Side slope deformation monitoring identification method based on 4D imaging millimeter wave radar

Abstract

The invention relates to the technical field of intelligent traffic and road safety monitoring, and discloses a 4D imaging millimeter wave radar-based slope deformation monitoring and identification method, which comprises the following steps of: acquiring a continuous point cloud; carrying out voxel modeling on the point cloud in a three-dimensional space; selecting three continuous frames of point clouds, and identifying mutant voxels by calculating the point number variation of each voxel in adjacent frames; performing spatial clustering on the mutant voxels to obtain an obstacle region; performing feature extraction on the obstacle area, wherein the extracted features comprise the height difference, the average reflection intensity and the centroid drift of the obstacle area; carrying out linear weighted fusion on the extracted features, and constructing a landslide mutation index; and judging whether an obstacle target exists in an obstacle area or not according to a preset condition, and performing risk grading identification according to the landslide sudden change index. According to the invention, all-weather, structured and high-reliability perception of major traffic potential safety hazards such as landslide / rockfall is realized, and a new basic capability is provided for a traffic safety system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent transportation and road safety monitoring, and particularly relates to a slope deformation monitoring and identification method based on 4D imaging millimeter wave radar. BACKGROUND

[0002] The existing road disaster monitoring methods mainly include manual inspection, video monitoring, laser radar detection, etc. However, these methods have certain limitations: manual inspection has poor real-time performance and is obviously affected by environmental conditions; video monitoring relies on light and sight distance, and its detection capability significantly decreases at night or in bad weather conditions; although laser radar and remote sensing technology have high accuracy, they have high equipment cost and complex deployment, and it is difficult to realize continuous monitoring in a dynamic environment. In recent years, 4D imaging millimeter wave radar has gradually been applied to traffic target detection due to its advantages of all-weather, anti-interference and high frequency output. However, the existing systems mainly focus on vehicle identification, and the three-dimensional contour modeling and risk assessment capabilities for natural obstacles still have deficiencies, mainly in the following aspects.

[0003] 1. Insufficient obstacle area identification capability: the existing radar processing methods mainly focus on single-point target tracking, lack spatial continuity analysis of regional obstacles, and it is difficult to timely find newly added rockfalls or slowly deformed landslide areas.

[0004] 2. Lack of perfect risk judgment mechanism: the existing methods mainly rely on a single index for judgment, and do not comprehensively consider time and physical characteristics such as obstacle appearance speed, range expansion trend, reflection intensity and stability, so it is difficult to quantify the risk level.

[0005] In summary, the main bottlenecks in risk identification and assessment of landslide / rockfall disasters are: insufficient obstacle area identification capability; lack of perfect risk judgment mechanism; these deficiencies result in that the monitoring system cannot stably operate in all-weather and complex environments, and affect the accuracy and reliability of risk identification and assessment of landslide / rockfall disasters. SUMMARY

[0006] In order to solve the above technical problems, the present application provides a slope deformation monitoring and identification method based on 4D imaging millimeter wave radar, which solves the problem that the existing traffic monitoring system cannot accurately identify sudden natural obstacles on the road slope in bad weather, at night or in low visibility scenes. The present application can realize real-time identification of structural static natural obstacles appearing in the road slope area under different meteorological conditions, and the specific steps include: S1. acquiring continuous point clouds; S2. modeling voxels in the three-dimensional space for the point clouds; S3. selecting three continuous frames of point clouds, and identifying mutant voxels by calculating the point number change of each voxel in adjacent frames; S4. performing spatial clustering on the mutant voxels to obtain an obstacle region; S5. performing feature extraction on the obstacle region, the extracted features including height difference, average reflection intensity and centroid drift of the obstacle region; S6. performing linear weighted fusion on the extracted features to construct a landslide mutation index; S7. determining whether the obstacle region contains an obstacle target according to a preset condition, and performing risk classification identification according to the landslide mutation index.

[0007] Preferably, the modeling step of the voxels comprises: dividing a three-dimensional space where the point cloud is located into a plurality of voxels of equal size; determining the voxel index based on the spatial position of different discrete points in the point cloud, and assigning the discrete points to the voxel identified by the voxel index.

[0008] Preferably, the step of identifying the mutant voxels by calculating the point number change amount of each voxel in adjacent frames comprises: if the maximum value of the point number change amount of the continuous three frames of point clouds is greater than a preset point number change amount, the corresponding voxel is a mutant voxel.

[0009] Preferably, the spatial clustering method comprises: clustering the mutant voxels in six directions of up and down, front and back, and left and right to obtain at least one obstacle region.

[0010] Preferably, the method for constructing the landslide mutation index comprises: ; wherein, is the maximum value of the point number change amount of the mutant voxels in the obstacle region, is the maximum value of the height difference change amount in the obstacle region, is the maximum value of the centroid drift change amount in the obstacle region, , , respectively are preset feature weights.

[0011] Preferably, the preset condition comprises simultaneously satisfying the following conditions: the landslide mutation index is greater than or equal to a preset landslide mutation index threshold value; the average reflection intensity is within a preset average reflection intensity range; the maximum value of the centroid drift change amount is less than or equal to a preset centroid drift threshold value.

[0012] Compared with the prior art, the present application has the following beneficial effects: The application realizes high-precision positioning of sudden static natural obstacles such as landslides and falling rocks by recognizing new mutant voxels through continuous multi-frame point cloud differential analysis and forming a complete obstacle region by combining spatial clustering; further, multi-dimensional features such as height difference, average reflectivity and centroid drift are extracted, the landslide mutation index is constructed, the obstacle region is quantitatively risk graded, and automatic early warning and linkage response are supported; the method overcomes the false positives and difficulty in quantification of existing traffic monitoring systems in bad weather, night or low visibility scenes, and improves the accuracy, real-time performance and environmental adaptability of road slope disaster identification. BRIEF DESCRIPTION OF DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, a brief introduction will be given below to the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings described below are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0014] Figure 1 A flowchart of a slope deformation monitoring and identification method based on 4D imaging millimeter wave radar is provided in a preferred embodiment of the present application. DETAILED DESCRIPTION

[0015] The technical solutions in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0016] Embodiment one At present, the road monitoring method based on 4D imaging millimeter wave radar mainly focuses on vehicle detection and moving target tracking. For sudden landslides, falling rocks and other natural obstacles on road slopes, traditional methods mostly rely on cameras or laser radars for identification, but such methods are limited by light, weather conditions and are difficult to maintain stable identification effect in night, rain, fog and other harsh environments. In addition, the risk determination of the existing method for the obstacle region mostly relies on a single feature such as height or reflectivity, which is difficult to effectively distinguish between road new obstacles and point cloud changes, resulting in false positives and false negatives.

[0017] As shown in Figure 1 The present embodiment provides a slope deformation monitoring and identification method based on 4D imaging millimeter wave radar, comprising the steps of: S1. Obtain continuous point cloud.

[0018] In some preferred embodiments, the point cloud refers to a large number of discrete three-dimensional space points generated after scanning a target area by a device such as a 4D imaging millimeter wave radar.

[0019] In some preferred embodiments, a certain highway slope is selected, and the rock-soil body is mainly composed of argillaceous sandstone, which is prone to shallow sliding in the rainy season. A 4D imaging millimeter wave radar is used to monitor the target slope area, and point cloud data is collected at a sampling frequency of 10 s / frame. Each frame of point cloud contains the three-dimensional coordinates (x, y, z) of each discrete point in the slope area, x is along the direction of the highway, y is perpendicular to the direction of the highway, and z is the vertical protrusion degree of the obstacle. The generated set of all discrete points is the point cloud.

[0020] S2. Model the point cloud in three-dimensional space by voxels.

[0021] In some preferred embodiments, a voxel is a basic discrete unit in three-dimensional space, which can be compared to the extension of a pixel in a two-dimensional plane in three dimensions. Its essence is a regular small cube with a fixed size in three-dimensional space. Each frame of point cloud is divided into a regular three-dimensional voxel grid, and the number of points in each voxel is counted.

[0022] S3. Select three consecutive frames of point cloud, and identify the mutant voxel by calculating the point number change of each voxel in adjacent frames.

[0023] In some preferred embodiments, a mutant voxel refers to a voxel in which the number of point clouds or the point number changes significantly and abnormally between consecutive frames of point cloud (adjacent two frames). Three consecutive frames of point cloud are selected, and the point number change in the three frames is compared. If the maximum point number change is greater than the preset point number change, the corresponding voxel is a mutant voxel.

[0024] S4. Spatially cluster the mutant voxel to obtain an obstacle region.

[0025] In some preferred embodiments, the method of spatially clustering the mutant voxel includes: judging the voxel connectivity based on the six-neighborhood, or reasonably designing according to the actual situation or the needs of the site, clustering the mutant voxel in the six directions of up, down, front, back, left and right, and obtaining at least one obstacle region.

[0026] S5. Extract features from the obstacle region, including the height difference, average reflection intensity, and centroid drift of the obstacle region.

[0027] In some preferred embodiments, the height difference refers to the height difference between the highest point and the lowest point in the obstacle region in the z-axis direction, reflecting the vertical protrusion degree of the obstacle; the average reflection intensity refers to the average radar reflection intensity value of all point clouds in the obstacle region, used to distinguish metal targets from natural objects such as rocks and mudslides; and the centroid drift represents the centroid position change of the target in two consecutive frames, reflecting whether it is a stationary body.

[0028] In some preferred embodiments, the volume cluster The height difference, the average reflection intensity, and the centroid drift features are extracted, and the volume cluster represents a set of spatially connected mutant voxels. Each volume cluster corresponds to an obstacle region. The calculation formulas of the features are as follows: Height difference : Let p be a point in the volume cluster , and denote the height of point p in the z-axis direction, then ; Average reflection intensity R m : Let reflect(p) denote the reflection intensity of point p, and denote the number of mutant voxels, then ; Centroid drift : Let and be the centroid positions of the obstacle region in the t-th frame and the t-1-th frame, respectively, then ; Through the above feature extraction, the system converts the obstacle region with mutant voxels into a candidate target with structural description, providing basic data support for the next landslide index calculation and risk assessment.

[0029] S6. Linearly weighting and fusing the extracted features to construct a landslide mutation index.

[0030] In some preferred embodiments, the landslide mutation index is a comprehensive index obtained by linearly weighting and fusing the features such as the height difference, the average reflection intensity, and the centroid drift of the obstacle region, used to represent the risk level of the landslide or rockfall.

[0031] S7. Determining whether the obstacle region contains an obstacle target according to a preset condition, and identifying the risk level according to the landslide mutation index.

[0032] When the obstacle region meets the preset determination condition, such as the voxel number threshold, the reflection intensity threshold, and the centroid drift threshold, it is determined that the region contains an obstacle target; at the same time, according to the size of the landslide mutation index, the risk level is divided into three levels: low, medium, and high.

[0033] The embodiment can detect and identify new natural obstacles such as slope landslides and rockfalls in non-illumination conditions, rainy and foggy weather. The use of the mutation voxel combined with feature fusion to construct the landslide mutation index reduces the misjudgment caused by a single feature and improves the accuracy and stability of detection. Experimental results show that the method can maintain a high recognition rate in different environments and can effectively classify obstacle risks to facilitate road safety management departments to take timely measures.

[0034] Embodiment two On the basis of embodiment one, the embodiment is used for modeling voxels.

[0035] Voxel modeling is a key step to connect original point cloud data and deformation analysis. Common voxel modeling methods include regular voxel grid modeling, adaptive voxel modeling and octree voxel modeling. These methods have high computational cost, complex process, or over-segmentation or under-segmentation problems in areas with large differences in point cloud density due to improper parameter settings. In order to solve the above problems, in some preferred embodiments, the three-dimensional space where the point cloud is located is divided into voxels of equal size; the voxel index of different discrete points in the point cloud is determined based on the spatial position of the discrete points, and the discrete points are distributed to the voxels identified by the voxel index. The voxel index (i, j, k) calculation method is: ; Wherein, x, y, z are the spatial coordinates of the discrete points, Vx, Vy, Vz represent the size of the voxel in x, y, z directions respectively.

[0036] For example, in some preferred embodiments, the highway slope area is divided into voxels with size Vx x Vy x Vz, the number of point clouds in each voxel is counted, and the voxel modeling is performed. For example, if the voxel size is 0.5m x 0.5m x 0.5m, voxel V(10, 5, 3) contains 8 point clouds, and voxel V(10, 6, 3) contains 0 point clouds.

[0037] In the embodiment, the original point cloud is converted into a voxel structure, and the number of points corresponding to each voxel is obtained. Compared with the commonly used technology, the method of the embodiment is more accurate and reliable in multi-frame point cloud data comparison.

[0038] Embodiment three On the basis of embodiment one, the embodiment is used for identifying mutation voxels.

[0039] The existing point cloud mutation detection method usually judges the obstacle area by comparing the point difference frame by frame, but the simple two-frame comparison is easily disturbed by instantaneous noise, which leads to the random noise points being mistaken for mutation targets, thereby reducing the identification accuracy. In order to solve the above problems, in some preferred embodiments, the mutation voxel is identified by comparing the point number change in three frames, and if the maximum point number change is greater than the preset point number change, the corresponding voxel is a mutation voxel.

[0040] In some preferred embodiments, the specific steps of the mutation voxel identification method of the present application include: using the continuous three frames of point cloud of the sliding window: t-2, t-1, t frame, counting the point number of the point cloud in each voxel, and identifying the mutation voxel of the shallow sliding layer by calculating the point number change of each voxel in the adjacent frames. Firstly, the point number change ΔN1 of the same voxel in t-1 and t-2 frames is calculated. Then, the point number change ΔN2 of the same voxel in t and t-1 frames is calculated.

[0041] For example, the voxel V(10, 5, 3) contains 8 point clouds in the t-2 frame, and contains 2 point clouds in the t-1 frame, so ΔN1 = 8-2 = 6. The same voxel V(10, 5, 3) contains 2 point clouds in the t-1 frame, and contains 1 point cloud in the t frame, so ΔN2 = 2-1 = 1. The larger value is taken as the point number change max(ΔN1, ΔN2) of the voxel, and the point number change of the voxel is 6. The threshold is set to 2: the point number change of the voxel is 6> The threshold is set to 2, so the voxel V(10, 5, 3) is determined as a mutation voxel.

[0042] Embodiment four On the basis of embodiment one, this embodiment is used for spatial clustering of the mutation voxel.

[0043] The commonly used clustering methods include the application space clustering based on the Euclidean distance or the three-dimensional clustering based on connectivity. However, these methods have large calculation amount and are sensitive to parameters, and are prone to cause low efficiency in large-scale point cloud real-time monitoring. In some preferred embodiments, the mutation voxel is clustered in six directions of up, down, front, back, left and right, and the spatial clustering process includes: obtaining a series of mutation voxel sets by the method of embodiment three; for each mutation voxel, checking whether there is other mutation voxel in the six directions (up, down, front, back, left and right). If there is, they are considered to belong to the same connected region; recursively traversing all adjacent mutation voxels to aggregate them into an obstacle region; if there is no direct adjacent relationship between the mutation voxels, multiple independent obstacle regions are formed, and each region corresponds to a potential landslide or rockfall target.

[0044] The embodiment can accurately identify different obstacle regions while ensuring computing efficiency.

[0045] Embodiment five On the basis of embodiment one, the embodiment constructs a landslide mutation index.

[0046] Existing road slope monitoring relies on a single feature for judgment, which is prone to missed detection or misjudgment due to environmental interference, and it is difficult to accurately assess landslide or rockfall risk. For example, the height difference is large, but it is actually a vehicle parking point, or the reflection intensity is high, but it is only a metal object, not a real landslide risk point. To solve this problem, in some preferred embodiments, the method for constructing a landslide mutation index provided by the present application comprises: ; Among them, is the maximum value of the point number change of the mutant voxel in the obstacle region, is the maximum value of the height difference change in the obstacle region, is the maximum value of the centroid drift change in the obstacle region, , , respectively, are preset feature weights.

[0047] In some preferred embodiments, , , The preset feature weights can be 1.0, 1.5, and 2.0, respectively, or can be reasonably designed by those skilled in the art according to actual conditions or field needs. For example, when =11, =1.1m, =0.015m, the landslide mutation index is: =1.0x11+1.5x1.1-2.0x0.015=12.62; By linearly weighting and fusing multiple features, the landslide mutation index proposed in the embodiment can comprehensively reflect the characteristics of the obstacle region and effectively distinguish between real landslides / rockfalls and ordinary environmental changes.

[0048] Embodiment six On the basis of embodiment one, the embodiment provides a method for determining obstacle targets and classifying risks.

[0049] Traditional slope monitoring methods often rely on a single indicator to determine the presence of obstacles, which is easily affected by environmental interference or abnormal points, leading to false positives or false negatives. At the same time, there is a lack of comprehensive judgment basis, and it is difficult to effectively classify obstacle risks. To solve the above problems, in some preferred embodiments, the present application provides a classification strategy, comprising the following steps: First, for each obstacle region, it is judged whether the following conditions are met at the same time: The landslide mutation index is greater than or equal to a preset landslide mutation index threshold ; The average reflection intensity is within a preset average reflection intensity range The maximum value of the change amount of the centroid drift is less than or equal to a preset centroid drift threshold ; In some preferred embodiments, the preset landslide mutation index threshold can be set to = 10; the preset average reflection intensity range can be set to ∈ (-15, -8) dB, which can effectively exclude non-natural objects such as metal vehicles; and the preset centroid drift threshold can be set to = 0.1, which is used to distinguish dynamic traffic flow from static obstacles. It can also be a reasonable design according to the field situation or actual needs of those skilled in the art.

[0050] When the above three conditions are met at the same time, it is determined that the obstacle region contains an obstacle (landslide or rockfall) target.

[0051] Then, according to the size of the landslide mutation index , the determined obstacle region is divided into risk levels, for example: 5≤ <10, low risk 10≤ <20, medium risk 20≤ , high risk In this embodiment, the higher the value, the closer the target is to a sudden large amount of static natural obstacles. The point cloud mutation, height difference, and centroid drift in the index are respectively derived from the data processing modules at the voxel level, cluster body level, and time sequence level, which use the index as a fusion scoring function for landslide events, allowing the system to comprehensively consider from multiple dimensions and improve the overall accuracy of target judgment.

[0052] The above shows and describes the basic principles, main features, and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above embodiments, and the above embodiments and descriptions in the specification are only to illustrate the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A slope deformation monitoring and identification method based on 4D imaging millimeter wave radar, characterized in that: include: S1. Obtain continuous point cloud; S2. voxel modeling of the point cloud in three-dimensional space; S3. Select three consecutive frames of point cloud and identify mutant voxels by calculating the change in the number of points of each voxel in adjacent frames; S4. performing spatial clustering on the mutant voxels to obtain obstacle regions; S5. Extract features of the obstacle area, where the extracted features include the height difference, average reflection intensity, and centroid drift of the obstacle area; S6. performing linear weighted fusion on the extracted features to construct a landslide mutation index; S7. Determine whether there is an obstacle target in the obstacle area according to preset conditions, and perform risk classification identification according to the landslide mutation index.

2. The slope deformation monitoring and identification method based on 4D imaging millimeter wave radar according to claim 1 is characterized in that: The voxel modeling step includes: dividing the three-dimensional space where the point cloud is located into a number of voxels of equal size; determining the voxel index based on the spatial position of different discrete points in the point cloud, and allocating the discrete points to the voxels identified by the voxel index.

3. The slope deformation monitoring and identification method based on 4D imaging millimeter wave radar according to claim 1 is characterized in that: The step of identifying mutant voxels by calculating the point number change of each voxel in adjacent frames includes: if the maximum point number change of the three consecutive frames of point cloud is greater than a preset point number change, the corresponding voxel is a mutant voxel.

4. The slope deformation monitoring and identification method based on 4D imaging millimeter wave radar according to claim 1 is characterized in that: The spatial clustering method includes clustering mutant voxels in six directions: up and down, front and back, and left and right, to obtain at least one obstacle area.

5. The slope deformation monitoring and identification method based on 4D imaging millimeter wave radar according to claim 1 is characterized in that: The method for constructing the landslide mutation index includes: ； in, is the maximum value of the number of mutant voxels in the obstacle area, is the maximum value of the height difference change in the obstacle area, is the maximum value of the center of mass drift change in the obstacle area, , , are the preset feature weights respectively.

6. The slope deformation monitoring and identification method based on 4D imaging millimeter wave radar according to claim 1 is characterized in that: The preset conditions include simultaneously meeting the following conditions: The landslide mutation index is greater than or equal to a preset landslide mutation index threshold; The average reflection intensity is within a preset average reflection intensity range; The maximum value of the change in the center of mass drift is less than or equal to a preset center of mass drift threshold.

Images