Road safety detection method and system based on internet of things, and storage medium

Abstract

This application relates to the field of intelligent transportation and road safety detection technology, and discloses a method, system, and storage medium for highway safety detection based on the Internet of Things (IoT). The method includes: deploying multiple sensor nodes on the highway surface to collect road surface state data in real time, including point cloud data, texture information, and vibration data; using the point cloud data to identify road areas; associating the road surface state data of the road areas with the real-time location data of the sensor nodes to identify road deformation areas; using data analysis algorithms to intelligently assess the road surface state and generate a risk level report; constructing a road deformation prediction model based on regression models and machine learning, and generating a road deformation trend map; and generating repair suggestions and a visualization report based on the risk level report and the road deformation trend map, which are then pushed to highway maintenance personnel via the IoT system. This application solves the technical problem of low efficiency and accuracy in highway safety detection and reduces the decision-making cost of road maintenance.

Description

Technical Field

[0001] This application relates to the field of intelligent transportation and road safety detection technology, and in particular to Internet of Things-based highway safety detection methods, systems and storage media. Background Technology

[0002] With the continuous expansion of highway networks and the sustained growth of traffic flow, road surfaces are subject to long-term effects from vehicle loads and natural environmental factors (such as temperature changes, precipitation, and ultraviolet radiation), making them prone to deformation and damage issues such as cracks, settlement, bulging, and potholes. If these problems are not detected and repaired in a timely manner, they will not only reduce highway traffic efficiency but may also cause safety accidents such as vehicle bumps and loss of control, posing a threat to the lives and property of drivers and passengers.

[0003] Traditional highway safety inspection methods rely heavily on manual inspections or periodic sampling, resulting in low efficiency, limited coverage, delayed data collection, and strong subjectivity, making it difficult to grasp the road surface condition in real time and comprehensively. Furthermore, traditional methods lack the ability to accurately predict road surface deformation trends, failing to anticipate potential risks and leading to maintenance work often being reactive, increasing maintenance costs and safety hazards. In addition, while some highway safety inspection methods incorporate sensor technology, they still suffer from the following technical problems: sensor deployment lacks scientific optimization, failing to comprehensively consider cost, data transmission latency, and detection coverage, leading to wasted detection resources or missed detections in critical areas, easily causing redundancy or insufficient coverage; low accuracy in road surface area identification and data processing, with point cloud data processing lacking noise reduction methods, making it difficult to accurately identify road areas; risk assessment relies on human experience, lacks unified standards, and lacks the ability to predict road surface deformation trends, resulting in a lack of foresight in maintenance decisions. Summary of the Invention

[0004] This application provides a method, system, and storage medium for highway safety detection based on the Internet of Things, which addresses the technical problem that the low efficiency and accuracy of highway safety detection in the prior art leads to high costs for road maintenance decisions.

[0005] Firstly, this application provides a highway safety detection method based on the Internet of Things, comprising the following steps: Step S1: Deploy multiple sensor nodes on the road surface and collect road surface status data in real time through the sensor nodes. The road surface status data includes point cloud data, texture information and vibration data. Step S2: Use the point cloud data to identify road areas, associate the road surface state data of the road areas with the real-time location data of the sensor nodes, identify road surface deformation areas, and obtain the spatial location and depth information of the road surface deformation areas; Step S3: Based on the spatial location and depth information, use data analysis algorithms to intelligently assess the road surface condition and generate a risk level report according to preset risk assessment standards; Step S4: Construct a road surface deformation prediction model based on regression model and machine learning, output the road surface deformation trend through the road surface deformation prediction model and generate a road surface deformation trend map; Step S5: Based on the risk level report and the road surface deformation trend map, generate repair suggestions and a visualization report, and push them to highway maintenance personnel through the Internet of Things system.

[0006] Furthermore, the deployment of multiple sensor nodes on the highway surface includes: dividing the highway into several detection units according to a preset division standard based on highway mileage, number of lanes, and terrain features; configuring sensor groups consisting of lidar sensors, image sensors, and vibration sensors in each detection unit; constructing a multi-objective optimization model that includes sensor deployment cost, data transmission latency, and detection coverage; generating an initial sensor deployment scheme using a greedy algorithm; iteratively optimizing the initial sensor deployment scheme using a simulated annealing algorithm; outputting the optimal coordinate positions of the sensor nodes within each detection unit; deploying the sensor nodes according to the optimal coordinate positions; and establishing a network connection between the sensor nodes and the server via a wireless communication protocol.

[0007] Further, the step of using the point cloud data to identify road areas includes: delineating multiple rectangular areas on the highway corresponding to the point cloud data, with the long side of each rectangular area parallel to the direction of travel of the moving body; extracting any rectangular area and calculating the median elevation of all point cloud data within that rectangular area, defining it as the reference elevation; calculating the difference between the elevation of each point cloud data within the rectangular area and the reference elevation; if the difference is greater than a predetermined difference threshold, it is identified as an interference point and removed to obtain optimized point cloud data; fitting the optimized point cloud data using the least squares method to obtain the plane with the smallest sum of squared elevation errors; if the sum of squared errors of the plane is less than a predetermined error threshold, the rectangular area is identified as a road area; and integrating all the rectangular areas identified as road areas to complete the identification of the road area.

[0008] Further, the step of associating the road surface state data of the road area with the real-time location data of the sensor nodes to identify road surface deformation areas and obtain the spatial location and depth information of the road surface deformation areas includes: synchronizing the road surface state data in time; binding the synchronized road surface state data with the real-time location data of the sensor nodes to obtain integrated time-location-state data; calculating the surface curvature, normal direction deviation, and elevation change rate at each location within the road area based on the point cloud data in the integrated data; if the surface curvature exceeds a set curvature threshold, or... If the deviation of the normal direction exceeds the normal deviation threshold, or the elevation change rate exceeds the elevation change rate threshold, the road area is determined to be a suspected road surface deformation area. Texture information and vibration data of the suspected road surface deformation area are extracted; edge detection and grayscale analysis are performed on the texture information, and spectral analysis and temporal analysis are performed on the vibration data to determine the road surface deformation area; point cloud data of the road surface deformation area is extracted, and the boundary coordinates and center coordinates of the road surface deformation area are calculated; the elevation of the point cloud data within the road surface deformation area is compared with the preset road benchmark elevation to determine the maximum height and maximum depth of the road surface deformation area.

[0009] Further, step S3 includes: extracting road damage features using a convolutional neural network based on the boundary coordinates, the center coordinates, the maximum height, and the maximum depth, wherein the road damage features include damage type, spatial coordinates, and depth data; classifying the road damage features using a data analysis algorithm and assigning them risk levels according to a preset risk assessment standard; and generating a risk level report containing the damage type, the spatial location, the depth information, and the risk level.

[0010] Further, the extraction of road damage features using a convolutional neural network includes: inputting the point cloud data into a PointNet++ network to extract geometric features; inputting the texture information into a 2D convolutional neural network to extract visual features; inputting the vibration data into a 1D convolutional network after short-time Fourier transform to extract spectral features; weighting and fusing the geometric features, visual features, and spectral features through an attention mechanism in the intermediate fusion layer of the multimodal feature fusion network to obtain a fused feature vector; and inputting the fused feature vector into a classification and regression network to output damage type, spatial coordinates, and depth data corresponding to the boundary coordinates, center coordinates, maximum height, and maximum depth.

[0011] Further, step S4 includes: extracting damage type, spatial coordinates, depth data, and risk level from the risk level report, and acquiring road surface condition data at different times to construct a multidimensional dataset; using the multidimensional dataset as training samples, fitting the deformation pattern of the road surface deformation area over time based on a time series regression model, and using a machine learning algorithm to train a road surface deformation prediction model that matches the damage type of the road surface deformation area; inputting the road surface condition data of the road surface deformation area into the road surface deformation prediction model, and outputting the spatial location change, depth change, and damage type evolution of the deformation area within a future preset time period; and constructing the road surface deformation trend map based on the spatial location change, the depth change, the damage type evolution, and a preset deformation warning threshold.

[0012] Further, step S5 includes: determining repair priorities and recommended repair methods based on the risk level in the risk level report and the deformation warning threshold in the road deformation trend map, and generating repair suggestions; integrating the repair suggestions, the risk level report, and the road deformation trend into a visualization report, the visualization report including the location of the deformation area, damage type, depth, and recommended repair methods; and pushing the visualization report to highway maintenance personnel in real time through an Internet of Things system.

[0013] Secondly, this application provides an IoT-based highway safety inspection system, comprising: a sensor deployment module for deploying multiple sensor nodes on the highway surface; a data acquisition module for real-time acquisition of road surface condition data through the sensor nodes, the road surface condition data including point cloud data, texture information, and vibration data; a data preprocessing and fusion module for using the point cloud data to identify road areas, associating the road surface condition data of the road areas with the real-time location data of the sensor nodes, identifying road surface deformation areas, and obtaining the spatial location and depth information of the road surface deformation areas; an intelligent assessment module for intelligently assessing the road surface condition based on the spatial location and depth information using data analysis algorithms, and generating a risk level report according to preset risk assessment standards; a prediction module for constructing a road surface deformation prediction model based on regression models and machine learning, outputting the road surface deformation trend through the road surface deformation prediction model, and generating a road surface deformation trend map; and a visualization report and decision support module, whereby users generate repair suggestions and visualization reports based on the risk level report and the road surface deformation trend map, and push them to highway maintenance personnel through the IoT system.

[0014] A third aspect of this application provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement an Internet of Things-based highway safety detection method.

[0015] The technical solution provided in this application divides the detection units according to highway mileage, number of lanes, and terrain features. It configures sensor nodes using a combination of lidar, image sensors, and vibration sensors, and completes the sensor node deployment using a multi-objective optimization model. The sensor nodes collect road surface status data in real time, effectively solving the problems of single data source and limited coverage in traditional detection methods. A rectangular area is delineated on the highway corresponding to the point cloud data. The difference between the elevation of each point cloud data point and the reference elevation is calculated. If the difference is greater than a predetermined difference threshold, it is identified as an interference point and removed, resulting in optimized point cloud data. The optimized point cloud data is then fitted using the least squares method to obtain the plane with the smallest sum of squared elevation errors. If the sum of squared errors of the plane is less than a predetermined error threshold, the rectangular area is determined to be a road area. By introducing dual constraints of elevation difference threshold and sum of squared error threshold, the shortcomings of traditional point cloud processing—ignoring interference points and low accuracy in road area identification—can be overcome, making road area division more accurate and reliable. Simultaneously, by synchronizing road surface condition data with real-time sensor location data, the problem of loose correlation between road data and location in traditional detection methods is solved, ensuring the acquisition of high-quality integrated "time-location-state" data. Based on the point cloud data in the integrated data, the surface curvature, normal direction deviation, and elevation change rate at each location within the road area are calculated. Suspected road surface deformation areas are identified based on surface curvature, normal direction deviation, and elevation change rate. The texture information and vibration data of suspected road surface deformation areas are analyzed to confirm the road surface deformation areas. That is, through multi-dimensional index collaborative verification, the accuracy and comprehensiveness of road surface deformation area identification are improved. The damage type, location, depth, and other road surface damage features of the road surface deformation areas are extracted by convolutional neural networks, and risk levels are classified according to preset risk assessment standards, ensuring the objectivity and accuracy of the risk level report generation. A road surface deformation prediction model is constructed by using regression and machine learning. The model outputs the road surface deformation trend and generates a road surface deformation trend map. Based on the risk level report and the road surface deformation trend map, repair suggestions and visualization reports are generated and pushed to highway maintenance personnel through an Internet of Things system. This ensures that potential risks can be predicted in advance, promotes the shift of maintenance work from post-remediation to pre-emptive prediction, solves the problem of lack of foresight in maintenance decisions, and reduces long-term maintenance costs and safety hazards. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of the Internet of Things-based highway safety detection method in the embodiments of this application;

[0018] Figure 2 This is a structural block diagram of the Internet of Things-based highway safety detection system in the embodiments of this application. Detailed Implementation

[0019] This application provides a method, apparatus, device, and medium for highway safety detection based on the Internet of Things. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0020] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the Internet of Things-based highway safety detection method in this application includes: Step S1: Deploy multiple sensor nodes on the road surface and collect road surface status data in real time through the sensor nodes. The road surface status data includes point cloud data, texture information and vibration data.

[0021] It is understood that the executing entity of this application can be an IoT-based highway safety detection system, or it can be a terminal or a server; the specific implementation is not limited here. This application's embodiments use a server as an example for illustration.

[0022] Specifically, deploying multiple sensor nodes on the road surface and collecting road surface condition data in real time through these nodes is a fundamental step in realizing road structure safety inspection within the Internet of Things (IoT) framework. Each sensor node consists of a lidar unit, an image acquisition unit, and a vibration detection unit. Each type of sensor is responsible for collecting road surface condition information from different dimensions. Point cloud data, acquired by lidar, characterizes the spatial geometry of the road surface; texture information, acquired by the image acquisition unit, reflects the material structure and subtle damage characteristics of the road surface layer; and vibration data, acquired by embedded accelerometers or microelectromechanical systems (MEMS), records the dynamic response to vehicle passage or environmental disturbances. Each sensor node maintains real-time data transmission with the server via a wireless communication protocol, forming road surface condition data that includes point cloud data, texture information, and vibration data.

[0023] Step S2: Use the point cloud data to identify road areas, associate the road surface state data of the road area with the real-time location data of the sensor nodes, identify road surface deformation areas, and obtain the spatial location and depth information of the road surface deformation areas.

[0024] Specifically, the point cloud data is preprocessed to remove irrelevant and noise points. Segmentation algorithms based on clustering or geometric fitting are used to distinguish road areas from non-road areas, thus obtaining pavement state data containing only the road surface. Due to discrepancies in the acquisition times of different sensors, a unified time reference is introduced to align data from different sensors according to timestamps, avoiding asynchronous errors caused by differences in sampling frequencies. After time synchronization, the pavement state data in each time slice is bound to the real-time location data output by the corresponding sensor nodes. By establishing a one-to-one mapping relationship, the state characteristics at the same moment are combined with spatial coordinates to generate integrated "time-location-state" data.

[0025] After obtaining the integrated data, features of potentially deformable areas are extracted by calculating height differences, curvature distribution, and normal vector changes at various locations along the road. For example, by comparing the residuals of local surface heights with the globally fitted surface, different types of anomalies such as bulges, depressions, or cracks can be identified. In this process, there is a close correspondence between road surface condition data and sensor location data. The location data ensures the spatial localization of anomalies, while the point cloud data provides the specific manifestations of the anomalies. The combination of these two factors allows deformed areas not only to be detected but also to be assigned clear spatial coordinates.

[0026] To further obtain depth information of the deformed area, local surface reconstruction needs to be performed within the identified deformed area and compared with the fitted ideal flat surface. The distribution of deformation depth can be obtained by calculating the vertical distance between each deformed point and the fitted reference surface. When sensor nodes continuously collect data at different locations, multiple observation points are formed in the same area. By weighted fusion of these observation points, the impact of single measurement errors can be reduced, thus obtaining stable deformation depth values. This multi-source data fusion-based processing method allows the algorithm features and sensor data processing techniques to work together to achieve accurate detection and quantitative characterization of road surface deformation areas. For example, in a road inspection scenario, sensor nodes are installed on a mobile platform. When they pass through a section of road surface with depressions, the point cloud data shows that the height of this area is significantly lower than the fitted road surface. Through real-time location data registration, the latitude and longitude coordinates and depth information of the depression point can be accurately marked on the road map. This method can achieve automated identification and location of large-scale road deformations without relying on manual inspection, not only improving detection efficiency but also providing quantitative basis for subsequent maintenance.

[0027] Step S3: Based on the spatial location and depth information, use data analysis algorithms to intelligently assess the road surface condition and generate a risk level report according to preset risk assessment standards.

[0028] Specifically, spatial location data and depth information are fused and analyzed as core input data sources. Spatial location information identifies the specific geographic coordinates of the detected object within the road environment, while depth information characterizes the geometric differences between the road surface and subsidence areas. Therefore, there is a correspondence between these two types of data; the coordinates of a given point are necessarily related to the depth changes at that location. Data analysis algorithms construct a spatiotemporal joint model, mapping the depth values ​​corresponding to different coordinate points to a three-dimensional road surface state matrix. Interpolation and filtering operations are used to eliminate sampling noise, ensuring the continuity and stability of road surface geometric features. After forming a complete road surface geometry dataset, feature extraction methods are used to identify local abrupt changes and continuous abnormal areas, extracting features such as deformation amplitude, deformation range, and depth gradient. These features are then correlated with preset risk assessment standards. Risk assessment standards typically include parameters such as depth threshold, deformation area threshold, and cumulative deformation threshold. When any feature exceeds the threshold, the area is marked as high-risk; when the feature is within the normal range, it corresponds to a low-risk level. To ensure the accuracy of risk classification, the algorithm incorporates neighborhood clustering analysis in addition to single-point feature judgment. This involves aggregating spatially adjacent depth anomaly points and using the aggregation degree and anomaly density to further confirm risk areas, thereby avoiding misjudgments caused by isolated noise points. For example, during the detection of a main urban road, if a depth depression of 8 centimeters is detected at a certain spatial coordinate point, and the anomaly area exceeds 0.5 square meters, the algorithm determines that the point is a high-risk area based on a threshold and generates a high-risk label for that location in the risk level report. Maintenance personnel can then quickly locate and take measures based on this information.

[0029] Step S4: Construct a road surface deformation prediction model based on regression model and machine learning, output the road surface deformation trend through the road surface deformation prediction model and generate a road surface deformation trend map.

[0030] Specifically, the construction of the pavement deformation prediction model relies on the synergy of regression models and machine learning methods. By fusing and extracting features from previously collected spatial location, depth information, and pavement condition assessment results, the raw data is transformed into a multi-dimensional feature set suitable for training and prediction. That is, normalization and denoising algorithms are used to preprocess the pavement sampling data, ensuring comparability of data from different sources and with different dimensions within the same feature space. These processed features are then correlated with historical pavement deformation data, creating a mappable input-output pair between spatial displacement, depth changes, load information, and time-series data, thus providing training samples for the regression model and machine learning. At the regression model level, multiple linear regression or nonlinear regression methods are used to fit the relationship between pavement deformation and input features, establishing a preliminary prediction framework. At the machine learning level, neural network methods are introduced to train and learn complex nonlinear mapping relationships, overcoming the limitations of traditional regression models under high-dimensional and noisy data. During the prediction model's operation, the regression model provides a basic fit to the deformation trend, while the machine learning method continuously updates the model and optimizes parameters based on newly collected pavement monitoring data, enabling the prediction results to dynamically reflect the actual stress and deformation state of the pavement. The data correspondence is manifested in multi-dimensional input features corresponding to multiple sets of predicted outputs. Spatial location changes correspond to longitudinal settlement trends, depth information corresponds to transverse cracking trends, and load and traffic flow data correspond to long-term fatigue deformation trends. These individual prediction results are integrated by the model to form an overall pavement deformation trend output. In generating the pavement deformation trend map, the time-series prediction data output by the prediction model is transformed into two-dimensional or three-dimensional curves through visualization algorithms, allowing for a visually intuitive presentation of settlement changes, crack expansion, or rutting deepening trends at different locations. The trend map not only reflects the current quantitative predictions but also reveals the future deformation development speed and risk level through the curve slope and fluctuation amplitude.

[0031] Step S5: Based on the risk level report and the road surface deformation trend map, generate repair suggestions and a visualization report, and push them to highway maintenance personnel through the Internet of Things system.

[0032] Specifically, risk level reports and road surface deformation trend maps are used as core input data. The two form a closely coupled correspondence through data processing and algorithm modeling, thereby generating repair suggestions and visualization reports, which are then pushed to highway maintenance personnel in real time through the Internet of Things system.

[0033] Specifically, the risk level report is generated based on modeling of sensor-collected data. It includes analysis results of multi-dimensional parameters such as pavement crack depth and settlement amplitude, and classifies road segments into different risk levels using a grading algorithm. The pavement deformation trend map, on the other hand, is generated through fitting and predictive models of time-series data, presenting the evolution trends of pavement crack length, pothole expansion rate, and settlement trajectory. Logically, these two reports are not independent but rather map the state of the same road segment through the correspondence between data identifiers and spatial locations. This involves matching the change curves in the trend map with the corresponding risk level segments, thus forming cross-validation of multi-dimensional information to ensure the remediation recommendations are targeted. During the generation of remediation recommendations, different remediation strategy databases are invoked based on the risk level. For example, when the risk level is moderate and the trend map shows a slow crack expansion rate, surface crack sealing is recommended. Conversely, when the risk level is severe and the trend curve shows an accelerating settlement rate, base course reinforcement or partial repaving is recommended. Here, the risk level provides a qualitative classification standard, while the trend chart provides a quantitative rate of change. The two interact to jointly determine the specific level and method of the remediation recommendations, ensuring that the generated results rely not only on single data points but also on a comprehensive judgment based on the fusion of multi-source information. This fusion process is achieved through a machine learning model, which performs pattern recognition and rule matching on the input level and trend data. While generating remediation recommendations, it also generates structured data for further visualization reports. The visualization report, generated by a graphical rendering engine, marks the risk level distribution on the road electronic map using color gradients. It also overlays cracks or settlement curves from the trend chart at corresponding locations and combines this with the remediation recommendations to create a graphical and textual presentation. This allows maintenance personnel to easily see the current risk level of the road surface and intuitively grasp the deterioration trend and corresponding remediation plans over a future period, thereby improving decision-making efficiency.

[0034] In this embodiment, multiple sensor nodes are deployed on the road surface to collect road surface condition data in real time, avoiding the problems of single data source and limited coverage in traditional detection methods. Point cloud data is used to identify road areas, reducing data interference. By associating the road surface condition data of the road area with the real-time location data of the sensor nodes, the loose correlation between road data and location in traditional detection methods is solved, ensuring high-quality integrated "time-location-state" data. Integrated data is used to identify road surface deformation areas, ensuring the accuracy of road detection. Data analysis algorithms are used to intelligently assess the road surface condition and generate a risk level report. A road surface deformation prediction model is constructed based on regression models and machine learning, generating a road surface deformation trend map. Based on the risk level report and the road surface deformation trend map, repair suggestions and a visualization report are generated and pushed to road maintenance personnel through an IoT system, ensuring that potential risks can be predicted in advance, solving the problem of lack of foresight in maintenance decisions, and reducing long-term maintenance costs and safety hazards.

[0035] In one specific embodiment, step S1 involves deploying multiple sensor nodes on the road surface, specifically including the following steps: (1) Based on the pre-set division criteria of highway mileage, number of lanes and terrain features, the highway is divided into several detection units; (2) The detection unit is configured with a sensor group consisting of a lidar sensor, an image sensor and a vibration sensor; (3) Construct a multi-objective optimization model that includes sensor deployment cost, data transmission delay and detection coverage. Use a greedy algorithm to generate an initial sensor deployment scheme. Use a simulated annealing algorithm to iteratively optimize the initial sensor deployment scheme and output the optimal coordinate position of the sensor node in each detection unit. (4) Deploy sensor nodes according to the optimal coordinate positions and realize network connection between sensor nodes and server through wireless communication protocol.

[0036] Specifically, the deployment of road surface sensor nodes is not a simple physical placement, but rather a data-driven optimization process. Roads are divided into several detection units based on mileage, number of lanes, and terrain features. This division essentially maps a continuous road space into several independent regions, each undertaking local monitoring tasks in subsequent data acquisition and processing. To ensure the completeness and diversity of data coverage within each detection unit, a sensor ensemble needs to be built within each unit. This sensor ensemble consists of LiDAR, image sensors, and vibration sensors. LiDAR generates 3D point cloud data of the road surface, image sensors provide texture and color information, and vibration sensors collect dynamic response signals during vehicle movement. These three types of data correspond to spatial morphology, surface visual features, and dynamic characteristics, respectively. Only through the joint deployment of these three types of sensors can a complete representation of the road surface state be obtained.

[0037] The specific locations of sensor nodes need to be determined by an optimization model. This model's objective function comprehensively considers three factors: deployment cost, data transmission latency, and detection coverage. Cost and coverage constitute constraints, while latency and data integrity constitute performance requirements. To solve this multi-objective optimization model, a greedy algorithm is introduced to generate an initial deployment plan. Based on the principle of local optima, the greedy algorithm gradually places sensor nodes according to coverage requirements, thus quickly obtaining a better solution in each iteration. Since the greedy algorithm is prone to getting trapped in local optima, a simulated annealing algorithm is introduced for secondary optimization. During iteration, the simulated annealing algorithm gradually searches for a better solution space by accepting suboptimal solutions with a certain probability. The control parameter, temperature, gradually decreases with the number of iterations, making the sensor location distribution tend towards the optimal globally. At this point, the optimal coordinates within each detection unit are automatically output by the algorithm. The algorithm's characteristics are closely integrated with the physical deployment process of the sensors, ensuring deployment efficiency while improving coverage comprehensiveness and transmission reliability. After the sensor nodes are physically deployed according to the optimization results, a wireless communication protocol is used to connect the nodes to the server. Data collected by each node is uploaded to the central server via a self-organizing network or cellular network. The server binds the data to the node coordinates according to the detection unit number, so that the collected data can be directly correlated with the spatial location in subsequent status analysis.

[0038] In one specific embodiment, step S2, which uses the point cloud data to identify road regions, specifically includes the following steps: (1) On the highway corresponding to the point cloud data, multiple rectangular areas are delineated, and the long side of each rectangular area is parallel to the direction of travel of the moving body; (2) Extract any rectangular region, calculate the median elevation of all point cloud data within the rectangular region, and define it as the reference elevation; calculate the difference between the elevation of each point cloud data within the rectangular region and the reference elevation. If the difference is greater than a predetermined difference threshold, it is determined to be an interference point and removed to obtain optimized point cloud data. (3) The optimized point cloud data is fitted by the least squares method to obtain the plane with the smallest sum of squared elevation errors; (4) If the sum of squared errors of the plane is less than a predetermined error threshold, the rectangular area is determined to be a road area; (5) Integrate all the rectangular areas that are identified as road areas to complete the road area identification.

[0039] Specifically, the collected 3D point cloud data is spatially limited by defining a rectangular computational region within the highway area where the moving object is located. The longer side of this rectangle aligns with the direction of the moving object's movement. This spatial limitation maintains data continuity while reducing interference from redundant point clouds. The rectangular region contains a large amount of elevation information. These point cloud elevations are statistically processed, and the median is extracted as the baseline elevation. This processing method avoids overall deviation caused by local outliers. After obtaining the baseline elevation, the difference between the elevation of each point cloud data point and the baseline elevation is calculated and compared with a preset difference threshold. If the difference exceeds the threshold, it is identified as an interference point and removed. This generates an optimized point cloud dataset. The removal of interference points directly reduces interference from non-road targets such as vehicles, guardrails, or vegetation on road recognition. The optimized point cloud data is then fitted to a plane using the least squares method. Utilizing the principle of minimizing the sum of squared errors, a planar model that effectively reflects the overall smoothness of the road is fitted. The sum of squared errors of this plane is calculated and compared to a predetermined error threshold. If the error value is below the threshold, it indicates a high degree of match between the fitted result and the road features, thus classifying the rectangular region as a road area. All rectangular regions identified as road areas are integrated to complete the road area identification. Taking a highway detection scenario as an example, when sensor nodes collect continuous point clouds during vehicle movement, if there are interfering factors such as protruding vehicles or bridge shadows in a certain area, the difference calculation and threshold determination mechanism can effectively eliminate these interfering points, ensuring the accuracy of the fitted plane. This allows for correct identification of road areas even in complex traffic environments. This processing flow solves the problems of blurred road area boundaries and susceptibility to external object interference in traditional point cloud data identification, improving the robustness of road area extraction and providing a reliable spatial benchmark for subsequent road surface condition analysis. Ultimately, it demonstrates significant application value in intelligent road surface detection and traffic safety assurance.

[0040] In one specific embodiment, step S2 involves associating the road surface condition data of the road area with the real-time location data of the sensor nodes to identify road surface deformation areas and obtain the spatial location and depth information of the road surface deformation areas. This specifically includes the following steps: Specifically, the steps include the following: (1) Synchronize the road surface status data in time, and bind the synchronized road surface status data with the real-time location data of the sensor node to obtain integrated time-location-status data; (2) Based on the point cloud data in the integrated data, calculate the surface curvature, normal direction deviation and elevation change rate at each location within the road area; (3) If the surface curvature exceeds the set curvature threshold, or the normal direction deviation exceeds the normal deviation threshold, or the elevation change rate exceeds the elevation change rate threshold, then the road area is determined to be a suspected road surface deformation area. (4) Extract the texture information and vibration data of the suspected road deformation area, perform edge detection and grayscale analysis on the texture information, and perform spectrum analysis and time domain analysis on the vibration data to determine the road deformation area; (5) Extract the point cloud data of the road deformation area and calculate the boundary coordinates and center coordinates of the road deformation area; (6) Compare the elevation of the point cloud data within the road deformation area with the preset road reference elevation to determine the maximum height and maximum depth of the road deformation area.

[0041] Specifically, through a time synchronization mechanism, road surface condition data and location data are correlated within a unified time dimension, forming a three-dimensional data structure of "time-location-state". This data structure provides a foundation for subsequent identification calculations because multi-source data such as point cloud data, texture features, and vibration features are fused within the same coordinate and time frame, allowing for one-to-one correspondence and cross-validation of data from different dimensions. This binding relationship enables the tracing of the source and scope of each data point in the spatiotemporal dimension, ensuring the reliability of subsequent judgment results. After synchronization and binding, geometric parameters such as curvature, normal direction deviation, and elevation change rate are calculated and extracted from the point cloud data. These parameters directly reflect the morphological characteristics of the road surface. When the curvature exceeds a set threshold, it indicates the presence of bulges or depressions in a local area; when the normal direction deviation exceeds the limit, it indicates abnormal tilting of the road surface; when the elevation change rate exceeds the limit, it indicates a sudden change in height in the longitudinal direction. These three types of indicators complement each other, forming a multi-dimensional basis for judging potential deformation areas, thereby improving the accuracy and stability of identification. The data input for this calculation process comes from LiDAR point clouds, while location data provides a reference for the spatial coordinates of the point clouds. These two types of data support each other, forming a complete deformation detection logic. To further eliminate environmental noise and false anomalies, texture information and vibration signals from suspected areas are jointly analyzed. Texture information provided by the image sensor, after edge detection and grayscale analysis, can reveal subtle morphological features such as cracks and potholes, while the spectral and temporal signals provided by the vibration sensor reflect the dynamic response characteristics when a vehicle passes over it. The combination of these two types of data allows for cross-verification of spatial geometric anomalies and dynamic characteristics. If the texture edges exhibit obvious crack structures and high-frequency energy peaks appear in the vibration spectrum, the area is more likely to be a genuine deformation region. This multi-source fusion not only improves the reliability of identification but also reduces the false alarm rate, demonstrating the functional interaction between algorithm features and sensor configuration.

[0042] After identifying the actual deformation area, point cloud data of that area is extracted, and boundary and center coordinates are calculated to obtain the spatial extent and location information of the area. By comparing this with the road benchmark elevation, the maximum height and maximum depth of the deformation area can be obtained. This process effectively combines three-dimensional geometric analysis with benchmark data from road engineering, allowing the detection results to be directly mapped into evaluation indicators usable in engineering. In this way, the identification results not only remain at the data level but can also directly serve road maintenance and construction planning. For example, at the bridgehead junction of a highway, point cloud data collected by sensor nodes shows that the area has a high curvature value and obvious elevation abrupt changes. At the same time, texture analysis reveals transverse crack structures, and the vibration spectrum also shows an energy peak shift. After fusing these features, the area is identified as the actual deformation area, and the boundary extent and center point location of the area are calculated, with a maximum subsidence depth of 4.3 cm. This result can be directly used by road maintenance departments to formulate repair plans, thereby solving the problem that traditional manual inspections are unable to detect deep structural deformations in a timely manner.

[0043] In one specific embodiment, the process of performing step S3 may specifically include the following steps: (1) Based on the boundary coordinates, the center coordinates, the maximum height and the maximum depth, road surface damage features are extracted by a convolutional neural network, wherein the road surface damage features include damage type, spatial coordinates and depth data; (2) The road damage characteristics are classified by data analysis algorithms, and the risk level is determined according to the preset risk assessment standards; (3) Generate a risk level report that includes the damage type, the spatial location, the depth information and the risk level.

[0044] Specifically, point cloud data is processed using the PointNet++ network to extract road surface geometric features, resulting in a 3D shape description including potholes, cracks, and subsidence. Image texture information is extracted using a 2D convolutional neural network to obtain features of surface texture distribution, crack orientation, and damage edge morphology. Vibration data is converted into a time-frequency spectrum through short-time Fourier transform and then input into a 1D convolutional neural network to extract frequency features, reflecting road surface unevenness and impact intensity. Multimodal features are weighted and fused in the intermediate fusion layer of the multimodal feature fusion network using an attention mechanism. This gives higher weights to features highly correlated with boundary coordinates, center coordinates, maximum height, and maximum depth, achieving complementarity and synergy among spatial geometric features, visual texture features, and spectral features, thereby improving the accuracy and stability of damage identification. The fused feature vector is input into a classification and regression network, outputting damage type labels while retaining the corresponding spatial coordinates and depth information, forming a complete set of road surface damage features. Data analysis algorithms are used to cluster and classify the feature set, distinguishing different damage types such as potholes, cracks, and spalling. Spatial location parameters are used to pinpoint the precise location of the damaged road surface, supporting subsequent maintenance work. The correlation between depth information and maximum height and depth is used to determine the severity of the damage. Combined with preset risk assessment standards, risk levels are classified as low, medium, or high risk. The generated risk level report includes damage type, spatial location, depth information, and risk level, providing decision-making support for highway maintenance departments and assisting in maintenance scheduling and resource allocation.

[0045] By employing the aforementioned multi-source data fusion and convolutional neural network feature extraction techniques, geometric, visual, and spectral features complement each other, enabling accurate identification of damage types even under conditions of insufficient lighting, occlusion, or noise interference, thus improving the robustness of pavement damage identification. The one-to-one correspondence between spatial coordinate parameters and feature outputs allows risk assessment results to be directly correlated with the road's spatial location, avoiding the problem of repeated detections caused by inaccurate location data in traditional methods. For example, when a pothole deeper than 50 mm is detected on a pavement section, the system can determine its high-risk level using the fused multimodal features and generate a report containing specific GPS coordinates and depth values, directly pushing it to maintenance personnel and shortening fault response time. This technical solution, where algorithmic and technical features support each other, jointly addresses the problems of insufficient accuracy in pavement damage identification and incomplete risk assessment in existing technologies, improving the intelligence and automation level of highway safety monitoring. It achieves closed-loop control across the entire chain from perception and identification to decision-making, possessing high practical value and significant potential for wider application.

[0046] In one specific embodiment, the process of performing step S4 may specifically include the following steps: (1) Extract damage type, spatial coordinates, depth data and risk level from the risk level report, and obtain road surface condition data at different times to construct a multidimensional dataset; (2) Using the multidimensional dataset as training samples, the deformation pattern of the road deformation area over time is fitted based on the time series regression model, and a road deformation prediction model matching the damage type of the road deformation area is trained using machine learning algorithms. (3) Input the pavement state data of the pavement deformation area into the pavement deformation prediction model, and output the spatial location change, depth change and damage type evolution of the deformation area within a future preset time period; (4) Construct the road surface deformation trend map based on the spatial location change, the depth change, the damage type evolution and the preset deformation warning threshold.

[0047] Specifically, the risk level report is analyzed to extract damage type, spatial coordinates, depth data, and risk level. This data is then fused with pavement condition data collected at different time points to construct a multidimensional dataset containing damage category, spatial distribution, depth information, and time series. This multidimensional dataset is used as training samples, and a time series regression model is introduced to fit the deformation pattern of the deformed area over time. This model can capture the dynamic trend of damage gradually expanding from initial small cracks to obvious depressions or potholes. These trends are used as input to a machine learning algorithm, and through training, a predictive model that can match different damage types is obtained. Furthermore, using the predicted spatial displacement, depth aggravation, and damage type evolution trends, combined with a preset deformation warning threshold, a pavement trend map is constructed. The trend map marks the change trajectory of the predicted location on a two-dimensional spatial coordinate system, while using color or grayscale levels to indicate the depth change magnitude, and using symbols or labels to distinguish the evolution of different damage types. In this way, algorithmic features and technical features are closely integrated. The former provides the ability to fit and predict time-series data, while the latter provides structured data input from sensors and detection systems. Together, they address the problems of insufficient prediction accuracy and difficulty in intuitively presenting damage evolution trends in traditional methods. For example, in highway bridge deck monitoring, this method can predict in the early stages that a longitudinal crack may evolve into a transverse through crack within three months, and show its expansion direction and degree of depth intensification, thus providing maintenance units with a time-window-based maintenance decision-making basis. This synergistic effect of algorithm and technical features not only improves prediction accuracy and visualization effects but also significantly enhances the foresight and scientific nature of road operation and maintenance.

[0048] In one specific embodiment, the process of performing step S5 may specifically include the following steps: (1) Based on the risk level in the risk level report and the deformation warning threshold in the road deformation trend diagram, determine the repair priority and recommended repair methods, and generate repair suggestions; (2) The repair recommendations, the risk level report and the road deformation trend are integrated into a visualization report, which includes the location of the deformation area, the damage type, the depth and the recommended repair method; (3) The visualization report is pushed to highway maintenance personnel in real time through the Internet of Things system.

[0049] Specifically, the risk level information in the risk level report is mapped to the deformation warning threshold in the pavement deformation trend map, forming a comprehensive assessment system for each pavement damage area. Algorithms analyze the risk level and predicted deformation trend to determine the repair priority for each area and match appropriate repair methods, generating structured repair recommendations. The repair recommendations, risk level report, and pavement deformation trend data are integrated to construct a visual report. This report presents the spatial location, damage type, depth, and recommended repair methods of the deformed area through a graphical interface, achieving a multi-dimensional correspondence between data and spatial information, damage characteristics, and construction plans. This allows maintenance personnel to intuitively understand the repair priorities and urgency levels. The visual report is pushed to the terminal devices of highway maintenance personnel in real time via an IoT system, ensuring information synchronization across different times and spaces. Simultaneously, the data structure in the report is consistent with the sensor-collected data and the output of the prediction model, enabling a closed loop between prediction results, risk assessment, and repair recommendations, thus achieving a continuous process from damage identification and risk assessment to maintenance decision-making.

[0050] The above describes the Internet of Things (IoT)-based highway safety detection method in the embodiments of this application. The following describes the IoT-based highway safety detection system in the embodiments of this application. Please refer to [link / reference]. Figure 2 One embodiment of the Internet of Things-based highway safety detection system in this application includes the following modules.

[0051] The sensor deployment module is used to deploy multiple sensor nodes on the road surface.

[0052] The data acquisition module is used to collect road surface condition data in real time through the sensor nodes. The road surface condition data includes point cloud data, texture information and vibration data.

[0053] The data preprocessing and fusion module is used to identify road areas using the point cloud data, associate the road surface state data of the road area with the real-time location data of the sensor nodes, identify road surface deformation areas, and obtain the spatial location and depth information of the road surface deformation areas.

[0054] The intelligent assessment module is used to intelligently assess the road surface condition based on the spatial location and depth information using data analysis algorithms, and generate a risk level report according to preset risk assessment standards.

[0055] The prediction module is used to construct a road surface deformation prediction model based on regression and machine learning, and output the road surface deformation trend and generate a road surface deformation trend map through the road surface deformation prediction model.

[0056] The visualization report and decision support module allows users to generate repair suggestions and visualization reports based on the risk level report and the road surface deformation trend map, which are then pushed to highway maintenance personnel via the Internet of Things system.

[0057] Through the collaborative efforts of the aforementioned components, multiple sensor nodes are deployed on the highway surface to collect road surface condition data in real time, avoiding the problems of single data source and limited coverage in traditional detection methods. Point cloud data is used to identify road areas, reducing data interference. By correlating the road surface condition data of these areas with the real-time location data of the sensor nodes, the loose correlation between road data and location in traditional detection methods is resolved, ensuring high-quality integrated "time-location-state" data. Integrated data identifies road deformation areas, ensuring the accuracy of road detection. Data analysis algorithms are used to intelligently assess the road surface condition and generate risk level reports. A road surface deformation prediction model is constructed based on regression models and machine learning, generating a road surface deformation trend map. Based on the risk level report and the road surface deformation trend map, repair suggestions and a visualization report are generated and pushed to highway maintenance personnel through an IoT system, ensuring that potential risks can be predicted in advance, addressing the lack of foresight in maintenance decisions, and reducing long-term maintenance costs and safety hazards.

[0058] This application also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein computer instructions are stored in the computer storage medium, and when the computer instructions are executed on the computer, the computer performs the steps of the Internet of Things-based highway safety detection method.

[0059] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0060] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A highway safety detection method based on the Internet of Things, characterized in that, Includes the following steps: Step S1: Deploy multiple sensor nodes on the road surface and collect road surface status data in real time through the sensor nodes. The road surface status data includes point cloud data, texture information and vibration data. Step S2: Use the point cloud data to identify road areas, associate the road surface state data of the road areas with the real-time location data of the sensor nodes, identify road surface deformation areas, and obtain the spatial location and depth information of the road surface deformation areas; Step S3: Based on the spatial location and depth information, use data analysis algorithms to intelligently assess the road surface condition and generate a risk level report according to preset risk assessment standards; Step S4: Construct a road surface deformation prediction model based on regression model and machine learning, output the road surface deformation trend through the road surface deformation prediction model and generate a road surface deformation trend map; Step S5: Based on the risk level report and the road surface deformation trend map, generate repair suggestions and a visualization report, and push them to highway maintenance personnel through the Internet of Things system.

2. The Internet of Things-based highway safety detection method according to claim 1, characterized in that, The deployment of multiple sensor nodes on the highway surface includes: Based on the pre-defined criteria of highway mileage, number of lanes, and terrain features, the highway is divided into several detection units; The detection unit is configured with a sensor group consisting of a lidar sensor, an image sensor, and a vibration sensor; A multi-objective optimization model is constructed, which includes sensor deployment cost, data transmission latency and detection coverage. An initial sensor deployment scheme is generated using a greedy algorithm. The initial sensor deployment scheme is iteratively optimized using a simulated annealing algorithm, and the optimal coordinate positions of the sensor nodes in each detection unit are output. Sensor nodes are deployed at the optimal coordinate locations, and a network connection between the sensor nodes and the server is established through a wireless communication protocol.

3. The Internet of Things-based highway safety detection method according to claim 1, characterized in that, The process of identifying road regions using the point cloud data includes: Multiple rectangular regions are delineated on the highway corresponding to the point cloud data, with the long side of each rectangular region parallel to the direction of travel of the moving body. Extract any rectangular region, calculate the median elevation of all point cloud data within the rectangular region, and define it as the reference elevation; calculate the difference between the elevation of each point cloud data within the rectangular region and the reference elevation. If the difference is greater than a predetermined difference threshold, it is identified as an interference point and removed to obtain optimized point cloud data. By fitting the optimized point cloud data using the least squares method, a plane with the minimum sum of squared elevation errors is obtained. If the sum of squared errors of the plane is less than a predetermined error threshold, the rectangular area is determined to be a road area; All rectangular areas identified as road areas are integrated to complete the road area identification.

4. The Internet of Things-based highway safety detection method according to claim 3, characterized in that, The step of associating the road surface condition data of the road area with the real-time location data of the sensor nodes to identify road surface deformation areas and obtain the spatial location and depth information of the road surface deformation areas includes: The road surface condition data is synchronized in time, and the synchronized road surface condition data is bound with the real-time location data of the sensor nodes to obtain integrated time-location-state data. Based on the point cloud data in the integrated data, calculate the surface curvature, normal direction deviation, and elevation change rate at each location within the road area; If the surface curvature exceeds a set curvature threshold, or the normal direction deviation exceeds a normal deviation threshold, or the elevation change rate exceeds an elevation change rate threshold, then the road area is determined to be a suspected road surface deformation area. The texture information and vibration data of the suspected road deformation area are extracted. Edge detection and grayscale analysis are performed on the texture information, and spectrum analysis and time domain analysis are performed on the vibration data to determine the road deformation area. Extract the point cloud data of the road deformation area, and calculate the boundary coordinates and center coordinates of the road deformation area; By comparing the elevation of the point cloud data within the road deformation area with the preset road benchmark elevation, the maximum height and maximum depth of the road deformation area are determined.

5. The Internet of Things-based highway safety detection method according to claim 4, characterized in that, Step S3 includes: Based on the boundary coordinates, the center coordinates, the maximum height, and the maximum depth, road surface damage features are extracted using a convolutional neural network. The road surface damage features include damage type, spatial coordinates, and depth data. The road surface damage characteristics are classified using data analysis algorithms, and their risk levels are determined according to preset risk assessment standards. Generate a risk level report that includes the damage type, the spatial location, the depth information, and the risk level.

6. The Internet of Things-based highway safety detection method according to claim 5, characterized in that, The extraction of road surface damage features using a convolutional neural network includes: The point cloud data is input into a PointNet++ network to extract geometric features, the texture information is input into a 2D convolutional neural network to extract visual features, and the vibration data is input into a 1D convolutional network after short-time Fourier transform to extract spectral features. In the intermediate fusion layer of the multimodal feature fusion network, the geometric features, visual features, and spectral features are weighted and fused through an attention mechanism to obtain a fused feature vector. The fused feature vector is then input into a classification and regression network to output damage type, spatial coordinates, and depth data corresponding to the boundary coordinates, center coordinates, maximum height, and maximum depth.

7. The Internet of Things-based highway safety detection method according to claim 5, characterized in that, Step S4 includes: Damage type, spatial coordinates, depth data and risk level are extracted from the risk level report, and road surface condition data at different times are obtained to construct a multidimensional dataset; Using the multidimensional dataset as training samples, the deformation pattern of the road deformation area over time is fitted based on a time series regression model, and a road deformation prediction model matching the damage type of the road deformation area is trained using a machine learning algorithm. The road surface condition data of the road surface deformation area is input into the road surface deformation prediction model, and the spatial location change, depth change and damage type evolution of the deformation area within a future preset time period are output. The road surface deformation trend map is constructed based on the spatial location change, the depth change, the damage type evolution, and the preset deformation early warning threshold.

8. The Internet of Things-based highway safety detection method according to claim 7, characterized in that, Step S5 includes: Based on the risk level in the risk level report and the deformation warning threshold in the road deformation trend map, the repair priority and recommended repair methods are determined, and repair suggestions are generated. The repair recommendations, the risk level report, and the road deformation trend are integrated into a visualization report, which includes the location of the deformation area, the damage type, the depth, and the recommended repair method. The visualized report is pushed to highway maintenance personnel in real time through an Internet of Things (IoT) system.

9. A highway safety detection system based on the Internet of Things, characterized in that, For implementing the Internet of Things (IoT)-based highway safety detection method as described in any one of claims 1-8, the IoT-based highway safety detection system comprises: A sensor deployment module for deploying multiple sensor nodes on the road surface; The data acquisition module is used to collect road surface condition data in real time through the sensor nodes. The road surface condition data includes point cloud data, texture information and vibration data. The data preprocessing and fusion module is used to identify road areas using the point cloud data, associate the road surface state data of the road area with the real-time location data of the sensor nodes, identify road surface deformation areas, and obtain the spatial location and depth information of the road surface deformation areas. The intelligent assessment module is used to intelligently assess the road surface condition based on the spatial location and depth information using data analysis algorithms, and generate a risk level report according to the preset risk assessment standards. The prediction module is used to construct a road surface deformation prediction model based on regression models and machine learning, and output the road surface deformation trend and generate a road surface deformation trend map through the road surface deformation prediction model. The visualization report and decision support module allows users to generate repair suggestions and visualization reports based on the risk level report and the road surface deformation trend map, which are then pushed to highway maintenance personnel via the Internet of Things system.

10. A computer-readable storage medium storing computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the Internet of Things-based highway safety detection method as described in any one of claims 1-8.

Images