A method and system for highway traffic flow prediction and accident identification

Abstract

The application relates to a highway traffic flow prediction and accident identification method and system, and relates to the technical field of intelligent detection, which comprises the following steps: firstly, a mobile robot is used to monitor a highway in real time, traffic flow characteristic data are acquired, feature extraction analysis and fusion are carried out, and traffic flow prediction data of a specified time period are obtained. Then, according to the traffic flow prediction data and in combination with real-time acquired traffic flow data, accident identification and accident classification are carried out, when an accident on the highway is identified, an algorithm is used, real-time identified data are combined, dynamic path planning is carried out, and based on the planned path, the mobile robot is controlled to perform a dispersing task on the lane of the highway. Through the automatic monitoring of the robot and the intelligent algorithm, the application realizes the whole-process unmanned of traffic flow prediction, accident identification and disposal, reduces manual intervention, improves monitoring efficiency and staff safety, and reduces operation cost.

Description

Technical Field

[0001] This application relates to the field of intelligent detection technology, and in particular to a method and system for predicting traffic flow and identifying accidents on highways. Background Technology

[0002] With the large-scale construction of highway networks and the continuous growth of traffic flow, the safe operation and efficient management of highways face increasingly severe challenges. As a crucial backbone of the national comprehensive three-dimensional transportation network, the efficiency and safety of highways directly impact socio-economic development and the public's travel experience. Traditional highway management models primarily rely on manual inspections, fixed surveillance cameras, and limited traffic guidance facilities, resulting in insufficient real-time performance, delayed emergency response, and incomplete traffic situation awareness, making it difficult to meet the demands of refined management in complex traffic environments. Therefore, developing highway robot systems with intelligent monitoring and dynamic traffic management capabilities has become a key direction for improving traffic management efficiency and ensuring safe and smooth road traffic.

[0003] In recent years, the rapid development of technologies such as artificial intelligence, the Internet of Things, autonomous driving, and 5G communication has injected new momentum into the intelligent management of highways. By integrating multimodal sensing devices, intelligent decision-making algorithms, and autonomous execution mechanisms, intelligent monitoring and traffic management robots on highways can achieve real-time collection and processing of road conditions, traffic flow information, and emergencies. Combined with cloud-based traffic big data analysis, they can dynamically generate optimal traffic management strategies and execute them automatically, thereby constructing an integrated intelligent traffic management system of "perception-decision-execution." The application of this intelligent equipment can effectively compensate for the shortcomings of traditional management models and improve the management efficiency and emergency response capabilities of highways in scenarios such as daily operation, accident handling, and severe weather response.

[0004] Current highway traffic flow monitoring typically employs a combination of fixed cameras and traditional algorithm models. Cameras are mounted on either side of the road or on gantries to record road conditions and detect vehicles, then traffic flow is calculated based on the corresponding algorithms. Furthermore, some highways utilize accident identification methods, primarily relying on designated accident detection and handling systems. These systems depend on manual patrols, requiring staff to conduct scheduled inspections at fixed locations to manage traffic and handle accidents upon discovery. This approach cannot automate the accurate identification of accidents. Therefore, existing technologies suffer from low efficiency in automated highway traffic flow monitoring and an inability to automatically identify and handle accidents. Summary of the Invention

[0005] This application provides a method and system for highway traffic flow prediction and accident identification, addressing the limitations of existing technologies, such as the limited endurance and anti-interference capabilities of drones, which make it difficult for them to operate stably in adverse weather conditions. This application integrates multi-modal sensors, including rain- and fog-resistant cameras and millimeter-wave radar, into a ground mobile robot for automated monitoring. Combined with data fusion algorithms, it significantly improves monitoring reliability in complex environments. Building upon traffic flow prediction, it further enables fully automated accident identification and route planning, enhancing monitoring efficiency and solving the problems of low efficiency in existing highway automation traffic flow monitoring and the inability to automatically identify / handle accidents.

[0006] Firstly, this application provides a method for predicting highway traffic flow and identifying accidents, including:

[0007] The acquired highway traffic flow characteristic data is normalized to obtain preprocessed target traffic flow characteristic data. The traffic flow characteristic data is obtained by the mobile robot monitoring the highway traffic flow in real time during dynamic patrol.

[0008] Temporal feature convolution processing is performed on the target traffic flow feature data to obtain temporal feature information, and spatial feature convolution processing is performed on the acquired highway segment data to obtain spatial feature information.

[0009] Based on the temporal and spatial feature information, traffic flow prediction data for a specified time period is obtained by splicing and fusing them.

[0010] Accident identification and accident classification assessment are performed based on the traffic flow prediction data and the real-time traffic flow data acquired in real time to determine accident information, which includes accident analysis results, accident impact range, and accident level.

[0011] Using a dynamic path planning algorithm, based on the traffic flow prediction data, the real-time traffic flow data, and accident information, dynamic path planning is performed to obtain target path planning information. The mobile robot is then controlled to perform traffic management tasks on the highway lanes based on the target path planning information.

[0012] Optionally, temporal feature convolution processing is performed on the target traffic flow feature data to obtain temporal feature information, and spatial feature convolution processing is performed on the acquired highway segment data to obtain spatial feature information, including:

[0013] Using the target traffic flow characteristic data as input, according to Δ d The expression Q(t) = Q(t) - Q(td) is differenced to eliminate non-stationarity, and then input into a preset autoregressive model. Perform convolution processing to extract temporal feature information;

[0014] Analyze the monitoring points and connection relationships of highway segments, and perform abstract processing to construct an abstract segment graph G=(V,E);

[0015] Using nodes V and edges E in the road segment abstract graph as input, according to Perform graph convolution operations to obtain spatial feature information;

[0016] Among them, Q(t) is the preprocessed target traffic flow characteristic data, α p Autoregressive coefficient, β p Here are the moving average coefficients, and ∈(t) represents white noise. It is an adjacency matrix with self-loops. W is the degree matrix. (l) Let X be the weight of the l-th layer. (l) These are node features.

[0017] Optionally, accident identification and accident classification assessment are performed based on the traffic flow prediction data and the real-time traffic flow data acquired in real time to determine accident information, including:

[0018] Based on the traffic flow prediction data, combined with real-time traffic flow data, traffic flow abrupt change detection is performed to obtain abrupt change analysis results.

[0019] When the mutation analysis result is a confirmed mutation result, video data of each lane of the highway is acquired for visual feature analysis to determine the accident analysis result.

[0020] When the accident analysis result indicates that an accident has occurred, the scope of impact and the accident level are analyzed and assessed based on the traffic flow prediction data and the real-time traffic flow data to obtain the scope of the accident's impact and the accident level.

[0021] Optionally, based on the traffic flow prediction data, combined with real-time traffic flow data, traffic flow abrupt change detection is performed to obtain abrupt change analysis results, including:

[0022] Based on the traffic flow prediction data and the real-time traffic flow data, the average speed v(t) of each vehicle in the lane and the vehicle density ρ(t) in the lane at time t are analyzed, and the abrupt change judgment formula is applied. Calculate the results of the mutation analysis;

[0023] Where μ is the historical mean, σ is the standard deviation, and w1, w2 and w3 are all weighting coefficients. When the score is greater than the preset threshold, it is determined that an accident may have occurred, and a definite mutation result is obtained.

[0024] Optionally, video data from each lane of the highway can be acquired for visual feature analysis to determine the accident analysis results, including:

[0025] The acquired video data is detected using an object detection algorithm, resulting in a vehicle bounding box set B = {b1,…,b...}. n};

[0026] Based on the vehicle bounding box set B = {b1, ..., b} n Trajectory analysis is performed to obtain the vehicle attitude angle set S = {s1, ..., s2}. n};

[0027] The vehicle attitude angle set S = {s1, ..., s2} n} is the input, according to Analyze key anomaly features to obtain the variance of collision angles.

[0028] Based on the vehicle bounding box set B = {b1, ..., b} n} Analyze the pixels of the vehicle bounding box to obtain the total number of pixels b, and mark the pixels covered by the vehicle bounding box as the vehicle region and the pixels not covered as the non-vehicle region to obtain the number of pixels a in the non-vehicle region;

[0029] Using the total number of pixels b and the number of pixels a as input, according to R debris =a / b analysis of key anomaly features to obtain the proportion R of fragmented regions debris ;

[0030] Based on collision angle variance and fragmented area ratio R debris According to the decision function Conduct accident assessment and obtain accident information;

[0031] Among them, P accident When the value is ≥0.8, an accident is determined to have occurred, and the accident occurrence result is obtained. k1 and k2 are both weighting coefficients, b i Let s represent the coordinates of the rectangular bounding box of the vehicle in the i-th frame of the image. i This represents the driving direction or body posture angle of the i-th vehicle in the image.

[0032] Optionally, based on the traffic flow prediction data and the real-time traffic flow data, an impact range analysis and accident level assessment are performed to obtain the accident impact range and accident level, including:

[0033] Based on the traffic flow prediction data and the real-time traffic flow data, analyze the upstream traffic flow q1 and downstream traffic flow q2 of the highway.

[0034] Based on upstream traffic flow q1 and downstream traffic flow q2, according to vw =q2-q1 / ρ2-ρ1, calculate the congestion propagation speed v w ;

[0035] Based on the congestion propagation speed v w According to L max =v w ·T response Calculate the maximum impact distance as the scope of the accident's impact;

[0036] Based on the vehicle bounding box set, analyze the covered lane area, determine the number of blocked lanes, and obtain the probability of personnel injury;

[0037] Based on the traffic flow prediction data and the real-time traffic flow data, the driving speed v of each vehicle in the congested section of the highway is analyzed. i Driving distance l i and the free-flow velocity v free ;

[0038] Based on driving speed v i Driving distance l i and the free-flow velocity v free According to V delay =∑ i (v free -v i )·l i Calculate the total delay kilometers;

[0039] Accident levels are determined using the fuzzy comprehensive evaluation method based on total delay kilometers.

[0040] Optionally, dynamic path planning is performed based on the traffic flow prediction data, the real-time traffic flow data, and accident information using a dynamic path planning algorithm, including:

[0041] Based on the traffic flow prediction data, the real-time traffic flow data, the highway road data, and the accident information, a road network model is constructed, and a cost function is used for path optimization to obtain rescue path optimization information. The rescue path optimization information is used to plan the optimal path from the starting point to the accident point for the robot.

[0042] Based on the accident points corresponding to the accident information, the gradient descent method of potential field theory is used to plan the diversion path and generate the target diversion path.

[0043] Based on the optimal path and the target diversion path, target path planning information is generated.

[0044] Optionally, a road network model is performed based on the traffic flow prediction data, the real-time traffic flow data, the highway road data, and the accident information, and path optimization is performed using a cost function to obtain rescue path optimization information, including:

[0045] Based on the traffic flow prediction data and the real-time traffic flow data, the real-time traffic flow q is determined. ij And, determining road capacity C based on highway road data. ij ;

[0046] With real-time traffic q ij and road capacity C ij For input, according to Road network modeling is performed, and the cost function f(n) = g(n) + h(n) + γVar(c) is used in the improved algorithm. ij ) Perform route optimization to obtain rescue route optimization information;

[0047] Based on the accident points corresponding to the accident information, a diversion path is planned using the gradient descent method of potential field theory to generate a target diversion path. This includes: analyzing the accident points using the gradient descent method of potential field theory, and based on... Perform traffic splitting path planning and generate target traffic splitting paths;

[0048] Among them, the starting point i and ending point j and d of the rescue route are identified through accident information. ij v represents the distance of the road segment. ij (t) represents the real-time average speed of the road segment at time t, λ is the weighting coefficient, g(n) represents the actual cost from the starting point to the current node, h(n) represents the heuristically estimated cost from the current node to the destination, γ represents the path stability factor, and Var(c ij The ) represents the variance of the cost of each segment in the path. Let Q be the gradient of the potential field. i The repulsive force intensity at the accident point is represented by r, and the current vehicle position coordinates are also represented by r. i Indicates the coordinates of the accident point. This represents a unit vector pointing from the point of the accident to the vehicle.

[0049] Optionally, after obtaining the target path planning information, the following may also be included:

[0050] When dispatching the mobile robot to the accident point corresponding to the accident information, dispatch information x is obtained. ij and scheduling time t ij Define decision variables

[0051] Construct the objective function based on the decision variables. And construct constraints based on the decision variables.

[0052] Resource scheduling optimization for mobile robots based on objective functions and constraints;

[0053] After dispatching mobile robots to perform traffic control tasks, the handling information is obtained, including the handling time and the scope of the accident's impact.

[0054] Based on the information provided, according to Calculate the recovery efficiency η;

[0055] The reward value r is determined based on the recovery efficiency η, according to... Construct a reward function and continuously iterate and update the learning mechanism to update the mobile robot's execution strategy when an accident occurs;

[0056] Among them, w j For the weights determined based on the accident level, T actual T represents the actual processing time. predict To estimate the processing time, L actual L represents the actual scope of the accident's impact. actual Let α be the expected impact range of the accident, γ be the learning rate, γ be the discount factor, (s,a) represent the current state and action, and s' represent the new state after the action is performed.

[0057] Secondly, this application provides a highway traffic flow prediction and accident identification system, including:

[0058] The traffic flow characteristic data processing module is used to normalize the acquired highway traffic flow characteristic data to obtain preprocessed target traffic flow characteristic data. The traffic flow characteristic data is obtained by the mobile robot in real time monitoring the highway traffic flow during dynamic patrols. The mobile robot can be equipped with an anti-interference multimodal sensor array, including an anti-rain and fog camera, millimeter-wave radar, lidar, and an inertial navigation system. It fuses multi-source data through an adaptive Kalman filter algorithm to suppress environmental noise (such as rain, snow, and strong light interference) and ensure stable collection of traffic flow data in extreme environments such as temperature ranges of -40℃ to 70℃ and heavy rainfall to determine the traffic flow characteristic data.

[0059] The feature convolution extraction module is used to perform temporal feature convolution processing on the target traffic flow feature data to obtain temporal feature information, and to perform spatial feature convolution processing on the acquired highway segment data to obtain spatial feature information.

[0060] The feature splicing and fusion module is used to splice and fuse the time feature information and the spatial feature information to obtain traffic flow prediction data for a specified time period;

[0061] The accident identification and classification module is used to identify and classify accidents based on the traffic flow prediction data and the real-time traffic flow data acquired in real time, and to determine accident information, which includes accident analysis results, accident impact range and accident level.

[0062] The path planning module is used to perform dynamic path planning based on the traffic flow prediction data, the real-time traffic flow data, and accident information using a dynamic path planning algorithm to obtain target path planning information. It then controls the mobile robot to perform traffic management tasks on the highway lanes based on the target path planning information. It can support multi-robot collaborative operations. Through a distributed collaborative scheduling algorithm, combined with real-time traffic flow data and robot location information, it dynamically allocates traffic management tasks and optimizes the paths of each robot to avoid path conflicts, ensuring that multiple robots arrive at the accident point quickly and orderly, and minimizing interference with normal traffic flow.

[0063] In summary, this application first utilizes a mobile robot to monitor highways in real time, acquiring traffic flow characteristic data. Through feature extraction, analysis, and fusion, traffic flow prediction data for a specified time period is obtained. Then, based on the traffic flow prediction data and the real-time acquired traffic flow data, accident identification and classification are performed. When an accident is identified on the highway, an algorithm is used, combined with the real-time identified data, to perform dynamic path planning. Based on the planned path, the mobile robot is controlled to perform traffic management tasks on the highway lanes. On the one hand, this application, through a mobile monitoring and traffic management robot, combined with multi-source sensors (such as radar and video) and a dynamic deployment strategy, achieves dynamic coverage and real-time monitoring of the entire highway section, eliminating blind spots of fixed equipment and improving the comprehensiveness of road condition information collection. On the other hand, through automated monitoring based on the collected video data by the robot, combined with intelligent algorithms, an accurate traffic flow prediction, automated accident identification, and handling process is achieved, reducing manual intervention, improving monitoring efficiency and staff safety, and reducing operating costs. Existing technologies have problems with low efficiency in automated traffic flow monitoring on highways and the inability to automatically identify / handle accidents. Attached Figure Description

[0064] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0065] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0066] Figure 1 A flowchart illustrating a method for predicting highway traffic flow and identifying accidents, provided in an embodiment of this application;

[0067] Figure 2 This is a flowchart illustrating the steps of an optional embodiment of a highway traffic flow prediction and accident identification method.

[0068] Figure 3 This is a structural block diagram of a highway traffic flow prediction and accident identification system provided in an embodiment of this application. Detailed Implementation

[0069] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0070] To facilitate understanding of the embodiments of this application, further explanations and descriptions will be provided below in conjunction with the accompanying drawings and specific embodiments. These embodiments do not constitute a limitation on the embodiments of this application.

[0071] Figure 1 This is a flowchart illustrating a method for predicting highway traffic flow and identifying accidents, provided as an embodiment of this application. Figure 1 As shown in the embodiments of this application, the highway traffic flow prediction and accident identification method may specifically include the following steps:

[0072] Step 110: Normalize the acquired highway traffic flow characteristic data to obtain preprocessed target traffic flow characteristic data.

[0073] The traffic flow characteristic data is obtained by mobile robots monitoring the highway traffic flow in real time during dynamic patrols.

[0074] In this embodiment, the mobile robot is typically a mobile robot deployed on the highway. This mobile robot platform is equipped with multiple sensors, including rain- and fog-resistant cameras, radar, video sensors, and lidar, and integrates high-precision cameras, millimeter-wave radar, infrared sensors, weather detectors, and other multimodal sensing devices to form a fusion architecture. The mobile robot continuously monitors the highway in real time, providing comprehensive coverage of the entire highway section. By adjusting the robot's patrol path and sensor collaboration strategies in real time, it eliminates the blind spots of traditional fixed cameras, significantly improving the completeness and timeliness of road condition data collection.

[0075] In this embodiment, the mobile robot monitors the highway to obtain relevant data, including but not limited to historical traffic flow, real-time traffic flow, time characteristics (such as hourly / weekly / holiday), weather index, and traffic flow of adjacent road segments, forming traffic flow feature data. Then, the traffic flow feature data undergoes a series of preprocessing steps, including normalization, which uniformly scales the features of different data to a closed interval, eliminating differences in the feature scales of different data, to obtain the target traffic flow feature data.

[0076] Step 120: Perform temporal feature convolution processing on the target traffic flow feature data to obtain temporal feature information, and perform spatial feature convolution processing on the acquired highway segment data to obtain spatial feature information.

[0077] Step 130: Based on the time feature information and the spatial feature information, the traffic flow prediction data for the specified time period is obtained by splicing and fusing them.

[0078] Steps 120-130 are described uniformly as follows:

[0079] In this embodiment, road segment data includes, but is not limited to: each road segment, monitoring point, road segment connection relationship between monitoring points, lanes, length, etc.

[0080] In this embodiment, when performing temporal feature convolution on the target traffic flow characteristic data, the target traffic flow data is differentially processed to eliminate non-stationarity. Then, it is input into a model (such as an autoregressive model, ARIMA) to analyze temporal correlation. This includes capturing the trend (e.g., weekday / weekend differences) and periodicity (e.g., daily peaks) of traffic flow over time through historical traffic flow and time feature data, and extracting time series features through graph convolution operations to obtain temporal feature information. ARIMA eliminates the non-stationarity of traffic flow data through differential processing, and combines autoregression and moving average to model temporal correlation, thus providing reliable temporal dimension analysis support for full-segment traffic flow prediction and accident early warning. This technology, in conjunction with the subsequent spatial modeling capabilities of GCN, jointly constructs the "spatiotemporal perception" foundation of the highway intelligent monitoring system.

[0081] When performing spatial feature convolution on road segment data, the data is abstracted into a graph and input into a Graph Convolutional Network (GCN) for convolution operations. This captures the spatial correlations of local (adjacent road segments) and global (cross-regional road segments) data, extracting spatial feature information. The GCN abstracts highways into a graph network and utilizes the normalization operations of the adjacency matrix and degree matrix to achieve efficient modeling of spatial correlations between monitoring points. Its core value lies in: geometric invariance: regardless of the road segment layout, spatial features can be adaptively learned through the graph structure; end-to-end training: combined with ARIMA, it forms a spatiotemporal joint model, providing crucial support for full-segment traffic flow prediction and accident identification; engineering interpretability: the physical meaning of nodes and edges is clear, facilitating model optimization in conjunction with actual highway scenarios (e.g., adjusting the adjacency matrix to reflect real-time construction and diversion).

[0082] This embodiment uses ARIMA to process the trend and period of time, and GCN to process the neighbor influence of space. Then, the two are combined to perform feature fusion, which obtains traffic flow prediction data for a specified time period (such as the next h hours), realizing "spatiotemporal dual flow" modeling and improving the accuracy of traffic flow prediction.

[0083] Step 140: Based on the traffic flow prediction data and the real-time traffic flow data, perform accident identification and accident classification assessment to determine accident information.

[0084] The accident information includes accident analysis results, the scope of the accident's impact, and the accident level.

[0085] In this embodiment, the accident analysis results are divided into accident occurrence results and accident non-occurrence results.

[0086] In its implementation, this embodiment primarily determines whether an accident has occurred by analyzing changes in traffic flow and video data. Specifically, it combines traffic flow prediction data and real-time traffic flow data to analyze the average speed of each lane and other factors to determine if there are sudden traffic changes. This includes comparing and analyzing changes in vehicle speeds upstream and downstream of a lane, as well as changes in vehicle density within the lane, to analyze the circumstances of sudden traffic changes and determine whether a traffic accident has occurred. Furthermore, when a sudden traffic flow change is identified, visual feature analysis can be performed using corresponding video data. By analyzing vehicle changes through video data and combining this with the sudden traffic change, it can be determined whether a traffic accident has occurred. If a traffic accident is determined to have occurred, an accident occurrence result is generated; otherwise, an accident non-occurrence result is generated. This multi-source data analysis avoids misjudgments from a single sensor. For example, sudden traffic flow changes may be caused by weather, construction, etc. (e.g., heavy rain causing a decrease in vehicle speed), requiring visual evidence to eliminate interference; visual features may be affected by image blur (e.g., at night), requiring the assistance of traffic flow change signals to trigger the change.

[0087] In practice, traffic flow changes can be analyzed using traffic flow prediction data and real-time traffic flow data, including analyzing traffic flow upstream and downstream of lanes to determine the scope of accident impact.

[0088] In this embodiment, when an accident is determined to have occurred, it can be classified into different levels, primarily based on the severity of the accident (classified into levels I-IV). Accident classification can be based on changes in traffic flow to determine the degree of congestion, or by combining video data analysis to assess the degree of congestion. Furthermore, the probability of personal injury can be analyzed (typically based on predictions using onboard sensors (such as collision accelerometers), visual image analysis (person fall detection), or historical accident models). Then, the total delay kilometers, i.e., the total delay kilometers for all vehicles caused by the congestion due to the accident, are further analyzed to differentiate the accident level.

[0089] Furthermore, dynamic route planning can be implemented based on accident levels for traffic management and rescue. In addition, resource scheduling optimization can be achieved based on accident levels, that is, different levels of resources can be allocated according to different accident levels.

[0090] Step 150: Using a dynamic path planning algorithm, dynamic path planning is performed based on the traffic flow prediction data, the real-time traffic flow data, and accident information to obtain target path planning information. The mobile robot is then controlled to perform traffic management tasks on the lanes of the highway based on the target path planning information.

[0091] In this embodiment, the target path planning information includes the target planned path, which includes two optimized planned paths: the rescue path and the diversion path.

[0092] The dynamic path planning algorithm in this embodiment is an improved algorithm, mainly composed of two algorithms: ① An improved A* algorithm (heuristic path search algorithm), which uses the accident point and traffic flow data corresponding to the accident information as a basis to perform efficient optimal path search, thereby planning an efficient optimal path for rescue path optimization; ② A vehicle detour path generation algorithm (implemented by gradient descent method based on potential field theory), which mainly focuses on the analysis of the accident point and generates detour paths in reverse based on the accident point as diversion paths, so that vehicles can travel along the diversion paths.

[0093] In this specific implementation, after determining the target planning path, the robot can move on the highway lanes (such as the emergency lane) according to the target planning path to perform traffic control tasks.

[0094] As can be seen, existing technologies have limitations in monitoring range: fixed cameras can only shoot from a single direction, resulting in blind spots; and existing technologies rely on manual patrols, leading to significant manpower and time consumption, and affecting patrol efficiency and safety in inclement weather or complex road conditions. This embodiment first uses a mobile robot for dynamic patrol and real-time monitoring on highways to overcome the limitations of monitoring range. Then, it uses multi-source data acquired through real-time monitoring for feature extraction and fusion processing to accurately predict future traffic flow, achieving accurate traffic flow prediction. Addressing the shortcomings of existing technologies in real-time performance and response speed in accident detection and handling (specifically, from accident detection to information transmission and personnel arrival at the scene), the mobile robot uses real-time analyzed data for accident identification and classification to determine the accident situation. Then, an improved dynamic path planning algorithm is used to dynamically plan rescue and diversion routes. Finally, the mobile robot executes traffic management tasks according to the planned routes. Therefore, this embodiment achieves fully unmanned operation of traffic flow prediction, accident identification, and handling through robotic automated monitoring and intelligent algorithms, reducing human intervention, improving monitoring efficiency and staff safety, and lowering operating costs.

[0095] Reference Figure 2 This illustration shows a flowchart of a highway traffic flow prediction and accident identification method according to an optional embodiment of this application. The method may specifically include the following steps:

[0096] Step 210: Normalize the acquired highway traffic flow characteristic data to obtain preprocessed target traffic flow characteristic data.

[0097] The traffic flow characteristic data is obtained by mobile robots monitoring the highway traffic flow in real time during dynamic patrols.

[0098] In related technologies, to address the shortcomings of insufficient monitoring coverage and manual patrol methods, some existing technologies employ drone-based monitoring systems. Drones, equipped with cameras and other devices, take off along preset routes or as needed to monitor road conditions from the air. This existing technology has the following main drawbacks: ① Limited battery life: Drone batteries have short flight times, limiting the duration of a single monitoring session and making it difficult to continuously and stably monitor highways for extended periods; ② Weak anti-interference capability: In complex electromagnetic environments or adverse weather conditions, the flight stability and data transmission reliability of drones may be affected, leading to data loss or transmission failure; ③ Flight safety risks: Flying over highways with heavy traffic and complex environments poses certain safety risks for drones, such as collisions with other objects or crashes, potentially threatening the normal operation of highways and the safety of personnel.

[0099] This embodiment primarily utilizes an intelligent mobile robot. Based on its integrated architecture and combining spatiotemporal registration and data fusion algorithms, it can achieve comprehensive real-time monitoring of factors such as road surface damage, icing and water accumulation, traffic density, vehicle speed, and abnormal events. For example, by modeling with LiDAR point cloud data, road obstacles can be accurately identified, and by combining visual image analysis technology, dynamic classification of vehicle types and driving states can be achieved, providing multi-dimensional data support for subsequent decision-making.

[0100] In this specific implementation, a mobile robot monitors the highway to obtain historical traffic flow Q(t), time characteristics (hour / week / holiday T(t)), weather index W(t), and traffic flow Q of adjacent road segments. adj (t), etc. Then, the acquired traffic flow feature data is uniformly normalized to obtain the target traffic flow feature data.

[0101] For example, normalization can be achieved using a formula: Where X represents the input data. For example, when the input normalized data is historical traffic flow Q(t), then... By normalizing the data to the [0,1] interval, we ensure that all input features meet the model's uniform requirements for data format.

[0102] This embodiment eliminates significant differences in the original units and numerical ranges of different feature data through normalization, thus preventing the model from being dominated by high-value features. Secondly, it maps all features to the same interval, unifying the data scale, which facilitates the model (such as ARIMA, graph convolutional networks) to learn the weight relationships between features, thereby improving training efficiency and prediction accuracy. Finally, it maintains the data distribution, that is, it preserves the relative distribution characteristics of the original data, only adjusting the numerical range without changing the inherent laws of the data.

[0103] In practical applications, mobile robots are highly adaptable to complex environments, ensuring reliable and stable data. They utilize onboard high-precision equipment to achieve large-scale continuous monitoring, acquiring various types of data within the monitoring range. Combined with algorithms, they can solve the problem of detection failure of single devices in harsh weather or complex terrain.

[0104] Step 220: Perform temporal feature convolution processing on the target traffic flow feature data to obtain temporal feature information, and perform spatial feature convolution processing on the acquired highway segment data to obtain spatial feature information.

[0105] In practical implementation, considering that highway traffic flow data is usually non-stationary (e.g., trend-based or seasonal), direct modeling may lead to prediction bias. Therefore, differencing is introduced. Difference processing eliminates long-term trends and seasonality by calculating the differences between adjacent values ​​in the sequence, transforming it into a stationary sequence that meets the prerequisites of the ARIMA model. Then, the model extracts time-series features, capturing the time dependencies in the stationary sequence, thus achieving modeling of the time-series characteristics of traffic flow.

[0106] For spatial characteristics, the main focus is on capturing the spatial dependencies between different sections of the highway, such as the impact of upstream congestion on downstream traffic flow.

[0107] In an optional embodiment, the above-mentioned temporal feature convolution processing based on the target traffic flow feature data to obtain temporal feature information, and spatial feature convolution processing based on the acquired highway segment data to obtain spatial feature information, may include: taking the target traffic flow feature data as input, and according to Δ... d The expression Q(t) = Q(t) - Q(td) is differenced to eliminate non-stationarity, and then input into a preset autoregressive model. Convolution processing is performed to extract temporal feature information; the monitoring points and segment connections of highway sections are analyzed and abstracted to construct a segment abstract graph G = (V, E); nodes V and edges E in the segment abstract graph are used as input, according to... Graph convolution is performed to obtain spatial feature information; where Q(t) is the preprocessed target traffic flow feature data, and α... p Autoregressive coefficient, β p Here are the moving average coefficients, and ∈(t) represents white noise. It is an adjacency matrix with self-loops. W is the degree matrix. (l) Let X be the weight of the l-th layer. (l) These are node features.

[0108] A detailed description of the modeling and extraction of temporal features is provided:

[0109] In this embodiment, firstly, in the difference processing formula: d is the difference order, Δ d Q(t) represents the stationary sequence after differentiating the original sequence Q(t). For example, the first difference d = 1 is Q(t) - Q(td), which is used to eliminate linear trends. By differentiating, the non-stationary traffic flow sequence is transformed into a stationary sequence, making it easier to capture time correlations.

[0110] Then, time features are extracted from the input model. The model formula consists of: an autoregressive term (AR part): The main method is to predict the current value using the difference sequence values; the moving average term (MA part) is also used. Error correlation is modeled using past white noise (prediction error) sequences. By linearly combining AR and MA terms, the time dependence in stationary sequences is captured, enabling the modeling of traffic flow temporal characteristics.

[0111] Among them, parameter order determination is as follows: the moving average term Q is determined by observing the autocorrelation function (ACF), the order P of the autoregressive term is determined by observing the partial autocorrelation function (PACF) plot, and the difference order d is confirmed by the ADF test.

[0112] A detailed description of the modeling and extraction of spatial features is provided:

[0113] In the abstract diagram of a highway, node V represents each monitoring point (such as sensor nodes set up every kilometer), and each node contains feature data. Edge E represents the road segment connection relationship, that is, the physical connectivity between adjacent monitoring points (such as upstream and downstream lanes). In the self-loop adjacency matrix, A is the original adjacency matrix (node ​​connectivity matrix), and I is the identity matrix (allowing nodes to retain their own features). The node feature matrix of layer l contains multi-dimensional data such as historical traffic flow, time, and weather; the weight matrix of layer l is mainly determined by training and learning the transformation method of node features; σ is the activation function. Through the standardization operation of the adjacency matrix and degree matrix, the weighted aggregation of adjacent node features is realized, thereby capturing the spatial dependencies between road segments.

[0114] Step 230: Based on the time feature information and the spatial feature information, the traffic flow prediction data for the specified time period is obtained by splicing and fusing them.

[0115] In this embodiment, temporal and spatial features are concatenated dimensionally to form a joint feature vector containing spatiotemporal information. A fully connected neural network is then used to perform a nonlinear transformation on the joint features to learn the interaction relationships between the spatiotemporal features, ultimately outputting traffic flow prediction data for the next h time period.

[0116] Therefore, this embodiment utilizes spatiotemporal joint modeling to simultaneously capture the temporal patterns of traffic flow (such as morning and evening rush hours) and spatial propagation characteristics (such as congestion diffusion), overcoming the limitations of traditional models that only consider a single dimension of time or space. By fusing multi-dimensional information, the bias of a single model is reduced, prediction accuracy is improved, and it is suitable for complex traffic scenarios.

[0117] In this embodiment, when splicing and fusing two types of features to output predicted data, considering that traffic flow prediction is a continuous numerical prediction (regression task), the MSE loss function is used. The loss function measures the average squared deviation between the predicted value and the true value; the smaller the value, the higher the prediction accuracy. The squaring operation amplifies the weight of larger errors, forcing the model to prioritize reducing severely deviated predictions.

[0118] For example, the loss function could be

[0119] Step 240: Based on the traffic flow prediction data, combined with real-time traffic flow data, traffic flow mutation detection is performed to obtain mutation analysis results.

[0120] In this embodiment, traffic flow abrupt change detection and visual feature analysis are used to determine whether a traffic accident has occurred. Traffic flow abrupt change detection uses traffic flow data as a benchmark and analyzes three key parameters in real time to capture abnormal dynamics within the lanes. Key parameters include: average lane speed (a sudden drop in the average speed of all vehicles in a lane at the same time is a direct signal of traffic congestion or an accident), lane density (the number of vehicles within a certain length of a lane at the same time; a surge in density reflects traffic convergence or stagnation), and acceleration (calculated by speed; a sudden increase in deceleration may indicate emergency braking by vehicles, a precursor to an accident). This embodiment identifies the form of vehicles within the lanes by real-time monitoring and analysis of key parameters, capturing abnormal dynamics to analyze abrupt changes. When no abrupt change is determined, the non-abrupt result is used as the abrupt change analysis result; when an abrupt change is determined, the determined abrupt change result is used as the abrupt change analysis result.

[0121] Optionally, based on the traffic flow prediction data, combined with real-time traffic flow data, traffic flow abrupt change detection is performed to obtain abrupt change analysis results, including: based on the traffic flow prediction data and the real-time traffic flow data, analyzing the average speed v(t) of each vehicle in the lane and the vehicle density ρ(t) in the lane at time t, and applying the abrupt change determination formula. Calculate the mutation analysis results; where μ is the historical mean, σ is the standard deviation, and w1, w2, and w3 are all weighting coefficients. When the score is greater than the preset threshold, it is determined that an accident may have occurred, and a definite mutation result is obtained.

[0122] In this embodiment, μ and σ are updated in real time based on historical data from a sliding time window (e.g., the past 30 minutes) to adapt to the periodic changes in traffic flow.

[0123] The weighting coefficients satisfy w1+w2+w3=1, and the weights can be adjusted according to different scenarios such as urban highways or mountain highways.

[0124] When the score exceeds a preset threshold (determined through empirical analysis), the system determines that a sudden change in traffic flow has occurred, triggering the subsequent accident identification process, namely the visual feature analysis process. This embodiment calculates the score in real time based on edge computing, enabling anomaly identification within seconds, thus buying valuable time for emergency response.

[0125] Therefore, this embodiment combines data from different dimensions to fuse parameters and achieve traffic flow mutation analysis. This embodiment's traffic flow mutation analysis can enable accident precursor detection and congestion early warning.

[0126] Step 250: When the mutation analysis result is a confirmed mutation result, video data of each lane of the highway is acquired for visual feature analysis to determine the accident analysis result.

[0127] In this embodiment, when a sudden change in traffic flow is detected, a visual feature analysis process is performed, such as calling the acquired video data to analyze whether there are vehicle collision debris, abnormal vehicle postures, etc. Specifically, video streams are acquired by cameras deployed on the robot, and object detection algorithms (such as YOLO and Faster R-CNN) are used to analyze the images in real time, outputting two types of key information: vehicle location and vehicle posture. Based on this, basic parameters such as vehicle density and spacing are analyzed and calculated, as well as whether the vehicle has experienced abnormal deviations (such as yaw or rollover after a collision), thereby determining whether an accident has occurred. When an accident is determined to have occurred, the result of determining that an accident has occurred is used as the accident analysis result; when an accident is determined not to have occurred, the result of determining that no accident has occurred is used as the accident analysis result.

[0128] In one optional embodiment, this application embodiment acquires video data of each lane of a highway for visual feature analysis to determine the accident analysis result. Specifically, this may include: detecting the acquired video data using a target detection algorithm to obtain a vehicle bounding box set B = {b1,…,b...} n}; Based on the vehicle bounding box set B = {b1,…,b n Trajectory analysis is performed to obtain the vehicle attitude angle set S = {s1, ..., s2}. n}; The set of vehicle attitude angles S = {s1, ..., s n} is the input, according to Analyze key anomaly features to obtain the variance of collision angles. Based on the vehicle bounding box set B = {b1, ..., b} n Analyze the pixels of the vehicle bounding box to obtain the total number of pixels b. Mark the pixels covered by the vehicle bounding box as vehicle regions and the pixels not covered as non-vehicle regions to obtain the number of pixels a in the non-vehicle regions. Using the total number of pixels b and the number of pixels a as input, according to R... debris =a / b analysis of key anomaly features to obtain the proportion R of fragmented regions debris Based on collision angle variance and fragmented area ratio R debris According to the decision function Accident determination is carried out to obtain accident information; among which, P accident When the value is ≥0.8, an accident is determined to have occurred, and the accident occurrence result is obtained. k1 and k2 are both weighting coefficients, b i Let s represent the coordinates of the rectangular bounding box of the vehicle in the i-th frame of the image. iThis represents the driving direction or body posture angle of the i-th vehicle in the image.

[0129] In this embodiment, by analyzing video data, two key pieces of information can be obtained: a set of vehicle bounding boxes and a set of vehicle attitude angles. Each bounding box in the vehicle bounding box set can represent the position and size of the corresponding vehicle in the image, used to locate the vehicle and calculate its spatial distribution; the attitude angle represents the angle between the vehicle's driving direction and the lane centerline, estimated by the bounding box tilt angle or the vehicle's tire direction.

[0130] This embodiment primarily analyzes the collision angle variance when performing visual analysis of key anomaly features. and fragmented area ratio R debris Two types of abnormal features.

[0131] The variance of the collision angle is mainly based on Analyze the collision angle. This represents the mean of the attitude angles of all vehicles in the current lane, while its variance reflects the dispersion of these attitude angles. When vehicles are driving normally, their attitude angles are concentrated around 0°. Approaching 0; when an accident occurs, the vehicle may deviate at multiple angles due to the collision. Significantly increased.

[0132] Fragmentation area percentage R debris The determination is primarily based on pixel analysis, including analyzing the number of non-vehicle pixels (a) and the total number of pixels (b) in the image. The number of non-vehicle pixels can be determined by identifying pixels other than tires and vehicle bodies (such as scattered debris, liquids, and mud) using image segmentation algorithms (such as Mask R-CNN). Among these, the normal road surface (R) debris Approximately 0; after an accident, debris, oil stains, etc., will appear on the road surface. debris Significantly increased.

[0133] This embodiment will R debris As input, calculate the probability P of the accident occurring. accident First, R debris Transform into a linear combination The weighting coefficients k1 and k2 are primarily determined through training on historical accident data, limiting their use in adjusting the importance of the two types of features. Then, a sigmoid function is used to map the linear combination result to the [0,1] interval, transforming it into the accident probability P. accident P accident The closer it is to 1, the higher the probability of an accident.

[0134] Therefore, this embodiment achieves multi-dimensional data fusion, combining vehicle posture anomalies (collision angle variance) and environmental anomalies (fragmentation), avoiding misjudgment based on a single feature (such as temporary obstacles in the construction area will not trigger both types of features at the same time).

[0135] It should be noted that in this embodiment, localized edge computing nodes are used for the entire process, from traffic flow prediction and accident identification to accident handling. It integrates a real-time processing and decision-making framework for multimodal data, and through a lightweight convolutional neural network and spatiotemporal sequence analysis model, it reduces the accident detection response time to the second level, forming a low-latency closed loop of "perception-decision-handling".

[0136] Step 260: When the accident analysis result indicates that an accident has occurred, an impact range analysis and accident level assessment are performed based on the traffic flow prediction data and the real-time traffic flow data to obtain the accident impact range and accident level.

[0137] In practice, once an accident occurs, the scope of its impact can be further analyzed, and the accident level can be assessed, serving as an important basis for subsequent path planning, resource scheduling / decision-making.

[0138] Specifically, the impact range analysis mainly includes analyzing the congestion situation upstream and downstream, as well as the impact of the accident on rescue efforts and the spread of congestion.

[0139] Accident severity assessment primarily aims to determine the severity of an accident, which can be based on three core indicators: the injury rate (i.e., estimating the number of injuries), the number of blocked lanes (statistically counting the number of lanes that are impassable due to the accident), and the total delay kilometers (i.e., the cumulative delay time of affected vehicles).

[0140] In an optional embodiment, the above-mentioned analysis of the impact range and assessment of the accident level based on the traffic flow prediction data and the real-time traffic flow data to obtain the accident impact range and accident level may include: analyzing the upstream traffic flow q1 and downstream traffic flow q2 of the highway based on the traffic flow prediction data and the real-time traffic flow data; and based on the upstream traffic flow q1 and downstream traffic flow q2, determining the impact range and accident level according to v w =q2-q1 / ρ2-ρ1, calculate the congestion propagation speed v w Based on the congestion propagation speed v w According to L max =v w ·T response Calculate the maximum impact distance as the accident impact range; analyze the covered lane area based on the vehicle bounding box set, determine the number of blocked lanes, and obtain the probability of personal injury; based on the traffic flow prediction data and the real-time traffic flow data, analyze the driving speed v of each vehicle in the congested section of the highway. iDriving distance l i and the free-flow velocity v free Based on driving speed v i Driving distance l i and the free-flow velocity v free According to V delay =∑ i (v free -v i )·l i Calculate the total delay kilometers; based on the total delay kilometers, determine the accident level using the fuzzy comprehensive evaluation method.

[0141] In analyzing the impact range of an accident, this embodiment uses the congestion propagation speed to reflect the rate of spread of traffic anomalies between upstream and downstream road segments. When an accident occurs, downstream vehicles decelerate, leading to increased density and decreased flow. The ratio of the flow difference to the density difference between upstream and downstream segments can quantify the speed at which congestion spreads upstream. This embodiment combines traffic flow data to calculate the flow rates of upstream and downstream road segments using the average lane speed v and the upstream lane densities ρ1 and ρ2, with the formula q = v·ρ.

[0142] Once the congestion propagation speed is determined, it can be used as one of the parameters, combined with the response time T. response (Refers to the time from the occurrence of the accident to the arrival of the robot or rescue force at the scene and the start of traffic control), calculate the maximum upstream distance that the accident may affect, and thus obtain the scope of the accident's impact.

[0143] For accident severity assessment, three core indicators are mainly defined to evaluate the severity of the accident: probability of injury P. inj Number of blocked lanes L block Total Delay Mileage V delay The core indicators are constructed into a three-dimensional indicator space (P). inj ,L block V delay The accident level (Level I-IV) is determined by fuzzy comprehensive evaluation method.

[0144] Among them, the probability of personnel injury P inj By analyzing camera images, estimating sensor data, or using passive detection driven by V2X data, and by receiving data actively sent by the vehicle from the OBU (On-Board Unit) (such as speed, braking signals, and steering angle), combined with the communication coverage of the roadside RSU (Roadside Unit), a vehicle driving status network is constructed to determine the situation of vehicles and personnel at the time of an accident.

[0145] Number of blocked lanes L block This can be determined by directly counting the number of lanes that are impassable due to accidents.

[0146] Total Delay Mileage Vdelay This reflects the cumulative delay time of all affected vehicles; the larger the value, the more severe the impact of traffic disruption on traffic efficiency.

[0147] Because the three indicators have different dimensions (probability, number of lanes, and vehicle kilometers), fuzzy mathematics is used to transform them into a unified membership function. By calculating a weighted comprehensive score, the accident levels are finally classified into I-IV levels.

[0148] Therefore, this embodiment standardizes the accident classification process, providing a quantitative basis for resource scheduling and traffic management strategies. Furthermore, LoRa low-power wide-area network nodes can be deployed along both sides of the highway, allowing robots to transmit monitoring data in real time via LoRa communication modules and receive instructions from the cloud. The fuzzy comprehensive evaluation results can be automatically synchronized to the cloud management platform, triggering emergency plans corresponding to the accident level.

[0149] Step 270: Dynamic path planning is performed based on the traffic flow prediction data, the real-time traffic flow data, and accident information using a dynamic path planning algorithm.

[0150] For a description of step 270, please refer to step 150, which will not be described in detail in this embodiment.

[0151] Optionally, the above-mentioned dynamic path planning algorithm, based on the traffic flow prediction data, the real-time traffic flow data, and accident information, may include the following sub-steps:

[0152] Sub-step 2701 involves performing road network modeling based on the traffic flow prediction data, the real-time traffic flow data, the highway road data, and the accident information, and then using a cost function to optimize the path to obtain rescue path optimization information.

[0153] The rescue path optimization information is used to plan the optimal path for the robot from the starting point to the accident site.

[0154] In this implementation, the dynamic path planning algorithm is improved. Using the aforementioned acquired data as input, the algorithm plans the rescue path. Specifically, when planning the rescue path, firstly, a corresponding road network model is constructed based on highways and real-time data using a road network modeling approach. Then, a cost function is introduced into the road network model. This cost function is implemented based on the improved A* algorithm, and an optimized rescue path is generated through the cost function.

[0155] In an optional embodiment, road network modeling is performed based on the traffic flow prediction data, the real-time traffic flow data, highway road data, and the accident information, and path optimization is performed using a cost function to obtain rescue path optimization information. Specifically, this may include: determining the real-time traffic flow q based on the traffic flow prediction data and the real-time traffic flow data.ij And, determining road capacity C based on highway road data. ij ; with real-time traffic q ij and road capacity C ij For input, according to Road network modeling is performed, and the cost function f(n) = g(n) + h(n) + γVar(c) is used in the improved algorithm. ij ) Optimize the route to obtain rescue route optimization information.

[0156] Among them, the starting point i and ending point j and d of the rescue route are identified through accident information. ij v represents the distance of the road segment. ij (t) represents the real-time average speed of the road segment at time t, λ is the weighting coefficient, g(n) represents the actual cost from the starting point to the current node, h(n) represents the heuristically estimated cost from the current node to the destination, γ represents the path stability factor, and Var(c ij The ) represents the variance of the cost of each segment in the path. Let Q be the gradient of the potential field. i The repulsive force intensity at the accident point is represented by r, and the current vehicle position coordinates are also represented by r. i Indicates the coordinates of the accident point. This represents a unit vector pointing from the point of the accident to the vehicle.

[0157] In optimizing rescue routes: First, road network modeling is performed. The starting point i and ending point j of the rescue are identified using accident information. Road network modeling is then based on a time-dependent cost function. The input parameters for road network modeling include: real-time traffic flow q. ij and road capacity C ij Distance d of road segment ij (representing the physical distance from road segment i to j), the real-time average speed v of road segment at time t. ij (t) and the weighting coefficient λ (used to balance the effects of velocity and flow rate).

[0158] Furthermore, in road network modeling, the time cost term d is constructed. ij / v ij (t) and congestion cost q ij (t) / C ij Among these factors, the time cost represents the travel time for a road segment; the lower the speed, the higher the time cost. The congestion cost, specifically the ratio of flow rate to capacity, reflects the degree of congestion; a higher ratio (closer to 1) indicates more congested road segments and a higher cost. Therefore, by dynamically updating road segment costs using real-time data, the system avoids selecting severely congested or slow-moving routes.

[0159] Then, based on the improved A* algorithm, the actual cost (cumulative segment cost) and the estimated cost (such as Euclidean distance) are analyzed, and the path stability factor γVar(c) is introduced.ij ), where Var(c ij ) represents the variance of the cost of each segment in the path, which measures the degree of fluctuation in the path cost; γ represents the adjustment coefficient, which is used to control the preference for stability.

[0160] Therefore, by improving the A* algorithm, it tends to select paths with low total cost and small cost fluctuations across road segments, reducing the risk of path failure due to real-time traffic changes (such as temporary accidents) and addressing the problem of the traditional A* algorithm ignoring the volatility of path costs (e.g., a path containing segments with a high risk of sudden congestion). This embodiment is based on a hybrid path planning method combining the improved A* algorithm and potential field theory, combined with real-time traffic flow prediction data, to dynamically generate optimal traffic diversion paths. Through a distributed cooperative scheduling algorithm, multi-robot task allocation and path collision avoidance are achieved, ensuring rapid arrival at the accident point and minimizing interference with normal traffic flow.

[0161] Sub-step 2702: Based on the accident point corresponding to the accident information, the diversion path is planned using the gradient descent method of potential field theory to generate the target diversion path.

[0162] In this specific implementation, the potential field gradient descent method is introduced to analyze the accident point, and then the diversion path is planned to generate the target diversion path.

[0163] In an optional embodiment, the above-mentioned process of planning diversion paths and generating target diversion paths based on the accident points corresponding to the accident information and using the gradient descent method of potential field theory may include: analyzing the accident points using the gradient descent method of potential field theory, and based on... Perform traffic diversion path planning and generate target traffic diversion paths.

[0164] in, Let Q be the gradient of the potential field. i The repulsive force intensity at the accident point is represented by r, and the current vehicle position coordinates are also represented by r. i Indicates the coordinates of the accident point. This represents a unit vector pointing from the point of the accident to the vehicle.

[0165] In optimizing the diversion path, the accident point is first treated as a source point with "repulsive force" based on the gradient descent method of potential field theory, and the normally traveling vehicles are treated as particles affected by the repulsive force. A detour path is generated using the gradient descent method, and the potential field gradient is determined by analyzing the direction pointing towards the decrease of repulsive force (i.e., guiding vehicles away from the accident point). To avoid entering the accident-affected area, the robot dynamically adjusts its driving path by iteratively calculating the gradient direction, thus forming a detour trajectory. This enables real-time traffic diversion guidance; when the robot detects an accident, it sends the detour path calculated by the potential field model to vehicles via roadside variable message signs or V2X technology, preventing traffic from converging into the accident area.

[0166] Sub-step 2703: Based on the optimal path and the target diversion path, generate target path planning information.

[0167] In this embodiment, the mobile robot generates a target planning path based on the generated rescue path and diversion path. When an accident occurs, on the one hand, the generated optimized rescue path ensures the robot's rapid response to the accident; on the other hand, the generated diversion path simultaneously guides the surrounding traffic flow, reducing the risk of secondary accidents and forming a closed loop of "perception-response-diversion". This ensures that rescue vehicles arrive at the scene in the shortest possible time and avoids delays in response due to sudden congestion during the rescue process.

[0168] Step 280: Control the mobile robot to perform traffic management tasks on the lanes of the highway based on the target path planning information.

[0169] Addressing the issue of time delays in the current technology's process from accident discovery to information transmission and response, hindering timely traffic management, this embodiment utilizes edge computing and real-time data processing technologies to shorten traffic flow prediction cycles and accident identification response times. Furthermore, through localized decision-making and coordinated scheduling using mobile robots, a low-latency closed loop of "detection-analysis-response" is achieved, improving highway emergency response efficiency. This embodiment leverages mobile robots for unmanned intelligent operation, reducing reliance on human labor and safety risks: through automated robot detection and autonomous algorithmic decision-making, the entire process from traffic flow prediction and accident identification to traffic management is automated, replacing manual patrols and on-site intervention. This avoids human intervention in high-risk scenarios such as severe weather and high traffic volumes, significantly reducing manpower input; simultaneously, it shortens the time cycle from accident discovery to response, improving monitoring efficiency and emergency response speed, reducing the safety hazards of manual inspections from the source.

[0170] In an optional embodiment, after obtaining the target path planning information, the method may further include: when scheduling the mobile robot to the accident point corresponding to the accident information, obtaining scheduling information x. ij and scheduling time t ij Define decision variables Construct the objective function based on the decision variables. And construct constraints based on the decision variables. Resource scheduling optimization for mobile robots is performed based on an objective function and constraints. After scheduling the mobile robots to perform traffic control tasks, handling information is obtained, including handling time and the scope of accident impact. Based on the handling information, according to... Calculate the recovery efficiency η; determine the reward value r based on the recovery efficiency η, according to... A reward function is constructed, and the learning mechanism is continuously iterated and updated to update the mobile robot's execution strategy when an accident occurs; where T actualT represents the actual processing time. predict To estimate the processing time, L actual L represents the actual scope of the accident's impact. actual Let α be the expected impact range of the accident, γ be the learning rate, γ be the discount factor, (s,a) represent the current state and action, and s' represent the new state after the action is performed.

[0171] For resource scheduling optimization, this embodiment can select the optimal combination from multiple available resources when an accident is detected. For example, a multi-objective resource scheduling model can be used to schedule a designated mobile robot to handle the accident. This model is used for resource scheduling optimization of a highway intelligent monitoring and traffic control robot system, aiming to solve the problem of how to rationally allocate resources such as robots (e.g., patrol vehicles, rescue equipment) when an accident occurs, in order to achieve the dual objectives of fastest response speed and lowest scheduling cost. Wherein, scheduling information x ij This indicates whether resource i is dispatched to incident j. The parameter x... ij Binary variables can simplify the "choice-not-choose" logic of resource allocation, facilitating mathematical modeling. For example, x ij =1 Resource i is scheduled to incident j; x ij =0 indicates that resource i is not scheduled to incident j; scheduling time t ij The time it takes for resource i to reach incident j can be calculated during dynamic path planning by taking the shortest time from the current location of the resource to the incident point.

[0172] An objective function is constructed using decision variables, resulting in an objective function model. This model, by quantifying decision variables and constraints, achieves scientific and intelligent emergency resource scheduling on highways and is a core component of the system's "perception-decision-execution" closed loop. The model includes two minimization optimization objectives: the weighted response time min∑w. j t ij x ij The total scheduling cost min∑c ij x ij Weighted response time represents the time it takes for all scheduled resources to arrive at each incident point, with weights w determined according to incident priority. j The total scheduling cost is obtained by weighted summation; the total scheduling cost is the sum of the total costs of scheduling all resources, c. ij This represents the operating cost of resource i. An optimal scheduling scheme is generated by balancing response time and scheduling cost.

[0173] Then, constraints for resource scheduling are constructed that satisfy the minimum resource requirement ∑x. ij ≥N j and reaching the time limit t ij ≤T maxThis ensures that resources can be used to handle incidents while also ensuring timely response to emergencies, thus preventing secondary incidents or increased congestion caused by slow dispatch.

[0174] In this embodiment, after each accident identification and detection indicates the completion of the evacuation task, the accident handling efficiency can be analyzed. The effectiveness of the scheduling scheme / strategy can be evaluated through the recovery efficiency η, so as to update the parameters of the resource scheduling model. Specifically, in the formula for calculating the recovery efficiency η, T... actual / T predict Used to measure the deviation between the actual processing time and the expected time, reflecting whether the system response efficiency meets the standard; L actual / L predict Measuring the deviation between the actual and projected impact range reflects the system's ability to control the spread of the incident. Recovery efficiency is determined by comprehensively evaluating the system's performance in terms of both time efficiency and impact control. Based on recovery efficiency, the update path algorithm, the incident severity assessment algorithm, and the resource scheduling model can be dynamically optimized.

[0175] Furthermore, this embodiment can employ a learning mechanism (such as the Q-learning algorithm) to dynamically optimize the handling strategy of the highway intelligent monitoring and traffic management robot system. In the reward function, the Q-value (Q(s,a)) obtained from the previous scheduling is combined with the immediate reward r and the discounted future reward as inputs to update the Q-value. Here, Q(s,a) represents the long-term expected value of performing action a in state s; a higher Q-value indicates a better action. The reward r can be calculated based on the recovery efficiency index η, serving as direct feedback after action execution to guide the strategy towards short-term optimization. The discounted future reward... This represents the maximum expected value of considering all possible actions under the new state s', multiplied by a discount factor, as the long-term reward, guiding the strategy toward long-term optimality.

[0176] Therefore, this embodiment constructs a self-learning framework for intelligent traffic management strategies to adapt to the characteristics of highway scenarios in complex environments, such as high vehicle speeds, variable weather conditions (e.g., heavy rain, fog, snow), and complex road conditions. This addresses the difficulty of traditional fixed signal control or manual intervention modes in adapting to real-time changes in traffic demands.

[0177] This embodiment utilizes an intelligent decision-making system based on deep learning and reinforcement learning algorithms. It can process massive amounts of perceived data in real time, predict congestion points and trends using traffic flow prediction models, and generate personalized traffic management plans by combining strategies such as shortest path planning, dynamic lane control, and emergency lane opening. Simultaneously, the robot supports real-time communication with the cloud-based traffic management platform, surrounding intelligent devices (such as variable message signs and traffic lights), and passing vehicles (via V2X technology), achieving vehicle-road-cloud-human collaborative control and forming a fully interconnected intelligent traffic management network.

[0178] Therefore, this embodiment realizes a closed-loop system for evaluating treatment effectiveness and optimizing strategies based on reinforcement learning. It constructs a reward function using historical treatment data and real-time feedback, continuously iteratively updating the decision matrix to achieve autonomous evolution and scene adaptation capabilities of the evacuation strategy. Specifically, the reward value r is determined using the recovery efficiency η, forming an "evaluation-learning-optimization" closed loop to improve system efficiency. In complex environments, the learning mechanism can improve data reliability by adjusting sensor fusion weights (such as the priority of rain and fog-resistant cameras).

[0179] Its advantages include: A) Autonomous strategy evolution without human intervention: the system automatically optimizes its response strategies through continuous interaction with the environment, adapting to dynamic changes in highway scenarios (such as traffic flow fluctuations and sudden weather changes); B) Enhanced generalization ability: by accumulating experience in different scenarios, it can generate more universal traffic management solutions, reducing dependence on specific conditions; C) Enhanced robustness: when encountering new types of accidents or sudden disturbances, the learning mechanism can gradually find reliable solutions by exploring new actions (such as trying to combine different traffic management strategies).

[0180] Furthermore, building upon mobile robots, a drone-ground sensor collaborative solution can be constructed: low-cost sensor nodes (geomagnetic coils, RFID tags) are deployed on the ground to monitor basic traffic flow data, while drones periodically take off to conduct high-altitude video inspections. LoRa wireless sensor networks: LoRa low-power wide-area network nodes are deployed along both sides of the highway, allowing robots to transmit monitoring data in real time via LoRa communication modules and receive instructions from the cloud.

[0181] In summary, this application first utilizes a mobile robot to monitor highways in real time, acquiring traffic flow characteristic data. Through feature extraction, analysis, and fusion, traffic flow prediction data for a specified time period is obtained. Then, based on the traffic flow prediction data and the real-time acquired traffic flow data, accident identification and classification are performed. When an accident is identified on the highway, an algorithm is used, combined with the real-time identified data, to perform dynamic path planning. Based on the planned path, the mobile robot is controlled to perform traffic management tasks on the highway lanes. On the one hand, this application, through a mobile monitoring and traffic management robot, combined with multi-source sensors (such as radar and video) and a dynamic deployment strategy, achieves dynamic coverage and real-time monitoring of the entire highway section, eliminating blind spots of fixed equipment and improving the comprehensiveness of road condition information collection. On the other hand, through automated monitoring based on the collected video data by the robot, combined with intelligent algorithms, an accurate traffic flow prediction, automated accident identification, and handling process is achieved, reducing manual intervention, improving monitoring efficiency and staff safety, and reducing operating costs. Existing technologies have problems with low efficiency in automated traffic flow monitoring on highways and the inability to automatically identify / handle accidents.

[0182] It should be noted that, for the sake of simplicity, the method embodiments are described as a series of actions. However, those skilled in the art should know that the embodiments of this application are not limited to the described order of actions, because according to the embodiments of this application, some steps may be performed in other orders or simultaneously.

[0183] like Figure 3 As shown in the figure, this application embodiment also provides a highway traffic flow prediction and accident recognition system 300, including:

[0184] The traffic flow feature data processing module 310 is used to perform normalization processing on the acquired traffic flow feature data of the highway to obtain preprocessed target traffic flow feature data. The traffic flow feature data is the data obtained by the mobile robot in real time monitoring the traffic flow of the highway during dynamic patrol.

[0185] The feature convolution extraction module 320 is used to perform temporal feature convolution processing on the target traffic flow feature data to obtain temporal feature information, and to perform spatial feature convolution processing on the acquired highway segment data to obtain spatial feature information.

[0186] The feature splicing and fusion module 330 is used to splice and fuse the time feature information and the spatial feature information to obtain traffic flow prediction data for a specified time period.

[0187] The accident identification and classification module 340 is used to identify and classify accidents based on the traffic flow prediction data and the real-time traffic flow data acquired in real time, and to determine accident information, which includes accident analysis results, accident impact range and accident level.

[0188] The path planning module 350 is used to perform dynamic path planning based on the traffic flow prediction data, the real-time traffic flow data and accident information through a dynamic path planning algorithm to obtain target path planning information, and to control the mobile robot to perform traffic management tasks on the lanes of the highway based on the target path planning information.

[0189] Optionally, the accident identification and classification module is specifically used to realize accident identification and classification. During the accident identification process, V2X data (such as braking signals and steering angles actively sent by the vehicle) can be integrated, and data from the on-board unit can be received through the roadside RSU to construct a vehicle driving status network, assist in judging abnormal driving behavior, and improve the accuracy of accident detection. The accident identification and classification module specifically includes:

[0190] The mutation detection submodule is used to detect traffic flow mutations based on the traffic flow prediction data and real-time traffic flow data, and to obtain mutation analysis results.

[0191] The visual analysis submodule is used to acquire video data of each lane of the highway for visual feature analysis and determine the accident analysis result when the mutation analysis result is a confirmed mutation result.

[0192] The accident determination submodule is used to perform impact range analysis and accident level assessment based on the traffic flow prediction data and the real-time traffic flow data when the accident analysis result indicates that an accident has occurred, so as to obtain the accident impact range and accident level.

[0193] Optional, the route planning module includes:

[0194] The first path planning submodule is used to perform road network modeling based on the traffic flow prediction data, the real-time traffic flow data, the highway road data and the accident information, and to perform path optimization using a cost function to obtain rescue path optimization information. The rescue path optimization information is used to plan the optimal path from the starting point to the accident point for the robot.

[0195] The second path planning submodule is used to plan the diversion path based on the accident point corresponding to the accident information and the gradient descent method of potential field theory to generate the target diversion path.

[0196] The third path planning submodule is used to generate target path planning information based on the optimal path and the target diversion path.

[0197] It should be noted that the highway traffic flow prediction and accident identification system provided in this application embodiment can execute the highway traffic flow prediction and accident identification method provided in any embodiment of this application, and has the corresponding functions and beneficial effects of the execution method.

[0198] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0199] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A method for predicting traffic flow and identifying accidents on highways, characterized in that, include: The acquired highway traffic flow characteristic data is normalized to obtain preprocessed target traffic flow characteristic data. The traffic flow characteristic data is obtained by the mobile robot monitoring the highway traffic flow in real time during dynamic patrol. Temporal feature convolution processing is performed on the target traffic flow feature data to obtain temporal feature information, and spatial feature convolution processing is performed on the acquired highway segment data to obtain spatial feature information. Based on the temporal and spatial feature information, traffic flow prediction data for a specified time period is obtained by splicing and fusing them. Accident identification and accident classification assessment are performed based on the traffic flow prediction data and the real-time traffic flow data acquired in real time to determine accident information, which includes accident analysis results, accident impact range, and accident level. The mobile robot is controlled to perform traffic management tasks on the highway lanes based on the traffic flow prediction data, the real-time traffic flow data, and accident information through a dynamic path planning algorithm. The process of identifying and classifying accidents based on traffic flow prediction data and real-time traffic flow data, and determining accident information, includes: detecting sudden changes in traffic flow based on the traffic flow prediction data and real-time traffic flow data to obtain a sudden change analysis result; when the sudden change analysis result indicates a confirmed sudden change, acquiring video data of each lane of the highway for visual feature analysis to determine the accident analysis result; and when the accident analysis result indicates an confirmed accident, performing an impact range analysis and accident level assessment based on the traffic flow prediction data and real-time traffic flow data to obtain the accident impact range and accident level. Based on the traffic flow prediction data, combined with real-time traffic flow data, traffic flow abrupt change detection is performed to obtain abrupt change analysis results, including: based on the traffic flow prediction data and the real-time traffic flow data, analyzing the average speed v(t) of each vehicle in the lane and the vehicle density ρ(t) in the lane at time t, and applying the abrupt change judgment formula. Calculate the mutation analysis results; where μ is the historical mean, σ is the standard deviation, and w1, w2, and w3 are all weighting coefficients. When the score is greater than the preset threshold, it is determined that an accident may have occurred, and a definite mutation result is obtained.

2. The method according to claim 1, characterized in that, Temporal feature convolution processing is performed on the target traffic flow feature data to obtain temporal feature information, and spatial feature convolution processing is performed on the acquired highway segment data to obtain spatial feature information, including: Using the target traffic flow characteristic data as input, according to Δ d The expression Q(t) = Q(t) - Q(td) is differenced to eliminate non-stationarity, and then input into a preset autoregressive model. Perform convolution processing to extract temporal feature information; Analyze the monitoring points and connection relationships of highway segments, and perform abstract processing to construct an abstract segment graph G=(V,E); Using nodes V and edges E in the road segment abstract graph as input, according to Perform graph convolution operations to obtain spatial feature information; Among them, Q(t) is the preprocessed target traffic flow characteristic data, α p Autoregressive coefficient, β q Here are the moving average coefficients, and ∈(t) represents white noise. It is an adjacency matrix with self-loops. W is the degree matrix. (l) Let X be the weight of the l-th layer. (l) These are node features.

3. The method according to claim 1, characterized in that, Video data from each lane of the highway is acquired and visual feature analysis is performed to determine the accident analysis results, including: The acquired video data is detected using an object detection algorithm, resulting in a vehicle bounding box set B = {b1,…,b...}. n }; Based on the vehicle bounding box set B = {b1, ..., b} n Trajectory analysis is performed to obtain the vehicle attitude angle set S = {s1, ..., s2}. n }; The vehicle attitude angle set S = {s1, ..., s2} n } is the input, according to Analyze key anomaly features to obtain the variance of collision angles. Based on the vehicle bounding box set B = {b1, ..., b} n } Analyze the pixels of the vehicle bounding box to obtain the total number of pixels b, and mark the pixels covered by the vehicle bounding box as the vehicle region and the pixels not covered as the non-vehicle region to obtain the number of pixels a in the non-vehicle region; Using the total number of pixels b and the number of pixels a as input, according to R debris =a / b analysis of key anomaly features to obtain the proportion R of fragmented regions debris ; Based on collision angle variance and fragmented area ratio R debris According to the decision function Conduct accident assessment and obtain accident information; Among them, P accident When the value is ≥0.8, an accident is determined to have occurred, and the accident occurrence result is obtained. k1 and k2 are both weighting coefficients, b i Let s represent the coordinates of the rectangular bounding box of the vehicle in the i-th frame of the image. i This represents the driving direction or body posture angle of the i-th vehicle in the image.

4. The method according to claim 1, characterized in that, Dynamic route planning is performed using a dynamic route planning algorithm based on the traffic flow prediction data, the real-time traffic flow data, and accident information, including: Based on the traffic flow prediction data, the real-time traffic flow data, the highway road data, and the accident information, a road network model is constructed, and a cost function is used for path optimization to obtain rescue path optimization information. The rescue path optimization information is used to plan the optimal path from the starting point to the accident point for the robot. Based on the accident points corresponding to the accident information, the gradient descent method of potential field theory is used to plan the diversion path and generate the target diversion path. Based on the optimal path and the target diversion path, target path planning information is generated.

5. A highway traffic flow prediction and accident identification system, characterized in that, include: The traffic flow feature data processing module is used to normalize the acquired traffic flow feature data of the highway to obtain preprocessed target traffic flow feature data. The traffic flow feature data is obtained by the mobile robot in real time monitoring the traffic flow of the highway during dynamic patrol. The feature convolution extraction module is used to perform temporal feature convolution processing on the target traffic flow feature data to obtain temporal feature information, and to perform spatial feature convolution processing on the acquired highway segment data to obtain spatial feature information. The feature splicing and fusion module is used to splice and fuse the time feature information and the spatial feature information to obtain traffic flow prediction data for a specified time period; The accident identification and classification module is used to identify and classify accidents based on the traffic flow prediction data and the real-time traffic flow data acquired in real time, and to determine accident information, which includes accident analysis results, accident impact range and accident level. The path planning module is used to perform dynamic path planning based on the traffic flow prediction data, the real-time traffic flow data and accident information through a dynamic path planning algorithm, obtain target path planning information, and control the mobile robot to perform traffic management tasks on the lanes of the highway based on the target path planning information. The accident identification and classification module specifically includes: a mutation detection submodule, used to detect traffic flow mutations based on the traffic flow prediction data and real-time traffic flow data to obtain mutation analysis results; a visual analysis submodule, used to acquire video data of each lane of the highway for visual feature analysis when the mutation analysis result indicates a confirmed mutation, to determine the accident analysis result; and an accident determination submodule, used to perform impact range analysis and accident level assessment based on the traffic flow prediction data and real-time traffic flow data when the accident analysis result indicates an confirmed accident, to obtain the accident impact range and accident level. Specifically, based on the traffic flow prediction data and combined with real-time traffic flow data, traffic flow abrupt change detection is performed to obtain abrupt change analysis results. This includes: based on the traffic flow prediction data and the real-time traffic flow data, analyzing the average speed v(t) of each vehicle in the lane and the vehicle density ρ(t) in the lane at time t, and applying the abrupt change determination formula. Calculate the mutation analysis results; where μ is the historical mean, σ is the standard deviation, and w1, w2, and w3 are all weighting coefficients. When the score is greater than the preset threshold, it is determined that an accident may have occurred, and a definite mutation result is obtained.

Images