as the successfully matched feature point pair, thus selecting the feature point.

[0050] Step S2-5) The sparse optical flow of the feature points is calculated using the pyramid-based Lucas-Kanade method to obtain the pixel displacement of the feature points.

[0051] The Lucas-Kanade (LK) method is a sparse optical flow estimation method based on the assumptions of constant brightness and temporal persistence (small motion). It calculates optical flow only for feature points extracted from the image, rather than calculating optical flow for every pixel in the image.

[0052] The present invention performs step S2-3) to ensure the assumption of constant brightness, and performs step S2-4) to ensure the assumption of time continuity (micro-motion).

[0053] Fundamental equation of optical flow: I x u+I y v+I t =0, where , which represents the spatial gradient in the x-direction. , which represents the spatial gradient in the y-direction. , which represents the time gradient, u=dx / dt, which represents the velocity in the x-direction (the optical flow component in the x-direction), v=dy / dt, which represents the velocity in the y-direction (the optical flow component in the y-direction).

[0054] Solving for the two unknowns u and v using the fundamental equations of optical flow, assuming that all pixels within a small neighborhood window move in the same way, and taking the neighborhood of a feature point (e.g., a 5x5 window), yields multiple equations, which are represented in matrix form:

[0055]

[0056] The least squares method is used to solve for the optical flow components, namely [u, v], the velocities in the x and y directions.

[0057] OpenCV provides the cv2.calcOpticalFlowPyrLK function for calculating sparse optical flow.

[0058] The Lucas-Kanade method, based on image pyramids, enables accurate tracking of large displacements of feature points between consecutive frames. Large motions are processed at the top layer of the pyramid with the lowest resolution, and the result is then used as the initial estimate for the next layer with higher resolution. This process is iterated until the bottom layer with the highest resolution handles small motions. This way, only minute motions need to be processed at each layer. Generally, five layers are sufficient. For example, assuming the original image is the bottom layer with a resolution of 640*480, the resolutions of the layers from bottom to top are 320*240, 160*120, 80*60, and 40*30, meaning the top layer has a resolution of 40*30.

[0059] Because the gradient at the outline of the floating object is obvious and meets the corner feature, the feature points selected by the ORB feature point detection method are located on the outline of the floating object, and the pixel displacement is calculated based on the optical flow component obtained in this step.

[0060] Specifically, the optical flow components [u, v] represent the number of pixels moved in the x and y directions. By tracking the optical flow components of feature points in several consecutive frames of images, the number of pixels moved by the feature point in the x and y directions can be accumulated, thus obtaining the pixel displacement of the feature point, which characterizes the motion of the feature point and represents the motion of the floating object.

[0061] Steps S2-6) Calculate the first water surface velocity using pixel displacement, pixel-to-actual length conversion coefficient, and time interval.

[0062] The first water surface velocity is obtained by multiplying the pixel-to-actual length conversion factor and the pixel displacement, and then dividing by the time interval. That is, the first water surface velocity = pixel-to-actual length conversion factor × pixel displacement / time interval.

[0063] The method for obtaining the pixel-to-actual-length conversion coefficient is as follows: First, a water level radar device is used to measure the vertical distance from the video acquisition device (e.g., an industrial camera or webcam) to the water surface. Then, the video acquisition device is positioned vertically downwards to capture images of the water surface, and the actual physical distance corresponding to the captured image size is calculated.

[0064] L = 2 * h * tan(α / 2) (1)

[0065] Where L represents the actual physical distance corresponding to the size of the captured image, h represents the vertical distance from the video acquisition device to the water surface, and α represents the field of view of the lens in the video acquisition device.

[0066] Then, the quotient of the actual physical distance L divided by the vertical resolution of the captured image is the pixel-to-actual-length conversion factor. For example, if the resolution of the captured image is 1920 x 1080, then L / 1080 is the pixel-to-actual-length conversion factor.

[0067] The time interval is the time difference between capturing two consecutive video frames. For example, when images are captured using a high-definition industrial camera or webcam at a frame rate of 25 fps, the time interval can be derived from the 25 fps frame rate.

[0068] Because the water level radar device and the video acquisition device in the vision system are set to have the same vertical distance to the water surface in advance, for example, the water level radar device is installed next to the video acquisition device, and is as close as possible to the video acquisition device and at the same height without affecting the rotation of the video acquisition device (e.g., a smart camera that can rotate in all directions). In this way, the vertical distance between the two and the water surface is equal. Thus, in step S1, the vertical distance between the water level radar device and the water surface is also the vertical distance h between the video acquisition device and the water surface. The vertical distance value received by the vision system can be directly used as the vertical distance h between the video acquisition device and the water surface in the above formula (1).

[0069] Furthermore, before calculating the water flow velocity, given the relatively fixed direction of water flow, the optical flow components corresponding to feature points with large deviations in their direction of motion can be removed to eliminate interference from abnormal optical flow data. Eliminating abnormal optical flow data avoids the adverse effects of non-water-flow-driven moving targets such as wind-blown debris and water surface bubbles on the accuracy of the water surface velocity, ensuring that the calculated result reflects the true water surface velocity.

[0070] Furthermore, the local Gaussian threshold segmentation parameters are dynamically adjusted to adapt to different real-world scenarios with uneven lighting.

[0071] Since the threshold for each pixel is the Gaussian weighted average of the pixels in the neighborhood window minus the threshold bias, the size of the neighborhood window and the threshold bias can be dynamically adjusted to achieve dynamic adjustment of the local Gaussian threshold.

[0072] For neighborhood windows, if the window size is too small, it cannot cover enough background area, and the threshold is easily affected by noise; if the window size is too large, it may mix part or all of the foreground image in the background area, causing the threshold to fail. The size of the neighborhood window can be adaptively adjusted according to the detection quality of feature points and the stability of optical flow.

[0073] The threshold bias value can be dynamically optimized according to the accuracy requirements of the first water surface velocity, ensuring that the foreground and background can be accurately distinguished under different uneven lighting conditions, thus ensuring the accuracy of measuring the first water surface velocity.

[0074] Step S3) Based on the first water surface velocity, a velocity measurement interval is set, and the Doppler frequency shift interval corresponding to the velocity measurement interval is calculated using a flow velocity radar device. The frequency reflecting the water flow velocity is found in the Doppler frequency shift interval, and the final water surface velocity is calculated based on the frequency.

[0075] The present invention uses a first water surface velocity as a reference value V. ref A water surface velocity variation value ΔV is set, thereby defining a velocity measurement range: [V ref –ΔV,V ref +ΔV).

[0076] The Doppler frequency shift interval corresponding to the speed measurement interval is calculated, effectively filtering out interference components outside the speed measurement interval.

[0077] The Doppler frequency shift interval corresponding to this velocity measurement range The calculation method is as follows:

[0078]

[0079]

[0080] Where f0 is the frequency of the radar wave emitted by the current velocity radar device, c is the speed of light, θ is the angle between the line connecting the moving target and the current velocity radar device and the actual direction of velocity, and V ref ΔV is the first surface velocity, and ΔV is the preset variation value of the surface velocity.

[0081] Typically, the radar wave frequency emitted by a flow velocity radar device is a relatively large value, usually 24 GHz or higher, and the frequency of the reflected wave is also a relatively large value of the same order of magnitude. Based on the well-known Doppler frequency calculation formula in this field, it can be known that the Doppler frequency shift is the difference between the radar wave frequency and the reflected wave frequency, usually several hundred hertz, which is proportional to the water flow velocity.

[0082] Therefore, the analysis of the Doppler frequency shift interval Find the frequency value within the range that reflects the water flow velocity; this frequency is the frequency shift value. Then, the final water surface velocity can be calculated:

[0083]

[0084] in The frequency reflecting the water flow velocity is obtained within the Doppler frequency shift range, v is the final water surface velocity, and the meanings of other parameters are the same as in the above formula.

[0085] The method for finding the frequency reflecting water flow velocity within the Doppler frequency shift interval is as follows: A flow velocity radar device emits radar waves towards the water surface and receives the reflected waves. The difference between the frequency of the emitted radar wave and the frequency of the reflected wave is used as the acquired offset frequency (referred to as frequency shift). Multiple offset frequencies are subjected to FFT (Fast Fourier Transform) calculations, resulting in multiple FFT results. These FFT results are N complex numbers, each corresponding to a frequency value, its amplitude, and phase. Frequency values ​​falling within the Doppler frequency shift interval are recorded or saved. When multiple frequency values ​​exist, the frequency with the largest amplitude (i.e., the highest energy) is selected as the frequency reflecting the water flow velocity. When there is only one frequency value, that frequency is selected as the frequency reflecting the water flow velocity. .

[0086] The above description, in conjunction with specific embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, several simple deductions or substitutions can be made without departing from the concept of the present invention, and all such deductions or substitutions should be considered to fall within the scope of protection defined by the claims submitted herein.

Claims

1. A video-assisted radar flow measurement method, characterized in that, include: S1) Use a water level radar device to measure the vertical distance between itself and the water surface, and send the vertical distance value to a vision system with a video acquisition device; S2) Using the vision system to acquire a first water surface flow velocity, and transmitting the first water surface flow velocity to the flow velocity radar device, wherein the step of acquiring the first water surface flow velocity includes: S2-1) Collect video data of the water surface, convert it into single-channel image data in grayscale form, and use the mask method to extract the single-channel image data of the region of interest; S2-2) Perform smoothing filtering on the single-channel image data of the region of interest to remove abnormal noise; S2-3) Adaptive local Gaussian thresholding is performed on the single-channel image data after filtering to remove abnormal noise to obtain a binarized image; S2-4) Feature points are selected from the binarized image using the ORB feature point detection method; S2-5) The sparse optical flow of the feature points is calculated using the pyramid-based Lucas-Kanade method to obtain the pixel displacement of the feature points; S2-6) Calculate the first water surface velocity using pixel displacement, pixel-to-actual-length conversion coefficient, and time interval; S3) Based on the first water surface velocity, a velocity measurement interval is set, and the Doppler frequency shift interval corresponding to the velocity measurement interval is calculated using a flow velocity radar device. The frequency reflecting the water flow velocity is found in the Doppler frequency shift interval, and the final water surface velocity is calculated based on the frequency.

2. The video-assisted radar flow measurement method according to claim 1, wherein the method for extracting single-channel image data of the region of interest using a masking method includes: A binary image of the same size as the original video is used to perform masking operations on the single-channel image. In each masking operation, one frame of video data is processed as a group of data, where the white area with a value of 255 corresponds to the region of interest, and the black area with a value of 0 corresponds to the background area.

3. The video-assisted radar flow measurement method according to claim 1, wherein Gaussian fuzzy filtering is used for smoothing.

4. The video-assisted radar flow measurement method according to claim 1, wherein steps S2-3 include: First, a neighborhood window and a threshold bias value are preset. The difference between the Gaussian weighted average value of the pixels in the neighborhood window and the threshold bias value is used as the threshold value of a pixel. Then, the gray value of the current pixel is compared with the threshold value of the pixel. If the gray value of the pixel is greater than or equal to the threshold value of the pixel, the pixel is set to 255 in the grayscale image. Otherwise, it is set to 0, thus converting the grayscale image into a binary image.

5. The video-assisted radar flow measurement method according to claim 1, wherein the process of selecting feature points using the ORB feature point detection method includes: The FAST algorithm is used to detect corner points in the image as key points. By constructing an image pyramid, key points are detected on each layer of the pyramid. The principal direction of the key point is calculated, and the pixels around the key point are rotated to the principal direction. By comparing the brightness of randomly selected pixel pairs around the key point, a binary string is generated as a descriptor. Then, a BRIEFF descriptor is generated for each key point. Hamming distance is used to measure the similarity between two BRIEFF descriptors, and the closest point pair is taken as the successfully matched feature point pair.

6. The video-assisted radar flow measurement method according to claim 1, wherein steps S2-6 include: The first water surface flow velocity is obtained by multiplying the pixel-to-actual-length conversion coefficient and the pixel displacement, and then dividing by the time interval. The water level radar device and the video acquisition device in the vision system are pre-set to be at the same vertical distance from the water surface. The vertical distance between the water level radar device and the water surface is the same as the vertical distance between the video acquisition device and the water surface. The actual physical distance corresponding to the water surface image captured by the video acquisition device is calculated. The quotient of the actual physical distance divided by the vertical resolution of the image is the pixel-to-actual-length conversion coefficient. The time interval is the time difference between the capture of two consecutive video images.

7. The video-assisted radar flow measurement method according to claim 1, wherein the method for calculating the Doppler frequency shift interval corresponding to the velocity measurement interval is: using the first water surface velocity as the reference value V ref A water surface velocity variation value ΔV is set, thereby defining a velocity measurement range: [V ref –ΔV,V ref +ΔV], calculate the Doppler frequency shift interval corresponding to this velocity measurement interval. ,in: ； ； Where f0 is the frequency of the radar wave emitted by the current velocity radar device, c is the speed of light, θ is the angle between the line connecting the moving target and the current velocity radar device and the actual direction of velocity, and V ref ΔV is the first surface velocity, and ΔV is the preset variation value of the surface velocity.

8. The video-assisted radar flow measurement method according to claim 7, wherein the method for finding the frequency reflecting the water flow velocity within the Doppler frequency shift interval is as follows: the flow velocity radar device emits radar waves to the water surface and receives reflected waves; the difference between the frequency of the emitted radar wave and the frequency of the reflected wave is used as the acquired offset frequency; a fast Fourier transform operation is performed on the acquired multiple offset frequencies to obtain multiple complex numbers, each of which corresponds to a frequency value and a corresponding amplitude and phase; the frequency values ​​falling within the Doppler frequency shift interval are recorded; when multiple frequencies exist, the frequency with the largest amplitude is selected as the frequency reflecting the water flow velocity; when there is only one frequency value, that frequency is selected as the frequency reflecting the water flow velocity.

9. The video-assisted radar flow measurement method according to claim 1, wherein the video acquisition device has a lens capable of 360° omnidirectional rotation.

Images