Car-road cooperation traffic management method, system, device and storage medium

Abstract

The application provides a vehicle-road cooperation traffic management method, system, device and storage medium, relates to the technical field of traffic management, and comprises the following steps: receiving traffic event information and an instruction message from a traffic management cloud control platform, and controlling the vehicle-road cooperation traffic management system to move to a target point according to the instruction message; obtaining observation data of each resource node in a target service area according to the traffic event information and the instruction message; adopting a preset fusion algorithm to fuse the observation data of each resource node in the target service area, and generating a state estimation sequence of each traffic participant; adopting a preset trajectory prediction model to perform trajectory prediction based on the state estimation sequence of each traffic participant, and outputting a multi-modal prediction trajectory cluster with a probability weight; and sending a decision suggestion to a corresponding traffic participant based on the multi-modal prediction trajectory cluster of each traffic participant. The vehicle-road cooperation traffic management method can realize dynamic traffic management.

Description

Technical Field

[0001] This application relates to the field of traffic management technology, and in particular to a vehicle-road cooperative traffic management method, system, device and storage medium. Background Technology

[0002] Intelligent Transportation Systems (ITS) represent the future direction of transportation systems. They effectively integrate advanced information technology, data communication and transmission technology, electronic sensing technology, control technology, and computer technology into the entire ground traffic management system. This results in a comprehensive, real-time, accurate, and efficient transportation management system that operates across a wide area, effectively addressing modern traffic congestion, optimizing transportation routes, and improving road network capacity. Among these technologies, Vehicle-to-Everything (V2X) technology plays an increasingly important role in enhancing traffic safety, improving road flow, and supporting autonomous driving.

[0003] However, traditional vehicle-road cooperative technologies often rely on fixed design systems, requiring significant time for installation and debugging, and are ill-suited for scenarios with frequent changes in location or configuration. Moreover, when dealing with complex scenarios, their efficiency and accuracy in real-time processing of multi-source data remain insufficient, making it difficult to achieve the desired traffic management results. Summary of the Invention

[0004] In view of this, this application proposes a vehicle-road cooperative traffic management method, system, device and storage medium.

[0005] Firstly, this application provides a vehicle-road cooperative traffic management method, which is applied to a vehicle-road cooperative traffic management system with mobility cooperative capabilities; the vehicle-road cooperative traffic management method includes:

[0006] The system receives traffic incident information and instruction messages from the traffic management cloud control platform, and controls the vehicle-road cooperative traffic management system to move to the target point according to the instruction messages; wherein, the traffic incident information includes at least one of the real-time traffic event list of the authorized traffic information service platform and the monitoring information of the vehicle-road cooperative traffic management system.

[0007] Based on the traffic incident information and the instruction message, obtain the observation data of each resource node within the target service area;

[0008] A preset fusion algorithm is used to fuse the observation data of each resource node in the target service area to generate a state estimation sequence for each traffic participant. The preset fusion algorithm is an adaptive unscented Kalman filter algorithm or an unscented Kalman filter algorithm.

[0009] Based on the state estimation sequences of each traffic participant, a preset trajectory prediction model is used to predict the trajectory, and a multimodal predicted trajectory cluster with probability weights is output.

[0010] Decision suggestions are sent to the corresponding traffic participants based on the multimodal predicted trajectory clusters of each traffic participant.

[0011] In one embodiment, obtaining observation data of each resource node within the target service area based on the traffic incident information and the instruction message includes:

[0012] The target service area is determined based on the traffic incident information and the instruction message, and a dynamic resource map of the target service area is constructed, wherein the dynamic resource map includes at least one resource node;

[0013] Based on the instruction message and the dynamic resource map, the execution tasks of each resource node are allocated, and the observation data obtained by each resource node in executing the allocated tasks are acquired.

[0014] In one embodiment, before constructing a dynamic resource map of the target service area, the method further includes:

[0015] Broadcast collaborative service signaling and receive response information from communication terminals within the service area;

[0016] Each communication terminal is registered as a resource node based on its response information, and the attribute information of each resource node is recorded, including ID, location information, and dynamic capability set.

[0017] The process of allocating execution tasks for each resource node based on the instruction message and the dynamic resource map includes:

[0018] The current macro-level task is determined based on the instruction message;

[0019] The current macro task is decomposed into multiple micro sub-tasks, and each micro sub-task is optimally matched with the capabilities of each resource node. The execution tasks of each resource node are then assigned according to the matching results.

[0020] In one embodiment, the step of fusing the observation data of each resource node within the target service area using a preset fusion algorithm to generate a state estimation sequence for each traffic participant includes:

[0021] The observation data of each resource node is synchronized to a unified spatiotemporal reference to obtain spatiotemporally aligned multi-source observation data;

[0022] Based on the spatiotemporally aligned multi-source observation data, high-dimensional state vectors are established for each traffic participant in the target service area, wherein the high-dimensional state vectors include position, velocity, heading angle and yaw rate.

[0023] Sigma point sampling and nonlinear prediction are performed on each of the high-dimensional state vectors to obtain state prediction values. Based on the state prediction values, the covariance of the observed information sequence is calculated for matching. An adaptive unscented Kalman filter algorithm is used to dynamically adjust the process noise matrix and the observation noise matrix, and key filter parameters are calculated.

[0024] The spatiotemporally aligned multi-source observation data is mapped to a unified state space, and the optimal Kalman gain is calculated by combining the adaptively adjusted process noise matrix, observation noise matrix and key filtering parameters.

[0025] Based on the optimal Kalman gain, the spatiotemporally aligned multi-source observation data and the state prediction values ​​are optimally fused to generate a state estimation sequence for each traffic participant.

[0026] In one embodiment, the step of predicting trajectories based on the state estimation sequences of each traffic participant using a preset trajectory prediction model, and outputting a multimodal predicted trajectory cluster with probability weights, includes:

[0027] The state estimation sequence of the target traffic participant is used as the observation sequence to input the pre-trained driving behavior intention inference model to obtain the probability distribution of each behavior intention of the target traffic participant at the current time. The driving behavior intention inference model is a hidden Markov model.

[0028] Based on the state estimation sequence of the target traffic participant, the historical trajectory of the target traffic participant is obtained. The historical trajectory of the target traffic participant is matched with the preset typical trajectory template corresponding to each behavioral intention. Based on the matching result and the probability distribution of each behavioral intention of the target traffic participant at the current time, a multimodal predicted trajectory cluster with probability weights is output.

[0029] In one embodiment, sending decision suggestions to the corresponding traffic participants based on the multimodal predicted trajectory clusters of each traffic participant includes:

[0030] Traffic conflict risk information is determined based on the multimodal predicted trajectory clusters of each traffic participant.

[0031] Traffic planning strategies are generated based on traffic conflict risk information, and decision-making suggestions are sent to relevant traffic participants based on the traffic planning strategies.

[0032] In one embodiment, determining traffic conflict risk information based on the multimodal predicted trajectory clusters of each traffic participant includes:

[0033] Obtain multimodal predicted trajectory clusters for target traffic participant combinations;

[0034] Determine the predicted trajectory combinations of the target traffic participant pairs, and determine whether there are any spatiotemporal overlaps between the predicted trajectory pairs within a preset time period in the future.

[0035] Predicted trajectory pairs with spatiotemporal overlap points are identified as conflict trajectory pairs. The joint conflict probability of each conflict trajectory pair is calculated. The joint probabilities of each conflict trajectory pair are added together to obtain the conflict probability of the target traffic participant pair.

[0036] Risk warning information of corresponding levels is generated based on the conflict probability of the target traffic participant pair.

[0037] Secondly, this application also provides a vehicle-road cooperative traffic management system, including: a mobile vehicle, a sensing module, and a processing module;

[0038] The mobile vehicle is used to carry the sensing module and the processing module;

[0039] The sensing module is used to scan the surrounding wide-area environment to obtain monitoring information;

[0040] The processing module is connected to the mobile vehicle and the sensing module respectively. The processing module is used to control the movement of the mobile vehicle and to execute the vehicle-road cooperative traffic management method as described in the first aspect.

[0041] Thirdly, this application also provides an electronic device, including a processor and a memory; the memory has a computer program stored thereon, wherein the computer program, when executed by the processor, implements the vehicle-road cooperative traffic management method as described in the first aspect.

[0042] Fourthly, this application also provides a non-transitory computer storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the vehicle-road cooperative traffic management method as described in the first aspect.

[0043] The vehicle-road cooperative traffic management method proposed in this application has the following advantages over related technologies:

[0044] 1. The vehicle-road cooperative traffic management method of this application is applied to a vehicle-road cooperative traffic management system with mobile cooperative capabilities. By receiving traffic event information and instructions from the cloud control platform and dynamically moving to the target point, it breaks through the limitations of traditional fixed vehicle-road cooperative systems that rely on fixed deployment, have long installation and debugging times, and are difficult to adapt to scenarios with frequent changes in location or configuration. By dynamically mobilizing discrete sensing resources within the target service area, it effectively expands the system's sensing and control range and significantly improves the scenario adaptability.

[0045] 2. An adaptive unscented Kalman filter (AKF) or unscented Kalman filter (UKF) algorithm is employed to fuse multi-source observation data, generating a high-precision traffic participant state estimation sequence. This addresses the shortcomings of traditional techniques in real-time processing efficiency and accuracy of multi-source data in complex traffic scenarios, particularly improving position and heading angle estimation accuracy in highly nonlinear and high-mobility scenarios. Building upon this, the traffic participant state estimation sequence is combined with a pre-defined trajectory prediction model to output a multimodal predicted trajectory cluster with probability weights, further enhancing the accuracy of long-term trajectory prediction. This prediction more closely reflects the actual intentions and future movement range of traffic participants. Decision suggestions are then sent to the corresponding traffic participants based on these predictions, enabling earlier identification of potential traffic risks. This provides a more sufficient and reliable decision-making window for autonomous driving systems and traffic participants, thereby optimizing traffic management and achieving more comprehensive collaborative management and more efficient risk warning. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a flowchart illustrating a vehicle-road cooperative traffic management method in one embodiment of this application;

[0048] Figure 2 This is a flowchart illustrating step S103 in one embodiment of this application;

[0049] Figure 3 This is a schematic diagram of the vehicle-road cooperative traffic management system in one embodiment of this application;

[0050] Figure 4 This is a schematic diagram of the vehicle-road cooperative traffic management system in another embodiment of this application. Detailed Implementation

[0051] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0052] In some embodiments, such as Figure 1 As shown, this application provides a vehicle-road cooperative traffic management method, which is applied to a vehicle-road cooperative traffic management system with mobility cooperative capabilities. The vehicle-road cooperative traffic management method includes the following steps S101 to S105.

[0053] S101: Receives traffic incident information and instruction messages from the traffic management cloud control platform, and controls the vehicle-road cooperative traffic management system to move to the target point according to the instruction messages. The traffic incident information includes at least one of the real-time traffic event list from the authorized traffic information service platform and the monitoring information from the vehicle-road cooperative traffic management system.

[0054] The command message may include information such as event type, GPS coordinates, recommended deployment point, and expected service radius. The real-time traffic event list can be periodically obtained from an authorized traffic information service platform. This list can cover key events affecting traffic flow within the target area, such as traffic congestion, road accidents, temporary construction, and traffic control, along with their specific locations, impact areas, and estimated durations. It should be noted that the above description is merely illustrative; the information included in the command message and real-time traffic event list can be flexibly configured according to actual needs.

[0055] It is understandable that after receiving traffic incident information and instruction messages from the traffic management cloud control platform, the target point parameters in the instruction messages can be parsed, and the movement path can be dynamically adjusted in combination with the real-time traffic conditions in the traffic incident information to avoid the impact of existing traffic incidents and ensure efficient and safe movement to the designated target point.

[0056] S102: Obtain observation data of each resource node within the target service area based on traffic incident information and instruction messages.

[0057] The target service area can be a managed area such as an accident site or a congested section. The target service area can be determined by geofencing matching and credibility filtering based on a real-time traffic event list. Alternatively, while in a moving or standby state, sensors on the vehicle-road cooperative traffic management system can scan (detect) the surrounding wide-area environment, and identify signs of sudden congestion or accidents using a sudden change detection algorithm based on average vehicle speed and density, thereby determining the target service area. For example, upon receiving a command message from the traffic management cloud control platform, the vehicle-road cooperative traffic management system can be controlled to move to the target point, and the target service area can be automatically calculated and determined based on a preset strategy (such as covering 200 meters of road) centered on the target point.

[0058] It is understandable that after locating the target service area based on traffic incident information and command messages, all heterogeneous resource nodes within that target service area can be dynamically identified and networked together. These resource nodes can include local sensing devices such as millimeter-wave radar, high-definition cameras, and inertial measurement units mounted on the system itself, as well as distributed sensing resources such as fixed roadside facilities connected to the vehicle-road cooperative network, vehicle-mounted sensing systems of nearby connected vehicles, and mobile cooperative nodes within the target service area. Furthermore, based on the event type in the traffic incident information and the task requirements of the command messages, effective resource nodes matching the task can be selected, thereby improving the quality of the observation data. After determining each resource node within the target service area, observation data from each resource node within the target service area can be received, thus acquiring the observation data of each resource node within the target service area.

[0059] S103: The observation data of each resource node in the target service area are fused using a preset fusion algorithm to generate the state estimation sequence of each traffic participant. The preset fusion algorithm is either an adaptive unscented Kalman filter algorithm or an unscented Kalman filter algorithm.

[0060] In the application, after acquiring observation data of each resource node within the target service area, the observation data can be preprocessed. Preprocessing may include spatiotemporal synchronization and outlier removal, and then initial weights are assigned based on data reliability. The pre-defined unscented Kalman filter algorithm addresses the nonlinear characteristics of traffic participant motion states by selecting a set of Sigma points to approximate the probability distribution of state variables, thus avoiding the errors caused by the linearization of traditional Kalman filters. Through an iterative process of state prediction and observation updates, the influence of noise is gradually reduced, yielding the optimal state estimate for each time step.

[0061] The Adaptive Unscented Kalman Filter (AUKF) algorithm, building upon UKF, adds an adaptive noise covariance adjustment mechanism. This mechanism can perceive dynamic changes in the traffic scene in real time and automatically correct process noise covariance and observation noise covariance. When the reliability of data from a resource node decreases due to occlusion or other reasons, the algorithm dynamically reduces its weight and increases the fusion weight of data from other reliable nodes, thereby enhancing the robustness of state estimation in complex traffic environments. Finally, after multi-source fusion of all valid observation data using one of the above two algorithms, the optimal state estimates of traffic participants at each time point are arranged in chronological order to form a complete state estimation sequence. This sequence accurately depicts the evolution of the motion state of traffic participants within the target service area, providing core data support with both accuracy and continuity for subsequent trajectory prediction.

[0062] S104: Based on the state estimation sequence of each traffic participant, a preset trajectory prediction model is used to predict the trajectory, and a multimodal predicted trajectory cluster with probability weights is output.

[0063] Understandably, given the continuous nature of traffic participants' trajectories, a refined analysis of the state estimation sequence is performed to extract real-time dynamic features across continuous time steps. Subsequently, a pre-defined trajectory prediction model adapted to complex vehicle-road cooperative scenarios is introduced. Addressing the inherent uncertainty of traffic participant behavior, the model generates multiple possible trajectories that conform to physical constraints and behavioral logic, forming a multimodal predicted trajectory cluster. During this process, the model assigns a corresponding probability weight to each predicted trajectory based on feature matching degree and the probability of the behavior occurring. The weight value directly reflects the likelihood of the trajectory occurring, ultimately outputting a multimodal predicted trajectory cluster with probability weights.

[0064] S105: Send decision suggestions to the corresponding traffic participants based on the multimodal predicted trajectory clusters of each traffic participant.

[0065] It is understandable that after acquiring multimodal predicted trajectory clusters with probability weights for each traffic participant, the system can identify high-probability potential risk scenarios by comparing the spatiotemporal overlap areas between multiple predicted trajectories of a particular traffic participant and the trajectories of surrounding participants. Simultaneously, based on the probability weights of each predicted trajectory, the system distinguishes the risk levels corresponding to core high-probability trajectories and low-probability sudden trajectories. Subsequently, based on the risk assessment results and optimal traffic management strategies, the system generates targeted and differentiated decision-making suggestions for the corresponding traffic participants. The suggestions are tailored to the traffic participant's type and current movement status. Finally, the system accurately distributes the decision-making suggestions in standardized message format to the corresponding traffic participants' in-vehicle terminals or personal mobile devices through V2X communication links, roadside broadcasting equipment, and other channels. Suggestions with high-risk levels are prioritized and marked with reminders to ensure that traffic participants receive and adopt them promptly, thereby effectively avoiding potential traffic conflicts and improving the overall safety and orderliness of traffic flow.

[0066] The vehicle-road cooperative traffic management method of this application is applied to a vehicle-road cooperative traffic management system with mobile cooperative capabilities. By receiving traffic event information and instructions from the cloud control platform and dynamically moving to the target point, it breaks through the limitations of traditional fixed vehicle-road cooperative systems, which rely on fixed deployment, have long installation and debugging times, and are difficult to adapt to scenarios with frequent changes in location or configuration. By dynamically mobilizing discrete sensing resources within the target service area, it effectively expands the system's perception and control range and significantly improves scenario adaptability. An adaptive unscented Kalman filter algorithm or an unscented Kalman filter algorithm is used to fuse multi-source observation data to generate a high-precision traffic participant state estimation sequence. This specifically addresses the problem of insufficient efficiency and accuracy in real-time processing of multi-source data in complex traffic scenarios using traditional technologies, especially improving the accuracy of position and heading angle estimation in highly nonlinear and highly maneuverable scenarios. Based on this, by combining the state estimation sequence of traffic participants, a multimodal predicted trajectory cluster with probability weights is output through a preset trajectory prediction model, which further improves the accuracy of long-term trajectory prediction and better matches the actual intentions and future movement range of traffic participants. Decision suggestions are sent to the corresponding traffic participants based on the prediction results, which can identify potential traffic risks earlier and provide more sufficient and reliable decision-making time windows for autonomous driving systems and traffic participants, thereby optimizing traffic management effectiveness and achieving more comprehensive collaborative management and more efficient risk warning.

[0067] In some embodiments, step S102, obtaining observation data of each resource node within the target service area based on traffic event information and instruction messages, includes: determining the target service area based on traffic event information and instruction messages, constructing a dynamic resource map of the target service area, allocating execution tasks to each resource node based on instruction messages and the dynamic resource map, and obtaining observation data obtained by each resource node from executing the allocated tasks. The dynamic resource map includes at least one resource node.

[0068] The traffic incident information clarifies the core impact area of ​​the incident and the key road sections or areas requiring coverage, while the instruction message further clarifies the task boundaries, core monitoring targets, and other requirements. The two are combined to form the specific coordinate range of the target service area. Subsequently, a network discovery mechanism is activated to collect resource nodes and their attribute information within the target service area, constructing a dynamic resource map. This dynamic resource map clearly presents the distribution, capability adaptability, and real-time availability of available resources within the area, providing precise resource basis for subsequent task allocation.

[0069] Then, based on the instruction message, the macro task is determined, and based on the macro task and the dynamic resource map, the macro task is decomposed into several executable micro sub-tasks, such as "filling the (X,Y) blind spot monitoring", "continuously tracking the trajectory of the target vehicle", and "collecting traffic flow speed data at the intersection". Each micro sub-task is matched with a corresponding resource node, and then the corresponding micro sub-task is assigned to each resource node for execution. Each resource node will obtain observation data by executing the assigned task, and this observation data is the data that is finally needed.

[0070] In some embodiments, before constructing a dynamic resource map of the target service area, the vehicle-road cooperative traffic management method further includes: broadcasting cooperative service signaling and receiving response information from each communication terminal within the service area; registering each communication terminal as a resource node based on the response information of each communication terminal, and recording the attribute information of each resource node. The attribute information includes ID, location information, and dynamic capability set.

[0071] Resource nodes can include communication terminals such as fixed RSUs, smart roadside cameras, connected vehicles (OBUs), engineering vehicles, and pedestrian smart terminals.

[0072] In the application, by activating the network discovery and capability registration mechanism, the host broadcasts "cooperative service" signaling to the designated target service area and surrounding area, triggering all potential cooperative resource nodes in the area to respond. When responding, each node will actively register its own key information. After collecting the registration information of all resource nodes, the information is integrated to form a dynamically updated dynamic resource map.

[0073] During the resource node registration phase, three core types of information can be recorded: identity and location, capabilities and expertise, and real-time status. Among them, identity and location include device number and real-time coordinates, and capabilities and expertise are clearly presented through standardized skill cards. For example, the capabilities of a smart camera may be: [Type: Video surveillance, Viewing angle: 120 degrees, Maximum clear distance: 100 meters, Processing capacity: Medium]; the capabilities of a connected car may be: [Type: Vehicle sensing, Provides: Front radar data, Communication bandwidth: High].

[0074] After receiving an instruction message, the system determines the current macro-task based on the message. This macro-task is then broken down into multiple micro-subtasks. Each micro-subtask is optimally matched with the capabilities of each resource node, and the execution tasks are assigned to each resource node based on the matching results. For example, upon receiving a macro-task instruction like "full coverage monitoring of the east-to-west intersection," it is broken down into multiple executable, specific "subtasks." For instance: Subtask A (area patrol): requires continuous monitoring of a specific area east of the intersection, recommended to be performed by a high-definition camera; Subtask B (blind spot coverage): requires monitoring a blind spot location, recommended to be performed by radar equipment capable of penetrating obstructions; Subtask C (focused tracking): requires continuous tracking of a specific suspicious vehicle, requiring equipment capable of continuous mobile tracking.

[0075] Subsequently, the core scheduling algorithm matches the optimal device for each subtask. The matching process considers three dimensions of matching degree: skill matching degree, which refers to whether the device's skill card meets the requirements of the subtask, such as patrolling requiring a device with a wide field of view; location matching degree, which refers to whether the distance between the device and the task location is close enough to ensure good execution results; and load matching degree, which refers to whether the device is currently idle and what its load status is, taking into account overall load balancing. After calculating and forming a globally optimal dispatch scheme, the devices undertaking the subtasks and the mobile platform itself together form a customized collaborative execution network.

[0076] In some embodiments, such as Figure 2 As shown, in step S103, a preset fusion algorithm is used to fuse the observation data of each resource node in the target service area to generate the state estimation sequence of each traffic participant, including the following steps S201 to S205.

[0077] S201: Synchronize the observation data of each resource node to a unified spatiotemporal reference to obtain spatiotemporally aligned multi-source observation data.

[0078] One approach is to select sensors from the vehicle-road cooperative traffic management system as spatiotemporal references, and then use a combination of hardware and software to synchronize the observation data of each resource node to a unified spatiotemporal benchmark. Alternatively, the timing signal of the traffic management cloud control platform can be used as a benchmark to calibrate the clocks of all resource nodes, assigning a standardized timestamp to each observation data point to ensure that data on the same event collected by different nodes remains consistent in the time dimension; then, through unified spatial benchmark processing, the observation data of all resource nodes can be converted to a preset unified coordinate system.

[0079] S202: Based on spatiotemporally aligned multi-source observation data, high-dimensional state vectors are established for each traffic participant in the target service area. The high-dimensional state vectors include position, velocity, heading angle, and yaw rate.

[0080] This allows for the creation of a high-dimensional state vector for each traffic participant, containing information such as position, velocity, heading angle, and yaw rate. X=[x,y,v,ψ,ω] T .

[0081] S203: Perform Sigma point sampling and nonlinear prediction on each high-dimensional state vector to obtain the state prediction value. Based on the state prediction value, match the covariance of the observation information sequence by calculating it. Use an adaptive unscented Kalman filter algorithm to dynamically adjust the process noise matrix and the observation noise matrix, and calculate the key filter parameters.

[0082] S204: Map the spatiotemporally aligned multi-source observation data to a unified state space, and calculate the optimal Kalman gain by combining the adaptively adjusted process noise matrix, observation noise matrix, and key filter parameters.

[0083] The key filtering parameters include the state prediction covariance and the global observation matrix.

[0084] The high-dimensional state vectors of each traffic participant are sampled using Sigma points and nonlinearly predicted. This involves using the UKF (Unscented Kalman Filter) framework, selecting a set of Sigma points through lossless transformation, and directly substituting them into the nonlinear vehicle motion model for prediction, thereby avoiding the linearization error of the EKF (Extended Kalman Filter). The nonlinear vehicle motion model can employ the existing "Constant Yaw Rate and Acceleration Model (CTRA)".

[0085] The adaptive unscented Kalman filter algorithm captures the probability distribution of the state vector through lossless transformation, and its core lies in a deterministic sampling strategy. The specific implementation steps are as follows:

[0086] Let the posterior estimate of the state at time k-1 be... Its error covariance matrix is First, calculate a statistically representative set of sampling points (i.e., the Sigma point set). :

[0087]

[0088]

[0089]

[0090] Where n is the dimension of the state vector; It is a composite scaling parameter used to adjust the distance between the Sigma point and the mean point; α is a small positive number (usually 10). -4 The value of α (≤α≤1) determines the distribution range of the Sigma points around the mean. The smaller the value of α, the closer the Sigma points are to the mean. When α=1, the spread range of the point set is the largest. (Kappa) is a secondary scaling parameter, typically set to 3- Or 0, its function is to reduce the composite parameters when the state dimension n is small. Fine-tuning was performed to ensure the numerical properties of the covariance calculation; The square root of matrix (n+λ)P is represented by the first square root of matrix (n+λ). i This ensures that the sampling points can fully capture the covariance information of the state estimate.

[0091] This sampling method ensures that the mean and covariance of the Sigma points are consistent with the original state distribution, eliminating the need for Taylor expansion linearization of the nonlinear function. This is what distinguishes it from EKF and improves the accuracy of nonlinear processing. This embodiment employs a constant yaw rate and acceleration model (CTRA) combined with an AUKF filter capable of online noise estimation, integrated into the dynamic networking environment of a mobile cooperative platform. This addresses the insufficient accuracy and robustness of fixed-parameter filtering in real-world complex traffic scenarios.

[0092] Then, adaptive noise estimation is performed: the actual covariance of the innovation sequence is calculated in real time and matched with the theoretical covariance. Through maximum likelihood estimation or covariance matching, the process noise matrix Q and the observation noise matrix R are dynamically adjusted online to adapt the filter to changes in target maneuvering and sensor noise fluctuations. Here, the innovation sequence refers to the difference between the actual observed value at a certain moment and the optimal predicted value based on past observations in time series analysis and filtering theory. .

[0093] For example, predict ( The data consists of the predicted state at the current moment, calculated by a UKF / AUKF filter based on the vehicle's optimal estimated state (position, velocity, heading angle) from the previous second and a nonlinear motion model. (Observations) The data refers to the actual measurements of the target vehicle from multiple source sensors (its own radar, cameras, or data reported by other cooperating vehicles). (News) The sequence is: = Sensor measured value - Model predicted value.

[0094] To achieve online adjustment of the process noise covariance matrix Q and the observation noise covariance matrix R, this paper adopts the existing covariance matching method based on a sliding window. The specific steps are as follows: Step 1 to Step 5.

[0095] Step 1, Innovation Sequence Collection: During the filtering process, a fixed-length FIFO queue of length L is maintained to continuously store the innovation vectors of the most recent L time steps. .

[0096] Step 2, Actual Covariance Calculation: At each time k, calculate the actual sample covariance matrix of the innovation sequence within the window:

[0097]

[0098] Step 3, Obtaining the Theoretical Covariance: According to the Kalman filter framework, the theoretical covariance of the innovation sequence should be:

[0099]

[0100] Where H(k) is the observation matrix, P {k|k-1} Let R(k) be the state prediction covariance, and R(k) be the current observation noise covariance.

[0101] Step 4, Difference Analysis and Adjustment: Comparison and The diagonal elements. If the j-th diagonal element satisfies:

[0102]

[0103] This indicates that the actual noise level of the j-th observation dimension exceeds expectations, and R(k)[j,j] needs to be increased; conversely, it should be decreased. The parameter α is the adjustment threshold, which is usually taken as 0.2 to 0.5.

[0104] Step 5, Recursive Adjustment Formula: To avoid abrupt parameter changes, an exponentially weighted moving average is used for smoothing adjustment.

[0105]

[0106]

[0107] Where β and γ are learning rate parameters (0 < β, γ < 1). This is the Kalman gain matrix. This adjustment allows the filter to automatically adapt to changes in sensor performance and target maneuverability.

[0108] S205: Based on the optimal Kalman gain, perform optimal fusion of spatiotemporally aligned multi-source observation data and state prediction values ​​to generate state estimation sequences for each traffic participant.

[0109] In vehicle-road cooperative scenarios, observation data comes from multiple heterogeneous sources: the mobile platform's own sensors (cameras, radar, etc.), fixed roadside facilities, and other connected vehicles. Each data source has its own observation noise characteristics, and these characteristics may change over time.

[0110] Suppose we fuse three data sources: our own millimeter-wave radar (source 1), fixed roadside cameras (source 2), and V2X reports from nearby vehicles (source 3), and the observation noise covariance matrix R is a block diagonal matrix:

[0111]

[0112] R1, R2, and R3 are the observation noise covariance submatrices of the three data sources, respectively.

[0113] First, perform the initial settings: R1 is preset to the nominal noise level of the radar itself; R2 is preset to the nominal noise level of the roadside camera; R3 is preset to the typical noise level of V2X data.

[0114] Next, online adaptive adjustment: by covariance matching, the information sequence of each data source is monitored. If the information covariance of source 2 (roadside camera) is found to be consistently large, R2 is increased accordingly; if source 1 (its own radar) is working stably, R1 is kept unchanged; if the quality of source 3 (V2X data) is inconsistent, R3 is dynamically adjusted.

[0115] Finally, the optimal fusion: the adjusted R' reflects the current actual credibility of each data source, and for sources with decreased credibility (R'... i Increase the weight of the corresponding part in the Kalman gain matrix K, and decrease its weight (reducing the corresponding part in the Kalman gain matrix K); maintain the weight of sources with stable confidence; all data are weighted and fused in a unified state space according to their respective adjusted confidence levels. Finally, a global, high-refresh-rate, high-precision state estimation sequence of dynamic traffic targets is generated and continuously updated. The sequence for each target... It contains its historical and current state vectors. This sequence will serve as input data for trajectory prediction in the next step.

[0116] In some embodiments, step S104 involves using a preset trajectory prediction model to predict trajectories based on the state estimation sequences of each traffic participant, and outputting a multimodal predicted trajectory cluster with probability weights. This includes: inputting the state estimation sequence of the target traffic participant as an observation sequence into a pre-trained driving behavior intention inference model to obtain the probability distribution of each behavioral intention of the target traffic participant at the current time; obtaining the historical trajectory of the target traffic participant based on the state estimation sequence of the target traffic participant; matching the historical trajectory of the target traffic participant with the preset typical trajectory templates corresponding to each behavioral intention; and outputting a multimodal predicted trajectory cluster with probability weights based on the matching results and the probability distribution of each behavioral intention of the target traffic participant at the current time.

[0117] The driving behavior intention inference model employs a Hidden Markov Model (HMM). Based on this, the hidden state set S = {lane keeping, left lane change, right lane change, sudden deceleration}. The observation sequence O is taken from the S103 high-precision state estimation sequence (e.g., vehicle state characteristics such as lateral displacement, yaw angle, and acceleration). The probability distribution P of each behavioral intention at the current moment is calculated using a forward-backward algorithm. I .

[0118] The following explanation uses a vehicle as an example to illustrate the probabilistic inference of driving intentions: First, the system categorizes the vehicle's macroscopic driving intentions into several discrete typical patterns, such as lane keeping, left lane change, right lane change, and emergency braking. Then, it receives the target vehicle's state estimation sequence over a recent period (e.g., 2-3 seconds). This sequence contains detailed information on the vehicle's position, speed, heading angle, and yaw rate as they evolve over time. The driving intention inference model learns from a large amount of historical driving data, mastering the statistical patterns of vehicle state changes under different driving intentions. Based on this, the aforementioned state estimation sequence is input into the model as an observation sequence. The model then uses its internal forward-backward algorithm (two core computational methods in Hidden Markov Models, used together to solve the evaluation problem of Hidden Markov Models, calculating the probability of the observation sequence occurring given model parameters and the observation sequence) to analyze the degree of matching between the sequence and various driving intention patterns. The output is the probability value of the target vehicle belonging to each preset driving intention at the current moment, forming a probability distribution vector, denoted as [missing information - likely a typo]. P I For example, a possible output would be: [Left lane change: 0.7, Lane keeping: 0.25, Right lane change: 0.05].

[0119] Multiple typical trajectory templates are pre-defined for each intent. Since the state estimation sequence includes historical trajectories, after obtaining the historical trajectory of the target traffic participant based on the state estimation sequence, the coordinate sequence of the target traffic participant's historical trajectory is matched with the pre-defined typical trajectory coordinate templates under each behavioral intent using the Dynamic Time Warping (DTW) algorithm. The shape similarity between the two is calculated to obtain a similarity score S. T Based on the current state, multiple continuous trajectories for the next 5-8 seconds are generated for each high-probability intention.

[0120] To transform discrete intentions into concrete spatial trajectories, the system performs the following parallel processing: trajectory template matching, where the system pre-stores multiple typical trajectory templates for each behavioral intention, defining common spatial paths to complete that intention; and employing Dynamic Time Warping (DTW) to compare the target vehicle's actual historical driving trajectory with all trajectory templates for each intention. The DTW algorithm elastically aligns two trajectories of potentially different lengths, calculates their shape similarity, and outputs a similarity score. S T The higher the score, the more closely the vehicle's recent driving pattern matches the typical pattern of that intention; future trajectory projections, for each high-probability (e.g., P I For an intention (greater than a set threshold), numerical simulation is performed using the vehicle's current precise state (position, speed, heading, etc.) as initial conditions, combined with a vehicle kinematics model matching the intention (e.g., a polynomial lane-changing model for a lane-changing intention). This generates multiple possible trajectories for the next 5-8 seconds, each representing a possible route for the vehicle under that intention. Multiple trajectories with slightly different parameters can be generated for the same intention to cover the uncertainties of driving behavior.

[0121] The confidence level of each candidate trajectory is determined by (Dynamic feasibility constraints, such as maximum velocity constraints, maximum curvature constraints, and maximum acceleration constraints) are comprehensively evaluated. The final output is a cluster of possible future trajectories with probability weights, characterizing the uncertainty of future behavior.

[0122] For the k Candidate trajectories Its overall confidence score The calculation is as follows:

[0123]

[0124] in, The HMM probability corresponding to the intent. To determine the similarity with the template trajectory, A dynamic feasibility score is given.

[0125] Then, the scores of all candidate trajectories are normalized to obtain the final probability weight for each trajectory:

[0126]

[0127] The multiple possible trajectories generated in the previous step for the next 5-8 seconds are comprehensively evaluated and screened, including dynamic feasibility review, comprehensive confidence calculation, normalization and output.

[0128] Dynamics Feasibility Review: Each possible trajectory must pass a check by a dynamics feasibility constraint module. This module scores the smoothness and executability of the trajectory based on vehicle physical limits (such as maximum acceleration, maximum curvature, and friction circle limits), denoted as _____. Trajectories that do not meet the basic physical constraints will be assigned extremely low scores or eliminated.

[0129] Overall confidence score calculation: For each candidate trajectory, its overall confidence score (Score) is determined by the following formula:

[0130]

[0131] That is, the score combines the prior probability of its corresponding intention, the similarity to historical driving patterns, and its own physical rationality.

[0132] Normalization and Output: Normalize the scores of all candidate trajectories so that the sum of their weights is 1, thus obtaining the final probability weight of each trajectory.

[0133] Ultimately, the system outputs a set of trajectory-probability pairs, called the multimodal trajectory family. The mathematical representation of this set is:

[0134]

[0135] in, It represents a specific future trajectory (a series of spatial coordinate points). The corresponding occurrence probability weight.

[0136] Through the above steps, a leap from high-precision state perception to probabilistic behavior prediction has been achieved. The output trajectory cluster not only indicates the most likely future scenario but also quantifies the probability of other risk scenarios, providing crucial prior information for subsequent collaborative decision-making.

[0137] In some embodiments, step S105, sending decision suggestions to the corresponding traffic participants based on the multimodal predicted trajectory clusters of each traffic participant, includes: determining traffic conflict risk information based on the multimodal predicted trajectory clusters of each traffic participant; generating a traffic planning strategy based on the traffic conflict risk information; and sending decision suggestions to the corresponding traffic participants based on the traffic planning strategy.

[0138] It is understandable that after obtaining the multimodal predicted trajectory clusters of each traffic participant with probability weights, traffic conflict risk information will be determined through spatiotemporal trajectory cross-analysis, taking into account environmental constraints such as road topology, traffic signal status, and lane traffic rules in the target service area.

[0139] Specifically, the system compares the multimodal predicted trajectories of two or more traffic participants one by one, focusing on the possibility of overlap of trajectories in spatial coordinates at different future time points. At the same time, it quantifies the probability of conflict occurrence by combining the probability weight of each predicted trajectory. For example, if the trajectory of vehicle A going straight (probability weight 80%) and the trajectory of vehicle B turning left (probability weight 75%) overlap in the central area of ​​the intersection and the overlap time difference is less than the safety threshold, it is judged as a high-risk conflict. If the trajectory overlap probability is less than 30% or there is a sufficient safety buffer distance, it is judged as a low-risk conflict. Finally, it forms complete traffic conflict risk information that includes the objects involved in the conflict, the location and time of the conflict, the risk level, and the type of conflict.

[0140] Based on this risk information, targeted traffic planning strategies are further generated, adhering to the principle of "prioritizing the avoidance of high-risk conflicts while balancing traffic efficiency and order." For high-risk conflict scenarios, differentiated avoidance strategies are formulated, such as planning deceleration and avoidance paths for one party in the conflict, and adjusting the driving trajectory or changing lanes for the other party to ensure safe isolation between the two parties in the spatiotemporal dimensions. For medium-risk conflict scenarios, strategies to optimize traffic timing are generated, such as suggesting that a participant slightly increase or decrease speed to avoid conflict time windows. For low-risk conflict scenarios, strategies to maintain the current driving status but enhance dynamic monitoring are formulated. Finally, the system uses precise channels such as V2X vehicle-to-infrastructure communication links, roadside broadcasting equipment, and in-vehicle terminal prompts to send decision suggestions based on traffic planning strategies to the corresponding traffic participants.

[0141] In some embodiments, determining traffic conflict risk information based on the multimodal predicted trajectory clusters of each traffic participant includes: acquiring the multimodal predicted trajectory clusters of target traffic participant pairs; determining the predicted trajectory combinations of the target traffic participant pairs, and determining whether there are spatiotemporal overlap points for each predicted trajectory combination pair within a preset time period in the future; identifying the predicted trajectory combination pairs with spatiotemporal overlap points as conflict trajectory combination pairs, calculating the joint conflict probability corresponding to each conflict trajectory combination pair, summing the joint probabilities of each conflict trajectory combination pair to obtain the conflict probability of the target traffic participant pair; and generating risk warning information of a corresponding level based on the conflict probability of the target traffic participant pair.

[0142] First, the conflict risk probability is calculated: For any two traffic participants (e.g., vehicle A and vehicle B), the system iterates through all predicted trajectories of A and B. For each trajectory pair, it checks whether there is a spatiotemporal overlap point (i.e., the distance between them is less than a preset safety threshold) in the future time period. If so, the joint probability of the combination causing a conflict is calculated (which is the product of the individual probabilities of the two trajectories). The joint probabilities of all overlapping trajectory combinations are summed to obtain the total conflict probability between A and B.

[0143] For example, suppose the system predicts that vehicle A has three possible trajectories with probabilities of 50% (going straight), 30% (turning left), and 20% (turning right); pedestrian B has two possible trajectories with probabilities of 70% (crossing the road) and 30% (waiting). After checking, the system finds that their trajectories will only intersect after 2 seconds if A "turns left" and B "crosses the road." Therefore, the probability of conflict between A and B is: 30% × 70% = 21%.

[0144] Tiered Warning Generation: Three risk probability thresholds (e.g., 20% and 60%) are preset. Based on the real-time conflict probability calculated in the previous step, different levels of warning information are generated and sent through V2X communication: Low risk (≤20%): only recorded, no active intervention; Medium risk warning: when the conflict probability is between 20% and 60%, a prompt warning is sent to relevant vehicles, suggesting that the driver or autonomous driving system take preventive measures such as attention and deceleration; High risk emergency warning: when the conflict probability is greater than 60%, an emergency braking or emergency avoidance warning is sent to relevant vehicles, requiring immediate strong intervention measures.

[0145] Finally, macro-level traffic coordination and control are implemented. Based on a comprehensive prediction of the future state of all traffic participants in the region, global optimization is carried out: dynamic speed guidance, calculating and broadcasting recommended speeds upstream of congestion-prone or accident-prone areas to smoothly decelerate traffic flow and improve traffic safety and efficiency; optimal detour route planning, planning and pushing alternative routes for affected vehicles when road interruptions or severe congestion are detected; and cooperative right-of-way allocation, dynamically allocating right-of-way at unsignalized intersections based on the predicted arrival time of vehicles in each direction and vehicle type priority, and using V2X communication commands to instruct relevant vehicles to "pass" or "yield," thereby achieving efficient and safe coordination at intersections in a digital manner.

[0146] In some embodiments, please refer to Figure 3 This application provides a vehicle-road cooperative traffic management system 100, which includes: a mobile vehicle 101, a sensing module 102 and a processing module 103.

[0147] The mobile vehicle 101 carries the sensing module 102 and the processing module 103. The sensing module 102 scans the surrounding wide-area environment to obtain monitoring information. The processing module 103 is connected to both the mobile vehicle 101 and the sensing module 102. The processing module 103 controls the movement of the mobile vehicle 101 and executes the vehicle-road cooperative traffic management method as described in the first aspect.

[0148] For example, such as Figure 4 As shown, the vehicle-road cooperative traffic management system can be deployed on a self-driving / towing mobile vehicle. This mobile vehicle may include: a load-bearing and leveling unit to ensure stable deployment on non-level roads; a liftable multi-functional mast with an integrated cluster of sensing sensors on top; and a comprehensive power supply and management unit supporting multi-source power supply and intelligent switching between mains power, generators, and energy storage batteries. The sensing module includes at least LiDAR, millimeter-wave radar, and high-definition intelligent cameras, with redundantly configurable sensor parameters. The processing module may embed: a main control and edge computing unit, which can be equipped with a high-performance GPU and AI acceleration chip; a dynamic networking coordination unit; an adaptive fusion sensing unit (AUKF core); a multimodal behavior prediction unit; a traffic decision and service generation unit; and a multi-layered communication subsystem that can simultaneously support C-V2X (vehicle-to-infrastructure, vehicle-to-vehicle), 4G / 5G (backhaul), and Wi-Fi / Bluetooth (near-field device access).

[0149] It should be noted that the vehicle-road cooperative traffic management system provided in this application embodiment and the vehicle-road cooperative traffic management method provided in this application embodiment are based on the same inventive concept. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned vehicle-road cooperative traffic management method, and the repeated parts will not be described again.

[0150] In some embodiments, an electronic device provided in this application includes a processor and a memory; the memory stores a computer program, wherein the computer program, when executed by the processor, implements the above-described vehicle-road cooperative traffic management method.

[0151] Specifically, the processor may include, for example, a general-purpose microprocessor, an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor may also include onboard memory for caching purposes. The processor may be a single processing unit or multiple processing units for performing different actions of the method flow according to embodiments of this application.

[0152] Memory can be any medium capable of containing, storing, transmitting, propagating, or transmitting instructions. For example, memory can include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, instruments, or propagation media. Specific examples of memory include: magnetic storage devices such as magnetic tape or hard disk drives (HDDs); optical storage devices such as optical discs (CD-ROMs); and also random access memory (RAM) or flash memory; and / or wired / wireless communication links.

[0153] This application also provides a non-transitory computer storage medium storing a computer program thereon, which, when executed by a processor, implements the vehicle-road cooperative traffic management method described above. This computer-readable medium may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into that device / apparatus / system. The aforementioned computer-readable medium carries one or more programs, which, when executed, implement the method as described in the embodiments of this application.

[0154] According to embodiments of this application, a computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wired, optical fiber, radio frequency signals, etc., or any suitable combination thereof.

[0155] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0156] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application. Therefore, the scope of this application should not be limited to the above embodiments. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A vehicle-road cooperative traffic management method, characterized in that, The vehicle-road cooperative traffic management method is applied to a vehicle-road cooperative traffic management system with mobile cooperative capabilities; The vehicle-road cooperative traffic management method includes: The system receives traffic incident information and instruction messages from the traffic management cloud control platform, and controls the vehicle-road cooperative traffic management system to move to the target point according to the instruction messages; wherein, the traffic incident information includes at least one of the real-time traffic event list of the authorized traffic information service platform and the monitoring information of the vehicle-road cooperative traffic management system. Broadcast collaborative service signaling and receive response information from communication terminals within the service area; Each communication terminal is registered as a resource node based on its response information, and the attribute information of each resource node is recorded, including ID, location information, and dynamic capability set. The target service area is determined based on the traffic incident information and the instruction message, and a dynamic resource map of the target service area is constructed. The target service area is the area to be managed at the accident point or congestion section. The dynamic resource map includes at least one resource node. The resource node includes the local sensing device on the system itself and the distributed sensing resources in the target service area that have been connected to the vehicle-road cooperative network. The current macro task is determined according to the instruction message, the current macro task is decomposed into multiple micro sub-tasks, each micro sub-task is optimally matched with the capabilities of each resource node, the execution tasks of each resource node are assigned according to the matching results, and the observation data obtained by each resource node in executing the assigned tasks are obtained. A preset fusion algorithm is used to fuse the observation data of each resource node in the target service area to generate a state estimation sequence for each traffic participant. The preset fusion algorithm is an adaptive unscented Kalman filter algorithm or an unscented Kalman filter algorithm. Based on the state estimation sequence of each traffic participant, a preset trajectory prediction model is used to predict the trajectory and output a multimodal predicted trajectory cluster with probability weights, wherein the weight value of the probability weight reflects the probability of the corresponding trajectory occurring. Decision suggestions are sent to the corresponding traffic participants based on the multimodal predicted trajectory clusters of each traffic participant.

2. The vehicle-road cooperative traffic management method as described in claim 1, characterized in that, The step involves fusing the observation data of each resource node within the target service area using a preset fusion algorithm to generate a state estimation sequence for each traffic participant, including: The observation data of each resource node is synchronized to a unified spatiotemporal reference to obtain spatiotemporally aligned multi-source observation data; Based on the spatiotemporally aligned multi-source observation data, high-dimensional state vectors are established for each traffic participant in the target service area, wherein the high-dimensional state vectors include position, velocity, heading angle and yaw rate. Sigma point sampling and nonlinear prediction are performed on each of the high-dimensional state vectors to obtain the state prediction value. Based on the state prediction value, the covariance of the observed innovation sequence is calculated for matching. An adaptive unscented Kalman filter algorithm is used to dynamically adjust the process noise matrix and the observation noise matrix, and key filter parameters are calculated. The innovation sequence is the difference between the actual observation value and the predicted observation value. The spatiotemporally aligned multi-source observation data is mapped to a unified state space, and the optimal Kalman gain is calculated by combining the adaptively adjusted process noise matrix, observation noise matrix and key filtering parameters. Based on the optimal Kalman gain, the spatiotemporally aligned multi-source observation data and the state prediction values ​​are optimally fused to generate a state estimation sequence for each traffic participant.

3. The vehicle-road cooperative traffic management method as described in claim 1, characterized in that, The process involves using a preset trajectory prediction model to predict trajectories based on the state estimation sequences of each traffic participant, outputting a multimodal predicted trajectory cluster with probability weights, including: The state estimation sequence of the target traffic participant is used as the observation sequence to input the pre-trained driving behavior intention inference model to obtain the probability distribution of each behavior intention of the target traffic participant at the current time. The driving behavior intention inference model is a hidden Markov model. Based on the state estimation sequence of the target traffic participant, the historical trajectory of the target traffic participant is obtained. The historical trajectory of the target traffic participant is matched with the preset typical trajectory template corresponding to each behavioral intention. Based on the matching result and the probability distribution of each behavioral intention of the target traffic participant at the current time, a multimodal predicted trajectory cluster with probability weights is output.

4. The vehicle-road cooperative traffic management method as described in claim 1, characterized in that, The step of sending decision suggestions to the corresponding traffic participants based on the multimodal predicted trajectory clusters of each traffic participant includes: Traffic conflict risk information is determined based on the multimodal predicted trajectory clusters of each traffic participant. Traffic planning strategies are generated based on traffic conflict risk information, and decision-making suggestions are sent to relevant traffic participants based on the traffic planning strategies.

5. The vehicle-road cooperative traffic management method as described in claim 4, characterized in that, The determination of traffic conflict risk information based on the multimodal predicted trajectory clusters of each traffic participant includes: Obtain multimodal predicted trajectory clusters for target traffic participant combinations; Determine the predicted trajectory combinations of the target traffic participant pairs, and determine whether there are any spatiotemporal overlaps between the predicted trajectory pairs within a preset time period in the future. Predicted trajectory pairs with spatiotemporal overlap points are identified as conflict trajectory pairs. The joint conflict probability of each conflict trajectory pair is calculated. The joint probabilities of each conflict trajectory pair are added together to obtain the conflict probability of the target traffic participant pair. Risk warning information of corresponding levels is generated based on the conflict probability of the target traffic participant pair.

6. A vehicle-road cooperative traffic management system, characterized in that, include: Mobile vehicle, sensing module, and processing module; The mobile vehicle is used to carry the sensing module and the processing module; The sensing module is used to scan the surrounding wide-area environment to obtain monitoring information; The processing module is connected to the mobile vehicle and the sensing module respectively. The processing module is used to control the movement of the mobile vehicle and to execute the vehicle-road cooperative traffic management method as described in any one of claims 1 to 5.

7. An electronic device, characterized in that, It includes a processor and a memory; the memory stores a computer program, wherein the computer program, when executed by the processor, implements the vehicle-road cooperative traffic management method as described in any one of claims 1 to 5.

8. A non-transitory computer storage medium, characterized in that, It stores a computer program, wherein the computer program, when executed by a processor, implements the vehicle-road cooperative traffic management method as described in any one of claims 1 to 5.

Images