A method and device for detecting road waterlogging based on a polarizer control

Abstract

The application discloses a kind of road waterlogging detection method and device based on polarizer control, it is related to road waterlogging detection field.The method of the application includes the following steps: obtaining the image of the region to be detected, preliminary judgment whether there is waterlogging;If suspected waterlogging area is detected, control polaroid rotates and collects N images;The N images collected are sequentially segmented, and the image corresponding to the suspected waterlogging area is segmented out;The average brightness of the image corresponding to the suspected waterlogging area is calculated, and if it is a real waterlogging area, an alarm is generated.The model established by the different image illumination change data of different waterlogging conditions when polaroid rotates can reflect the change characteristics of water reflection when polaroid rotates, which can distinguish waterlogging area from non-waterlogging area;The suspected area illumination change condition when rotating is judged by polarized light illumination change model, effectively avoiding the false alarm caused by pure image feature analysis, and improving the detection accuracy.

Description

Technical Field

[0001] This invention relates to the field of road water accumulation detection, and more specifically to a road water accumulation detection method and apparatus based on polarizing mirror control. Background Technology

[0002] Existing methods for detecting road water accumulation do not consider utilizing the characteristics of how reflected light from water changes under different angles of a polarizing lens to achieve water accumulation detection. This approach would ensure the detection of large-area water accumulation within the monitored area while avoiding false alarms from bright images resembling water accumulation in video image analysis. The advantages and disadvantages of several commonly used road water accumulation detection schemes are described below.

[0003] 1) Sensor-based water accumulation detection:

[0004] This method is a relatively universal detection method and can be widely applied to various scenarios. Examples include CN202111545707 "Vehicle Speed ​​Control Method, Device, Equipment and Medium Based on Water Accumulation Environment," CN202111405519 "An Emergency System for Underground Water Accumulation," and CN202121872432 "A Road Water Accumulation Detector," etc. The advantage of this method is its high detection sensitivity; the disadvantage is that because it uses sensors, several sensors need to be deployed at each desired monitoring point to ensure signal sampling. If water accumulation detection is performed on a large area of ​​roads, a large number of sensors need to be deployed, resulting in high deployment and maintenance costs.

[0005] 2) Water accumulation detection based on infrared images:

[0006] This method uses infrared images to express temperature information and utilizes temperature levels to determine the area of ​​water accumulation, as illustrated in CN202111139709, "Image Processing-Based Method and System for Detecting Water Accumulation on Urban Roads." The advantage of this method is that temperature information avoids interference from RGB images, filtering out some false alarms; the disadvantage is that the mapping between infrared imaging and temperature is significantly affected by ambient temperature.

[0007] 3) Water accumulation detection based on video images:

[0008] This method employs video image analysis based on surveillance cameras, such as CN202111131998 "Method, Device, and Storage Medium for Processing Waterlogged Areas Based on Swarm Intelligence," CN202110622559 "Road Waterlogging Detection System Based on Deep Convolutional Neural Networks," CN202011490946 "A Method and Device for Image Recognition," and CN202110223872 "A Rainfall Intensity Detection Method Based on Convolutional Neural Networks," etc. The advantage of this type of method is that the surveillance camera has a wide coverage area, allowing for the detection of road waterlogging over a large area; the disadvantage is that image features close to waterlogged areas can interfere with detection, leading to false alarms.

[0009] 4) Polarizing mirror control:

[0010] Current polarizing filter control is mainly used for video imaging control, such as CN202110704653 UAV airborne remote control polarizing filter system, CN202110437419 an automatic light adjustment device and method for a camera, CN202010835685 a method for acquiring surface texture of an object based on photometric stereo method and a photography studio, CN202010622342 a vision enhancement device and method for unmanned vehicles in extreme environments, CN201821875720 a lens adapter ring containing a filter system, etc. These methods adjust the rotation angle of the polarizing filter to improve detection results by eliminating reflections from the target object, to improve image quality by avoiding light interference, or to extract texture features by eliminating lighting factors.

[0011] Although the above schemes have achieved rotation control of the polarizing mirror, none of them have established a relationship model between the different imaging effects of the polarizing mirror at different rotation angles and the reflective characteristics of accumulated water, so as to achieve the purpose of detecting accumulated water by using the rotating imaging image sequence of the polarizing mirror. Summary of the Invention

[0012] In view of this, the present invention provides a method and apparatus for detecting road water accumulation based on polarizing mirror control, so as to solve the problems existing in the background art.

[0013] To achieve the above objectives, the present invention adopts the following technical solution:

[0014] A method for detecting road water accumulation based on polarizing mirror control includes the following steps:

[0015] Acquire images of the area to be detected to make a preliminary judgment on whether there is water accumulation;

[0016] If a suspected waterlogged area is detected, control the polarizing mirror to rotate and acquire N images;

[0017] The collected N images are sequentially segmented to extract images corresponding to suspected waterlogged areas.

[0018] Calculate the average image brightness corresponding to suspected water accumulation areas. If the area is indeed a water accumulation area, generate an alarm.

[0019] Optional, the specific steps for initially determining whether there is standing water are as follows:

[0020] The image of the area to be detected is matched with the water accumulation template image using a normalized correlation coefficient to obtain preliminary matching results;

[0021] Based on the texture features of water accumulation in the image, roughness and contrast are used to calculate the similarity of texture matching, and secondary matching is performed.

[0022] The extent of water accumulation is determined by combining the initial and secondary matching results.

[0023] Optionally, the specific calculation method for color matching is as follows:

[0024] Where x is the x-coordinate of a pixel, y is the y-coordinate of a pixel, I(x,y) represents the image to be matched, T(x,y) represents the water accumulation template image, T'(x,y) represents the template image minus its own mean divided by its own variance, I'(x,y) represents the image to be matched minus its own mean divided by its own variance, w represents the width of the template image, and h represents the height of the template image. For each pixel coordinate (x,y), x' is the x-coordinate of the template matching region, ranging from 0 to w-1; y' is the y-coordinate of the template matching region, ranging from 0 to h-1. The similarity R(x,y) of the template matching is calculated as follows:

[0025]

[0026] in:

[0027]

[0028]

[0029] When the similarity R(x,y) of the matched regions exceeds the set threshold, the matched region is counted as a color matching region.

[0030] Alternatively, the roughness can be calculated using the following method:

[0031] Calculate the average intensity value of the active window in the image, where the window size is represented as 2. k ×2 k Where k ranges from 0 to 5, x is the x-coordinate of a pixel in the entire image, and y is the y-coordinate of a pixel in the entire image. i is the x-coordinate of the active window region, j is the y-coordinate of the active window region, and g(i,j) is the grayscale value within the active window region. Then, the average intensity value A of the active window is... k (x,y) is:

[0032]

[0033] Calculate the average intensity difference, E, between non-overlapping windows in the horizontal and vertical directions for each pixel. k,h (x,y) represents the average horizontal intensity difference, E k,v (x,y) represents the vertical average intensity difference:

[0034] E k,h (x,y)=|A k (x+2 k-1 ,y)-Ak (x-2 k-1 ,y)|;

[0035] E k,v (x,y)=|A k (x,y+2 k-1 )-A k (x,y-2 k-1 )|;

[0036] For each pixel coordinate (x, y), the value of k that maximizes the average intensity difference corresponds to the optimal size S of the active window. best (x,y)=2 k Where m is the image width and n is the image height, calculate the optimal size S of the active window corresponding to each pixel in the image. best The mean value is used to obtain the roughness feature F. crs :

[0037]

[0038] Optionally, the contrast ratio can be calculated as follows:

[0039] Let μ4 be the fourth center distance of the image, and σ be the standard deviation of the image gray values. 2 Let α be the variance of the image grayscale values, and α4 be the kurtosis of the image grayscale values. The formula for calculating α4 is α4 = μ4σ. 4 Image contrast F con The calculation is as follows:

[0040]

[0041] For color-matching regions detected by the color matching submodule, detection is performed by the texture matching submodule, and the calculated roughness F... crs and contrast F con When all values ​​reach the set threshold, the matching area is determined to be a waterlogged area.

[0042] Optionally, the acquired N images can be sequentially segmented to extract images corresponding to suspected waterlogged areas. The specific steps are as follows:

[0043] Acquire image sequence I n The data are sequentially fed into the YOLOv4 network, and the corresponding waterlogged region sequence R is detected. n and the corresponding confidence sequence C n ;

[0044] For the confidence sequence C n The highest confidence level C is obtained by sorting and calculating. max The corresponding waterlogged area is R. max ;

[0045] For image sequence I n , with R max Each image in the sequence is segmented to obtain the sequence W of suspected waterlogged areas. n .

[0046] Optionally, the average brightness of the image corresponding to the suspected water accumulation area is calculated. Specifically, the average brightness of each water accumulation area in the sequence is calculated to obtain the average brightness vector to be identified. The average brightness vector to be identified is then fed into a pre-trained polarization illumination change model. If the time for continuous detection of water accumulation areas exceeds the set duration threshold T, then it is judged as a water accumulation area.

[0047] Optionally, this also includes constructing a polarization illumination variation model, specifically as follows: Several water accumulation scenes are pre-selected. For each scene, 36 images are collected by sampling once every 10 degrees of rotation. After labeling the water accumulation area in each image, the average brightness *l* of the water accumulation area is calculated. For each scene, a set of average brightness vectors L = {l1, l2, ..., l...} varies with the polarizing mirror angle. t}, where t is the sampling sequence number of the polarizing mirror rotation interval. N sets of water accumulation scene samples are collected as training samples for the polarization illumination change model. A one-dimensional convolution is performed on this data in time to obtain the polarization illumination change model.

[0048] A road water accumulation detection device based on polarizing mirror control includes a water accumulation image pre-detection module, a polarizing mirror rotation control module, an image sequence water accumulation region segmentation module, and a water accumulation image sequence analysis module;

[0049] When the water accumulation image pre-detection module detects a suspected water accumulation area, the polarizing mirror rotation control module controls the polarizing mirror to rotate and acquire images; the image sequence water accumulation area segmentation module performs sequence segmentation on the acquired images to segment out the suspected water accumulation areas in the images; and the water accumulation image sequence analysis module is used to determine the suspected water accumulation areas.

[0050] As can be seen from the above technical solution, compared with the prior art, the present invention provides a road water accumulation detection method and device based on polarizing mirror control, which has the following beneficial effects:

[0051] 1. The polarizing mirror is driven to rotate by a motor. The polarizing mirror is then driven to rotate after the video pre-inspection is carried out to the suspected area, which can avoid the polarizing mirror rotating continuously and extend the service life of the equipment. One picture is captured every 10 degrees of rotation. This can ensure that the different imaging effects obtained by the polarizing mirror at each rotation angle are captured, while avoiding excessive sampling and increasing the amount of calculation. In addition, the polarizing mirror rotation is started to judge the change of illumination after the pre-inspection, which can also effectively reduce the amount of calculation and reduce the load on the equipment.

[0052] 2. The model established by this invention based on the changes in image illumination under different water accumulation conditions when the polarizing mirror is rotated can reflect the changes in the reflection of water accumulation as the polarizing mirror rotates. These characteristics can distinguish between water accumulation areas and non-water accumulation areas. By judging the changes in illumination in suspected areas during rotation through the polarization illumination change model, false alarms caused by simple image feature analysis are effectively avoided, and the detection accuracy is improved. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0054] Figure 1 This is a technical roadmap of the present invention;

[0055] Figure 2 This is a diagram of the YOLOv4 network structure of the present invention;

[0056] Figure 3 This is a diagram of the one-dimensional convolutional network structure of the present invention. Detailed Implementation

[0057] 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 some embodiments of the present invention, and not all embodiments. 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.

[0058] This invention discloses a method for detecting road water accumulation based on polarizing mirror control, such as... Figure 1 As shown, it includes the following steps:

[0059] S1: Acquire an image of the area to be detected to make a preliminary judgment on whether there is water accumulation;

[0060] S2: If a suspected waterlogged area is detected, control the polarizing mirror to rotate and acquire N images;

[0061] S3: Sequentially segment the N collected images to extract images corresponding to suspected waterlogged areas;

[0062] S4: Calculate the average brightness of the images corresponding to the suspected waterlogged area sequence to obtain the average brightness vector; input the average brightness vector into the pre-trained polarization illumination change model to identify whether it matches the polarization illumination change of the waterlogged area. If it matches, it is a real waterlogged area, and an alarm is generated.

[0063] Specifically, in S1, when the camera acquires an image, if the polarizing filter is not rotating, the image is sent to the water accumulation image pre-detection module for pre-detection to roughly determine whether there is water accumulation in the current area. If no suspected water accumulation area is detected, the camera continues to acquire images.

[0064] In S2, if the water accumulation image pre-detection module detects a suspected water accumulation area, the polarizing mirror rotation control module starts to control the polarizing mirror to rotate 360 ​​degrees. The polarizing mirror collects one image for every 10 degrees of rotation, and a total of 36 images are collected for each rotation.

[0065] In S3, when the polarizing mirror completes a 360-degree rotation, the 36 image sequences acquired are sent to the water accumulation image sequence segmentation module to segment out the suspected water accumulation areas in the images.

[0066] In S4, the segmentation results of the image sequence are sent to the water accumulation image sequence analysis module to analyze whether a water accumulation area actually exists. If a water accumulation area exists, an alarm is generated.

[0067] Furthermore, the purpose of the water accumulation image pre-detection module in S1 is to improve detection speed and filter out most cases where there is obviously no water accumulation. Therefore, the detection conditions for pre-detection are relatively broad, and subsequent modules will perform detailed secondary judgment and confirmation. There are many methods for water accumulation image detection, such as deep learning-based detection, image histogram-based detection, edge-based detection, etc. Here, a color matching plus texture matching method is used to achieve rapid pre-detection of suspected water accumulation areas in the image. The water accumulation image pre-detection module consists of a color matching submodule and a texture matching submodule.

[0068] Color matching employs a normalized correlation coefficient (NRC) matching method. The template image is slid across the entire image, and the overlapping areas of the template image and the image to be matched are compared using NRC matching. Let x be the x-coordinate of a pixel, y be the y-coordinate of a pixel, I(x,y) be the image to be matched, T(x,y) be the template image, T'(x,y) be the template image minus its own mean divided by its own variance, I'(x,y) be the image to be matched minus its own mean divided by its own variance, w be the width of the template image, and h be the height of the template image. For each pixel coordinate (x,y), x' is the x-coordinate of the template matching region, ranging from 0 to w-1; y' is the y-coordinate of the template matching region, ranging from 0 to h-1. The similarity R(x,y) of the template matching is calculated as follows:

[0069]

[0070] in:

[0071]

[0072]

[0073] When the similarity R(x,y) of the matched regions exceeds the set threshold, the matched region is counted as a color matching region.

[0074] Color matching cannot reflect the detailed texture features of water accumulation, so texture features need to be introduced for secondary matching of color matching regions. There are various methods to describe texture features, and here the Tamura algorithm is used for texture feature matching. Tamura texture features include six features: roughness, contrast, orientation, line density, regularity, and coarseness. Here, we mainly use roughness and contrast to calculate the similarity of texture matching.

[0075] For the roughness feature of the image, the average intensity value of the active window of the image is first calculated, and the size of the window is represented as 2. k ×2 k Where k ranges from 0 to 5, x is the x-coordinate of a pixel in the entire image, and y is the y-coordinate of a pixel in the entire image. i is the x-coordinate of the active window region, j is the y-coordinate of the active window region, and g(i,j) is the grayscale value within the active window region. Then, the average intensity value A of the active window is... k (x,y) is:

[0076]

[0077] Calculate the average intensity difference, E, between non-overlapping windows in the horizontal and vertical directions for each pixel. k,h (x,y) represents the average horizontal intensity difference, E k,v (x,y) represents the vertical average intensity difference:

[0078] E k,h (x,y)=|A k (x+2 k-1 ,y)-A k (x-2 k-1 ,y)|;

[0079] E k,v (x,y)=|A k (x,y+2 k-1 )-A k (x,y-2 k-1 )|;

[0080] For each pixel coordinate (x, y), the value of k that maximizes the average intensity difference corresponds to the optimal size S of the active window. best (x,y)=2 kWhere m is the image width and n is the image height, calculate the optimal size S of the active window corresponding to each pixel in the image. best The mean value is used to obtain the roughness feature F. crs :

[0081]

[0082] The contrast of an image can be calculated by statistically analyzing the distribution of pixels. Let μ4 be the fourth-order center distance of the image, and σ be the standard deviation of the image's grayscale values. 2 Let α be the variance of the image grayscale values, and α4 be the kurtosis of the image grayscale values. The formula for calculating α4 is α4 = μ4σ. 4 Image contrast F con The calculation is as follows:

[0083]

[0084] For the color matching region detected by the color matching submodule, further detection is performed by the texture matching submodule. When the calculated roughness F crs and contrast F con When all values ​​reach the set threshold, the area is identified as a suspected water accumulation area detected by the water accumulation image pre-detection module.

[0085] Furthermore, in S2, when a suspected water accumulation area is detected by the water accumulation image pre-detection module, the polarizing mirror rotation control module sends a polarizing mirror rotation command to the monitoring camera equipped with polarizing mirror rotation control function, driving the polarizing mirror to rotate 360 ​​degrees. One image is captured for every 10 degrees of rotation, for a total of 36 images captured per rotation. After the polarizing mirror has rotated 360 degrees, it stops rotating and waits to receive the next drive command from the water accumulation image pre-detection module.

[0086] Furthermore, in S3, since the monitored area includes a large number of non-waterlogged areas, uniformly extracting features from these areas and sending them to the subsequent sequence analysis module would introduce significant errors in the calculation results. Therefore, it is necessary to segment the waterlogged areas in the image. Because the polarizing mirror rotates 360 degrees, at certain angles, the imaging effect of the actual waterlogged area may not be detected by the segmentation module, or the segmentation effect may not be ideal. Therefore, the detection region with the highest confidence in the sequence image is selected as the unified segmentation region for the image sequence. There are many image segmentation methods; here we take the YOLOv4 method as an example. The YOLOv4 network structure is as follows: Figure 2 As shown.

[0087] The process for segmenting water accumulation regions in the image sequence is as follows:

[0088] S31: For image sequence I nThe value of n ranges from 1 to 36. A total of 36 images are fed into the YOLOv4 network in sequence, and the corresponding water accumulation region sequence R is detected. n and the corresponding confidence sequence C n .

[0089] S32: For the confidence sequence C n The highest confidence level C is obtained by sorting and calculating. max The corresponding waterlogged area is R. max .

[0090] S33: For image sequence I n , with R max Each image in the sequence is segmented to obtain the final water accumulation region sequence W. n .

[0091] Furthermore, in S4, to learn the characteristics of illumination changes in the water accumulation area as the polarizing mirror rotates, several water accumulation scenes were pre-selected. For each scene, 36 images were collected by sampling once every 10 degrees of rotation. After labeling the water accumulation area in each image, the average brightness l of the water accumulation area was calculated. For each scene, a set of average brightness vectors L={l1,l2,···,l...} as the polarizing mirror angle changes was obtained. t}, where t is the sampling sequence number of the polarizing mirror rotation interval, ranging from 1 to 36. N sets of water accumulation scene samples are collected as training samples for the polarization illumination change model. A one-dimensional convolution is performed on this data over time to obtain the polarization illumination change model. The one-dimensional convolutional network structure is as follows: Figure 3 As shown.

[0092] The water accumulation region sequence W obtained by the image sequence water accumulation region segmentation module n The system calculates the average brightness of each waterlogged area in the sequence to obtain the average brightness vector to be analyzed. This vector is then fed into a pre-trained polarized illumination variation model to determine whether it is a waterlogged area. When the time for continuously detecting a waterlogged area exceeds a set duration threshold T, the system generates an alarm.

[0093] This embodiment also discloses a road water accumulation detection device based on polarizing mirror control, including a water accumulation image pre-detection module, a polarizing mirror rotation control module, an image sequence water accumulation region segmentation module, and a water accumulation image sequence analysis module;

[0094] When the water accumulation image pre-detection module detects a suspected water accumulation area, the polarizing mirror rotation control module controls the polarizing mirror to rotate and acquire images; the image sequence water accumulation area segmentation module performs sequence segmentation on the acquired images to segment out the suspected water accumulation areas in the images; and the water accumulation image sequence analysis module is used to determine the suspected water accumulation areas.

[0095] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0096] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for detecting road waterlogging based on a control of a polarizer, characterized by, The method comprises the following steps: Obtaining an image of a region to be detected to preliminarily determine whether there is waterlogging; If a suspected waterlogging region is detected, rotating a polarizer and collecting N images; Sequentially segmenting the collected N images to segment out images corresponding to the suspected waterlogging region; Calculating the brightness mean value of the images corresponding to the suspected waterlogging region, and generating an alarm if it is a real waterlogging region; The calculation of the brightness mean value of the images corresponding to the suspected waterlogging region is specifically as follows: calculating the brightness mean value in each waterlogging region in the sequence to obtain a brightness mean value vector to be identified, and inputting the brightness mean value vector to be identified into a pre-trained polarized light illumination change model; if the time of continuously detecting the waterlogging region exceeds a set duration threshold T, it is determined that it is a waterlogging region; Further comprising constructing a polarization light illumination change model, specifically as follows: a plurality of water accumulation scenes are selected in advance, 36 pictures are collected for each water accumulation scene in a manner of sampling once every 10 degrees of rotation, the brightness mean value of the water accumulation area is calculated after calibrating the water accumulation area of each picture , and a set of brightness mean value vectors varying with the angle of the polarization mirror is obtained for each scene , wherein is the serial number of the polarization mirror rotation interval, N sets of water accumulation scene samples are collected as polarization light illumination change model training samples, and a one-dimensional convolution is performed on the data in time sequence to obtain the polarization light illumination change model.

2. The method of claim 1, wherein the method is based on a polarized mirror control. The specific steps of preliminarily determining whether there is waterlogging are as follows: Color matching is performed on the image of the region to be detected and a waterlogging template image by using a normalized correlation coefficient to obtain a preliminary matching result; Based on the texture features of the image waterlogging, the similarity of texture matching is calculated by using roughness and contrast to perform secondary matching; Combining the preliminary matching result and the secondary matching result to determine the waterlogging condition.

3. The method of claim 2, wherein the method is based on a polarized mirror control. The specific calculation method of color matching is as follows: in, The x-coordinate of the pixel is The ordinate of the pixel is Indicates the image to be matched. Represents a template image of accumulated water. This represents the result of subtracting the template image's mean from its variance. This represents the result of subtracting the mean of the image to be matched from its variance. Indicates the width of the template image. This represents the height of the template image; for each pixel coordinate... , The x-coordinate of the template matching region, with a value ranging from 0 to... ; The vertical coordinate of the template matching region, ranging from 0 to... Calculate the similarity of template matching as follows: ； Wherein: ； ； When the similarity of the matching regions When the set threshold is exceeded, the matching area will be counted as a color matching area.

4. The method of claim 2, wherein the method is characterized by, The calculation method of roughness is as follows: The average intensity value of the moving window of the image is calculated, and the size of the window is represented as wherein the value range of is 0-5, is the horizontal coordinate of the pixel point of the whole image, is the vertical coordinate of the pixel point of the whole image; is the horizontal coordinate in the moving window area, is the vertical coordinate in the moving window area, is the gray value in the moving window area, and the average intensity value of the moving window is : ； calculating the average intensity difference between each pixel and the non-overlapping windows in the horizontal and vertical directions, is the horizontal average intensity difference, is the vertical average intensity difference: ； ； For each pixel coordinate The k value that makes the average intensity difference reach the maximum corresponds to the optimal size of the active window , is the image width, is the image height, the optimal size of the active window corresponding to each pixel in the image is calculated The average value of the roughness feature : 。 5. The method of claim 4, wherein the method is based on a polaroid control. The calculation method of contrast is as follows: Let be the fourth-order central moment of the image, be the standard deviation of the image gray values, be the variance of the image gray values, be the kurtosis of the image gray values, and the formula is be the contrast of the image be calculated as follows: ； For the color matching area detected by the color matching sub-module, detection is made by the texture matching sub-module. When the calculated roughness feature and contrast all reach the set threshold, the matching area is determined as a waterlogging area.

6. The method of claim 1, wherein the method is based on a polarized mirror control. Sequentially segmenting the collected N images to segment out images corresponding to the suspected waterlogging region, and the specific steps are as follows: Acquiring an image sequence , sequentially input into a YOLO V4 network, detecting a corresponding waterlogging area sequence and a corresponding confidence sequence ; For the confidence sequence , the sorting calculation obtains the maximum confidence , and the corresponding water accumulation area is ; For image sequences ,by Each image in the sequence is segmented to obtain a sequence of suspected waterlogged areas. .

7. A device for detecting road waterlogging based on a control of a polarizer, characterized by, The system comprises a waterlogging image pre-detection module, a polarizer rotation control module, an image sequence waterlogging region segmentation module, and a waterlogging image sequence analysis module; When the waterlogging image pre-detection module detects a suspected waterlogging region, the polarizer rotation control module controls the rotation of the polarizer and collects images; the image sequence waterlogging region segmentation module sequentially segments the collected images to segment out the suspected waterlogging region in the images; and the waterlogging image sequence analysis module is used to determine the suspected waterlogging region; The calculation of the brightness mean value of the images corresponding to the suspected waterlogging region is specifically as follows: calculating the brightness mean value in each waterlogging region in the sequence to obtain a brightness mean value vector to be identified, and inputting the brightness mean value vector to be identified into a pre-trained polarized light illumination change model; if the time of continuously detecting the waterlogging region exceeds a set duration threshold T, it is determined that it is a waterlogging region; Also include the construction of polarization light illumination change model, as follows: pre-selected several water scenes, each scene according to every 10 degrees of rotation sampling once, for each water scene to collect 36 pictures, for each picture to calibrate the water area, then calculate the average brightness of the water area , each scene gets a set of brightness mean vector with the change of the polarization mirror angle , wherein is the sequence number of the polarization mirror rotation interval, collect N groups of water scene samples as the polarization light illumination change model training samples, do one-dimensional convolution on the data in time sequence, get the polarization light illumination change model.

Images