Highway traffic management method and system based on highway disease dynamic early warning

Abstract

The invention discloses a highway traffic management method and system based on highway disease dynamic early warning, and relates to the technical field of intelligent traffic. Comprising the steps that S1, an unmanned aerial vehicle carries sensing, communication and calculation integrated roadside equipment to inspect along an expressway, and road surface image data are collected in real time; s2, the roadside equipment performs real-time processing on the pavement image data through a built-in intelligent image recognition model, recognizes an abnormal target and generates a unique identifier; and S3, acquiring image coordinates of an abnormal target, and converting the image coordinates into latitude and longitude coordinates in real time in combination with the positioning information of the roadside equipment. According to the invention, by using the maneuverability of the unmanned aerial vehicle, the sensing, communication and calculation integrated roadside equipment continues to fly along with the unmanned aerial vehicle in the direction opposite to the traffic flow after identifying the disease, dynamically updates the event confidence and continuously broadcasts the early warning information to the rear vehicle at the same time, so that the early warning coverage distance is increased while vehicle finding at the roadside is realized, and the occurrence of accidents is reduced.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, specifically to a high-speed traffic management method and system based on dynamic early warning of highway defects. Background Technology

[0002] With the continuous growth of highway mileage, road surface inspection and beyond-line-of-sight early warning of abnormal events have become increasingly prominent challenges. Traditional manual inspections are inefficient, costly, and risky. In existing technologies, fixed roadside sensing systems consist of discrete devices, resulting in high data processing latency. Furthermore, due to high costs and low deployment density, vehicles must enter the communication range of the roadside equipment to receive warnings, leading to severely insufficient driver reaction time in highway scenarios. While some existing technologies utilize drones for real-time road defect inspection, drones only serve as data collectors or communication relays, lacking edge computing capabilities and unable to achieve real-time identification and dynamic broadcasting of abnormal targets. This still fails to solve the problems of insufficient warning coverage and poor information timeliness.

[0003] For example, the invention patent with publication number CN120655646A discloses a method and system for real-time inspection of road surface defects based on unmanned aerial vehicles (UAVs). It uses a UAV to acquire and preprocess images of the road surface; an improved YOLOv5 algorithm is used to detect defect targets in the preprocessed images; an improved convolutional neural network is used to extract features from the detected defect targets; an improved Vision Transformer algorithm is used to classify and identify the extracted defect features to determine the type of defect; after the defect type is determined, a 3D reconstruction algorithm is used to assess the severity of the defect and generate a defect report.

[0004] However, when the above and similar technical solutions are actually applied to highway scenarios, they have the following shortcomings: the drones are only used as data collection or communication relay platforms, and the computing tasks such as image recognition still need to be completed by ground servers or the cloud, which makes it difficult to meet the real-time requirements of highway scenarios for abnormal event warnings; the drones only fly along preset routes to collect images, and the task is completed after the recognition results are generated, and their flight trajectory is disconnected from the warning broadcast. Summary of the Invention

[0005] The purpose of this invention is to provide a high-speed traffic management method and system based on dynamic early warning of highway defects, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a highway traffic management method based on dynamic early warning of highway defects, comprising: S1. The drone is equipped with an integrated roadside device that combines sensing, communication and computing to inspect the highway and collect road surface image data in real time. S2. The roadside equipment uses a built-in intelligent image recognition model to process the road surface image data in real time, identify abnormal targets, and generate unique identifiers. S3. Obtain the image coordinates of the abnormal target and, in conjunction with the positioning information of the roadside equipment, convert them into latitude and longitude coordinates in real time. S4. Calculate the event confidence level based on the real-time distance between the abnormal target and the roadside equipment and the time difference from the time the abnormal target was identified to the current time; S5. When the confidence level of the event is greater than the preset warning threshold, the abnormal target information is encapsulated into the vehicle network standard information format to obtain the warning information; it is dynamically broadcast to the vehicles behind through the long-term evolution vehicle network communication technology, while the drone continues to fly in the opposite direction of the traffic flow, so that the warning range is dynamically extended to the rear. S6. During the flight of the UAV, the real-time distance and time difference are updated in real time, the event confidence is dynamically adjusted, and the judgment and broadcast steps are repeatedly triggered until the preset termination conditions are met.

[0007] Furthermore, the intelligent image recognition model specifically includes: Real-time road surface image data is registered with a historical disease database. For the detected targets located within the historical disease location range, their current image coordinates and size are extracted and compared with previous size data to calculate the size expansion rate. Each disease record in the database has a unique identifier. Detection targets with a size expansion rate exceeding a preset expansion threshold are identified as expanding diseases and assigned a first priority label; detection targets with a size expansion rate below the preset expansion threshold are identified as stable diseases and assigned a second priority label; the identification results are updated in the historical record corresponding to the unique identifier of the disease. For detection targets located outside the historical disease location range, their texture features are extracted and compared to identify texture abrupt change areas; cross-frame morphological tracking is performed, and if the morphology shows a unidirectional extension trend, it is determined to be a new disease and assigned a third priority identifier; a new unique identifier is generated for the new disease and written into the database. For locations that exist in the historical defect database but are not detected in the current road surface image data, confirmation is performed across multiple frames; if no defect is detected at this location in multiple consecutive frames, and the image features at this location show that the road surface is smooth, it is determined that it has been repaired, and the defect record corresponding to the unique identifier is marked as removed; Output the exception target with a unique identifier and priority indicator.

[0008] Furthermore, the method for dynamically calculating event confidence is as follows:

[0009]

[0010]

[0011] Where C is the event confidence level, w d For distance weights, w t For time weighting, w p d represents the priority weight, d represents the real-time distance (in kilometers) between the abnormal target and the integrated sensing, communication, and computing roadside equipment, and t represents the time difference from the time the abnormal target was identified to the current time.

[0012] Furthermore, the method for calculating the real-time distance d is as follows:

[0013] Where r is the Earth's radius, (Ф1,λ1) are the latitude and longitude of the current location of the integrated sensing, communication, and calculation roadside equipment, and (Ф2,λ2) are the latitude and longitude of the location of the abnormal target.

[0014] Furthermore, the preset termination conditions include at least one of the following: the real-time distance is greater than the preset effective warning distance, the event confidence level is lower than the warning threshold for more than the preset duration, confirmation of receipt is received from a vehicle behind via the vehicle network, and an abnormality has been handled instruction is received from the operation platform.

[0015] Furthermore, the integrated sensing, communication, and computing roadside equipment adopts a core board and base board separation design, with the core board and base board connected through a high-speed backplane connector; the core board integrates a vehicle networking communication module and an intelligent computing module, and is connected through a serializer-deserializer interface.

[0016] Furthermore, it also includes coordinated early warning systems using multiple drones: The operation platform divides the highway into several continuous inspection sections with overlapping areas between adjacent sections. Each drone performs inspections within its assigned section. The first drone immediately established a tracking record after identifying the abnormal target; When the first drone needs to be decommissioned due to endurance or distance limitations, it sends a handover request to the operations platform. The operation platform dynamically calculates the rendezvous point location based on the real-time position, flight direction, and speed of each drone, and selects the second drone to rendezvous. After the first drone flies to the rendezvous point, it sends the tracking data to the second drone. The second drone continues to broadcast early warnings based on the received tracking records; The first drone ceased issuing warnings and returned to its nest.

[0017] Furthermore, the tracking record includes at least the unique identifier of the abnormal target, the historical confidence sequence, the current size expansion rate, and the latitude and longitude coordinates of the last detection; after receiving the tracking record, the second UAV performs a local search based on the latitude and longitude coordinates to determine the abnormal target that needs to be warned, and inherits the historical confidence based on its unique identifier.

[0018] A high-speed traffic management system based on dynamic early warning of highway defects uses any one of the above-mentioned high-speed traffic management methods based on dynamic early warning of highway defects.

[0019] Compared with the prior art, the beneficial effects of the present invention are: A highway traffic management method and system based on dynamic early warning of highway defects is proposed. By packaging defect information according to the V2X standard message set, the information is directly broadcast to vehicles behind using the sensing and communication computing equipment carried by UAVs. This enables the early warning information to reach every vehicle equipped with an on-board unit (OBU) point-to-point, achieving beyond-line-of-sight early warning. By having multiple UAVs fly in adjacent areas in a coordinated manner, the early warning mission can be seamlessly handed over, solving the problems of limited single-unit endurance and limited early warning distance.

[0020] Meanwhile, this invention employs an integrated roadside device that combines sensing, communication, and computing into a single device mounted on a drone. This allows for real-time local processing of collected road surface data during flight, solving the real-time problem of road defect early warning. By utilizing the drone's maneuverability, the integrated roadside device continues to fly along the opposite traffic flow after identifying defects, dynamically updating event confidence levels and continuously broadcasting warning information to vehicles behind. This achieves roadside vehicle location while increasing warning coverage distance and reducing the occurrence of accidents. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the hardware structure of the integrated sensing, communication, and computing roadside equipment of the present invention; Figure 2 This is a schematic diagram of the overall structure of the present invention; Figure 3 This is a schematic diagram of the integrated sensor-computer interface processing flow of the present invention; Figure 4 This is a schematic diagram of the intelligent image recognition model processing flow of the present invention; Figure 5 This is a schematic diagram of the multi-UAV collaborative relay early warning process of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Figure 1 This is a schematic diagram of the hardware structure of the integrated sensing, communication, and computing roadside device of the present invention. The integrated sensing, communication, and computing roadside device is an intelligent roadside unit designed based on the V2X communication standard. It adopts a modular architecture with a separate core board and baseboard, and the two parts are connected through a high-speed backplane connector, achieving a balance between high integration of core functions and maintainability. Specifically: The core board integrates the core processing and communication functions of the integrated sensing, communication, and computing roadside equipment, specifically including: a C-V2X communication module, integrating an LTE-V2X baseband processing unit, supporting PC5 mode direct communication and Uu mode network communication, enabling low-latency information interaction with the on-board unit (OBU), and supporting encoding and decoding of standard V2X information sets (RSI, RSM); an NPU computing module, integrating a neural network processing unit, used to run intelligent image recognition models and achieve real-time image inference; and local DDR storage for caching raw image frames from cameras, intermediate calculation results for intelligent inference, V2X message queues, and operation logs. The C-V2X communication module and the NPU computing module achieve high-speed serial communication via Xilinx's SerDes IP, meeting the bandwidth requirements for transmitting road image data from the NPU-processed system to the C-V2X module, ensuring low latency between perception, computation, and communication.

[0024] The baseboard houses peripheral interfaces and expansion modules: a GNSS module providing a precise spatial reference for locating defects and controlling UAV flight; it also supports PPS timing output for time synchronization between the integrated roadside equipment and the on-board unit (OBU). A 5G module transmits inspection data, equipment status, and log information back to the operations platform; it also serves as a backup communication link to ensure mission continuity when the UAV flies outside the coverage area of ​​its local area network. A security module handles V2X message signing and verification, and device authentication to prevent malicious attacks and information forgery. A radio frequency module includes an LTE-V2X transceiver, power amplifier, low-noise amplifier, filter, and duplexer to ensure reliable communication with following vehicles. A camera module uses dual cameras; debugging and expansion interfaces include a UART serial port, MIPI interface, JTAG, USB, SPI, Ethernet port, and CAN bus.

[0025] like Figures 2-3As shown, the present invention provides a technical solution: a highway traffic management method based on dynamic early warning of highway defects, comprising: S0, the operation platform sets drone routes and scheduled tasks, and the drone nests execute the tasks.

[0026] It is important to note that during the initial phase, the operations platform dynamically plans the inspection routes, flight altitudes, and patrol frequencies of the drones based on multi-source information such as highway road conditions, real-time weather data, and historical accident records. The operations platform then transmits task commands to drone nests deployed along the highway via a wireless communication network. These drone nests, serving as nodes for drone takeoff, landing, charging, and data relay, have the following functions: receiving inspection tasks from the operations platform and allocating tasks according to priority and drone status; monitoring the drone's flight status and equipment operation in real time; and optionally receiving road surface image data and road anomaly information transmitted back by the drones for local display or transmission back to the operations platform for statistical analysis.

[0027] S1. Drones equipped with integrated sensing, communication, and computing roadside equipment patrol along the highway and collect real-time road surface image data.

[0028] S2. The roadside equipment equipped with the drone, which integrates sensing, communication, and computing, processes the road surface image data in real time through its built-in intelligent image recognition model, identifies abnormal targets, and generates unique identifiers.

[0029] like Figure 4 As shown, the present invention provides an intelligent image recognition model processing method; Specifically: The first step is to register the real-time road surface image data with the historical disease database. For the detection targets located within the historical disease location range, extract their current image coordinates and size, and compare them with the size data from previous times to calculate the size expansion rate. Each disease record in the database has a unique identifier.

[0030] It should be noted that the historical disease database is stored in a relational database. Each disease record includes: a unique identifier, the time of first discovery, the time of last detection, the coordinates of the bounding box of each detection, the pixel area of ​​each detection, the texture feature vector (LBP histogram), the current priority (1-extended, 2-stable, 3-new disease), and the status (0-active, 1-removed).

[0031] Real-time acquired road surface image frames I t Compared with the baseline image of this road section stored in the historical disease database I ref(Delivered by the operating platform) Registration is performed. The Scale Invariant Feature Transform (SIFT) algorithm is used to extract feature points from the two images. Mismatched points are removed using a random sampling consensus algorithm. The homography matrix H is calculated, and I... t Mapping to I ref In the coordinate system, image offset caused by changes in the flight attitude of the UAV is eliminated.

[0032] Based on the registered image coordinates, I t All candidate targets detected (obtainable through the YOLOv8 target detection model) are matched against historical diseases in the database. If the distance between the center of the bounding box of a candidate target and the center of the bounding box of the most recently recorded historical disease is less than a threshold (e.g., Euclidean distance less than 50 pixels), they are determined to be the same disease, and their unique identifier and historical size sequence are obtained. The size expansion rate v s The calculation method for (unit: pixel area / day) is as follows:

[0033] Among them, S n S represents the current pixel area detected for the disease. n-1 For the historical pixel area of ​​the disease, t n For the current time, t n-1 This is the historical pixel area detection time. If t n -t n-1 If the time interval is less than 1 hour, the rate will not be updated temporarily, and the previous calculation result will be used to avoid short-term fluctuations.

[0034] The second step is to classify the detection targets whose size expansion rate exceeds the preset expansion threshold as expanding diseases and assign them a first priority identifier (priority set to 1); classify the detection targets whose size expansion rate is lower than the preset expansion threshold as stable diseases and assign them a second priority identifier (priority set to 2); and update the judgment results to the historical record corresponding to the unique identifier of the disease.

[0035] It should be noted that the extended threshold T 扩展 The setting is based on the type of highway pavement, such as T for asphalt pavement. 扩展 =30 pixels / day. The updated historical data for this defect includes the bounding box coordinates, pixel area, and detection time.

[0036] The third step is to extract and compare the texture features of the detection targets located outside the historical disease location range to identify texture change areas; and to perform cross-frame morphological tracking. If the morphology shows a unidirectional extension trend, it is determined to be a new disease and assigned a third priority identifier; a new unique identifier is generated for the new disease and written into the database.

[0037] It is important to note that for detection targets that do not match any historical diseases, the following new disease identification process should be performed: Texture feature extraction involves resizing the image of the detected target region to a uniform size of 64×64 pixels, converting it to grayscale, and extracting LBP texture features. A circular neighborhood LBP with a radius of 2 and 16 neighboring points is used to generate a uniform pattern LBP histogram. This histogram is then used as the texture feature vector for the detected target.

[0038] Texture similarity comparison involves calculating the distance between the texture feature vector and the texture feature vectors of all active diseases in the historical disease database, such as chi-square distance.

[0039] Among them, F q To query the feature vector (i.e., the LBP texture feature vector extracted from the currently detected target region), F dp Let F be the database feature vector (i.e., the LBP texture feature vector stored in a specific disease record in the historical disease database), where i is the dimension index of the feature vector, n is the total number of dimensions of the feature vector, and F is the feature vector dimension. q (i) represents the component value of the query vector in the i-th dimension, F db (i) represents the component value of the database feature vector in the i-th dimension, and ε is the minimum value to prevent zero (e.g., take 10). -6 Let the texture matching threshold T be... 纹理 =0.3, if it exists <T 纹理 If a match is found, it indicates that the texture features of the detected target are similar to a known disease type in the database, but are located outside the historical disease location range. This is then marked as a suspected newly emerging disease of a known type, and the matching disease type (such as transverse cracks, pits, etc.) is recorded for cross-frame morphological tracking. If a match exists... ≥T 纹理 If a match is found, it indicates that the texture features of the detected target are not similar to any known disease type in the database, and it may be a new disease or a false detection. It is then marked as a suspected unknown type anomaly, and cross-frame morphological tracking continues.

[0040] Cross-frame morphological tracking involves tracking the detected target across N consecutive frames (e.g., 5 frames). A KLT optical flow tracker is used, employing corner points within the target's bounding box of the previous frame as feature points. The positions of these points are tracked in the current road surface image frame to update the bounding box. The morphological change trend index is defined as the length change rate r. l :

[0041] Among them, l tl is the pixel length of the longer side of the target bounding box in the current road surface image frame. t-1 This represents the pixel length of the longer side of the target bounding box in the previous frame. Width change rate r w :

[0042] Among them, w t w is the pixel length of the shorter side of the target bounding box in the current road surface image frame. t-1 Let r be the pixel length of the shorter side of the target bounding box in the previous frame. If any of the following conditions are met in m consecutive frames (e.g., 3 frames), the detected target is determined to have a unidirectional extension trend: In the 3 consecutive frames, r in each frame... l >0.1 and r w <0.05 (long side dominates extension); in 3 consecutive frames, r in each frame w >0.1 and r l <0.05 (short side dominates extension). Here, 0.1 is the threshold for the length or width growth rate, which can be adjusted according to the highway pavement type and damage characteristics; 0.05 is the suppression threshold for the rate of change in another dimension, ensuring the uniformity of the extension direction. The bounding box update rules include: calculating the minimum bounding rectangle of the successfully tracked feature points in the current pavement image frame, using this rectangle as the target bounding box for the current pavement image frame; if the feature points are too scattered (e.g., the area of ​​the bounding rectangle exceeds twice the area of ​​the bounding box in the previous frame), the bounding box is updated by adding the average optical flow displacement to the previous frame's bounding box position to avoid excessive bounding box enlargement.

[0043] Based on texture similarity comparison results and cross-frame morphological tracking results, a comprehensive determination is made as to whether it is a new disease: if the texture matches and shows a unidirectional extension trend, it is confirmed as a new disease and assigned the third priority; if the texture matches but does not show a unidirectional extension trend, no determination is made for now, and observation continues; if the texture does not match but shows a unidirectional extension trend, it is confirmed as a new disease and assigned the third priority; if the texture does not match and does not show a unidirectional extension trend, it is directly excluded. For detection targets confirmed as new diseases: a unique identifier is generated; if it is a texture match, the matching disease type is associated in the database; if it is a texture mismatch, the disease type is marked as unknown and awaits manual review; the first detection image, bounding box, texture features, and detection time of the disease are written to the database, the status is set to active, and the priority is set to 3.

[0044] The fourth step is to confirm locations in the historical defect database that are not detected in the current road surface image data across multiple frames. If no defects are detected at a location in multiple consecutive frames and the image features at that location show that the road surface is smooth, it is determined that the defect has been repaired, and the defect record corresponding to the unique identifier is marked as removed.

[0045] It is important to note that for defects that are active in the historical defect database but not detected in the current road surface image frame, a removal confirmation process is executed: The bounding box location of the defect, most recently recorded, is used for target detection in the subsequent M consecutive frames (e.g., 10 frames). If the defect is not detected in any of these 10 frames, it is determined that the defect has been repaired, the status of the defect record in the database is updated to "removed," and the time of discovery and removal is recorded. If the defect is detected again in any of the 10 frames, the count is reset, and normal tracking continues.

[0046] Step 5: Output the abnormal targets with unique identifiers and priority indicators.

[0047] It should be noted that after the above processing, the following information is output for all active defects in the current road surface image frame: unique identifier, current priority, image coordinates, center point pixel coordinates, and detection time.

[0048] S3. Obtain the image coordinates of the abnormal target, and convert the image coordinates into latitude and longitude coordinates in real time by combining them with the positioning information of the roadside equipment.

[0049] It is important to note that collinearity equations are used for coordinate transformation. Utilizing precise positioning information (latitude, longitude, and altitude) provided by the GNSS module of the integrated roadside equipment and the UAV attitude angles acquired via UART, combined with camera intrinsic parameters (focal length, principal point, and distortion coefficients), pixel coordinates are converted to ground latitude and longitude using a perspective projection model. This perspective projection transformation involves continuous matrix operations from the camera coordinate system, the aircraft coordinate system, the local horizontal coordinate system, to the geographic coordinate system. These operations are accelerated in parallel by the NPU, and camera parameters and attitude matrices are temporarily stored in DDR memory to support high-frequency continuous computation.

[0050] S4. Calculate the event confidence level based on the real-time distance between the abnormal target and the roadside equipment and the time difference from the time the abnormal target was identified to the current time.

[0051] The method for dynamically calculating event confidence is as follows:

[0052]

[0053]

[0054] Where C is the event confidence level, w d For distance weights, w t For time weighting, w pd represents the priority weight (e.g., the first priority is 1.0, the second priority is 0.7, and the third priority is 0.9), d represents the real-time distance (in kilometers) between the abnormal target and the integrated sensing, communication, and computing roadside equipment, and t represents the time difference from the time the abnormal target was identified to the current time.

[0055] The method for calculating the real-time distance d is as follows:

[0056] Where r is the Earth's radius, (Ф1,λ1) are the latitude and longitude of the current location of the integrated sensing, communication, and calculation roadside equipment, and (Ф2,λ2) are the latitude and longitude of the location of the abnormal target.

[0057] S5. When the confidence level of the event is greater than the preset warning threshold, the abnormal target information is encapsulated into a standard V2X information set (RSI, RSM) to obtain the warning information; it is dynamically broadcast to the vehicles behind through LTE-V communication technology, while the drone continues to fly in the opposite direction of the traffic flow, so that the warning range is dynamically extended backward.

[0058] It should be noted that the warning threshold can be set based on the type of highway section, such as 0.4 for straight sections; 0.3 for tunnels and sharp curves; 0.35 for accident-prone sections; and 0.45 for areas near service areas and low-traffic sections. The warning information must be digitally signed by a security module before being broadcast. After receiving the information, the on-board unit (OBU) verifies the sender's identity and the integrity of the information to prevent forged warning information from interfering with driving. The dynamic broadcast uses a geofencing mechanism, sending warning information only to vehicles located within a certain range behind the abnormal target and traveling in the same direction, avoiding interference from invalid information received by vehicles in oncoming lanes.

[0059] S6. During the flight of the UAV, the real-time distance and time difference are updated in real time, the event confidence is dynamically adjusted, and the judgment and broadcast steps are repeatedly triggered until the preset termination conditions are met.

[0060] It is important to note that the GNSS module continuously outputs position update 1, and the NPU computing module dynamically calculates the event confidence level every 100 milliseconds. When the event confidence level is lower than the warning threshold, it sends a command to the flight controller via UART to adjust the UAV's flight speed or hovering position and optimize the broadcast range coverage. Key status parameters are transmitted back to the drone's nest in real time via the network port for monitoring by the operation platform.

[0061] The preset termination conditions include at least one of the following: the real-time distance is greater than the preset effective warning distance (e.g., 2 kilometers), the event confidence level is lower than the warning threshold for more than a preset duration (e.g., 10 seconds), receiving confirmation information returned by a vehicle behind via the vehicle network, and receiving an abnormality handling instruction sent by the operation platform.

[0062] Based on the aforementioned S1-S6 single-machine early warning, this invention further provides, as follows: Figure 5 The multi-drone collaborative early warning scheme shown addresses the issues of limited single-drone endurance and restricted early warning distance. Specifically, it includes: The first step is for the operation platform to divide the highway into several continuous inspection sections with overlapping areas between adjacent sections. Each drone performs inspections within its assigned section.

[0063] It is important to note that each section is routinely inspected by a single drone. The length of each section is determined based on the drone's endurance (allowing for a return-to-home margin), typically taken as follows: Range length = Flight time × Cruise speed × 0.7 The overlap length between adjacent sections is no less than 500 meters to ensure sufficient buffer time for handover. Each section corresponds to a drone nest, responsible for the take-off, landing, charging, and data relay of drones within that section. Initially, the operation platform assigns inspection tasks to drones in each section based on their status and current location. Each drone, equipped with integrated sensing, communication, and computing roadside equipment, conducts inspections within its assigned section according to a preset route.

[0064] The second step is for the first drone to immediately establish a tracking record after identifying the abnormal target.

[0065] It is important to note that the tracking record includes: a unique identifier for the abnormal target, a historical confidence sequence, the current size expansion rate, texture feature vectors, and the latitude and longitude coordinates of the last detection. The tracking record is updated every time a target is detected; the record retains the most recent N (e.g., 100) historical data entries, discarding the oldest record when this limit is exceeded; key content (such as event confidence) is smoothed using a moving average to avoid fluctuations in single frames.

[0066] The third step is to send a handover request to the operations platform when the first drone needs to be decommissioned due to limitations in endurance or distance.

[0067] It should be noted that the handover request information includes the first UAV identifier, current location (latitude and longitude), current heading, current flight speed, remaining battery power, unique identifier of the abnormal target, latitude and longitude of the abnormal target, and reason for handover.

[0068] The fourth step is for the operation platform to dynamically calculate the rendezvous point location based on the real-time position, flight direction and speed of each drone, and select the second drone to rendezvous.

[0069] It is important to note that after receiving the handover request, the operation platform will execute the successor drone selection process: from all drones that are idle or in inspection mode, select drones that meet the following conditions as candidate successor drones: located in front of the first drone (along the opposite direction of traffic flow); the distance to abnormal targets is less than a preset threshold (e.g., 3 kilometers); the remaining battery power is higher than the minimum task battery power (e.g., 50%); and the equipment is in normal condition.

[0070] The connection point position P h The calculation method is as follows: P h =P t +α×D f Among them, P t Location of the abnormal target; D f The fixed offset distance (e.g., 2 kilometers) along the highway ensures that the UAV flies in front of the target after the handover; α is the directional coefficient, which is 1 along the lane direction and -1 in the opposite direction.

[0071] For each candidate rendezvous drone, calculate its estimated time T to reach the rendezvous point. 到达 :

[0072] Among them, Distance(P d ,P h P represents the current location of the drone. d To the point of connection P h Spatial distance (unit: meters); V d The current flight speed of the drone (unit: meters per second).

[0073] Select T 到达 The smallest drone is used as the second drone, while ensuring that T arrives in less than the maximum waiting time (e.g., 60 seconds).

[0074] Step 5: After the first drone flies to the rendezvous point, it sends the tracking data to the second drone.

[0075] It is important to note that after receiving the handover confirmation, the first drone will fly to the handover point according to the following strategy: continue flying along the original route and maintain early warning broadcasts for abnormal targets; when the distance to the handover point is less than the preset distance (e.g., 200 meters), decelerate to the handover speed (e.g., 5 meters / second); after reaching the handover point, hover or circle at low speed, establish a point-to-point communication link between drones through the 5G module (or via relay from the operation platform), and transmit tracking records; the data is encrypted by the security module before transmission to ensure that the handover information is not tampered with.

[0076] If the second drone fails to connect successfully within the predetermined time (e.g., 30 seconds) (e.g., target loss, communication interruption), the operation platform will trigger the emergency plan: notify the first drone to extend its loiter time (if battery power allows), or dispatch the third drone to the rendezvous point, and send an alarm to the operation platform duty personnel, and manually intervene if necessary.

[0077] Step 6: The second drone continues to broadcast early warnings based on the received tracking records.

[0078] It is important to note that after receiving the tracking record, the second UAV executes the following follow-up process: data verification, checking data integrity, and requesting retransmission if verification fails; target localization, adjusting the camera of the integrated sensing, communication, and computing roadside device to align with the location based on the latitude and longitude of the last detected abnormal target in the tracking record, and performing a local search. The real-time acquired image is compared with the texture feature vector in the tracking record to calculate the similarity; if the similarity is higher than a threshold (e.g., 0.8), it is confirmed as the same target, and a unique target identifier is associated; a follow-up completion confirmation is sent to the operations platform, including: the target's unique identifier, the current UAV identifier, the confirmation time, and the target's current status. After receiving the confirmation, the operations platform sends an exit command to the first UAV.

[0079] Step 7: The first drone ceases its warning and returns to its nest.

[0080] It should be noted that the C-V2X module stops broadcasting V2X information, and the radio frequency module enters a low-power mode; the NPU computing module writes the complete task log to DDR and uploads it to the operation platform in batches via the network or 5G module; the GNSS module continues to work to provide navigation and positioning until the first UAV returns to the nest.

[0081] A high-speed traffic management system based on dynamic early warning of highway defects uses any one of the above-mentioned high-speed traffic management methods based on dynamic early warning of highway defects.

[0082] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended embodiments and their equivalents.

Claims

1. A highway traffic management method based on dynamic early warning of highway defects, characterized in that, include: S1. The drone is equipped with an integrated roadside device that combines sensing, communication and computing to inspect the highway and collect road surface image data in real time. S2. The roadside equipment uses a built-in intelligent image recognition model to process the road surface image data in real time, identify abnormal targets, and generate unique identifiers. S3. Obtain the image coordinates of the abnormal target and, in conjunction with the positioning information of the roadside equipment, convert them into latitude and longitude coordinates in real time. S4. Calculate the event confidence level based on the real-time distance between the abnormal target and the roadside equipment and the time difference from the time the abnormal target was identified to the current time; S5. When the confidence level of the event is greater than the preset warning threshold, the abnormal target information is encapsulated into the vehicle network standard information format to obtain the warning information; The warning is dynamically broadcast to vehicles behind by using long-term evolved vehicle-to-everything (V2X) communication technology, while the drone continues to fly in the opposite direction of traffic flow, thus dynamically extending the warning range backward. S6. During the flight of the UAV, the real-time distance and time difference are updated in real time, the event confidence is dynamically adjusted, and the judgment and broadcast steps are repeatedly triggered until the preset termination conditions are met.

2. The highway traffic management method based on dynamic early warning of highway defects according to claim 1, characterized in that, The intelligent image recognition model specifically includes: S21. Register the real-time road surface image data with the historical disease database. For the detection targets located within the historical disease location range, extract their current image coordinates and size, and compare them with the size data from previous times to calculate the size expansion rate. Each disease record in the database has a unique identifier. S22. Detection targets with a size expansion rate exceeding a preset expansion threshold are identified as expanding diseases and assigned a first priority label; detection targets with a size expansion rate below the preset expansion threshold are identified as stable diseases and assigned a second priority label; the identification results are updated to the historical record corresponding to the unique identifier of the disease. S23. For detection targets located outside the historical disease location range, extract their texture features and compare them to identify texture change areas; and perform cross-frame morphological tracking. If the morphology shows a unidirectional extension trend, it is determined to be a new disease and assigned a third priority identifier; generate a new unique identifier for the new disease and write it into the database. S24. For locations that exist in the historical defect database but are not detected in the current road surface image data, perform cross-frame confirmation; if no defect is detected at this location in multiple consecutive frames, and the image features at this location show that the road surface is smooth, then it is determined that it has been repaired, and the defect record corresponding to the unique identifier is marked as removed. S25. Output the abnormal target with a unique identifier and priority identifier.

3. The highway traffic management method based on dynamic early warning of highway defects according to claim 1, characterized in that, The method for dynamically calculating event confidence is as follows: ， ， ， Where C is the event confidence level, w d For distance weights, w t For time weighting, w p d represents the priority weight, d represents the real-time distance (in kilometers) between the abnormal target and the integrated sensing, communication, and computing roadside equipment, and t represents the time difference from the time the abnormal target was identified to the current time.

4. A high-speed traffic management method based on dynamic early warning of highway defects according to claim 3, characterized in that, The method for calculating the real-time distance d is as follows: ， Where r is the Earth's radius, (Ф1,λ1) are the latitude and longitude of the current location of the integrated sensing, communication, and calculation roadside equipment, and (Ф2,λ2) are the latitude and longitude of the location of the abnormal target.

5. A high-speed traffic management method based on dynamic early warning of highway defects according to claim 1, characterized in that, The preset termination conditions include at least one of the following: the real-time distance is greater than the preset effective warning distance, the event confidence level is lower than the warning threshold for more than the preset duration, the confirmation information returned by the vehicle behind via the vehicle network is received, and the abnormality has been handled instruction sent by the operation platform.

6. A high-speed traffic management method based on dynamic early warning of highway defects according to claim 1, characterized in that, The integrated roadside device adopts a core board and base board design, which are connected by a high-speed backplane connector. The core board integrates a vehicle network communication module and an intelligent computing module, and is connected through a serializer-deserializer interface.

7. A high-speed traffic management method based on dynamic early warning of highway defects according to claim 1, characterized in that, It also includes coordinated early warning systems using multiple drones: The operation platform divides the highway into several continuous inspection sections with overlapping areas between adjacent sections. Each drone performs inspections within its assigned section. The first drone immediately established a tracking record after identifying the abnormal target; When the first drone needs to be decommissioned due to endurance or distance limitations, it sends a handover request to the operations platform. The operation platform dynamically calculates the rendezvous point location based on the real-time position, flight direction, and speed of each drone, and selects the second drone to rendezvous. After the first drone flies to the rendezvous point, it sends the tracking data to the second drone. The second drone continues to broadcast early warnings based on the received tracking records; The first drone ceased issuing warnings and returned to its nest.

8. A high-speed traffic management method based on dynamic early warning of highway defects according to claim 7, characterized in that, The tracking record includes at least the unique identifier of the abnormal target, the historical confidence sequence, the current size expansion rate, and the latitude and longitude coordinates of the last detection; after receiving the tracking record, the second UAV performs a local search based on the latitude and longitude coordinates to determine the abnormal target that needs to be warned, and inherits the historical confidence based on its unique identifier.

9. A high-speed traffic management system based on dynamic early warning of highway defects, characterized in that, The method for high-speed traffic management based on dynamic early warning of highway defects, according to any one of claims 1-8, is used.

Images