Intelligent alarm method for vehicle intervention after disasters

Abstract

The present invention discloses an intelligent alarm method for post-disaster vehicle intervention, comprising: collecting visible light images and infrared thermal imaging images, and simultaneously scanning three-dimensional spatial coordinate data; inputting the visible light images into a convolutional neural network model to identify the type of road obstacles, and generating three-dimensional contour data of the obstacles in combination with the three-dimensional spatial coordinate data; when the height of the obstacle exceeds a height threshold and the distance from the vehicle is less than a distance threshold, uploading a feature dataset containing GPS coordinates, three-dimensional contour data, and obstacle type to a cloud server via a cellular network; based on the three-dimensional spatial coordinate data, the cloud server matches the feature dataset with a pre-stored digital elevation model of the disaster area, and calculates the ratio of the obstacle's projected area on the current lane cross-section to the lane width; when the ratio exceeds the ratio threshold, an alarm command is generated. The present invention can effectively reduce the risk of accidents caused by misjudgment of obstacles or improper path planning after a disaster.

Description

Technical Field

[0001] The present invention relates to the technical field related to the Internet of Things, and more specifically, to an intelligent alarm method for vehicle intervention after a disaster. Background Art

[0002] After a disaster, road conditions often become complex and dangerous, posing numerous challenges to driving safety. Traditional traffic safety warning systems have limitations when responding to post-disaster scenarios. Existing on-board obstacle detection systems primarily rely on single sensors, such as cameras or millimeter-wave radar. However, single sensors have significant shortcomings in disaster environments. For example, cameras are significantly affected by light, smoke, dust, and other factors, significantly reducing their detection capabilities at night or in inclement weather. While millimeter-wave radar can provide three-dimensional spatial coordinate data, its ability to identify obstacle types is limited, making it unable to accurately distinguish between obstacles of varying materials and characteristics. These limitations of single sensors result in insufficient reliability and accuracy in complex post-disaster environments. Furthermore, after a disaster, road geological conditions may change, such as due to landslides and ground subsidence. Traditional lane models are fixed and cannot adapt to this dynamic geological environment. When vehicles travel on roads with geological hazards, fixed lane models cannot provide accurate risk assessments, potentially leading to inappropriate avoidance path selection.

[0003] Therefore, it is necessary to design a technical solution that can overcome the above-mentioned defects. Summary of the Invention

[0004] One purpose of the present invention is to provide an intelligent alarm method for vehicle intervention after a disaster, which can effectively reduce the risk of accidents caused by misjudgment of obstacles or improper path planning after a disaster.

[0005] In order to achieve these objects and other advantages of the present invention, according to one aspect of the present invention, the present invention provides an intelligent alarm method for post-disaster driving intervention, including: S1: collecting visible light images and infrared thermal imaging images within a range of 0-150 meters in front of the vehicle through a vehicle-mounted multispectral camera, and scanning three-dimensional spatial coordinate data within a range of 0-150 meters in front through a vehicle-mounted millimeter-wave radar; S2: inputting the visible light image into a convolutional neural network model to identify the type of road obstacles, and performing temperature gradient analysis on the infrared thermal imaging image, and generating three-dimensional contour data of the obstacle in combination with the three-dimensional spatial coordinate data of the millimeter-wave radar; S3: when the height of the obstacle exceeds a height threshold and is consistent with the height threshold, the vehicle is automatically detected. When the distance of the vehicle is less than the distance threshold, the vehicle positioning module is called to obtain the current GPS coordinates, and the feature data set containing the GPS coordinates, the three-dimensional contour data and the obstacle type is uploaded to the cloud server through the cellular network; S4: based on the three-dimensional spatial coordinate data, the cloud server spatially matches the feature data set with the pre-stored digital elevation model of the disaster area, and calculates the ratio of the projected area of ​​the obstacle on the current lane cross-section to the lane width; S5: when the ratio exceeds the ratio threshold, an alarm instruction containing the three-dimensional coordinates of the obstacle, the obstacle type and the avoidance path parameters is generated, and the alarm instruction is sent to the vehicle through the cellular network.

[0006] Furthermore, the S2 also includes: S21: based on the temperature value of each pixel point in the infrared thermal imaging image, dividing multiple temperature intervals and marking them as different color channels; S22: identifying continuous temperature anomaly areas, if the temperature difference between a certain area and the surrounding area exceeds a dynamic threshold and lasts for more than 3 seconds, it is determined to be a thermal radiation disaster residue or a low-temperature freezing area; S23: spatially superimposing the obstacle boundary box identified by the convolutional neural network in the visible light image with the temperature anomaly area, activating the multimodal verification mechanism when the overlap exceeds 80%, and reconstructing the surface deformation characteristics of the obstacle through millimeter-wave radar point cloud data; S24: correcting the geometric distortion caused by optical occlusion in the three-dimensional contour data based on the correspondence between the surface deformation characteristics and the temperature distribution.

[0007] Furthermore, the spatial feature point set of the obstacle is located through millimeter-wave radar point cloud data. If the distance between the temperature anomaly area and the center of gravity of the feature point set in the infrared thermal imaging is less than 0.5 meters, it is marked as a credible thermal radiation area. Within the credible thermal radiation area, the boundary where the visible light image texture is continuous but the radar point cloud density drops sharply is identified as the optical occlusion boundary. According to the relationship between the direction of the temperature gradient and the change in the radar point cloud density, an interpolation algorithm is selected to reconstruct the three-dimensional contour within the occlusion boundary: when the high temperature extends in the direction of the point cloud sparseness, the concave repair algorithm is used to fill the contour; when the high temperature extends in the direction of the point cloud denseness, the convex repair algorithm is used to correct the contour.

[0008] Furthermore, the S3 also includes: S31: before calling the vehicle positioning module, the material properties of the obstacle are judged according to the reflection intensity data of the millimeter wave radar, and the priority of metal obstacles is set to the highest; S32: when the cellular network bandwidth is lower than the preset value, the data compression engine is started: differential coding compression is used for the three-dimensional contour data of continuous frames, and the original data of the highest temperature area of ​​the infrared image is retained while the background area is compressed; S33: the upload target server cluster is dynamically selected according to the disaster type database, and the landslide data is sent to the geological disaster monitoring server, and the debris flow data is sent to the hydrological monitoring server.

[0009] Furthermore, the S4 specifically includes: S41: extracting historical deformation parameters of the road ahead of the vehicle from the digital elevation model of the disaster area, and calculating the geological subsidence trend coefficient of the current obstacle projection position; S42: establishing a virtual lane extension model, superimposing the influence of the geological subsidence trend coefficient on the lane width baseline value, and generating a dynamic lane width reference value; S43: performing stress deformation simulation on the three-dimensional contour of the obstacle along the driving direction, and predicting the maximum expansion value of the projection area within the next 10 seconds; S44: using the real-time ratio of the dynamic lane width reference value to the maximum expansion value as the spatial matching index.

[0010] Furthermore, the S41 specifically includes: S411: extracting the hourly surface elevation change data of the road ahead of the vehicle in the last 72 hours from the digital elevation model of the disaster area, and constructing a time series deformation map; S412: identifying periodic uplift or subsidence areas in the deformation map, and if the deformation rate of the same coordinate point exceeds the threshold for three consecutive hours and the change direction is consistent, it is marked as an active deformation point; S413: establishing a dynamic monitoring grid with a radius of 5 meters with the projection position of the obstacle as the center, and counting the proportion of active deformation points and the average deformation acceleration in the grid; S414: dynamically calculating the geological subsidence trend coefficient based on the weighted value of the proportion of active deformation points and the average deformation acceleration, wherein: when the proportion of active deformation points exceeds 40% and the acceleration is positive, the coefficient value is increased to 1.2-1.5 times the baseline value; when the proportion of active deformation points is less than 10% and the acceleration is negative, the coefficient value is reduced to 0.6-0.8 times the baseline value.

[0011] Furthermore, the S43 specifically includes: S431: identifying the surface texture features of the obstacle through the on-board multispectral camera, and combining the corrected three-dimensional contour data to classify it as a rigid material or a flexible material; S432: applying a virtual wind resistance load in the driving direction, and the wind resistance load value is dynamically calculated based on the ratio of the real-time speed of the vehicle to the obstacle height; S433: if it is a rigid material, monitoring the infrared temperature change rate of the edge of the obstacle base, activating the thermal expansion compensation model when the rate exceeds 2°C / second, and adding the axial extension margin in the projected area calculation; S434: if it is a flexible material, retrieving the current precipitation data and simulating the softening effect of rainwater penetration, and calculating the envelope range of deformation and expansion in the next 10 seconds based on the material porosity parameters; S435: taking the maximum superposition value of the thermal expansion compensation margin and the softening effect envelope range as the maximum expansion value.

[0012] Furthermore, the S5 includes: S51: retrieving the avoidance trajectory data of all vehicles on the current road section in the past 24 hours from the cloud server, and constructing a probability density distribution heat map; S52: screening candidate paths that meet the vehicle kinematic constraints in the heat map based on the vehicle's wheelbase parameters and the lateral extension length of the obstacle; S53: giving priority to the path with the highest overlap with the geological stability zone in the digital elevation model of the disaster area among the candidate paths; S54: decomposing the selected path into a longitudinal deceleration curve and a lateral offset instruction sequence, wherein the lateral offset instruction includes a steering angle safety margin value calculated in real time based on tire grip.

[0013] Furthermore, the S53 includes: S531: extracting rock fracture surface strike data from the digital elevation model of the disaster area, marking the fracture zone with an angle less than 30 degrees with the driving direction as a potential unstable area; S532: obtaining the ground vibration monitoring data of the current road section in the past 2 hours. If a candidate path passes through a grid unit with a vibration amplitude exceeding 0.3g and a duration exceeding 10 seconds, the geological stability coefficient of the path is downgraded; S533: calculating the surface bearing capacity margin of each candidate path based on the vehicle's full-load mass parameters, and eliminating paths with margin values ​​lower than the safety threshold; S534: sorting the remaining paths by the weighted fracture avoidance rate and the surface bearing capacity margin, and selecting the path with the highest comprehensive score.

[0014] Furthermore, the S54 includes: S541: real-time acquisition of tire pressure monitoring values, road infrared temperature values, and road roughness index detected by millimeter-wave radar; S542: when the tire pressure is 15% lower than the standard value and the road temperature is higher than 50°C, activating the high-temperature soft tire compensation mode to increase the basic safety margin by 20%; S543: if the road roughness index drops by more than 40% for five consecutive frames, it is determined to be a slippery road surface, and a sinusoidal damping factor is injected into the lateral offset command to suppress steering shake; S544: dynamically adjusting the safety margin based on the vehicle's real-time yaw rate feedback: when the yaw rate exceeds 0.5 rad / s, an additional 5% margin is added for every 0.1 rad / s increase; when the yaw rate is lower than 0.2 rad / s for three seconds, the baseline margin value is restored.

[0015] The present invention has at least the following beneficial effects:

[0016] The present invention improves the accuracy and reliability of post-disaster driving safety intervention through multi-sensor fusion and intelligent processing. Visible light and infrared thermal imaging images are collected by on-board multispectral cameras, and combined with millimeter-wave radar three-dimensional coordinate data, multi-dimensional perception of obstacles within 150 meters ahead is achieved, solving the problem of insufficient illumination and material recognition capabilities of a single sensor, and providing comprehensive data support for subsequent analysis. Obstacle types are identified through convolutional neural networks, and three-dimensional contours are generated by combining infrared thermal imaging temperature gradient analysis with radar data to achieve joint modeling of geometric features and physical properties, accurately distinguish disaster residues, avoid geometric distortion caused by optical occlusion, and improve the accuracy of obstacle feature extraction. The cloud-based processing module dynamically calculates the ratio of the obstacle projection area to the lane width based on the digital elevation model of the disaster area, breaking through the limitations of traditional fixed threshold assessments, adapting to complex scenarios such as road deformation and geological subsidence after disasters, and ensuring that risk judgments fit actual road conditions. When the risk exceeds the threshold, the system generates an alarm instruction containing the obstacle coordinates, type and avoidance path parameters, providing clear intervention guidance for driving, effectively reducing the risk of accidents caused by misjudgment of obstacles or improper path planning, and building an efficient and safe closed loop of "perception-analysis-decision-making".

[0017] Other advantages, objectives and features of the present invention will be reflected in part through the following description, and in part will be understood by those skilled in the art through study and practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a flowchart of an embodiment of the present application. DETAILED DESCRIPTION

[0019] The present invention will be described in further detail below in conjunction with the accompanying drawings so that those skilled in the art can implement the invention with reference to the description.

[0020] It should be understood that terms such as "having," "comprising," and "including" used in the embodiments of this application do not exclude the presence or addition of one or more other elements or combinations thereof. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of this application are intended only to explain the relative positional relationships and movement of components in a specific posture. If the specific posture changes, the directional indications will also change accordingly. When an element is referred to as being "fixed to" or "disposed on" another element, it can be directly on the other element or there may be an intervening element. When an element is referred to as being "connected to" another element, it can be directly connected to the other element or indirectly connected to the other element through an intervening element. References to "first," "second," etc. in the embodiments of this application are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features designated as "first" or "second" may explicitly or implicitly include at least one of such features.

[0021] It should be noted that the technical solutions between the various embodiments of the present application can be combined with each other, but it must be based on the fact that ordinary technicians in this field can implement it. When the combination of technical solutions is mutually contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by this application.

[0022] like Figure 1 As shown, an embodiment of the present application provides an intelligent alarm method for post-disaster driving intervention, including: collecting visible light images and infrared thermal imaging images within a range of 0-150 meters in front of the vehicle through a vehicle-mounted multispectral camera, and scanning three-dimensional spatial coordinate data within a range of 0-150 meters in front through a vehicle-mounted millimeter-wave radar; inputting the visible light image into a convolutional neural network model to identify the type of road obstacles, and performing temperature gradient analysis on the infrared thermal imaging image, and generating three-dimensional contour data of the obstacle in combination with the three-dimensional spatial coordinate data of the millimeter-wave radar; when the height of the obstacle exceeds the height threshold and the distance from the vehicle is less than the distance threshold, calling the vehicle positioning module to obtain the current GPS coordinates, and uploading a feature data set containing GPS coordinates, three-dimensional contour data and obstacle type to a cloud server through a cellular network; based on the three-dimensional spatial coordinate data, the cloud server spatially matches the feature data set with a pre-stored digital elevation model of the disaster area, and calculates the ratio of the projected area of ​​the obstacle on the current lane cross section to the lane width; when the ratio exceeds the ratio threshold, generating an alarm instruction containing the three-dimensional coordinates of the obstacle, the obstacle type and avoidance path parameters, and sending the alarm instruction to the vehicle through the cellular network.

[0023] For example, the FLIR Boson 320 multispectral camera can be used as an on-board multispectral camera. It features dual visible and infrared channels, a visible light resolution of 1920×1080, and an infrared thermal imaging resolution of 320×256. It is installed in the center above the vehicle's front windshield, with its optical axis aligned with the vehicle's centerline. The Bosch MRR evo2 mid-range radar can be used as a millimeter-wave radar, with a scanning range of 0-160 meters and an angular resolution of ±3.5°. It is installed in the center of the inner side of the front bumper, 0.8 meters above the ground. The convolutional neural network model uses the ResNet-50 architecture, with input images resized to 224×224. The training dataset contains 50,000 images of 20 types of obstacles in disaster scenarios, including fallen rocks, trees, accumulated water, and debris, after data augmentation. The model is fine-tuned for disaster scenarios based on ImageNet pre-training. Temperature gradient analysis of infrared thermal images calculates the absolute value of the temperature difference between each pixel and its eight neighboring pixels to generate a gradient matrix. Temperature ranges are divided into 0-40°C (10°C intervals), 40-80°C (20°C intervals), and above 80°C (30°C intervals), mapped to blue, yellow, and red channels, respectively. To generate 3D contour data, the Zhang Zhengyou calibration method is first used to obtain camera intrinsic parameters. Using timestamp synchronization data from the millimeter-wave radar, a hand-eye calibration is performed to establish a transformation between the radar coordinate system and the camera coordinate system. The radar point cloud is projected onto the image plane. Obstacle coordinates are extracted from the point cloud using semantic segmentation results from the visible light image, and a 3D surface mesh is constructed using the Delaunay triangulation algorithm. The digital elevation model of the disaster area is obtained through drone aerial surveys or geological surveys and stored in DEM format with a spatial resolution of 1 meter. It contains latitude and longitude, elevation values, and road centerline vector data. The coordinate system uses the WGS84 geographic coordinate system. The height threshold is set at 1.2 meters, the distance threshold at 50 meters, and the ratio threshold at 0.3. The cellular network module uses the Quectel M95 4G ​​module, which supports the TCP / IP protocol. Data is encapsulated in JSON format before upload, including timestamps, GPS coordinates, 3D contour vertex coordinates, and obstacle type tags. During the spatial matching process, the server extracts the road centerline within 150 meters of the vehicle's current position from the digital elevation model. A 10-meter section of road is cut along the road's direction, centered on the obstacle's projected location, to generate a cross-sectional plane perpendicular to the road centerline. For each vertex coordinate of the obstacle's 3D contour, its 2D projection point on the cross section is calculated using a plane projection transformation matrix, forming a closed polygonal contour. The projected area is calculated using Green's formula, integrating the polygon vertex coordinate sequence to obtain the obstacle's projected area on the cross section. Lane width data is derived from the road attribute table in the digital elevation model, with the baseline value being the national standard lane width of 3.5 meters.

[0024] In this embodiment, a multispectral camera and millimeter-wave radar synchronously collect data. After time alignment and spatial calibration, the visible light image is fed into a convolutional neural network to identify obstacle types. The infrared image uses temperature gradient analysis to assist in determining material properties, and the radar point cloud data constructs a three-dimensional profile. When the obstacle height and distance meet threshold conditions, the system packages the feature data and uploads it to the cloud. The cloud uses a digital elevation model to convert the three-dimensional coordinates into a projection of the road cross section and calculate the occupancy ratio. This method achieves efficient detection and risk assessment of post-disaster obstacles through multimodal data fusion and precise spatial matching, providing real-time, reliable intervention for driving safety.

[0025] In another embodiment, the processing of the infrared thermal imaging image in the above method further includes: dividing a plurality of temperature intervals based on the temperature value of each pixel in the infrared thermal imaging image and marking them as different color channels; identifying continuous temperature anomaly areas, and if the temperature difference between a certain area and the surrounding area exceeds a dynamic threshold and lasts for more than 3 seconds, it is determined to be a thermal radiation disaster residue or a low-temperature icing area; spatially superimposing the obstacle boundary box identified by the convolutional neural network in the visible light image with the temperature anomaly area, activating a multimodal verification mechanism when the overlap exceeds 80%, and reconstructing the surface deformation characteristics of the obstacle through millimeter-wave radar point cloud data; and correcting the geometric distortion caused by optical occlusion in the three-dimensional contour data based on the correspondence between the surface deformation characteristics and the temperature distribution.

[0026] For example, the infrared thermal imaging image is divided into temperature intervals using an equidistant segmentation method: 0-20°C (blue channel), 20-40°C (green channel), 40-60°C (yellow channel), and above 60°C (red channel). The temperature range corresponding to each channel is displayed in pseudo-color on the image. A dynamic threshold is calculated as the standard deviation between the temperature of the central pixel in the current region and the temperature of the surrounding 5×5 neighboring pixels. An abnormality is detected when the absolute value of the temperature difference exceeds 1.5 times the standard deviation. Continuous detection requires 15 consecutive frames (at a frame rate of 5 fps, corresponding to 3 seconds) to meet this condition. Temperature anomaly areas are identified using a connected component analysis algorithm, extracting the minimum enclosing rectangle as the region boundary. The obstacle bounding box of the visible light image is output by a convolutional neural network in the format (x1, y1, x2, y2). When spatially overlaying, the infrared and visible light images are first registered using SIFT feature matching. The intersection-over-union (IoU) ratio is calculated, and multimodal verification is triggered when the IoU ratio is ≥80%. When reconstructing surface deformation features from millimeter-wave radar point cloud data, voxel grid filtering is first used to reduce noise. The normal vector and curvature value of each point are then calculated, and edge points with sudden changes in curvature are extracted as deformation feature points. Geometric distortion correction utilizes a weighted moving least squares method. For missing points in occluded areas, the coordinates of five adjacent valid points are weighted averaged, guided by the temperature gradient. This process is iteratively optimized until the point cloud density in the reconstructed area reaches at least 80% of that of the surrounding area. The FLIR T400 infrared camera can be installed alongside a visible light camera (Sony IMX490) on the front of the vehicle, 10 cm apart. Exposure is synchronized via hardware triggering.

[0027] In this embodiment, the system divides infrared images into temperature ranges and uses dynamic thresholds to detect persistent temperature anomalies. Combined with spatial overlap of visible light obstacle bounding boxes, the system then uses radar point cloud data to finely reconstruct surface deformation. This process effectively identifies residual thermal radiation or iced areas. Through cross-validation of multimodal data, it corrects for outline distortion caused by optical occlusion, improves the integrity and accuracy of the obstacle's 3D model, and provides more reliable geometric and physical feature information for subsequent risk assessment.

[0028] In another embodiment, the spatial feature point set of the obstacle is located using millimeter-wave radar point cloud data. If the distance between the temperature anomaly area and the center of gravity of the feature point set in the infrared thermal imaging is less than 0.5 meters, it is marked as a credible thermal radiation area. Within the credible thermal radiation area, the boundary where the visible light image texture is continuous but the radar point cloud density drops sharply is identified as the optical occlusion boundary. Based on the relationship between the temperature gradient direction and the change in radar point cloud density, an interpolation algorithm is selected to reconstruct the three-dimensional contour within the occlusion boundary: when the high temperature extends in the direction of the point cloud sparseness, the concave repair algorithm is used to fill the contour; when the high temperature extends in the direction of the point cloud denseness, the convex repair algorithm is used to correct the contour.

[0029] For example, the Harris 3D corner detection algorithm is used to extract feature points from millimeter-wave radar point clouds, selecting the top 20% of points as feature points. The centroid is calculated as the average of the feature point coordinates. The centroid of an infrared temperature anomaly region is obtained by taking the weighted average of all pixel coordinates within the region (weighted by the temperature value). Three-dimensional distance is calculated by calculating the square root of the sum of the squared differences of each coordinate component. Regions with a distance less than 0.5 meters are marked as trustworthy. Texture continuity in visible light images is detected by calculating the variance of the gradient magnitude of pixels within a bounding box. Texture continuity is determined when the variance is less than a threshold (e.g., 15). A sudden drop in radar point cloud density is defined as a decrease of more than 60% in the number of points per unit volume (compared to the adjacent area) to determine occlusion boundaries. The dent repair algorithm uses surface reconstruction based on the Poisson equation, fitting the surface of the missing region with occlusion boundaries as constraints. The convex repair algorithm uses triangular mesh subdivision, adding vertices at the occlusion boundary and expanding the mesh vertices outward in the direction of the temperature gradient, with an expansion distance of 0.1 times the gradient magnitude. The time interval of millimeter-wave radar point cloud data is 20ms, which is synchronized with the camera frame rate through timestamp interpolation to ensure the consistency of spatial position.

[0030] In this example, the system first identifies credible regions based on the spatial correlation between feature point sets and areas of temperature anomaly. Within these credible regions, the system then locates occlusion boundaries by combining visible light texture and radar point cloud density variations. The system then reconstructs the contours using an appropriate interpolation algorithm based on the direction of the temperature gradient. This approach effectively addresses the issue of missing 3D models caused by optical occlusion under complex lighting conditions. By integrating temperature and geometric features, the system improves the accuracy of reconstructing obstacle surface details, providing critical data support for accurately assessing the extent of lane occupancy.

[0031] In another embodiment, the processing before calling the vehicle positioning module in the above method also includes: judging the material properties of the obstacle based on the reflection intensity data of the millimeter wave radar, and setting the priority of metal obstacles to the highest; when the cellular network bandwidth is lower than the preset value, starting the data compression engine: using differential coding compression for the three-dimensional contour data of continuous frames, retaining the original data of the highest temperature area of ​​the infrared image and compressing the background area; dynamically selecting the upload target server cluster according to the disaster type database, and sending landslide data to the geological disaster monitoring server, and sending mudslide data to the hydrological monitoring server.

[0032] For example, millimeter-wave radar reflection intensity data can be used to determine the material quality using preset thresholds. The reflection intensity threshold for metal materials is set to above -10dBm, while that for non-metal materials (such as trees and soil) is set to below -10dBm. The Bosch MRRevo2 radar model is used, with a reflection intensity measurement accuracy of ±3dB. The data compression engine uses differential coding compression, storing only the difference in vertex coordinates between two consecutive frames of 3D contour data, achieving a compression ratio of up to 5:1. Infrared image compression utilizes ROI (region of interest) technology, automatically identifying the top 20% of the hottest regions as the original data. The background area is compressed using the JPEG algorithm with a quality factor of 30. The default cellular network bandwidth is 100kbps, and bandwidth status is determined by real-time monitoring of signal strength and bit error rate. The disaster type database includes 10 common disaster categories (such as landslides, mudslides, and earthquake debris), with each disaster type corresponding to a dedicated server cluster address. The vehicle positioning module uses the u-blox NEO-M8N GPS module, with a positioning accuracy of 2.5 meters, and is mounted on the vehicle's roof antenna.

[0033] In this embodiment, the system first distinguishes obstacle materials based on radar reflection intensity, prioritizing high-risk metal obstacles (such as collapsed steel frames and scattered vehicle parts) to ensure real-time response to emergency data. When network bandwidth is insufficient, differential encoding and ROI compression technologies are used to reduce data transmission volume, ensuring the effective upload of critical information (such as peak temperature areas and 3D contour changes). Dynamic server selection, combined with a disaster type database, enables targeted processing of specialized data and improves cloud-based analysis efficiency. This process optimizes data processing priorities and transmission strategies, ensuring reliable communication in the complex network environment following a disaster, and providing timely and effective data support for subsequent risk assessments.

[0034] In another embodiment, the spatial matching of the cloud server in the above method specifically includes: extracting the historical deformation parameters of the road ahead of the vehicle from the digital elevation model of the disaster area, and calculating the geological subsidence trend coefficient of the current obstacle projection position; establishing a virtual lane expansion model, superimposing the influence of the geological subsidence trend coefficient on the lane width baseline value, and generating a dynamic lane width reference value; performing stress deformation simulation on the three-dimensional contour of the obstacle along the driving direction, and predicting the maximum expansion value of the projected area within the next 10 seconds; and using the real-time ratio of the dynamic lane width reference value to the maximum expansion value as the spatial matching indicator.

[0035] For example, the historical deformation parameters of the digital elevation model of the disaster area are extracted from surface elevation change data over the past 72 hours, with a temporal resolution of 1 hour. The data source is drone aerial surveys or ground subsidence monitoring stations. To calculate the geological subsidence trend coefficient, a dynamic monitoring grid with a radius of 5 meters is established, centered on the projected location of the obstacle. The deformation rate (unit: mm / h) of each coordinate point within the grid is calculated. If a point's deformation rate exceeds 10 mm / h for three consecutive hours and is in the same direction (uplift or subsidence), it is marked as an active deformation point. The trend coefficient formula is: coefficient = baseline value (1.0) + (percentage of active deformation points × 0.5 + average deformation acceleration × 0.3). When the proportion of active deformation points exceeds 40% and the acceleration is positive, the coefficient is increased to 1.2-1.5 times; when the proportion is less than 10% and the acceleration is negative, the coefficient is decreased to 0.6-0.8 times. In the virtual lane expansion model, the baseline lane width is the national standard value of 3.5 meters. The dynamic width after applying the trend coefficient is 3.5 times the coefficient. The stress-deformation simulation uses different models for obstacle materials (rigid / flexible): rigid materials take thermal expansion effects into account, with the axial extension margin increasing by 0.05 meters for every 10°C increase in temperature; flexible materials take rain softening into account, with the expansion envelope expanding by 0.1 meters for every 10mm increase in precipitation. The simulation time step is set to 0.1 seconds, with a total of 100 steps (the next 10 seconds) calculated.

[0036] In this embodiment, a cloud server analyzes historical elevation data to identify geologically active areas and dynamically adjusts lane widths to accommodate subsidence or uplift trends. It also predicts the deformation and expansion range based on the material properties of obstacles. This approach combines road geological conditions with the dynamic changes of obstacles, making spatial matching indicators more closely aligned with actual post-disaster road conditions. This avoids the limitations of traditional fixed lane models, provides a more accurate basis for lane occupancy assessments for risk assessment, and enhances the scientific nature and reliability of alarm decisions.

[0037] In another embodiment, the step of extracting historical deformation parameters in the above method specifically includes: extracting the surface elevation change data of the road ahead of the vehicle every hour in the past 72 hours from the digital elevation model of the disaster area, and constructing a time series deformation map; identifying periodic uplift or subsidence areas in the deformation map, and if the deformation rate of the same coordinate point exceeds the threshold for three consecutive hours and the change direction is consistent, it is marked as an active deformation point; establishing a dynamic monitoring grid with a radius of 5 meters with the obstacle projection position as the center, and counting the proportion of active deformation points and the average deformation acceleration in the grid; dynamically calculating the geological subsidence trend coefficient based on the weighted value of the proportion of active deformation points and the average deformation acceleration.

[0038] For example, the time series graph of surface elevation change data uses UTC time as the horizontal axis and elevation change (unit: mm) as the vertical axis, with a resolution of 1 meter x 1 meter grid. The deformation rate threshold is set to 10 mm / h, and the rate is calculated based on the elevation difference between two consecutive hours. Consistent direction refers to positive change (uplift) or negative change (subsidence) for three consecutive hours. The dynamic monitoring grid uses a square grid with a side length of 1 meter and a total of 25 grid points (within a radius of 5 meters). The proportion of active deformation points is the ratio of the number of marked points in the grid to the total number of points. The average deformation acceleration is obtained by fitting a quadratic polynomial to the three-hour deformation rate data, and calculating the derivative to obtain the acceleration value (unit: mm / h²). In the weighted calculation of the geological subsidence trend coefficient, the weight of the proportion is 0.6, the weight of the acceleration is 0.4, and the baseline value is 1.0. The specific formula is: coefficient = 1.0 + 0.6 × (proportion - 0.2) + 0.4 × (acceleration / 50). When the proportion exceeds 0.4, the upward adjustment mechanism is triggered, and when it is less than 0.1, the downward adjustment mechanism is triggered, ensuring that the coefficient is dynamically adjusted within the range of 0.6-1.5. The digital elevation model data is stored in a cloud database in GeoTIFF format, including UTM projection coordinates and elevation values, and is updated once an hour.

[0039] In this embodiment, the system constructs long-term deformation maps to accurately identify geologically active areas and utilizes dynamic grid statistical analysis to improve the accuracy of subsidence trend assessment. This process fully considers the spatiotemporal changes in road conditions following disasters, avoiding the incompleteness of single-point data. This allows the geological subsidence trend coefficient to reflect the stability of the current road section in real time, providing reliable basic parameters for the virtual lane expansion model. This in turn optimizes the calculation accuracy of obstacle projection areas and enhances the dynamic adaptability of risk assessment.

[0040] In another embodiment, the step of predicting the maximum expansion value of the projected area in the above method specifically includes: identifying the surface texture features of the obstacle through a vehicle-mounted multispectral camera, and classifying it as a rigid material or a flexible material in combination with the corrected three-dimensional contour data; applying a virtual wind resistance load in the driving direction, and the wind resistance load value is dynamically calculated based on the ratio of the real-time speed of the vehicle and the obstacle height; if it is a rigid material, monitoring the infrared temperature change rate of the base edge of the obstacle, activating the thermal expansion compensation model when the rate exceeds 2°C / second, and adding an axial extension margin to the projected area calculation; if it is a flexible material, retrieving current precipitation data and simulating the softening effect of rainwater penetration, and calculating the envelope range of deformation and expansion in the next 10 seconds based on the material porosity parameters; taking the maximum superposition value of the thermal expansion compensation margin and the softening effect envelope range as the maximum expansion value.

[0041] For example, obstacle material classification is achieved through texture analysis of visible light images. Rigid materials (such as rock and metal) have sharp texture edges and high gradient amplitudes, while flexible materials (such as soil and vegetation) have blurred textures and low gradient amplitudes. A support vector machine (SVM) classification model is used, and the training sample consists of 5,000 images of disaster scene materials. The virtual wind resistance load is calculated using the formula: load = 0.5 × air density × vehicle speed² × obstacle frontal area × drag coefficient, where the air density is 1.2 kg / m³, the drag coefficient is set to 0.8 for rigid materials, and 1.2 for flexible materials. Infrared temperature change rate monitoring measures the average temperature change in a 3×3 pixel area at the edge of the substrate, with a one-second interval. If the temperature exceeds 2°C / second three times in a row, the compensation model is activated, with an extension margin of 0.05 meters per 10°C temperature rise. Precipitation data is acquired via an onboard meteorological sensor with an accuracy of 0.1mm. The softening effect calculation assumes that for every 10% increase in porosity, the expansion coefficient increases by 0.08. The envelope is calculated as the expansion coefficient multiplied by the time (in seconds) for each vertex coordinate of the current contour along the normal. The FLIR Boson 320 multispectral camera, mounted above the windshield, is synchronized with the millimeter-wave radar (Bosch MRRevo2) to within 10ms.

[0042] In this embodiment, the system uses material classification and multi-physics coupling analysis to separately consider the effects of thermal expansion and rain softening on obstacle deformation, and dynamically adjusts the expansion prediction model based on wind load. This process refines the modeling based on the physical properties of different materials, avoiding the prediction bias of a unified model and ensuring that the maximum expansion value more closely matches the actual deformation trend. This in turn improves the accuracy of projected area calculations, providing critical data support for the subsequent precise assessment of lane occupancy ratios and the rational planning of avoidance paths.

[0043] In another embodiment, the step of generating an alarm instruction in the above method includes: retrieving the avoidance trajectory data of all vehicles on the current road section in the past 24 hours from a cloud server, and constructing a probability density distribution heat map; screening candidate paths that meet the vehicle kinematic constraints in the heat map based on the vehicle's wheelbase parameters and the lateral extension length of the obstacle; giving priority to the path with the highest overlap with the geological stability zone in the digital elevation model of the disaster area among the candidate paths; decomposing the selected path into a longitudinal deceleration curve and a lateral offset instruction sequence, wherein the lateral offset instruction includes a steering angle safety margin value calculated in real time based on tire grip.

[0044] For example, avoidance trajectory data is stored in a cloud database in the format of (timestamp, longitude, latitude, steering wheel angle, vehicle speed). Heatmap construction uses kernel density estimation (KDE) with a bandwidth parameter set to 5 meters, generating a probability distribution grid with a resolution of 0.5 meters. Darker colors indicate higher pass probability. Vehicle kinematic constraints are: the minimum turning radius of the candidate path is ≥ the vehicle's wheelbase × 1.5 (the wheelbase parameter is obtained through the OBD interface, with a typical value of 2.7 meters), and the lateral offset is ≤ 3.5 meters (lane width). Geological stability zone data is obtained from a digital elevation model, marking areas where the angle between the rock fracture strike and the driving direction is ≥30 degrees and the historical 72-hour deformation rate is less than 5 mm / h. The overlap is calculated as the ratio of the overlap length between the path and the stability zone to the total path length. Paths with an overlap greater than 60% are preferred. The longitudinal deceleration curve is fitted using a quadratic function, with the initial speed being the current vehicle speed and the target speed being 0. The braking distance meets the requirements of the national standard GB21670-2008. The steering angle safety margin for the lateral offset command is calculated based on the tire grip formula: margin = 1-(lateral acceleration / 0.8g), where the lateral acceleration is measured in real time by the vehicle's inertial sensor, and 0.8g is the tire's maximum grip threshold.

[0045] In this embodiment, the system analyzes high-frequency safe paths using historical trajectory heat maps, combines geological stability zone screening with vehicle kinematic constraints, and generates an avoidance path that aligns with actual operating conditions. This process leverages collective driving experience and geological stability data, avoiding the limitations of a single algorithm. Furthermore, through real-time grip calculations, it dynamically adjusts steering margins to ensure safe path planning that aligns with vehicle handling characteristics. This provides reliable intervention instructions to the driver or autonomous driving system, effectively reducing driving risks in complex post-disaster road conditions.

[0046] In another embodiment, the step of screening candidate paths in the above method specifically includes: extracting rock fracture surface strike data from the digital elevation model of the disaster area, marking the fracture zone with an angle less than 30 degrees with the driving direction as a potential unstable area; obtaining ground vibration monitoring data of the current road section in the past 2 hours, and if a candidate path passes through a grid unit with a vibration amplitude exceeding 0.3g and a duration exceeding 10 seconds, the geological stability coefficient of the path is downgraded; calculating the surface bearing capacity margin of each candidate path based on the vehicle's fully loaded mass parameters, and eliminating paths with margin values ​​lower than the safety threshold; weighted sorting of the remaining paths according to the fracture zone avoidance rate and the surface bearing capacity margin, and selecting the path with the highest comprehensive score.

[0047] For example, rock fracture strike data is obtained from a geological exploration database and stored as a vector file containing the fracture location and strike angle (angle with true north). Areas with an angle less than 30 degrees with the driving direction are marked as high-risk zones. Ground vibration monitoring data is collected by accelerometers deployed on the road at a frequency of 100Hz. Areas with vibration amplitudes exceeding 0.3g and lasting for more than 10 seconds are defined as active vibration zones and marked on a digital elevation model using a 10m x 10m grid. The surface bearing capacity margin is calculated as: margin = (pavement design bearing capacity - vehicle fully loaded mass / contact area) / pavement design bearing capacity. The fully loaded vehicle mass is obtained using an onboard load cell. The contact area is assumed to be the sum of the contact areas of four tires (typically 0.12 square meters). The safety threshold is set at 0.2. The fault zone avoidance rate is the ratio of the length of a path avoiding potential instability areas to the total path length. In the weighted ranking, the avoidance rate is weighted 0.6, and the bearing capacity margin is weighted 0.4. The overall score is 0.6 × avoidance rate + 0.4 × margin. The top three paths with the highest scores are selected for further optimization. The digital elevation model is integrated with the seismic data and fault zone data using a unified coordinate system (WGS84) to ensure spatial matching accuracy.

[0048] In this embodiment, the system integrates multi-source geological data to assess route safety based on three dimensions: fault zone orientation, ground vibration, and surface bearing capacity, to avoid entering potentially unstable areas. This process quantifies geological risk factors and optimizes routes through weighted ranking. This ensures that avoidance routes not only avoid immediate obstacles but also mitigate long-term geological disaster risks. This improves the comprehensiveness and reliability of post-disaster route planning and provides safer driving options for vehicles.

[0049] In another embodiment, the step of decomposing the path instruction in the above method specifically includes: obtaining tire pressure monitoring values, road infrared temperature values, and road roughness index detected by millimeter-wave radar in real time; when the tire pressure is 15% lower than the standard value and the road temperature is higher than 50°C, activating the high-temperature soft tire compensation mode to increase the basic safety margin by 20%; if the road roughness index drops by more than 40% for five consecutive frames, it is determined to be a slippery road surface, and a sinusoidal damping factor is injected into the lateral offset instruction to suppress steering jitter; the safety margin is dynamically adjusted according to the vehicle's real-time yaw rate feedback: when the yaw rate exceeds 0.5 rad / s, an additional 5% margin is added for every 0.1 rad / s increase; when the yaw rate is lower than 0.2 rad / s for three seconds, the baseline margin value is restored.

[0050] For example, tire pressure monitoring is obtained via the TPMS (Tire Pressure Monitoring System), with a standard value set according to the vehicle manual (typically 2.5 bar). The tire pressure sensor has an accuracy of ±0.1 bar. Road surface infrared temperature is collected by an onboard infrared camera (FLIRT400), with a resolution of 320×240 pixels and a temperature accuracy of ±2°C. The road surface roughness index is calculated based on the signal-to-noise ratio of the millimeter-wave radar echo signal using the formula: roughness = 10×log10 (signal power / noise power). The index ranges from 0 to 100, with higher values ​​indicating greater roughness. The high-temperature soft tire compensation mode is triggered when tire pressure is <2.125 bar and temperature is >50°C. The base safety margin is 10%, which increases to 12%. A slippery road condition is determined when the roughness index decreases by >40% over five consecutive frames (0.1 second per frame). The sine damping factor is set to 2 Hz with an amplitude of 0.1 rad to suppress high-frequency steering system jitter. The yaw rate is measured by the inertial measurement unit (IMU) with an accuracy of ±0.05 rad / s. The dynamic adjustment rule is: when the angular velocity is greater than 0.5 rad / s, the margin is calculated as follows: base value + 5% × (angular velocity - 0.5) / 0.1, with a maximum increase of 30%. If the angular velocity is less than 0.2 rad / s for 3 seconds, the margin returns to 10% of the base value.

[0051] In this embodiment, the system monitors tire condition, road surface temperature, and road roughness in real time, dynamically adjusting the steering safety margin for complex driving conditions such as high temperatures and slippery conditions. Steering judder is suppressed using a damping factor, and adaptive adjustment of the margin is achieved through the integration of yaw rate feedback. This process fully considers the interaction between the vehicle and the road surface, avoiding the limitations of fixed parameter control. This allows lateral offset commands to be more closely aligned with real-time driving conditions, improving vehicle handling stability and path tracking accuracy on complex post-disaster roads, and effectively reducing the risk of sideslip or loss of control due to changing road conditions.

[0052] Although the embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the description and implementation methods. They can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. Post-disaster driving intervention intelligent alarm method, characterized by: include: S1: The vehicle-mounted multispectral camera collects visible light images and infrared thermal imaging images within a range of 0-150 meters in front of the vehicle, and the vehicle-mounted millimeter-wave radar scans the three-dimensional spatial coordinate data within a range of 0-150 meters in front of the vehicle; S2: Inputting the visible light image into a convolutional neural network model to identify the type of road obstacles, and performing temperature gradient analysis on the infrared thermal imaging image, and combining the three-dimensional spatial coordinate data to generate three-dimensional contour data of the obstacle; S3: When the height of the obstacle exceeds the height threshold and the distance from the vehicle is less than the distance threshold, calling the vehicle positioning module to obtain the current GPS coordinates, and uploading the feature data set including the GPS coordinates, the three-dimensional profile data, and the obstacle type to the cloud server via the cellular network; S4: The cloud server spatially matches the feature dataset with a pre-stored digital elevation model of the disaster area based on the three-dimensional spatial coordinate data, and calculates a ratio of a projected area of ​​the obstacle on the current lane cross section to the lane width; S5: When the ratio exceeds a ratio threshold, generating an alarm instruction including the three-dimensional coordinates of the obstacle, the obstacle type, and avoidance path parameters, and sending the alarm instruction to the vehicle via a cellular network; The S4 specifically includes: S41: Extract historical deformation parameters of the road ahead of the vehicle from the digital elevation model of the disaster area and calculate the geological subsidence trend coefficient at the current obstacle projection position; S42: establishing a virtual lane extension model, superimposing the influence of the geological subsidence trend coefficient on the lane width reference value, and generating a dynamic lane width reference value; S43: Perform stress and deformation simulation on the three-dimensional outline of the obstacle along the driving direction to predict the maximum expansion value of the projected area in the next 10 seconds; S44: Using the real-time ratio of the dynamic lane width reference value to the maximum expansion value as a spatial matching index; The S41 specifically includes: S411: Extract hourly surface elevation change data of the road ahead of the vehicle over the past 72 hours from the digital elevation model of the disaster area and construct a time series deformation map; S412: Identify periodic uplift or subsidence areas in the deformation map. If the deformation rate of the same coordinate point exceeds a threshold for three consecutive hours and the change direction is consistent, mark it as an active deformation point. S413: Establish a dynamic monitoring grid with a radius of 5 meters, centered on the obstacle projection position, and calculate the proportion of active deformation points and the average deformation acceleration within the grid. S414: Dynamically calculate the geological subsidence trend coefficient based on the weighted value of the active deformation point ratio and the average deformation acceleration, wherein: when the active deformation point ratio exceeds 40% and the acceleration is positive, the coefficient value is increased to 1.2-1.5 times the baseline value; when the active deformation point ratio is less than 10% and the acceleration is negative, the coefficient value is decreased to 0.6-0.8 times the baseline value.

2. The intelligent alarm method for vehicle intervention after a disaster as claimed in claim 1, characterized in that: Said S2 further comprises: S21: Based on the temperature value of each pixel in the infrared thermal imaging image, multiple temperature intervals are divided and marked as different color channels; S22: Identify continuous temperature anomaly areas. If the temperature difference between a certain area and the surrounding area exceeds the dynamic threshold and lasts for more than 3 seconds, it is determined to be a thermal radiation disaster residue or a low-temperature icing area. S23: spatially superimposing the obstacle bounding box identified by the convolutional neural network in the visible light image with the temperature anomaly area, activating a multimodal verification mechanism when the overlap exceeds 80%, and reconstructing the surface deformation characteristics of the obstacle using the millimeter-wave radar point cloud data; S24: Correcting geometric distortion caused by optical occlusion in the three-dimensional contour data according to the corresponding relationship between the surface deformation characteristics and the temperature distribution.

3. The intelligent alarm method for vehicle intervention after a disaster as claimed in claim 2, characterized in that: The spatial feature point set of the obstacle is located using millimeter-wave radar point cloud data. If the temperature anomaly area in the infrared thermal image is less than 0.5 meters away from the center of gravity of the feature point set, it is marked as a credible thermal radiation area. In the credible thermal radiation region, identifying a boundary where the visible light image texture is continuous but the radar point cloud density drops sharply as an optical occlusion boundary; According to the relationship between the temperature gradient direction and the density change of the radar point cloud, an interpolation algorithm is selected to reconstruct the three-dimensional contour within the occlusion boundary: when the high temperature extends towards the sparse point cloud direction, the concave repair algorithm is used to fill the contour; when the high temperature extends towards the dense point cloud direction, the convex repair algorithm is used to correct the contour.

4. The intelligent alarm method for vehicle intervention after a disaster as claimed in claim 1, characterized in that: Said S3 further comprises: S31: Before calling the vehicle positioning module, the obstacle material properties are determined based on the reflection intensity data of the millimeter-wave radar, and the priority of metal obstacles is set to the highest; S32: When the cellular network bandwidth is lower than a preset value, the data compression engine is started: differential coding is used to compress the three-dimensional contour data of consecutive frames, and the original data of the highest temperature area of ​​the infrared image is retained while the background area is compressed; S33: Dynamically select the upload target server cluster according to the disaster type database, and send the landslide data to the geological disaster monitoring server, and the debris flow data to the hydrological monitoring server.

5. The intelligent alarm method for post-disaster vehicle intervention according to claim 1, characterized in that: The S43 specifically includes: S431: Identify the surface texture features of the obstacle through the on-board multispectral camera, and classify it into rigid material or flexible material in combination with the corrected three-dimensional contour data according to claim 2; S432: Applying a virtual wind resistance load in the driving direction, wherein the wind resistance load value is dynamically calculated based on the ratio of the vehicle's real-time speed to the obstacle height; S433: If the material is rigid, monitor the infrared temperature change rate at the edge of the obstacle base. When the rate exceeds 2°C / second, activate the thermal expansion compensation model and add an axial expansion margin to the projected area calculation. S434: If the material is flexible, retrieve the current precipitation data and simulate the softening effect of rainwater penetration. Calculate the envelope of deformation and expansion within the next 10 seconds based on the material porosity parameter. S435: Taking the maximum superposition value of the thermal expansion compensation margin and the softening effect envelope as the maximum expansion value.

6. The intelligent alarm method for vehicle intervention after a disaster as claimed in claim 1, characterized in that: The S5 includes: S51: Retrieve the avoidance trajectory data of all vehicles on the current road section in the past 24 hours from the cloud server and construct a probability density distribution heat map; S52: Based on the vehicle's wheelbase parameter and the lateral extension length of the obstacle, select candidate paths that meet the vehicle's kinematic constraints in the thermal map; S53: Among the candidate paths, the path with the highest overlap with the geological stability zone in the digital elevation model of the disaster area is preferentially selected; S54: Decomposing the selected path into a longitudinal deceleration curve and a lateral offset instruction sequence, wherein the lateral offset instruction includes a steering angle safety margin value calculated in real time based on tire grip.

7. The intelligent alarm method for vehicle intervention after a disaster as claimed in claim 6, characterized in that: The S53 includes: S531: Extract rock fracture strike data from the digital elevation model of the disaster area and mark fracture zones with an angle less than 30 degrees to the driving direction as potential instability areas; S532: Obtain ground vibration monitoring data for the current road section over the past two hours. If a candidate path passes through a grid cell with a vibration amplitude exceeding 0.3g and a duration exceeding 10 seconds, downgrade the geological stability factor of the path. S533: Calculate the surface bearing capacity margin of each candidate path based on the fully loaded vehicle mass parameter, and eliminate paths with margin values ​​lower than a safety threshold; S534: Sort the remaining paths by weighted ratio of fault zone avoidance rate and surface bearing capacity margin, and select the path with the highest comprehensive score.

8. The intelligent alarm method for vehicle intervention after a disaster as claimed in claim 7, characterized in that: The S54 includes: S541: Real-time acquisition of tire pressure monitoring values, road surface infrared temperature values, and road surface roughness index detected by millimeter-wave radar; S542: When tire pressure is 15% below the standard value and the road temperature is above 50°C, the high-temperature soft tire compensation mode is activated, increasing the basic safety margin by 20%; S543: If the road roughness index drops by more than 40% for five consecutive frames, the road is determined to be slippery, and a sinusoidal damping factor is injected into the lateral offset command to suppress steering judder. S544: Dynamically adjust the safety margin based on the vehicle's real-time yaw rate feedback: When the yaw rate exceeds 0.5 rad / s, an additional 5% margin is added for every 0.1 rad / s increase in yaw rate; when the yaw rate is below 0.2 rad / s for 3 seconds, the baseline margin value is restored.

Images