A dynamic safety early warning method and device for a highway

Abstract

This application discloses a dynamic safety early warning method and device for highways, relating to the field of intelligent transportation technology, to address the problems of low efficiency and high latency in existing early warning systems. The method includes: acquiring real-time traffic data of a target highway area using multi-dimensional detection equipment to preliminarily predict the traffic risk level of the target highway area; determining whether to trigger a drone inspection of the target highway area based on the traffic risk level and the equipment coverage of the target highway area to obtain a road surface point cloud of the target highway area; determining a set of obstacle parameter values ​​for the target highway area based on the road surface point cloud; and segmenting the target highway area according to the obstacle parameter value set to evaluate diversion early warning for each segment of the target highway area based on traffic flow-related data and obstacle parameter values.

Description

Technical Field

[0001] This application relates to the field of intelligent transportation technology, and in particular to a dynamic safety early warning method and device for highways. Background Technology

[0002] With the continuous expansion of the expressway network and the sustained increase in traffic flow, expressway operation and management face new challenges. On the one hand, the proportion of heavy-duty logistics vehicles has increased significantly, and the weight of special vehicles such as coal trucks or falling objects on special road sections may cause cumulative damage to the road structure, creating road obstacles that hinder normal traffic flow. On the other hand, during holidays and other special periods, traffic flow experiences explosive growth, and these road obstacles may easily form traffic bottlenecks in local sections. Therefore, timely and accurate assessment of road conditions in expressway areas and timely diversion and early warning for potential risk areas are crucial to ensuring the safe and efficient operation of expressways.

[0003] Traditional road inspection methods often rely on periodic manual patrols or fixed sensor detection on key road sections for traffic diversion. However, these methods are inefficient and suffer from drawbacks such as response lag and insufficient coverage, making it difficult to detect sudden road anomalies in a timely manner. Furthermore, existing highway safety warnings are typically based on traffic volume data collected over a certain historical period. This reliance on historical data is unsuitable for special circumstances, such as surges in traffic during holidays or heavy truck traffic. Moreover, existing warning systems primarily focus on road structural safety assessments and lack real-time response capabilities to dynamic changes in traffic flow. They are unable to quickly respond to traffic diversion designs when road anomalies occur, potentially leading to the spread of localized congestion and even secondary accidents. Summary of the Invention

[0004] To address the aforementioned technical problems, one or more embodiments of this application provide a dynamic safety early warning method and device for highways.

[0005] One or more embodiments of this application employ the following technical solutions:

[0006] One or more embodiments of this application provide a dynamic safety early warning method for highways, the method comprising:

[0007] Based on a multi-dimensional detection device for a target highway area, traffic data of the target highway area is acquired in real time to predict the traffic risk level of the target highway area; wherein, the traffic data includes: traffic flow related data and road network related data;

[0008] Based on the traffic risk level and the equipment coverage of the target highway area, it is determined whether to trigger a drone to inspect the target highway area in order to obtain the road surface point cloud of the target highway area.

[0009] Based on the local curvature of the road in the target high-speed area, the flat area and the obstruction area corresponding to the road surface point cloud are determined, so as to determine the set of obstacle parameter values ​​of the target high-speed area according to the road surface point cloud corresponding to the obstruction area and its adjacent platform area.

[0010] The target highway area is segmented based on the set of obstacle parameter values, and a diversion warning assessment is performed on each segment of the target highway area based on the traffic flow data and obstacle parameter values ​​corresponding to each segment.

[0011] Optionally, in one or more embodiments of this application, the multi-dimensional detection device based on the target highway area acquires traffic data of the target highway area in real time, specifically including:

[0012] Based on the equipment tags of the entrance and exit gantry equipment in the target highway area, determine the linked road network equipment corresponding to each entrance and exit gantry equipment within a preset range; wherein, the linked road network equipment includes: traffic monitoring equipment, toll collection related equipment, and road environment monitoring equipment;

[0013] Obtain the initial traffic flow data and initial road network data recorded by the entrance / exit gantry equipment and the linked road network equipment within the current detection period;

[0014] The initial traffic flow data and the initial road network data are time-domain aligned to determine the aligned traffic flow data and the aligned road network data, which are then used as the initial traffic data.

[0015] Obtain the real-time detection data uploaded by the vehicles corresponding to the initial traffic data, and determine the vehicle ID and passage time window corresponding to each real-time detection data.

[0016] By using the vehicle ID and passage time window, the real-time detection data supplements the initial traffic data, thereby obtaining traffic data for the target highway area.

[0017] Optionally, in one or more embodiments of this application, predicting the traffic risk level of the target highway area specifically includes:

[0018] The custom configuration information is matched with the traffic flow related data and road network related data to determine the configuration item data and non-configuration item data of the traffic flow related data and road network related data based on the matching results;

[0019] Based on the custom configuration information corresponding to the configuration item data, determine the risk level of the configuration item data;

[0020] Based on a pre-set database, detection cycle data corresponding to the non-configuration item data is obtained, and the risk level corresponding to the non-configuration item data is determined according to the difference between the detection cycle data and the non-configuration item data.

[0021] Based on the risk level of each configuration item data and the risk level corresponding to each non-configuration item data, the traffic risk level of the target highway area is predicted.

[0022] Optionally, in one or more embodiments of this application, determining whether to trigger a drone to inspect the target highway area based on the traffic risk level and the equipment coverage of the target highway area to obtain a road surface point cloud of the target highway area specifically includes:

[0023] If it is determined that the traffic risk level is greater than or equal to the preset traffic risk level, and the equipment coverage is less than the preset coverage, then the UAV is triggered to inspect the target highway area, so as to obtain the initial road acquisition image collected by the multi-view camera based on the multi-view camera carried by the UAV.

[0024] If it is determined that the traffic risk level is less than the preset traffic risk level, or the equipment coverage is greater than the preset coverage, then the initial road acquisition image collected by the preset acquisition equipment is obtained.

[0025] The initial road acquisition image is preprocessed to obtain the road acquisition image of the target highway area within the current period; wherein, the preprocessing includes: image filtering, image enhancement, and noise reduction;

[0026] The road acquisition images are used to perform three-dimensional reconstruction to generate three-dimensional point cloud data of the target highway area, and the road surface point cloud within the three-dimensional point cloud data is extracted based on a preset hierarchical segmentation strategy.

[0027] Optionally, in one or more embodiments of this application, the set of obstacle parameter values ​​for the target highway area is determined based on the road surface point cloud corresponding to the obstruction area and its adjacent platform areas, specifically including:

[0028] The height of the road obstacle is determined based on the vertical coordinate difference between the road 3D point cloud of the obstructing area and the road 3D point cloud of the adjacent flat area; wherein, the height of the road obstacle includes positive and negative values.

[0029] The three-dimensional point cloud of the obstruction area is projected onto a two-dimensional plane to determine the area of ​​the road defect based on the area corresponding to the closed region.

[0030] The defect density of the target high-speed region is determined based on the distance between each closed region.

[0031] Based on the road defect depth, road defect area, and defect density corresponding to each of the aforementioned obstruction areas, a set of defect parameter values ​​for the target highway area is determined.

[0032] Optionally, in one or more embodiments of this application, the target highway area is segmented according to the set of obstacle parameter values, so as to evaluate the diversion warning for each segment of the target highway area based on the traffic flow related data and obstacle parameter values ​​corresponding to each segment of the target highway area, specifically including:

[0033] Based on the set of obstacle parameter values, the segment boundary information of the target high-speed area is determined, and the target high-speed area is segmented based on the segment boundary information and the preset segment distance to obtain the target high-speed area segments;

[0034] The target highway area containing the obstruction area is taken as the initial diversion section. Based on the obstruction parameter value corresponding to the initial diversion section, the initial diversion section is added to the diversion section.

[0035] Based on the traffic flow data corresponding to the road segment to be diverted, and the traffic flow data corresponding to the road segments adjacent to the road segment to be diverted, the number of diverted traffic corresponding to each target road area is predicted.

[0036] Based on the warning level corresponding to the number of traffic diversions, a warning mechanism corresponding to the warning level is triggered.

[0037] Optionally, in one or more embodiments of this application, generating three-dimensional point cloud data of the target highway area by performing three-dimensional reconstruction on the road acquisition images specifically includes:

[0038] The road acquisition image is subjected to continuous Gaussian blurring to determine the road feature points and road feature descriptors of the road acquisition image. Based on the road feature descriptors, the road feature points are matched to obtain matching feature points.

[0039] Based on the pose of the multi-view camera and the matching feature points, the road acquisition image is matched with the three-dimensional actual structural model of the target highway area to construct a sparse point cloud of the road in the target highway area.

[0040] Interpolation processing is performed on the sparse point cloud of the road to update the sparse point cloud and obtain three-dimensional point cloud data of the target highway area.

[0041] Optionally, in one or more embodiments of this application, extracting road surface point clouds within the 3D point cloud data based on a preset hierarchical segmentation strategy specifically includes:

[0042] Based on the road structure features corresponding to the target high-speed area, the three-dimensional point cloud data is divided into spatial regions to obtain the three-dimensional point cloud data of the road reference area.

[0043] Local point sets of the road 3D point cloud data are obtained, and the road surface normal direction of each local point set is calculated based on principal component analysis;

[0044] Based on the angle threshold between the road surface normal direction and the vertical direction, the local point set is filtered to determine the road surface point cloud based on the filtered local point set.

[0045] Optionally, in one or more embodiments of this application, before acquiring traffic flow-related data and road network-related data of the target highway area in real time based on multi-dimensional detection information of the target highway area, the method further includes:

[0046] Based on the vehicle entry and exit data of the highway within a preset detection period, determine the vehicle entry and exit data corresponding to each section of the highway.

[0047] Based on the number of vehicle tags contained in the vehicle entry and exit data, the freight rate corresponding to each road segment is determined.

[0048] If the freight ratio is determined to be greater than the preset freight ratio threshold, the road segment area is determined to be a heavy-load vehicle concentration area, and the road segment area corresponding to the adjacent heavy-load vehicle concentration area is taken as the target highway area.

[0049] One or more embodiments of this application provide a dynamic safety early warning device for highways, the device comprising:

[0050] At least one processor; and,

[0051] A memory communicatively connected to the at least one processor; wherein,

[0052] The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform any of the methods described above.

[0053] The above-described technical solutions adopted in the embodiments of this application can achieve the following beneficial effects:

[0054] By acquiring real-time traffic flow and road network data through multi-dimensional detection equipment, abnormal conditions in high-speed areas can be dynamically captured, significantly improving the timeliness and accuracy of risk identification compared to traditional fixed-cycle detection methods. Furthermore, linking traffic risk levels with equipment coverage in drone inspection decisions avoids the resource waste of fixed inspections and can precisely trigger supplementary detections for blind spots or high-risk road sections, achieving a collaborative optimization strategy for both risk and coverage. Local curvature analysis based on point cloud data can distinguish between flat and obstructed areas. By combining point cloud features from adjacent platform areas, interference from regular road undulations can be eliminated, accurately extracting the three-dimensional parameters of real obstacles and improving the robustness of obstacle detection. Segmented matching of obstacle parameters with traffic flow allows for a quantitative assessment of the urgency of traffic diversion in each road segment, providing data support for differentiated traffic diversion design. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0056] Figure 1 A schematic flowchart of a dynamic safety early warning method for highways provided in this application embodiment;

[0057] Figure 2 A schematic diagram of the structure of a dynamic safety early warning device for highways provided in an embodiment of this application;

[0058] Figure 3 This is a schematic diagram of the structure of a non-volatile storage medium provided in an embodiment of this application. Detailed Implementation

[0059] This application provides a method, device, and medium for dynamic safety early warning of highways.

[0060] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application.

[0061] This application provides a flowchart illustrating a dynamic safety early warning method for highways, comprising... Figure 1As can be seen, in one or more embodiments of this application, a dynamic safety early warning method for highways includes the following steps:

[0062] S101: A multi-dimensional detection device based on a target highway area acquires traffic data of the target highway area in real time to predict the traffic risk level of the target highway area; wherein, the traffic data includes: traffic flow related data and road network related data.

[0063] In the intelligent traffic diversion decision-making process on highways, this application aims to identify potential congestion and accident risks caused by factors such as sudden changes in traffic flow, heavy vehicle aggregation, or road network bottlenecks in advance, providing reliable data support for subsequent precise decision-making. In this embodiment, multi-dimensional detection equipment deployed in the target highway area collects real-time traffic flow data such as vehicle volume, speed, and vehicle type composition, as well as road network data such as lane number and entrance / exit distribution. Through analysis of this traffic data, the traffic risk level of the target highway area is predicted. This process overcomes the shortcomings of low efficiency and delayed response of manual inspections, achieving preliminary real-time risk perception. Furthermore, by acquiring traffic flow and road network data, the fusion of dynamic traffic flow and static road network data is achieved, providing accurate and reliable data support for subsequent precise diversion assessments.

[0064] Furthermore, with the booming development of the logistics industry, the proportion of freight vehicles in highway traffic flow is increasing, especially the frequent passage of heavy-duty vehicles. On the one hand, due to their heavy-duty characteristics, they put enormous pressure on the road surface; on the other hand, there is a risk that some cargo may fall off and obstruct traffic. Therefore, in order to achieve more targeted and concentrated analysis of areas with concentrated heavy-duty vehicles, in one or more embodiments of this application, before acquiring traffic flow-related data and road network-related data of the target highway area in real time based on multi-dimensional detection information of the target highway area, the method further includes:

[0065] Based on vehicle entry and exit data of highways within a pre-set detection period, such as daily, weekly, or monthly, the vehicle entry and exit data corresponding to each highway segment is determined. This pre-set detection period can be determined based on historical risk records of the highway; if the number of risk events is high, the interval between detection periods will be correspondingly reduced. The vehicle entry and exit data acquired in this process typically comes from toll station records, ETC systems, or road monitoring equipment, enabling accurate statistics on vehicle traffic conditions at different time periods. Then, based on the number of vehicle tags included in the vehicle entry and exit data, the freight ratio corresponding to each highway segment is determined. That is, by identifying vehicle tags such as vehicle type classification, license plate information, or load markings, the distribution of passenger and freight vehicles can be further distinguished, thereby calculating the freight vehicle ratio of each highway segment. If the freight ratio is determined to be greater than a pre-set freight ratio threshold, the highway segment is identified as a concentrated area of ​​heavy-load vehicles, and this highway segment is designated as the target highway area. First, the mechanism of dynamically adjusting the detection cycle based on historical risk records breaks through the limitations of fixed detection frequencies, realizes intelligent matching between monitoring frequency and risk level, and significantly improves resource utilization efficiency. Second, by integrating multi-source data such as toll stations, ETC and monitoring equipment, it overcomes the shortcomings of incomplete coverage from a single data source, ensures the integrity and accuracy of vehicle statistics, and helps to achieve targeted monitoring and diversion early warning for high-risk road sections.

[0066] Specifically, in one or more embodiments of this application, the multi-dimensional detection device based on the target highway area acquires traffic data of the target highway area in real time, specifically including:

[0067] When acquiring traffic data for a target highway area, the system first identifies the corresponding traffic monitoring equipment, toll collection equipment, road environment monitoring equipment, and other linked road network equipment within the preset range of the gantry equipment at the entrances and exits of the target highway area, based on the equipment tags of these devices. This solves the problems of scattered data sources and poor equipment coordination in traditional detection methods. Furthermore, the system achieves unified access to multi-source data through equipment linkage, providing comprehensive data support for subsequent analysis. Next, the system acquires the initial traffic flow-related data and initial road network-related data recorded by the gantry equipment at the entrances and exits and the linked road network equipment within the current detection cycle. The initial traffic flow-related data and initial road network-related data are then time-domain aligned to ensure spatiotemporal consistency, and the aligned traffic flow-related data and aligned road network-related data are determined as the initial traffic data. Then, to further improve data accuracy, this embodiment also acquires real-time detection data uploaded by vehicles corresponding to the initial traffic data, such as vehicle GPS trajectories and axle load information. Based on the vehicle ID and passage time window corresponding to each real-time detection data, the real-time detection data supplements the initial traffic data, obtaining the traffic data for the target highway area. The supplementation of the initial traffic data with real-time detection data significantly improves the accuracy of the preliminary assessment of subsequent traffic risk levels. It also avoids the problem of insufficient detection accuracy when relying solely on fixed sensors for detection.

[0068] Furthermore, to address the pain points of heterogeneous data formats and inconsistent quality from multiple sources, this application, after supplementing the initial traffic data with real-time detection data based on the vehicle ID and passage time window corresponding to each real-time detection data, will further process the supplemented initial traffic data through format standardization rules to ensure data usability and provide high-quality input for subsequent analysis. Specifically: First, to address the low processing efficiency and lack of targeted data processing caused by the mixed types of multi-source data, the supplemented initial traffic data will be classified and identified based on various data sources in the target highway area, such as gantry equipment, vehicle terminals, and environmental sensors, to clarify its data type. This can be addressed by including: structured passage records, unstructured surveillance videos, or semi-structured sensor logs, etc. Subsequently, since traffic data in highway business scenarios does not exist in isolation. For example, truck traffic data is directly related to road load pressure data and indirectly related to road wear and maintenance cycles. Therefore, to form a complete business event chain, clearly present the causal and logical relationships between data, and clarify the flow and role of data in the business process, further processing is required. This application embodiment analyzes the relationships between supplemented initial traffic data in different business scenarios, such as toll collection and auditing, traffic dispatching, and maintenance decision-making, constructing a business event chain model to distinguish between directly and indirectly related data. This overcomes the limitations of traditional single-business data processing, revealing the potential impact patterns between data through cross-business correlation analysis and providing a multi-dimensional perspective for data value mining. Based on the business event chains involved in each road network-related data, it determines the data grading elements associated with each supplemented initial traffic data, such as data update frequency and the degree of impact on business decisions. The data grading corresponding to each axle load-related data is then determined by the frequency of occurrence of each data association element. Based on the data type and data grading, a pre-set data standard file is queried to obtain the format standardization processing rules corresponding to each supplemented initial traffic data. This solves the problem of inconsistent data formats, making integration, analysis, and sharing difficult. It avoids errors in risk prediction for target highway areas due to chaotic data formats, thus affecting the effective utilization of data.

[0069] Furthermore, in one or more embodiments of this application, predicting the traffic risk level of a target high-speed area specifically includes:

[0070] Traditional risk assessments use a uniform processing method for all data, making it impossible to distinguish between key and auxiliary indicators, resulting in low assessment efficiency. To improve the scenario adaptability of preliminary predictions, custom configuration information uploaded by users based on actual scenarios can be received. This custom configuration information can then be matched with traffic flow-related data and road network-related data to determine the configuration and non-configuration items for these data based on the matching results.

[0071] For categorized configuration item data, the risk level is determined based on the user-uploaded custom configuration settings. For example, a user might specify in the configuration that the risk level is high when the volume of heavy trucks exceeds a certain threshold; in this case, the risk level corresponding to heavy truck volume data meeting this condition is determined to be high. This method applies the user's personalized risk assessment criteria to the configuration item data, assigning it a corresponding risk level. For non-configuration item data, the system retrieves multiple related detection period data from a pre-set database. These detection period data record the performance and corresponding risk of similar data under different conditions. By analyzing the difference between these detection period data and the non-configuration item data, past experience and patterns can be referenced to determine the current risk level of the non-configuration item data. For example, the risk level of the current traffic volume data can be determined based on the traffic risk level corresponding to the same traffic volume data in the past. After determining the risk levels of both configuration item and non-configuration item data, a preliminary prediction of the traffic risk level for the entire target highway area is made by comprehensively considering the risk levels of all data; the weight of this comprehensive assessment is not limited here. The above process addresses the problem that a single data risk level cannot comprehensively reflect the overall risk situation of highway traffic. It avoids the issue of biased judgments about highway traffic risks caused by relying on only partial data or a single standard, thus failing to provide comprehensive and effective information for highway traffic management.

[0072] S102: Based on the traffic risk level and the equipment coverage of the target highway area, determine whether to trigger a drone to inspect the target highway area in order to obtain the road surface point cloud of the target highway area.

[0073] To avoid wasting inspection resources or missing key areas due to single-dimensional judgment, and to address the monitoring blind spot problem caused by insufficient equipment coverage, this embodiment of the application integrates traffic risk level and equipment coverage as dual-dimensional parameters to achieve dynamic allocation of inspection resources. Specifically, it determines whether the preset inspection triggering conditions are met based on the traffic risk level obtained in step S101 and the equipment coverage of the target highway area. By incorporating UAV point cloud data acquisition, it compensates for the insufficient monitoring accuracy of traditional monitoring methods, forming a data acquisition method that combines routine monitoring with emergency inspections, thereby improving the accuracy of subsequent traffic diversion. In this way, risk assessment results are closely linked to UAV inspection operations, ensuring that inspection work can be flexibly carried out according to the actual road risk conditions. The flexible inspection using UAVs avoids the problems of incomplete coverage and insufficient detection accuracy associated with existing methods that rely solely on setting up sensors on certain key road sections or structures on highways, due to the limited deployment locations of fixed sensors, which can only monitor local road conditions. It also avoids the high costs and poor real-time performance issues caused by relying solely on UAV inspections.

[0074] Specifically, in one or more embodiments of this application, determining whether to trigger a drone inspection of the target highway area is based on the traffic risk level and the equipment coverage of the target highway area, in order to obtain a road surface point cloud of the target highway area, specifically includes:

[0075] Traditional inspection methods rely on fixed cycles or manual triggering, failing to dynamically adjust to actual risks, leading to resource waste or monitoring blind spots. Full-process drone inspections also suffer from excessive costs. Therefore, this application addresses this issue by real-time monitoring of the traffic risk level and equipment coverage of the target highway area. If the traffic risk level is determined to be greater than or equal to a preset traffic risk level, and the equipment coverage is less than a preset coverage, a drone inspection of the target highway area is triggered. This allows the drone to acquire initial road images from its multi-view cameras. These multi-view cameras consist of multiple cameras capable of simultaneously capturing road conditions from different angles and positions. Conversely, if the traffic risk level is determined to be less than a preset traffic risk level, or the equipment coverage is greater than a preset coverage, initial road images acquired by pre-set acquisition equipment are obtained.

[0076] The initial road images acquired may contain poor-quality images due to factors such as poor shooting angle, poor lighting conditions, and equipment shake, and may also contain noise interference. Therefore, preprocessing of the initial images is necessary to obtain road images of the target highway area within the current period. Preprocessing includes image filtering, image enhancement, and noise reduction. Since road surface point clouds can clearly present features such as road surface conditions, this application performs 3D reconstruction of the road images to generate 3D point cloud data of the target highway area, and extracts road surface point clouds from the 3D point cloud data based on a preset hierarchical segmentation strategy. This solves the problems of insufficient comprehensive and accurate assessment of road traffic 3D conditions relying solely on 2D images, and the presence of a large amount of irrelevant information in the 3D point cloud data affecting the accuracy of road traffic feature extraction.

[0077] Specifically, in one or more embodiments of this application, the process of generating three-dimensional point cloud data of the target highway area by performing three-dimensional reconstruction on road acquisition images includes the following steps:

[0078] First, the road acquisition images are smoothed at multiple scales through continuous Gaussian blurring, effectively suppressing noise interference and enhancing the stability of key features. Based on an improved feature detection algorithm, scale- and rotation-invariant road feature points are extracted, generating highly discriminative feature descriptors. Cross-view feature point matching is achieved through descriptor similarity measurement, establishing a reliable correspondence foundation for subsequent 3D reconstruction. This process specifically optimizes the feature extraction strategy for low-texture highway surfaces, ensuring a sufficient number of stable feature points are obtained even on homogeneous surfaces such as asphalt. Then, to address the difficulty of directly integrating 2D images into 3D space and the inability to quickly construct a 3D road point cloud framework, and to avoid the chaotic and inefficient 3D point cloud construction caused by the lack of an effective matching method between images and 3D models during 3D reconstruction, this embodiment matches the road acquisition images with the actual 3D structural model of the target highway area to construct a sparse point cloud of the target highway area. The actual 3D structural model of the target highway area can be obtained based on the building data corresponding to the target highway area recorded in the database. Then, supplementary 3D points are generated in the empty areas of the sparse point cloud using an adaptive interpolation process. It should be noted that this process can combine the road surface curvature obtained in the above steps to constrain it, and at the same time perform continuous smoothing prior, so as to realize the geometric rationality verification of the interpolation points, avoid surface distortion caused by over-interpolation, and finally output complete three-dimensional point cloud data of the target high-speed area.

[0079] In one feasible embodiment of this specification, three-dimensional reconstruction of road acquisition images is performed to generate three-dimensional point cloud data of the target highway area, specifically including:

[0080] To reduce image noise, smooth the image, and make the features in the image more prominent and stable, avoiding the influence of overly complex image details on subsequent analysis, this application first performs continuous Gaussian blur processing on the road acquisition image to determine the road feature points and road feature descriptors. Based on the road feature descriptors, the road feature points are matched to obtain the first matching feature points, which provide the basic connection points for subsequent 3D model construction. Then, to address the problem that 2D images are difficult to directly integrate into 3D space and that it is impossible to quickly construct a 3D point cloud framework for the road, and to avoid the problem of chaotic and inefficient 3D point cloud construction due to the lack of an effective matching method between the image and the 3D model during 3D reconstruction, this application embodiment matches the road acquisition image with the actual 3D structural model based on the pose of the multi-view camera and the first matching feature points, constructing a sparse point cloud of the road in the target highway area. This sparse point cloud can reflect the general outline and key structure of the road surface. Next, the epipolar geometric constraints of the road-captured images are determined based on the pose of the multi-view camera. Then, based on the road segment information of the target highway area, the search range constraints of the road-captured images are determined. Based on the search range constraints and epipolar geometric constraints, road feature points in the road-captured images are filtered to obtain the areas to be thickened corresponding to the filtered road feature points. In this process, the epipolar geometric constraints limit the straight-line range within which the corresponding feature points should exist in images from different viewpoints, narrowing the search space for feature point matching. Simultaneously, combined with the road segment information of the target highway area, such as the length, width, and direction of the road, the search range constraints are determined, clarifying which areas belong to the road itself and which are irrelevant background areas. Through these two constraints, the road feature points in the road-captured images are filtered, eliminating feature points that do not meet the conditions. The areas corresponding to the remaining feature points are the areas to be thickened. By filtering feature points through dual constraints, the efficiency and accuracy of feature point matching are significantly improved, accurately locating the areas to be thickened that require further processing and reducing interference from invalid data.

[0081] Because multi-view cameras capture images from different angles, the position of the same object can vary in different images. Therefore, it is necessary to determine the viewing angle at which each road image in the area to be thickened was captured. By analyzing the matching relationships of filtered road feature points under different viewing angles, the relative positional changes of these feature points in different images are identified, thereby determining the disparity information between the second matching feature points. Disparity information reflects the depth variation of an object in three-dimensional space; for example, objects closer to the camera have greater disparity, while objects farther away have smaller disparity. It is key data for converting two-dimensional image information into three-dimensional depth information. Based on the obtained disparity information and the camera's imaging principle, the three-dimensional point data of the area to be thickened can be determined. This three-dimensional point data is then used to update the sparse road point cloud, obtaining the three-dimensional point cloud data of the target highway area. These newly generated three-dimensional point data are then integrated into the previously constructed sparse road point cloud, supplementing and updating the sparse point cloud, increasing its density and detail, ultimately resulting in complete and accurate three-dimensional point cloud data of the target highway area. This solves the problem of incomplete and detail-lacking sparse road point cloud data, which cannot meet the needs of accurate road condition analysis.

[0082] Furthermore, in one or more embodiments of this application, extracting road surface point clouds within 3D point cloud data based on a preset hierarchical segmentation strategy specifically includes:

[0083] Each highway has unique structural features, such as lane distribution, roadbed shape, and pavement slope. Based on these features, the space represented by the 3D point cloud data is rationally divided. During the division, areas closely related to the pavement structure are marked as pavement reference areas, from which the corresponding 3D pavement point cloud data is extracted. This solves the problem that 3D point cloud data contains a large amount of irrelevant information, making it difficult to quickly locate pavement areas, increasing data processing difficulty and computational resource consumption. It also avoids interference from excessive irrelevant data during the analysis process, which would reduce work efficiency. Then, local point sets of the pavement 3D point cloud data are obtained, and the surface normal direction of each local point set is calculated using principal component analysis. That is, principal component analysis performs mathematical operations on the coordinate data of points in the local point set to find the direction with the greatest data change. This direction corresponds to the surface normal direction of that local area. The surface normal direction reflects the inclination and orientation of the pavement in that local area. For example, the normal direction of a horizontal pavement is close to vertical, while the normal direction of a sloping pavement will have a corresponding inclination.

[0084] By calculating the surface normal direction of each local point set, the morphological characteristics of different locations on the road surface can be described in detail. Then, based on the actual road conditions, a threshold angle between the surface normal direction and the vertical direction is set. For each local point set, the calculated surface normal direction is compared with the set threshold angle to the vertical direction. If the angle is within the threshold range, it indicates that the area represented by the local point set conforms to the characteristics of a traffic road surface, such as a near-horizontal road surface or a road surface area with a reasonable slope; otherwise, the local point set is considered to possibly belong to a non-road surface area and is filtered out to obtain a road surface point cloud. By setting an angle threshold to filter local point sets, the road surface point cloud can be accurately extracted, effectively eliminating interference from non-road surface points and improving the accuracy and purity of the road surface point cloud data.

[0085] S103: Based on the local curvature of the road in the target high-speed area, determine the flat area and the obstruction area corresponding to the road surface point cloud, so as to determine the set of obstacle parameter values ​​of the target high-speed area according to the road surface point cloud corresponding to the obstruction area and its adjacent platform area.

[0086] Traditional methods based on height thresholds are prone to missing low-lying or locally protruding obstacles such as bricks and tire fragments. Therefore, this application determines the flat and obstructed areas corresponding to the road surface point cloud based on the local curvature of the target highway area. The local curvature is determined by selecting a certain number of neighboring points for each point cloud data point. The selection of these neighboring points can employ the K-nearest neighbor algorithm or the radius neighborhood search algorithm, which will not be elaborated here. Then, using the selected neighboring points, a local surface is fitted using the least squares method. Based on the fitted local surface, the local curvature of each point is calculated using Gaussian curvature. A reasonable curvature threshold is then set based on actual road conditions and experience. It is understood that flat areas have lower road surface curvature, while areas with obstacles or other defects have higher curvature. The calculated curvature value for each point is compared with the set threshold. If the curvature value of a point is less than the threshold, the area where that point is located is determined to be a flat area; if the curvature value of a point is greater than or equal to the threshold, the area where that point is located is determined to be an obstructed area. Furthermore, different colors or markers can be used to visually distinguish the point clouds of flat and obstructed areas for easier viewing and analysis later. After identifying the flat and obstructed areas, the set of obstacle parameter values ​​for the target highway area can be determined by comparing the 3D point cloud of the road in the obstructed area with the corresponding road surface point cloud of the adjacent platform area.

[0087] Specifically, in one or more embodiments of this application, the set of obstacle parameter values ​​for the target highway area is determined based on the road surface point cloud corresponding to the obstruction area and its adjacent platform areas, specifically including:

[0088] The vertical coordinates of the road obstacle's 3D point cloud and the adjacent flat area's 3D point cloud are extracted. The depth of the obstacle is determined by calculating the difference between their vertical coordinates. If the vertical coordinates of the obstacle's point cloud are lower than those of the adjacent flat area (a negative difference indicates a concave obstacle), and if the vertical coordinates are higher than those of the adjacent flat area (a positive difference indicates a convex obstacle), the length of the obstacle can be determined using the horizontal coordinates of both the obstacle and the adjacent flat area's 3D point clouds. The 3D point cloud data of the obstacle area is then projected onto a 2D plane, mapping points in 3D space to a 2D coordinate system according to certain rules. This transforms the complex shape of the obstacle in 3D space into a closed region on a 2D plane. The area covered by this closed region is calculated to determine the area of ​​the road obstacle. After identifying multiple 2D closed regions corresponding to obstacle areas, the distances between these closed regions are measured. By analyzing these distance data, the number or distribution of obstacle areas per unit area is calculated, thus determining the obstacle density of the road. For example, if multiple closely spaced closed obstacle zones exist within a small area, it indicates high obstacle density and poor road conditions in that section of road; conversely, if the closed obstacle zones are far apart and scattered, the obstacle density is low. The parameters obtained for each obstacle zone, such as road obstacle height, length, area, and density, are integrated to form a complete set of obstacle parameter values. This set comprehensively describes the characteristics of road obstacles from multiple dimensions, solving the problem that a single obstacle parameter cannot fully reflect the true road conditions.

[0089] S104: The target highway area is segmented according to the set of obstacle parameter values, so as to evaluate the diversion warning of each target highway area based on the traffic flow related data and obstacle parameter values ​​corresponding to each segment of the target highway area.

[0090] To avoid the inefficient approach of applying uniform warnings to the entire highway area using traditional methods, this application aims to achieve targeted control of high-risk sections. It also addresses the limitations of relying solely on obstacle parameters or traffic flow data, as small obstacles may pose a higher risk in high-traffic areas. This embodiment divides the target highway area into several segments based on a set of obstacle parameter values. Combining real-time traffic flow data for each segment, such as vehicle density, speed, and vehicle type distribution, a diversion warning level is determined for each segment to assist in generating dynamic diversion plans. Adjusting the warning strategy based on real-time traffic conditions improves the effectiveness of warnings and helps to implement diversion only on high-risk segments, reducing the impact on overall road network efficiency. Furthermore, by diverting high-traffic and high-obstacle-density segments in advance, the probability of rear-end collisions or loss of control accidents can be effectively reduced.

[0091] Specifically, in one or more embodiments of this application, the target highway area is segmented according to the set of obstacle parameter values, so as to evaluate the diversion warning for each segment of the target highway area based on the traffic flow related data and obstacle parameter values ​​corresponding to each segment of the target highway area, specifically including:

[0092] To address the issue of fixed segmentation failing to adapt to the random distribution of obstacles and improve the identification accuracy of high-risk road sections, this process first determines the segment boundary information of the target highway area based on the obstacle parameter value set. Then, based on the segment boundary information and the preset segment distance, the target highway area is segmented to obtain segmented sections. Next, the target highway area containing the obstructed area is used as the initial diversion segment. Based on the obstacle parameter values ​​corresponding to the initial diversion segment, it is added to the waiting diversion segment. Then, based on the traffic flow data corresponding to the waiting diversion segment and the traffic flow data corresponding to adjacent road segments, the diversion quantity corresponding to each target road area is predicted. Based on the warning level corresponding to the diversion quantity, a warning mechanism corresponding to the warning level is triggered. For example, if cargo falls and remains lodged in certain road sections, the segment boundary information of the target highway area is first determined based on the obstacle parameter value set. For example, if an obstacle is detected between kilometer 100 and kilometer 102, a comprehensive analysis of the terrain and traffic conditions of this area and adjacent road segments is conducted. K99 and K103 are then used as segment boundaries. Based on a pre-set segment distance of 2 kilometers per segment, K99-K101 and K101-K103 are divided into two target highway segments. Next, the target highway segment K99-K103, which includes the obstruction area K100-K102, is used as the initial diversion segment. Based on obstacle parameters such as obstacle height, defect area, and defect density, the risk level is assessed, and the initial diversion segment is added to the diversion segment. Then, traffic flow data for the diversion segment K99-K103, as well as traffic flow data for adjacent road segments K97-K99 and K103-K105, are obtained. If the current hourly traffic flow of the road segment to be diverted is 3,000 vehicles, while the hourly traffic flow of the adjacent road segment is only 1,500 vehicles, and the congestion index of the road segment to be diverted continues to rise, then based on historical diversion data, it can be predicted that 1,000 vehicles should be diverted from the current road segment to the adjacent road segment to alleviate traffic pressure. Based on the number of vehicles diverted, the system determines that the corresponding warning level is orange, and then triggers the warning mechanism corresponding to the orange warning, such as releasing real-time traffic information to drivers through traffic radio and navigation software, prompting drivers to plan their routes in advance and exit from the nearest exit or take an alternate route.

[0093] like Figure 2 As shown in the diagram, this application provides a structural schematic diagram of a dynamic safety early warning device for highways. Figure 2 As can be seen, in one or more embodiments of this application, a dynamic safety early warning device for highways includes:

[0094] At least one processor; and,

[0095] A memory communicatively connected to the at least one processor; wherein,

[0096] The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform any of the methods described above.

[0097] like Figure 3 As shown in the diagram, this application provides a structural schematic of a non-volatile storage medium. Figure 3 It is understood that, in one or more embodiments of this application, a non-volatile storage medium stores computer-executable instructions 301, which are capable of executing any of the methods described above.

[0098] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and non-volatile computer storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0099] The foregoing has described specific embodiments of this application. Other embodiments are within the scope of the appended specification. In some cases, the actions or steps described in the specification may be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0100] The above description is merely one or more embodiments of this application and is not intended to limit this application. For those skilled in the art, various modifications and variations can be made to one or more embodiments of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of one or more embodiments of this application should be included within the scope of the claims of this application.

Claims

1. A dynamic safety early warning method for highways, characterized in that, The method includes: Based on a multi-dimensional detection device for a target highway area, traffic data of the target highway area is acquired in real time to predict the traffic risk level of the target highway area; wherein, the traffic data includes: traffic flow related data and road network related data; Based on the traffic risk level and the equipment coverage of the target highway area, it is determined whether to trigger a drone to inspect the target highway area in order to obtain the road surface point cloud of the target highway area. Based on the local curvature of the road in the target high-speed area, the flat area and the obstruction area corresponding to the road surface point cloud are determined, so as to determine the set of obstacle parameter values ​​of the target high-speed area according to the road surface point cloud corresponding to the obstruction area and its adjacent platform area. The target highway area is segmented according to the set of obstacle parameter values, and a diversion warning assessment is performed on each segment of the target highway area based on the traffic flow data and obstacle parameter values ​​corresponding to each segment of the target highway area. Before acquiring real-time traffic flow-related data and road network-related data for the target highway area based on multi-dimensional detection information of the target highway area, the method further includes: Based on the vehicle entry and exit data of the highway within a preset detection period, determine the vehicle entry and exit data corresponding to each section of the highway. Based on the number of vehicle tags contained in the vehicle entry and exit data, the freight rate corresponding to each road segment is determined. If it is determined that the freight ratio is greater than the preset freight ratio threshold, the road segment area is determined to be a heavy-load vehicle concentration area, and the road segment area corresponding to the adjacent heavy-load vehicle concentration area is taken as the target highway area. By combining the traffic risk level with the equipment coverage of the target highway area, it is determined whether to trigger a drone inspection of the target highway area to obtain a road surface point cloud of the target highway area, specifically including: If it is determined that the traffic risk level is greater than or equal to the preset traffic risk level, and the equipment coverage is less than the preset coverage, then the UAV is triggered to inspect the target highway area, so as to obtain the initial road acquisition image collected by the multi-view camera based on the multi-view camera carried by the UAV. If it is determined that the traffic risk level is less than the preset traffic risk level, or the equipment coverage is greater than the preset coverage, then the initial road acquisition image collected by the preset acquisition equipment is obtained. The initial road acquisition image is preprocessed to obtain the road acquisition image of the target highway area within the current period; wherein, the preprocessing includes: image filtering, image enhancement, and noise reduction; The road acquisition images are reconstructed in three dimensions to generate three-dimensional point cloud data of the target highway area, and the road surface point cloud is extracted from the three-dimensional point cloud data based on a preset hierarchical segmentation strategy. The target highway area is segmented based on the set of obstacle parameter values. Based on the traffic flow data and obstacle parameter values ​​corresponding to each segment of the target highway area, a diversion warning assessment is performed for each segment, specifically including: Based on the set of obstacle parameter values, the segment boundary information of the target high-speed area is determined, and the target high-speed area is segmented based on the segment boundary information and the preset segment distance to obtain the target high-speed area segments; The target highway area containing the obstruction area is taken as the initial diversion section. Based on the obstruction parameter value corresponding to the initial diversion section, the initial diversion section is added to the diversion section. Based on the traffic flow data corresponding to the road segment to be diverted, and the traffic flow data corresponding to the road segments adjacent to the road segment to be diverted, the number of diverted traffic corresponding to each target road area is predicted. Based on the warning level corresponding to the number of traffic diversions, trigger the warning mechanism corresponding to the warning level; The process of reconstructing the road images to generate 3D point cloud data of the target highway area specifically includes: The road acquisition image is subjected to continuous Gaussian blurring to determine the road feature points and road feature descriptors of the road acquisition image. Based on the road feature descriptors, the road feature points are matched to obtain matching feature points. Based on the pose of the multi-view camera and the matching feature points, the road acquisition image is matched with the three-dimensional actual structural model of the target highway area to construct a sparse point cloud of the road in the target highway area. Interpolation processing is performed on the sparse point cloud of the road to update the sparse point cloud and obtain three-dimensional point cloud data of the target highway area.

2. The dynamic safety early warning method for highways according to claim 1, characterized in that, Based on a multi-dimensional detection device for the target highway area, traffic data of the target highway area is acquired in real time, specifically including: Based on the equipment tags of the entrance and exit gantry equipment in the target highway area, determine the linked road network equipment corresponding to each entrance and exit gantry equipment within a preset range; wherein, the linked road network equipment includes: traffic monitoring equipment, toll collection related equipment, and road environment monitoring equipment; Obtain the initial traffic flow data and initial road network data recorded by the entrance / exit gantry equipment and the linked road network equipment within the current detection period; The initial traffic flow data and the initial road network data are time-domain aligned to determine the aligned traffic flow data and the aligned road network data, which are then used as the initial traffic data. Obtain the real-time detection data uploaded by the vehicles corresponding to the initial traffic data, and determine the vehicle ID and passage time window corresponding to each real-time detection data. By using the vehicle ID and passage time window, the real-time detection data supplements the initial traffic data, thereby obtaining traffic data for the target highway area.

3. The dynamic safety early warning method for highways according to claim 1, characterized in that, Predicting the traffic risk level of the target highway area specifically includes: The custom configuration information is matched with the traffic flow related data and road network related data to determine the configuration item data and non-configuration item data of the traffic flow related data and road network related data based on the matching results; Based on the custom configuration information corresponding to the configuration item data, determine the risk level of the configuration item data; Based on a pre-set database, detection cycle data corresponding to the non-configuration item data is obtained, and the risk level corresponding to the non-configuration item data is determined according to the difference between the detection cycle data and the non-configuration item data. Based on the risk level of each configuration item data and the risk level corresponding to each non-configuration item data, the traffic risk level of the target highway area is predicted.

4. The dynamic safety early warning method for highways according to claim 1, characterized in that, Based on the road surface point cloud corresponding to the obstruction area and its adjacent platform areas, the set of obstacle parameter values ​​for the target highway area is determined, specifically including: The height of the road obstacle is determined based on the vertical coordinate difference between the road 3D point cloud of the obstructing area and the road 3D point cloud of the adjacent flat area; wherein, the height of the road obstacle includes positive and negative values. The three-dimensional point cloud of the obstruction area is projected onto a two-dimensional plane to determine the area of ​​the road defect based on the area corresponding to the closed region. The defect density of the target high-speed region is determined based on the distance between each closed region. Based on the road defect depth, road defect area, and defect density corresponding to each of the aforementioned obstruction areas, a set of defect parameter values ​​for the target highway area is determined.

5. The dynamic safety early warning method for highways according to claim 1, characterized in that, The road surface point cloud within the 3D point cloud data is extracted based on a pre-defined hierarchical segmentation strategy, specifically including: Based on the road structure features corresponding to the target high-speed area, the three-dimensional point cloud data is divided into spatial regions to obtain the three-dimensional point cloud data of the road reference area. Local point sets of the road 3D point cloud data are obtained, and the road surface normal direction of each local point set is calculated based on principal component analysis; Based on the angle threshold between the road surface normal direction and the vertical direction, the local point set is filtered to determine the road surface point cloud based on the filtered local point set.

6. A dynamic safety early warning device for highways, characterized in that, The device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method described in any one of claims 1-5.

Images