Guardrail deformation monitoring method, device and system based on feature level semantic consistency

Abstract

The invention discloses a guardrail deformation monitoring method, device and system based on feature level semantic consistency. The method comprises the following steps: acquiring image data of a to-be-monitored guardrail acquired by a camera and point cloud data of the to-be-monitored guardrail acquired by a millimeter wave radar within a time period; extracting data points representing a guardrail in multiple frames of point cloud data collected by the millimeter wave radar in the time period, and then superposing the data points to obtain a guardrail point cloud set; extracting guardrail curve characteristics in the single-frame image data collected by the camera in the time period and carrying out guardrail contour detection to obtain a guardrail contour; coordinates of all pixel points on the guardrail contour are determined, and data points in the guardrail point cloud set are corrected; and guardrail deformation is identified based on the corrected guardrail point cloud set. The image data collected by the camera is used as a reference standard, millimeter wave radar data is guided to approach the image data, the interference problem is solved, and the accuracy of guardrail deformation monitoring is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of guardrail deformation monitoring, and in particular relates to a guardrail deformation monitoring method, device and system based on feature-level semantic consistency. Background Art

[0002] Highway corrugated guardrails are important safety facilities used to prevent vehicles from running out of the lane when they lose control. However, due to various reasons, corrugated guardrails may deform, affecting their protective performance. Therefore, it is necessary to monitor corrugated guardrails in a timely manner to improve road operation safety.

[0003] There are many existing methods for monitoring guardrail deformation, such as using cameras or drones to collect image data, and extracting guardrail features such as edges, contours, curvatures, etc. through analyzing image processing algorithms, and then determining the degree of deformation and position of the guardrail.

[0004] However, the use of cameras to monitor guardrail deformation has the following problems: it is affected by factors such as lighting, weather, and occlusion, requires a large amount of image data and computing resources, and the accuracy and robustness of the image processing algorithm needs to be improved. Summary of the invention

[0005] In order to solve the problem of poor accuracy caused by interference in the existing guardrail deformation monitoring using camera data, the present invention proposes a guardrail deformation monitoring method, device and system based on feature-level semantic consistency.

[0006] The objective of the present invention is achieved through the following technical solutions:

[0007] The first aspect of the present invention discloses a guardrail deformation monitoring method based on feature-level semantic consistency, comprising: obtaining image data of a guardrail to be monitored collected by a camera and point cloud data of a guardrail to be monitored collected by a millimeter-wave radar within a time period, wherein the camera and the millimeter-wave radar are installed on a smart pole at the same detection point; extracting data points representing the guardrail from multiple frames of point cloud data collected by the millimeter-wave radar within the time period and superimposing them to obtain a guardrail point cloud set;

[0008] Extracting the guardrail curve features in the single-frame image data collected by the camera within the time period and performing guardrail contour detection to obtain the guardrail contour;

[0009] Determine the coordinates of all pixel points on the guardrail contour and correct the data points in the guardrail point cloud set;

[0010] The guardrail deformation is identified based on the corrected guardrail point cloud set.

[0011] The second aspect of the present invention discloses a guardrail deformation monitoring device based on feature-level semantic consistency, comprising a memory and a controller that are communicatively connected in sequence, wherein a computer program is stored in the memory, and the controller is used to read the computer program and execute the guardrail deformation monitoring method based on feature-level semantic consistency described in the first aspect.

[0012] The third aspect of the present invention discloses a guardrail deformation monitoring system based on feature-level semantic consistency, comprising:

[0013] At least one set of data collection terminals, the data collection terminals comprising a camera and a millimeter wave radar installed on the same smart pole on the road as the camera;

[0014] The guardrail deformation monitoring device based on feature-level semantic consistency described in the second aspect is signal-connected to the data acquisition terminal.

[0015] The beneficial effects of the present invention are:

[0016] The present invention uses the image data collected by the camera as a reference standard to guide the millimeter wave radar data to approach the image data, which not only solves the interference problem but also improves the accuracy of guardrail deformation monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 The figure is a flow chart of the guardrail deformation monitoring method of the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0020] 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 invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0022] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0023] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inside", "outside", etc. indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, or the positions or positional relationships in which the product of the invention is usually placed when in use, or the positions or positional relationships commonly understood by those skilled in the art, which are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0024] In the description of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "set", "install", "connect", and "connect" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0025] The first aspect of the present invention discloses a guardrail deformation monitoring method based on feature-level semantic consistency. The method can be, but is not limited to, performed by a guardrail deformation monitoring device. The guardrail deformation monitoring device can be software, or a combination of software and hardware. The guardrail deformation monitoring device can be integrated in smart devices such as smart mobile terminals, tablets, and computers. Specifically, Figure 1As shown, the guardrail deformation monitoring method based on feature-level semantic consistency includes the following steps S01 to S05. It should be noted that the step identifiers in this scheme are only for the convenience of explaining the method and do not constitute a limitation on the sequence. The sequence of each step is based on its language description and the sequence of each signal. In this scheme, feature-level semantic consistency refers to the use of guardrail contour pixel points collected by the camera to adjust and correct the guardrail point cloud collected by the millimeter-wave radar to achieve alignment.

[0026] Step S01, obtaining image data of the guardrail to be monitored collected by a camera and point cloud data of the guardrail to be monitored collected by a millimeter-wave radar within a time period, wherein the camera and the millimeter-wave radar are installed on a smart pole at the same detection point.

[0027] In this solution, the camera and millimeter-wave radar are installed on the smart pole at the same detection point, and the general installation distance between the two is less than 10 cm.

[0028] In order to improve the accuracy of guardrail deformation monitoring, the length of this time period should not be too long to avoid the influence of various natural and environmental factors on the accuracy of guardrail deformation monitoring. The length of the time period can be 10 to 50 times the millimeter wave radar data acquisition cycle.

[0029] After acquiring the point cloud data, it can be preprocessed, for example, by using dynamic data filtering and intensity information filtering methods to distinguish points on the guardrail and points on the road next to the guardrail.

[0030] Step S02: extracting data points representing the guardrail from the multi-frame point cloud data collected by the millimeter-wave radar within the time period and superimposing them to obtain a guardrail point cloud set.

[0031] The point cloud data includes multiple data points. Specifically, in this step, the data points with a speed greater than 0 in each frame of point cloud data are first deleted to obtain a multi-frame static point cloud set; the data points with a reflection intensity less than a first threshold in each frame of the static point cloud set are then deleted to obtain data points representing the guardrail; finally, the data points representing the guardrail are superimposed to obtain a guardrail point cloud set.

[0032] According to the speed data in the millimeter wave radar data, the data points of moving objects such as vehicles and pedestrians can be obtained. In static data, due to the difference in the material of the guardrail and the road surface, there will be differences in reflection intensity, and the road surface reflection intensity will be much lower than the guardrail reflection intensity. In this regard, the data points of the guardrail and other data points can be further distinguished and filtered out in the static point cloud set by the reflection intensity threshold. Here, the value of the first threshold can be set to any value from 19 to 20.5.

[0033] The guardrail point cloud is extracted only from the data collected by a single frame of millimeter-wave radar, and its point cloud is sparse, and the accuracy of guardrail deformation monitoring is low. This scheme uses multi-frame point cloud data for superposition to enhance the guardrail point cloud volume, so as to enhance the recognition accuracy of guardrail deformation. However, when using multi-frame point cloud data for superposition, there is a certain acquisition time difference between multi-frame point cloud data. The radar is installed on the pole. Due to strong winds or the passage of large vehicles, it will cause shaking. The slight shaking will be amplified after the calculation of the azimuth angle and pitch angle, which will cause the overall data to shift; and if the number of superimposed frames is too large, the waveform guardrail in the image will be too wide. In this regard, in order to further improve the accuracy of the superimposed guardrail point cloud, the number of frames of the superimposed point cloud data here is less than or equal to 20.

[0034] Step S03: extracting the guardrail curve features in the single-frame image data collected by the camera within the time period and performing guardrail contour detection to obtain the guardrail contour.

[0035] Specifically, this step includes step S031 to step S035.

[0036] Step S031, grayscale the image data to obtain a grayscale image.

[0037] Among them, the calculation method of grayscale processing is:

[0038] Gray(x,y)=0.299·R(x,y)+0.587·G(x,y)+0.114·B(x,y),

[0039] Where R(x,y), G(x,y), and B(x,y) represent the values ​​of the red, green, and blue channels at the coordinate (x,y) in the image data, respectively, and Gray(x,y) is the grayscale value at each coordinate after conversion.

[0040] Step S032: perform edge detection on the grayscale image to obtain an edged guardrail image.

[0041] Specifically, a Canny edge detector may be used to perform edge detection on the grayscale image to obtain an edged guardrail image Edge.

[0042] Canny edge detector is widely used in computer vision. It can detect changes in image shape density and find image boundaries.

[0043] Step S033: input the edged guardrail image, grayscale image, and image data collected by the camera into an adaptive threshold calculation module to calculate a binary threshold.

[0044] Specifically, the calculation method is:

[0045] F1=CNN([Edge,Gray,Picture]),

[0046] s=W2·sigmoid(W1·F1+b1)+b2,

[0047] Among them, F1 is the binarization threshold, Picture is the image data collected by the camera, that is, the original image, s is the adaptive threshold, W1 and W2 are weight matrices, b1 and b2 are bias items, Gray is the grayscale image, Edge is the edged guardrail image, CNN is the convolutional neural network operation, and Sigmoid is the sigmoid function operation.

[0048] Step S034: binarize the grayscale image according to the binarization threshold to obtain a binarized image.

[0049] Step S035: perform guardrail contour detection based on the binarized image and the image data collected by the camera to obtain the guardrail contour.

[0050] Specifically, the calculation method is:

[0051] C(x,y)=DeepLabV3(B,Picture),

[0052] Among them, C(x,y) is the guardrail contour point at each coordinate, C(x,y)∈{0,1}, 1 means there is a guardrail contour at (x,y), and 0 means there is no guardrail contour at (x,y); DeepLabV3() is the DeepLabV3 network operation; B is the binary image.

[0053] Step S04, determining the coordinates of all pixel points on the guardrail contour and correcting the data points in the guardrail point cloud set.

[0054] As mentioned above, strong winds or large vehicles passing by will cause shaking, and the tiny shaking will be amplified after the calculation of the azimuth and pitch angle, which will lead to the overall data offset. In this regard, this solution further uses the contour data of the image data to correct and adjust the radar point cloud data, further improving the accuracy of the guardrail point cloud.

[0055] Specifically, this step includes step S041 to step S045.

[0056] Step S041, determining the intrinsic parameter matrix and the extrinsic parameter matrix of the camera to determine the geographic coordinates of all pixel points on the guardrail contour.

[0057] Specifically, the geographic coordinates of the pixel points are calculated as follows:

[0058]

[0059] Where (u, v) is the horizontal and vertical coordinates of a point on the image data collected by the camera; K is the camera's intrinsic parameter matrix, [R t] is the camera's extrinsic parameter matrix, which consists of the rotation matrix R and the translation vector t, and (X, Y, Z) is the geographic coordinate point corresponding to the point (u, v) on the image data collected by the camera;

[0060]

[0061] Among them, f1 and f2 are the distances from the focus of the camera lens to the image plane in the X-axis and Y-axis directions of the image data collected by the camera respectively; c1 and c2 are the pixels of the optical center in the X-axis and Y-axis directions of the image data collected by the camera respectively.

[0062] Step S042: Match all pixel points on the guardrail contour with the data points in the guardrail point cloud set according to the nearest neighbor search algorithm to obtain a first matching point pair set.

[0063] Nearest neighbor search (NNS) is a form of proximity search that aims to find the point closest to a given point in a given set, such as a KD tree.

[0064] In this step, all pixel points on the guardrail contour are traversed in sequence to find the point with the smallest distance from it in the guardrail point cloud set, and use this as a pair of matching point pairs until matching data points are found for all pixel points to obtain the first matching point pair set D.

[0065] Step S043: Combine the guardrail point cloud set and the guardrail contour to delete the matching point pairs in the first matching point pair set whose distances between pixels and data points are greater than a second threshold, to obtain a second matching point pair set.

[0066] This step screens the first matching point pair set based on the distance data calculated in step S042. Specifically, the calculation method is as follows:

[0067] p=softmax(MLP(sigmoid(MLP(C,Radar)),D)),

[0068] Among them, p is the probability distribution of whether each pair of matching point pairs in the first matching point pair set D will be deleted, softmax() is the softmax function operation, MLP() is the multi-layer perceptron operation, sigmoid() is the sigmoid function, C is the guardrail contour image data, and Radar is the guardrail point cloud set.

[0069] Step S044: sort the matching point pairs in the second matching point pair set from small to large according to the ordinates of the geographic coordinates of the pixel points on the guardrail contour to obtain a sequential matching point pair set.

[0070] By sorting the geographic coordinates from small to large, the matching points can be guaranteed to be spatially ordered, facilitating subsequent feature extraction.

[0071] Step S045: setting a local sliding window to extract local features of the matching point pairs in the set of sequential matching point pairs.

[0072] Specifically, the local features are extracted as follows:

[0073] h = LSTM5(D(α,β)),

[0074] Among them, h is the local feature of the matching point pair, LSTM5() is an LSTM network operation with a length not greater than 5, and D(α,β) is a point in the second matching point pair set, where α is a pixel point on the guardrail contour and β is a data point in the guardrail point cloud set.

[0075] By setting a local sliding window and performing matching point feature extraction within the sliding window, and using the long short-term memory network (LSTM) for feature extraction, it is possible to capture the temporal or spatial dependencies between matching points, thereby extracting richer local features. This allows the subsequent transformer feature point alignment to use not only information at the data point level, but also feature-level information between data points.

[0076] Step S046: correct and adjust the data points in the guardrail point cloud set according to the local features and the sequential matching point pair set.

[0077] Specifically, the correction adjustment is feature point alignment, and the calculation method is:

[0078] E(γ)=Transformer(D(α,β),h),

[0079] Among them, E(γ) is the corrected guardrail point cloud set, γ is the data point in the adjusted guardrail point cloud set, and Transformer() is the Transformer block operation.

[0080] This step can ensure the accuracy of the data points in the guardrail point cloud set corresponding to the millimeter-wave radar and improve the quality of the radar data.

[0081] Step S05: identifying guardrail deformation based on the corrected guardrail point cloud set.

[0082] Specifically, this step first calculates the probability distribution p′ of the data points with deformation in the guardrail point cloud set, then screens out the data points whose probability distribution p′ is greater than or equal to the third threshold and solves their geographic coordinates; finally, the solved geographic coordinates are mapped to the coordinate system of the image data of the guardrail to be monitored collected by the camera, marked in the figure, and a feedback confirmation signal is generated.

[0083] Where p′=softmax(Transformer(E(γ))),

[0084] Where E(γ) is the corrected guardrail point cloud set, Transformer() is the Transformer block operation, and softmax() is the softmax function operation.

[0085] The third threshold is preferably set to be greater than or equal to 0.5, for example, can be set to 0.6.

[0086] The method for calculating geographic coordinates is detailed in step S041, and this step will not be described in detail.

[0087] After the feedback confirmation signal is generated, the monitoring personnel can make a second confirmation on whether the corrugated guardrail has been deformed based on the red marked position in the video camera screen.

[0088] Since the camera and millimeter-wave radar are installed on the smart pole at the same detection point, if the smart pole is offset, all the devices on the smart pole will be offset as a whole, that is, the camera and millimeter-wave radar will also be offset at this time, but the camera and millimeter-wave radar will remain relatively still to a certain extent. Since image data is usually more stable and can provide accurate geographic location information, the above solution uses the image data collected by the camera as a reference standard to guide the millimeter-wave radar data to approach the image data. It not only solves the problem of small single-frame millimeter-wave radar point cloud, but also solves the problem of multi-frame millimeter-wave radar being too scattered or broad after superposition visualization due to equipment jitter, environmental interference, etc.

[0089] The second aspect of the present invention discloses a guardrail deformation monitoring device based on feature-level semantic consistency, including a memory and a controller that are sequentially connected in communication, the memory stores a computer program, and the controller is used to read the computer program and execute the guardrail deformation monitoring method based on feature-level semantic consistency described in the first aspect. For example, the memory may include, but is not limited to, a random access memory (RAM), a read-only memory (ROM), a flash memory, a first-input first-output (FIFO), and / or a first-input last-output (FILO) memory, etc.; the controller may not be limited to a microcontroller of the STM32F105 series. In addition, the computer device may also include, but is not limited to, a power supply unit, a display screen, and other necessary components.

[0090] The third aspect of the present invention discloses a guardrail deformation monitoring system based on feature-level semantic consistency, comprising:

[0091] At least one set of data collection terminals, the data collection terminals comprising a camera and a millimeter wave radar installed on the same smart pole on the road as the camera;

[0092] The guardrail deformation monitoring device based on feature-level semantic consistency described in the second aspect is signal-connected to the data acquisition terminal.

[0093] The operating principles of the devices and systems disclosed in the second and third aspects of the present invention are detailed in the first aspect and will not be elaborated here.

[0094] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A guardrail deformation monitoring method based on feature-level semantic consistency, characterized in that: include: Obtain image data of the guardrail to be monitored collected by a camera and point cloud data of the guardrail to be monitored collected by a millimeter-wave radar within a time period, wherein the camera and the millimeter-wave radar are installed on a smart pole at the same detection point; extract data points representing the guardrail from multiple frames of point cloud data collected by the millimeter-wave radar within the time period, and then superimpose them to obtain a guardrail point cloud set; Extracting the guardrail curve features in the single-frame image data collected by the camera within the time period and performing guardrail contour detection to obtain the guardrail contour; Determine the coordinates of all pixel points on the guardrail contour and correct the data points in the guardrail point cloud set; The guardrail deformation is identified based on the corrected guardrail point cloud set.

2. The method for monitoring guardrail deformation based on feature-level semantic consistency according to claim 1 is characterized in that: The extracting of data points representing the guardrail from the multi-frame point cloud data collected by the millimeter wave radar within the time period and superimposing the data points to obtain the guardrail point cloud set comprises: Delete the data points with a speed greater than 0 in each frame of point cloud data to obtain a multi-frame static point cloud set; Deleting data points in each frame of the static point cloud set whose reflection intensity is less than a first threshold value, to obtain data points representing the guardrail; The data points representing the guardrail are superimposed to obtain a guardrail point cloud set.

3. The method for monitoring guardrail deformation based on feature-level semantic consistency according to claim 1 is characterized in that: The extracting of the guardrail curve features in the single-frame image data collected by the camera within the time period and performing guardrail contour detection to obtain the guardrail contour includes: Grayscale processing is performed on the single-frame image data collected by the camera during the time period to obtain a grayscale image; edge detection is performed on the grayscale image to obtain an edged guardrail image; The edged guardrail image, grayscale image, and image data collected by the camera are input into the adaptive threshold calculation module to calculate the binary threshold; Binarization processing is performed on the grayscale image according to the binarization threshold to obtain a binarized image; The guardrail contour is detected according to the binarized image and the image data collected by the camera to obtain the guardrail contour.

4. The method for monitoring guardrail deformation based on feature-level semantic consistency according to claim 3 is characterized in that: The edged guardrail image, grayscale image, and image data collected by the camera are input into the adaptive threshold calculation module to calculate the binarization threshold value: F1=CNN([Edge,Gray,Picture]), s=W2·sigmoid(W1·F1+b1)+b2, Among them, F1 is the binarization threshold, Picture is the image data collected by the camera, that is, the original image, s is the adaptive threshold, W1 and W2 are weight matrices, b1 and b2 are bias items, Gray is the grayscale image, Edge is the edged guardrail image, CNN is the convolutional neural network operation, and Sigmoid is the sigmoid function operation.

5. The method for monitoring guardrail deformation based on feature-level semantic consistency according to claim 3 is characterized in that: The guardrail contour detection is performed based on the binary image and the image data collected by the camera, and the guardrail contour is obtained as follows: C(x,y)=DeepLabV3(B,Picture), Among them, C(x,y) is the guardrail contour point at each coordinate, C(x,y)∈{0,1}, 1 means there is a guardrail contour at (x,y), and 0 means there is no guardrail contour at (x,y); DeepLabV3() is the DeepLabV3 network operation; B is the binary image.

6. The method for monitoring guardrail deformation based on feature-level semantic consistency according to claim 1, characterized in that: The step of determining the coordinates of all pixel points on the guardrail contour and correcting the data points in the guardrail point cloud includes: Determine the intrinsic parameter matrix and the extrinsic parameter matrix of the camera to determine the geographic coordinates of all pixel points on the guardrail contour; According to the nearest neighbor search algorithm, all pixel points on the guardrail contour are matched with the data points in the guardrail point cloud set to obtain a first matching point pair set; Combine the guardrail point cloud set and the guardrail contour to delete the matching point pairs in the first matching point pair set whose distances between the pixel points and the data points are greater than a second threshold, to obtain a second matching point pair set; Sorting the matching point pairs in the second matching point pair set according to the ordinates of the geographic coordinates of the pixel points on the guardrail contour from small to large to obtain a sequential matching point pair set; Set a local sliding window to extract local features of matching point pairs in the sequential matching point pair set; The data points in the guardrail point cloud set are corrected and adjusted according to local features and sequential matching point pairs.

7. The method for monitoring guardrail deformation based on feature-level semantic consistency according to claim 6, characterized in that: The identification of guardrail deformation based on the corrected guardrail point cloud set includes: Calculate the probability distribution p′ of the data points with deformation in the guardrail point cloud set, where p′=softmax(Transformer(E(γ))), Where E(γ) is the corrected guardrail point cloud set, Transformer() is the Transformer block operation, and softmax() is the softmax function operation; Screen out data points whose probability distribution p′ is greater than or equal to a third threshold and calculate their geographic coordinates; The solved geographic coordinates are mapped to the coordinate system of the image data of the guardrail to be monitored collected by the camera, marked in the image, and a feedback confirmation signal is generated.

8. A guardrail deformation monitoring device based on feature-level semantic consistency, comprising a memory and a controller that are sequentially connected in communication, wherein a computer program is stored in the memory, characterized in that: The controller is used to read the computer program and execute the guardrail deformation monitoring method based on feature-level semantic consistency described in any one of claims 1-7.

9. A guardrail deformation monitoring system based on feature-level semantic consistency, characterized in that: include: At least one set of data collection terminals, the data collection terminals comprising a camera and a millimeter wave radar installed on the same smart pole on the road as the camera; The guardrail deformation monitoring device based on feature-level semantic consistency described in claim 8 is signal-connected to the data acquisition terminal.

Images