Intelligent management method and device for traffic checkpoint based on vehicle state perception

Abstract

The application provides a traffic kiosk intelligent management method and device based on vehicle state perception, relates to the technical field of traffic kiosk management, and comprises the following steps: collecting vehicle multi-dimensional information through a front-end sensing device to obtain an initial vehicle state information flow; aligning and fusing the initial vehicle state information flow in time and space to construct a standardized vehicle state vector; performing preliminary behavior identification on the vehicle state according to the standardized vehicle state vector to obtain a high-frequency abnormal behavior vehicle; performing time series modeling on the short-time behavior of the vehicle through a deep learning model to identify an intention prediction label of the vehicle; comparing the intention prediction label with a vehicle behavior portrait library to evaluate behavior deviation detection and obtain a behavior deviation degree; and performing risk judgment according to the behavior deviation degree and executing a passing strategy optimization. The application can solve the technical problem that the intelligent response capability of vehicles to multi-kiosk traffic control strategies is poor in the prior art, and achieve the technical effect of improving vehicle management efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of traffic portal management, and particularly relates to a traffic portal intelligent management method and device based on vehicle state perception. BACKGROUND

[0002] In the current urban traffic management system, the traffic portal, as a key node of the road network, undertakes important tasks of traffic monitoring and illegal identification, and is widely used in scenes such as overspeed snapshot, illegal record, portal control, and traffic statistics.

[0003] At present, the existing system generally has problems such as fragmented data sources, incomplete perception information, and weak data fusion capability. Especially in multi-portal and multi-scene joint analysis, it is often impossible to realize continuous tracking of vehicle behavior trajectory and dynamic intention judgment, resulting in blind spots in the control of key vehicles. For example, in the face of complex traffic behaviors such as frequent detours, temporary lane changes, and fake license plates, the traditional portal system can only passively record static events, and lacks the ability to predict and intervene abnormal behavior development. In addition, most of the existing technologies are mainly based on rule matching, which is difficult to adapt to the changing traffic environment and behavior characteristics, and cannot dig out the behavior deviation trend of potential risk vehicles, nor can it realize accurate profiling and key control through historical behavior rules.

[0004] In summary, in the prior art, due to the dispersion of data acquisition equipment, lack of unified coding and real-time fusion of information, the behavior state of vehicles in multiple portals and multiple scenes cannot be continuously tracked and dynamically modeled, which further affects the timeliness of abnormal behavior identification, the accurate control of key vehicles, and the intelligent response capability of traffic control strategies. SUMMARY

[0005] The purpose of the present application is to provide a traffic portal intelligent management method and device based on vehicle state perception, to solve the technical problem in the prior art that due to the dispersion of data acquisition equipment, lack of unified coding and real-time fusion of information, the behavior state of vehicles in multiple portals and multiple scenes cannot be continuously tracked and dynamically modeled, which further affects the timeliness of abnormal behavior identification, the accurate control of key vehicles, and the intelligent response capability of traffic control strategies.

[0006] In view of the above problems, the present application provides a traffic portal intelligent management method and device based on vehicle state perception.

[0007] In a first aspect, the application provides a traffic checkpoint intelligent management method based on vehicle state perception, which is implemented by a traffic checkpoint intelligent management device based on vehicle state perception, and includes the following steps: collecting vehicle multi-dimensional information by a front-end perception device to obtain an initial vehicle state information flow; aligning and fusing the initial vehicle state information flow in time and space to construct a standardized vehicle state vector; performing preliminary behavior identification on the vehicle state according to the standardized vehicle state vector to obtain high-frequency abnormal behavior vehicles; performing time series modeling on the short-time behavior of the vehicle by a deep learning model to identify an intention prediction label of the vehicle; comparing the intention prediction label with a vehicle behavior portrait library to evaluate behavior deviation detection and obtain a behavior deviation degree; performing risk judgment according to the behavior deviation degree and executing a passing strategy optimization to generate a signal control scheme.

[0008] Preferably, the traffic checkpoint intelligent management method based on vehicle state perception further includes the following steps: deploying a camera to obtain vehicle image information and extracting license plates, vehicle models and colors based on the vehicle image information; setting up a geomagnetic coil to collect vehicle speed, position and passing time; using a wireless identification device to read vehicle identification information; collecting environmental state data at the checkpoint, including visibility, lighting conditions and meteorological factors; and encoding the license plates, vehicle models, colors, vehicle speed, position, passing time, vehicle identification information and environmental state data into the initial vehicle state information flow.

[0009] Preferably, the traffic checkpoint intelligent management method based on vehicle state perception further includes the following steps: synchronously processing the initial vehicle state information flow by timestamp calibration and performing spatial registration to form a unified vehicle state vector; performing labelization processing on the unified vehicle state vector and storing it in a cache area to construct the standardized vehicle state vector.

[0010] Preferably, the traffic checkpoint intelligent management method based on vehicle state perception further includes the following steps: determining whether there is a behavior trend that does not conform to a preset passing rule in the passing mode of the vehicle by the standardized vehicle state vector; and starting a local recording and highlighting identification mechanism for high-frequency abnormal behavior vehicles in the preliminary identification result.

[0011] Preferably, the traffic checkpoint intelligent management method based on vehicle state perception further includes the following steps: extracting a trajectory sequence of the high-frequency abnormal behavior vehicle in a continuous time period; performing time series modeling on the trajectory sequence by a recurrent neural network to obtain a vehicle time series model; and determining whether there is a trend intention of the vehicle by the vehicle time series model and outputting the intention prediction label.

[0012] Preferably, the traffic checkpoint intelligent management method based on vehicle state perception further includes the following steps: establishing a unique identification for each vehicle and associating it with its historical passing record, recording passing characteristics, time period habits and historical behaviors of each vehicle, and constructing the vehicle behavior portrait library.

[0013] Preferably, the traffic kiosk intelligent management method based on vehicle state perception further comprises: uploading the behavior deviation degree to a cloud platform for structured processing, and connecting with a city traffic database to fuse vehicle flow state data of each kiosk and road to build a global traffic state; based on the global traffic state, evaluating kiosk traffic pressure and traffic efficiency, and combining the high-frequency abnormal behavior vehicles to perform key vehicle control and traffic management.

[0014] In a second aspect, the application also provides a traffic kiosk intelligent management device based on vehicle state perception, which is used to execute the traffic kiosk intelligent management method based on vehicle state perception as described in the first aspect, comprising: an initial vehicle state information flow obtaining module for obtaining initial vehicle state information flow by collecting vehicle multi-dimensional information through a front-end perception device; a standardized vehicle state vector construction module for aligning and fusing the initial vehicle state information flow in time and space to construct a standardized vehicle state vector; a high-frequency abnormal behavior vehicle obtaining module for preliminarily identifying vehicle state according to the standardized vehicle state vector to obtain high-frequency abnormal behavior vehicles; a vehicle intention prediction label obtaining module for time-series modeling of vehicle short-time behavior through a deep learning model to identify a vehicle intention prediction label; a behavior deviation degree obtaining module for comparing and evaluating the behavior deviation by comparing the intention prediction label with a vehicle behavior portrait library to obtain a behavior deviation degree; and a signal control scheme generation module for risk judgment according to the behavior deviation degree and performing traffic strategy optimization to generate a signal control scheme.

[0015] The technical solutions provided in the application have at least the following technical effects or advantages: by achieving the technical target of multi-dimensional dynamic perception and intelligent risk control of vehicle behavior, the technical effects of improving traffic abnormal behavior identification accuracy, strengthening key vehicle management efficiency and optimizing overall traffic flow safety are achieved.

[0016] The above description is only a summary of the technical solutions of the application. In order to enable the technical means of the application to be more clearly understood, and in order to enable the above and other purposes, features and advantages of the application to be more apparent and easy to understand, the following specific embodiments of the application are described. It should be understood that the contents described in this part are not intended to identify the key or important features of the embodiments of the application, nor are they intended to limit the scope of the application. Other features of the application will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only exemplary, and for those skilled in the art, other drawings can be obtained without creative labor on the basis of the provided drawings.

[0018] Figure 1 The flowchart of the traffic portal intelligent management method based on vehicle state perception of the present application.

[0019] Figure 2 The structural schematic diagram of the traffic portal intelligent management device based on vehicle state perception of the present application.

[0020] Legend: initial vehicle state information flow obtaining module 11, standardized vehicle state vector construction module 12, high-frequency abnormal behavior vehicle obtaining module 13, vehicle intention prediction label obtaining module 14, behavior deviation degree obtaining module 15, signal control scheme generation module 16. DETAILED DESCRIPTION

[0021] The present application provides a traffic portal intelligent management method and device based on vehicle state perception, which solves the technical problem in the prior art that due to the dispersion of data acquisition equipment, lack of unified coding and real-time fusion of information, the behavior state of vehicles in multiple portals and multiple scenes cannot be continuously tracked and dynamically modeled, further affecting the timeliness of abnormal behavior identification, the precision of key vehicle control, and the intelligent response ability of traffic control strategies. The technical goal of multi-dimensional dynamic perception and intelligent risk control of vehicle behavior is achieved, and the technical effect of improving the accuracy of traffic abnormal behavior identification, enhancing the management efficiency of key vehicles, and optimizing the safety of overall traffic flow is achieved.

[0022] Hereinafter, the technical solutions in the present application will be described clearly and completely with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. It should be understood that the present application is not limited by the example embodiments described herein. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application. In addition, it should be noted that, for convenience of description, only parts related to the present application are shown in the drawings, not all.

[0023] Embodiment one, please refer to the accompanying Figure 1 The present application provides a traffic portal intelligent management method based on vehicle state perception, which is applied to a traffic portal intelligent management device based on vehicle state perception, and specifically includes the following steps:

[0024] S1: acquiring vehicle multi-dimensional information through a front-end perception device to obtain an initial vehicle state information flow.

[0025] Specifically, a series of perception terminals are deployed in scenarios such as traffic checkpoints, road sections or intersections, and the terminals include cameras, geomagnetic coils, wireless radio frequency identifiers, laser radars and other devices for real-time perception and data collection of vehicles from multiple dimensions. For example, the camera can identify the license plate, vehicle model, and vehicle color, the geomagnetic coil can detect whether the vehicle passes and its speed, and the wireless radio frequency identifier reads the identity code transmitted by the vehicle-mounted device to confirm the vehicle identity. Multi-dimensional information means that each vehicle is simultaneously collected in the passing process. Various attributes and behavior states, including static vehicle features, dynamic passing trajectories, behavior patterns, and environmental conditions, can accurately reflect the instantaneous state of the vehicle. With the fusion and real-time transmission of data between devices, the initial vehicle state information flow is finally formed, that is, a data set arranged in chronological order and describing the current state of a single vehicle.

[0026] S2: Aligning and fusing the initial vehicle state information flow in time and space to construct a standardized vehicle state vector.

[0027] Specifically, aligning and fusing the initial vehicle state information flow in time and space means matching and integrating data from different front-end perception devices under a unified time reference and spatial coordinate system, so that each piece of information can correspond to the true state of the same vehicle at a certain time and place. Time alignment is mainly achieved through timestamp calibration, that is, there may be millisecond or second differences in recording data by different devices. After synchronization by a unified time source, it is ensured that the image obtained from the camera, the passing signal detected by the geomagnetic coil, and the ID data identified by the wireless identification of a certain vehicle can be accurately matched. Spatial fusion requires projecting data from different positions or perspectives into a unified geographic coordinate system to eliminate errors caused by device position differences and ensure that the information reflects the accurate position and direction of the vehicle on the actual road after fusion. After alignment, the structured vehicle state data is encoded by field, unified format and converted into vector form, called standardized vehicle state vector. The standardized vehicle state vector is a set of fixed-length, fixed-order numerical representations, such as the first dimension representing vehicle speed, the second dimension representing passing direction, the third dimension representing vehicle type, and the fourth dimension representing light intensity. Through vectorization structure, machine learning model calling and subsequent behavior analysis can be facilitated. With the increase of vehicle passing times, the system can continuously enrich the vehicle state vector library, for example, 100,000 initial information flows are collected every day, and about 90,000 effective standardized vehicle state vectors can be generated after fusion. The data volume will further expand with the increase of device density and traffic flow, laying a data foundation for subsequent intelligent identification, prediction and warning.

[0028] S3: Preliminary behavior identification of vehicle state according to the standardized vehicle state vector to obtain high-frequency abnormal behavior vehicles.

[0029] Specifically, the vehicle state is preliminarily behavior-identified according to the standardized vehicle state vector, and the current behavior of the vehicle is classified and identified through rule judgment or intelligent algorithm, so as to judge whether there is a situation that does not conform to the regular traffic behavior. The vehicle state refers to the collection of the motion characteristics, traffic mode and environment-related characteristics of the vehicle at a certain moment, such as whether the speed of the vehicle is out of limit, whether the path is detoured, whether it appears in the time limit period, whether it violates the guide lane rule, etc., and the preliminary behavior identification is a rapid judgment on these characteristics, which belongs to the first stage of traffic behavior analysis. The identification process can use a threshold rule-based method, such as detecting that a vehicle has turned left in a non-guide lane for five times in a row, or it can combine a machine learning model to learn and identify unusual but potentially risky behavior patterns through historical data. The vehicles with high frequency of abnormal behavior in a period of time are counted and marked as high-frequency abnormal behavior vehicles. High-frequency abnormal behavior refers to the behavior of a vehicle repeatedly deviating from normal traffic rules at multiple time periods and multiple locations, such as continuously violating the speed limit for multiple days, frequently changing lanes or temporarily stopping, which usually has a certain behavior trend or potential risk.

[0030] S4: Time series modeling of vehicle short-time behavior is performed through a deep learning model to identify the intention prediction label of the vehicle.

[0031] Specifically, time series modeling of vehicle short-time behavior is performed through a deep learning model, and models with time series processing capability such as recurrent neural networks, long short-term memory networks or gated recurrent units are used to analyze the behavior characteristics of the vehicle occurring in a short time, so as to establish a dynamic relationship model of the evolution of vehicle behavior over time. Vehicle short-time behavior refers to a series of actions occurring within tens of seconds to several minutes, such as continuous lane changing, acceleration and deceleration, sudden stopping or hesitation near an intersection, which may not be abnormal individually, but when combined together, they can reflect the potential intention of the vehicle. Time series modeling is to input these standardized vehicle state vectors arranged in time sequence into the neural network, so that the model learns the causal and trend relationships between behaviors, such as decelerating first and then pulling over, which may mean that the vehicle will stop soon, and changing lanes first and then accelerating, which may indicate the intention to overtake. The intention prediction label of the vehicle identified by the model analysis is a label with practical significance assigned to the current short-time behavior of each vehicle, which can include prediction states such as imminent left turn, right turn, temporary stop, detour, and red light running.

[0032] S5: The intention prediction label is compared with the vehicle behavior portrait library to evaluate behavior deviation detection, and the behavior deviation degree is obtained.

[0033] Specifically, the behavior deviation detection is evaluated by comparing the intention prediction label with the vehicle behavior portrait library, which means matching and difference analysis between the current vehicle behavior intention identified by the deep learning model and the behavior portrait accumulated in the long-term traffic process of the vehicle, so as to judge whether the current behavior deviates from the historical traffic rule. The intention prediction label represents the possible actions of the vehicle in a short time, such as about to turn left, pull over to stop or try to bypass, etc., while the vehicle behavior portrait library is a behavior profile established for each vehicle based on historical data, including its typical traffic path, time preference, common behavior pattern and environmental adaptability characteristics, etc. For example, a vehicle enters the urban area in a fixed time period for most of the past month and never stops temporarily, but suddenly slows down and continues to change lanes during the peak period, so the difference between the intention prediction label "pull over to stop" and the historical portrait "commuter high-speed straight" is compared to identify the abnormal behavior deviation. The process of evaluating the behavior deviation detection can be quantitatively judged by using distance function, similarity calculation or rule reasoning method, and the result is the behavior deviation degree, which represents the deviation degree of the current behavior from the past characteristics. The greater the deviation degree, the more the behavior deviates from the behavior portrait of the vehicle. With the increase of the information dimension of the portrait library and the improvement of the recognition model accuracy, the calculation of the behavior deviation degree will be more accurate, for example, initially only based on path change to identify deviation, and later can be extended to multiple dimensions such as time period, traffic state, environmental response, etc., so as to form a point-to-surface, coarse-to-fine growth relationship, providing stronger data support for key vehicle control and abnormal behavior research and judgment.

[0034] S6: judging the risk according to the behavior deviation degree, and performing traffic strategy optimization to generate a signal control scheme.

[0035] Specifically, the risk judgment according to the behavior deviation degree refers to the comprehensive evaluation of whether the vehicle has potential risks after obtaining the deviation degree between the current behavior of the vehicle and the historical behavior portrait of the vehicle, combined with the traffic management rules and environmental context information. The basis for risk judgment not only includes the numerical value of the behavior deviation degree, but also involves the type of deviation behavior, the time period characteristics and the sensitivity of the region, for example, the risk level of the sudden detour behavior at the school gate region may be higher than that of the same behavior on ordinary road sections. When it is determined that there is a high level of risk, the traffic strategy optimization process will be started, that is, the traffic control means such as signal timing, lane guidance, warning prompt involving the vehicle will be adjusted to minimize the probability of potential interference or accidents. Traffic strategy optimization is a dynamic adjustment to local traffic state, for example, by extending the green light duration of the main road to relieve traffic flow, shortening the traffic time of high-risk direction or limiting abnormal vehicles from entering sensitive areas. After completing the strategy adjustment, a signal control scheme is generated, that is, the optimized decision results are converted into specific red-green light timing control logic, which is automatically issued and executed by the traffic signal control system. For example, if a vehicle has a temporary stop intention that is completely inconsistent with the history in the peak period, it can be judged as a potential violation or vehicle failure risk, and the traffic time of the direction of the road section is shortened and a voice warning is given in advance to ensure that the overall traffic order is not disturbed. With the refinement of urban perception system and the improvement of data processing capability, the response speed and strategy type of signal control scheme will continue to increase, realizing the growth change from passive response to active prevention, and further promoting the intelligent traffic from static rules to dynamic self-adaptation.

[0036] Further, the application also includes: arranging a camera to obtain vehicle image information, and extracting a license plate, a vehicle type and a color based on the vehicle image information; arranging a geomagnetic coil to collect vehicle speed, position and passing time; using a wireless identification device to read vehicle identification information; collecting environmental state data at the toll station, including visibility, lighting conditions and meteorological factors; and encoding the license plate, vehicle type, color, vehicle speed, position, passing time, vehicle identification information and environmental state data into the initial vehicle state information flow.

[0037] Specifically, by arranging a camera at the traffic toll station, image information of the vehicle when passing through the toll station can be obtained. These images include the front, side or rear of the vehicle, and after processing the images through computer vision technology, key information fields such as the vehicle license plate number, vehicle type (such as small cars, SUVs, trucks, etc.) and vehicle color can be extracted. The identification result will be used as the basic feature of the vehicle identity for subsequent analysis, comparison and behavior judgment.

[0038] Next, to acquire dynamic data of vehicles passing through checkpoints, geomagnetic coils are installed beneath the road surface. When a vehicle passes these coils, the timing of its entry and exit can be detected, and its speed at the checkpoint can be calculated. These coils can also pinpoint the vehicle's precise location within the checkpoint, helping to more accurately reconstruct the vehicle's trajectory and temporal characteristics, thus forming the spatiotemporal coordinate basis of the vehicle's status.

[0039] Meanwhile, to identify the unique information of special vehicles or registered vehicles, wireless identification devices, such as RFID readers or DSRC equipment, are installed at checkpoints. When a vehicle is equipped with a corresponding electronic tag (such as an ETC tag or emergency vehicle identification), its unique identification code can be read to determine whether the vehicle has special passage rights, is on a blacklist, or is a key monitoring target.

[0040] In addition, real-time environmental data is collected at the checkpoint, including visibility (whether there is fog, dust, or other obstructions), lighting conditions (daytime, nighttime, or backlighting), and meteorological factors (such as rain, snow, wind speed, and temperature). External conditions may affect the clarity of camera images or the stability of sensors, and the perception strategy and recognition algorithm thresholds are automatically adjusted according to environmental changes.

[0041] Finally, the acquired information items, including license plate, vehicle type, color, vehicle speed, location, passage time, vehicle identification information, and environmental status data, are uniformly encoded into an initial vehicle status information stream according to a predetermined format. This constitutes the original data foundation for vehicle traffic status, used for subsequent behavior modeling, traffic judgment, and strategy generation, and enables standardized cross-device data interaction and multi-source fusion processing.

[0042] Furthermore, this application also includes: using timestamp calibration to synchronize the initial vehicle state information stream and perform spatial registration to form a unified vehicle state vector; and tagged the unified vehicle state vector and storing it in a cache area to construct the standardized vehicle state vector.

[0043] Specifically, timestamp calibration technology is used for synchronization. Timestamp calibration refers to aligning data collected from different sensing devices (such as cameras, geomagnetic coils, and wireless identification devices) according to a unified time base to ensure that all information reflects the vehicle's status at the same moment or under the same event. Because the sampling frequency and internal clock of each device may differ, without calibration, a mismatch between a vehicle's speed and image information may occur, thus affecting the accuracy of recognition.

[0044] Subsequently, after completing time synchronization, spatial registration processing is performed. The core of spatial registration is to uniformly map the data coordinates from different spatial positions or perspectives. For example, there may be a physical offset between the image coordinates recorded by the camera and the ground coordinates recorded by the geomagnetic coil. Through the registration algorithm, these data can be transformed into the same coordinate system, thereby more accurately describing the actual position and motion trajectory of the vehicle, ensuring that the spatial description of the vehicle state remains consistent and comparable.

[0045] By completing time synchronization and spatial registration, all raw data is integrated to form a unified vehicle state vector. The vehicle state vector is a structured multi-dimensional data representation form that includes static attributes of the vehicle (such as license plate, vehicle model, color), dynamic attributes (such as speed, acceleration, position), and external environmental information (such as lighting, weather), to comprehensively express the comprehensive state of a vehicle at a certain time.

[0046] Subsequently, the unified vehicle state vector is labeled to assign standardized semantic labels to different data fields, facilitating subsequent storage, retrieval, and analysis. For example, the color field is labeled as "vehicle feature" and the speed field is labeled as "dynamic parameter", giving it a clear classification meaning in the database. In addition, the labeling process can also be used for data preprocessing in machine learning tasks, helping to improve model training efficiency.

[0047] Finally, the labeled vehicle state vector is stored in the cache area and further standardized vehicle state vectors are constructed. Standardization refers to the normalization of data structure, field order, unit, etc. to ensure that multiple vehicle state information has a unified format when compared and integrated. For example, speed is unified in "kilometers per hour", and time is converted to "year-month-day-hour-minute-second" format. The cache area provides temporary and efficient access space, allowing data to be quickly supplied to real-time analysis systems and model engines.

[0048] Further, the application also includes: determining whether the vehicle passing mode has a behavior trend that does not conform to the preset passing rule through the standardized vehicle state vector; starting a local recording and highlighting identification mechanism for vehicles with high-frequency abnormal behavior in the preliminary identification result.

[0049] Specifically, after forming the standardized vehicle state vector, the vector is used to determine whether the vehicle's passing mode has a behavior trend that is inconsistent with the preset passing rule. The passing mode refers to the typical behavior trajectory and action combination of the vehicle when passing through the toll gate or road, such as whether to remain in a fixed lane, whether to pass normally with deceleration, whether to comply with traffic signals, etc. The preset passing rule is a passing behavior standard set by the traffic management department according to traffic regulations or specific scenarios, for example, vehicles must not change lanes in the toll gate area, must not reverse, must not stagnate for a long time, etc. When the behavior mode reflected by the standardized vehicle state vector has a trend that is obviously inconsistent with these rules, it is determined that it has an abnormal tendency, such as frequent lane changing, sudden turning near the toll gate, or sudden acceleration in the speed limit area, etc., thereby issuing a risk warning.

[0050] For the initially identified vehicles, if a vehicle frequently exhibits the above abnormal behaviors within a short time or multiple time periods, it is marked as a high-frequency abnormal behavior vehicle. High frequency refers to the occurrence of inconsistent behaviors more than 3 times within a specific time window, for example, 10 minutes, or the occurrence of multiple different types of violations within 1 hour. In order to enhance management and intervention efficiency, a local recording mechanism will be started for these vehicles to save their passing trajectories, abnormal types, and time and location of behavior occurrence in real time. At the same time, a highlight identification mechanism will also be enabled, that is, the vehicle will be highlighted in color, icon marked or risk level reminded on the platform interface, so that the police or management personnel can quickly discover, track or intercept such high-risk vehicles on the monitoring terminal.

[0051] Through the continuous judgment and labeling mechanism, not only single abnormal behavior can be responded to, but also overall monitoring of behavior trends can be established. For example, a vehicle continuously passes the line at the same time period for 3 consecutive days, even if each time the line is pressed for a short time, but the repeated deviation behavior may mean potential dangerous driving or intentional rule-bending behavior. At this time, the vehicle will be automatically included in the list of key attention and enter the long-term behavior analysis sequence. With the increase of vehicle passing data accumulation, if 100 vehicles pass through the toll gate every day, and 10 of them are identified as high-frequency abnormal behavior, there will be 300 abnormal records in 1 month.

[0052] Further, the application also includes: extracting a trajectory sequence of the high-frequency abnormal behavior vehicle within a continuous time period; using a recurrent neural network to perform time series modeling on the trajectory sequence to obtain a vehicle time series model; determining whether the vehicle has a trend intention through the vehicle time series model, and outputting an intention prediction label.

[0053] Specifically, for the identified high-frequency abnormal behavior vehicles, their trajectory sequences in continuous time periods are extracted. The trajectory sequence refers to the set of motion trajectories formed by the vehicle continuously passing through different spatial positions within a certain time range, including time points, geographic coordinates, speed directions, etc. The definition of continuous time period refers to uninterrupted or short-interval multiple passing behavior records, such as a vehicle passing through the same checkpoint every morning from 8:00 to 9:00 for 5 days, and all its passing trajectories will be extracted continuously for analysis of its regularity or potential intent.

[0054] Then, the trajectory sequence is modeled using a recurrent neural network. The recurrent neural network is a deep learning model that is good at processing data with time dependence. Its internal structure can remember and pass on the information of the previous time to capture the regularity of data evolution over time. After the trajectory sequence is input into the recurrent neural network, the model can learn the behavior change trend of the vehicle at different time points, such as gradual acceleration or regular detour, and further refine the behavior pattern of the vehicle. Through modeling, it is no longer limited to judging single-point abnormalities, but can also understand the internal relationship between behaviors.

[0055] Subsequently, based on the constructed vehicle time sequence model, the future behavior of the vehicle is judged for trend and an intent prediction label is output. The trend intent refers to whether the vehicle is likely to be planning to perform a certain behavior, such as frequently detouring a certain area to evade monitoring, repeatedly approaching the checkpoint boundary to observe law enforcement rules, etc. The intent prediction label is the classification result of these trend behaviors, such as being labeled as “avoiding monitoring”, “possible to rush the checkpoint”, “illegal detouring” or “abnormal parking”, providing a basis for targeted disposal for the traffic supervision department.

[0056] By abstracting the trajectory information into time series data and then modeling and extracting deep rules by a neural network, a leap from current behavior recognition to future behavior prediction can be achieved. For example, among 20 vehicles identified as high-frequency abnormal vehicles within 1 week, 8 vehicles show obvious repeated trajectories. After modeling by a recurrent neural network, it is predicted that 3 of them have the intent to evade monitoring. The prediction result can be pushed to the front-end law enforcement system or an automatic warning platform in advance, so as to intervene before a specific violation event occurs. With the accumulation of data, the prediction capability will continue to improve. If 100 trajectory sequences are extracted daily, there will be more than 30,000 trajectory samples after one year, which is enough to train more accurate and efficient vehicle behavior prediction models.

[0057] Further, the application also includes: establishing a unique identifier for each vehicle and associating its historical passing records, recording the passing characteristics, time period habits and historical behaviors of each vehicle, and constructing a vehicle behavior portrait library.

[0058] Specifically, in order to realize the personalized management and identification of each vehicle, a unique identification is established for each vehicle. The unique identification refers to the identity code of each vehicle, which is based on the license plate number, wireless identification code or encrypted ID generated by multi-feature fusion, so that each vehicle can be accurately distinguished in a large amount of data, and its traffic situation can be continuously tracked. The unique identification is the basis for building a vehicle data model, ensuring the integrity and continuity of data association.

[0059] Next, the unique identification of each vehicle is associated with its historical traffic record. The historical traffic record includes various information collected by the traffic checkpoint at different times and different places, such as traffic time, traffic speed, stay duration, path trajectory, etc. By binding with the vehicle identification, the behavior log of each vehicle in different scenarios can be formed, providing data sources for subsequent modeling.

[0060] Further, the traffic characteristics of each vehicle are extracted. Traffic characteristics refer to the behavior of vehicles in multiple spatio-temporal dimensions, such as whether they often travel during peak hours, whether they prefer a particular route, whether they have lane changes, U-turns, etc. The characteristics are obtained by statistical and analytical historical data, and are important indicators reflecting the stability, compliance and risk level of vehicle behavior.

[0061] In addition, the time habits of the vehicle are analyzed, i.e. the time period when the vehicle often appears within a day or a week, such as a vehicle often appearing at the urban checkpoint between 7am and 8am, or frequently traveling at night on Fridays. The time regularity reflects the use mode of the vehicle, which is closely related to the driver's work and rest arrangement, and can also help to determine whether the vehicle is abnormal or sudden.

[0062] Finally, combined with the traffic characteristics and time habits, the historical behavior of the vehicle is recorded, including whether there are illegal records, whether the abnormal traffic rules are triggered multiple times, whether the traffic path is often changed, etc. Through multi-dimensional data fusion, a complete vehicle behavior portrait library is constructed. The portrait library is a dynamically updated data set that represents the behavior pattern and risk portrait of each vehicle, which not only helps traffic management personnel to control and manage vehicles, but also provides a basis for intelligent prediction and active intervention.

[0063] Further, the application also includes: uploading the behavior deviation degree to a cloud platform for structured processing, and interfacing with a city traffic database to fuse vehicle flow state data of each checkpoint and road to construct a global traffic state; based on the global traffic state, evaluating the checkpoint traffic pressure and traffic efficiency, and combining the high-frequency abnormal behavior vehicles to perform key vehicle control and traffic management.

[0064] Specifically, the behavior deviation of each vehicle is uploaded to the cloud platform. The behavior deviation refers to the difference between the actual traffic behavior of the vehicle and the normal or preset traffic mode, which may be reflected in traffic speed, path selection, traffic time anomaly or illegal behavior, etc. The greater the deviation, the more the vehicle behavior deviates from the norm or standard. The deviation data is uploaded to the cloud platform in batches, which facilitates unified storage, centralized processing, and supports large-scale vehicle behavior trend analysis and modeling.

[0065] Next, the deviation data uploaded to the cloud platform is structured. Structured processing refers to converting unstructured or semi-structured data (such as raw video analysis results, free-form recognized text) into queryable and computable standardized table data, including steps such as uniform field format, attribute classification, and time format conversion. For example, "3 times of detour and reverse driving during evening peak hours" is converted to "time period = evening peak, behavior type = detour and reverse driving, frequency = 3" for subsequent analysis and processing.

[0066] Then, the processed data is interfaced with the city traffic database. The city traffic database is a large comprehensive information platform that collects information on various roads, toll gates, vehicles, traffic lights, traffic events, and other information within the city. Integrating with this database allows vehicle behavior data to be integrated with more extensive traffic environment information, providing more accurate traffic evaluation capabilities. The interface is achieved through an interface protocol to ensure data interoperability and real-time updates.

[0067] In the integration process, vehicle flow state data from each toll gate and road is introduced to construct the global traffic state. Vehicle flow state data includes the number of vehicles passing through per unit time, average speed, queue length, waiting time, and other content, which comes from devices such as geomagnetic coils, video recognition systems, and traffic light controllers. The global traffic state refers to a collection of real-time traffic conditions based on the city or regional scope, which is a comprehensive reflection of the city's traffic operating state, helping to identify congestion hotspots, traffic bottlenecks, and the scope of abnormal events.

[0068] Based on the global traffic state, the traffic pressure and efficiency of each toll gate are further evaluated. Traffic pressure refers to the degree of vehicle load faced by the toll gate per unit time, reflecting whether the toll gate is close to saturation or already congested. Traffic efficiency reflects the proportion of vehicles that successfully pass through the toll gate per unit time, with higher efficiency indicating that the toll gate is well scheduled and traffic is smooth. Combining these two indicators, toll gate scheduling can be dynamically optimized, traffic light durations can be adjusted, or control measures can be initiated.

[0069] Finally, combined with the evaluation results and the identified high-frequency abnormal behavior vehicles, the key vehicle deployment and traffic management are performed. The key vehicle deployment refers to setting strategies to track, limit traffic, warn or intervene in specific vehicles in real time; traffic management may include vehicle traffic restrictions, traffic path recommendations, entry into restricted areas, and other measures. It can achieve closed-loop management from individual vehicle behavior analysis to city-level joint control.

[0070] In summary, the traffic checkpoint intelligent management method based on vehicle state perception provided by the present application has the following technical effects: by achieving the technical goal of multi-dimensional dynamic perception and intelligent risk control of vehicle behavior, the technical effects of improving the accuracy of traffic abnormal behavior identification, strengthening the efficiency of key vehicle management, and optimizing the overall traffic flow safety are achieved.

[0071] In the second embodiment, based on the same inventive concept as the traffic checkpoint intelligent management method based on vehicle state perception in the preceding embodiments, the present application also provides a traffic checkpoint intelligent management device based on vehicle state perception, please refer to the attached Figure 2 , including: an initial vehicle state information flow obtaining module 11 for collecting vehicle multi-dimensional information through a front-end perception device to obtain an initial vehicle state information flow; a standardized vehicle state vector construction module 12 for aligning and fusing the initial vehicle state information flow in time and space to construct a standardized vehicle state vector; a high-frequency abnormal behavior vehicle obtaining module 13 for preliminarily identifying the behavior of a vehicle according to the standardized vehicle state vector to obtain a high-frequency abnormal behavior vehicle; a vehicle intent prediction label obtaining module 14 for modeling the short-time behavior of a vehicle through a deep learning model to identify a vehicle intent prediction label; a behavior deviation degree obtaining module 15 for comparing and evaluating the behavior deviation between the intent prediction label and a vehicle behavior portrait library to obtain a behavior deviation degree; and a signal control scheme generation module 16 for judging the risk according to the behavior deviation and performing traffic strategy optimization to generate a signal control scheme.

[0072] Further, the traffic checkpoint intelligent management device based on vehicle state perception is also used to: deploy a camera to obtain vehicle image information, and extract license plates, vehicle models, and colors based on the vehicle image information; set a geomagnetic coil to collect vehicle speed, position, and passing time; use a wireless identification device to read vehicle identification information; collect environmental state data at the checkpoint, including visibility, lighting conditions, and meteorological factors; and encode the license plates, vehicle models, colors, vehicle speeds, positions, passing times, vehicle identification information, and environmental state data into the initial vehicle state information flow.

[0073] Further, the traffic kiosk intelligent management device based on vehicle state perception is further used for: synchronously processing the initial vehicle state information flow by using timestamp calibration, and performing spatial registration to form a unified vehicle state vector; performing label processing on the unified vehicle state vector and storing the unified vehicle state vector in a cache area to construct the standardized vehicle state vector.

[0074] Further, the traffic kiosk intelligent management device based on vehicle state perception is further used for: judging, by using the standardized vehicle state vector, whether a vehicle passing mode has a behavior trend that does not conform to a preset passing rule; and starting a local recording and highlighting identification mechanism for a vehicle with a high-frequency abnormal behavior in a preliminary identification result.

[0075] Further, the traffic kiosk intelligent management device based on vehicle state perception is further used for: extracting a trajectory sequence of the vehicle with the high-frequency abnormal behavior in a continuous time period; performing time series modeling on the trajectory sequence by using a recurrent neural network to obtain a vehicle time series model; judging, by using the vehicle time series model, whether the vehicle has a trend intention, and outputting an intention prediction label.

[0076] Further, the traffic kiosk intelligent management device based on vehicle state perception is further used for: establishing a unique identification for each vehicle and associating a historical passing record of the vehicle, recording passing features, time period habits and historical behaviors of each vehicle, and constructing a vehicle behavior portrait library.

[0077] Further, the traffic kiosk intelligent management device based on vehicle state perception is further used for: uploading the behavior deviation degree to a cloud platform for structured processing, and connecting with a city traffic database to fuse vehicle flow state data of each kiosk and road to construct a global passing state; evaluating kiosk passing pressure and passing efficiency based on the global passing state, and combining the vehicle with the high-frequency abnormal behavior to perform key vehicle control and passing management.

[0078] Each embodiment in the specification is described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The traffic kiosk intelligent management method based on vehicle state perception in the first embodiment and the specific examples are also applicable to the traffic kiosk intelligent management device based on vehicle state perception in the present embodiment. Based on the detailed description of the traffic kiosk intelligent management method based on vehicle state perception, those skilled in the art can clearly know the traffic kiosk intelligent management device based on vehicle state perception in the present embodiment. Therefore, for the sake of brevity of the specification, the traffic kiosk intelligent management device based on vehicle state perception in the present embodiment is not described in detail.

[0079] The foregoing description of the disclosed embodiments enables a person skilled in the art to make or use the application. Numerous modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein can be applied to other embodiments without the use of the inventive faculty. Therefore, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0080] It will be readily apparent to one skilled in the art that varying substitutions and modifications can be made to the application disclosed herein without departing from the scope and spirit of the application. Moreover, it is the intent that all such variations and modifications be considered as falling within the scope of the application, and that the application be limited only by the definitions contained in the appended claims.

Claims

1. A traffic checkpoint intelligent management method based on vehicle status perception, characterized in that, include: The initial vehicle status information stream is obtained by collecting multi-dimensional vehicle information through front-end sensing devices; The initial vehicle state information stream is aligned and fused in time and space to construct a standardized vehicle state vector; Based on the standardized vehicle state vector, preliminary behavior identification of the vehicle state is performed to identify vehicles with high-frequency abnormal behavior. This step includes: The standardized vehicle state vector is used to determine whether there are behavioral trends in vehicle traffic patterns that are inconsistent with preset traffic rules. The mechanism for local recording and highlighting of vehicles with high-frequency abnormal behavior identified in the preliminary identification results will be activated. The process of using a deep learning model to perform time-series modeling of short-term vehicle behavior and identifying vehicle intent prediction labels includes the following steps: Extract the trajectory sequences of the vehicles exhibiting high-frequency abnormal behavior over consecutive time periods. A recurrent neural network is used to perform time-series modeling on the trajectory sequence to obtain a vehicle time-series model. The vehicle time series model is used to determine whether the vehicle has a trend of intent, and the intent prediction label is output. The intent prediction labels are compared with the vehicle behavior profile database to evaluate behavior deviation detection and obtain the behavior deviation degree. Risk assessment is performed based on the behavioral deviation, and traffic strategy optimization is implemented to generate a signal control scheme. This step includes uploading the behavioral deviation to a cloud platform for structured processing, connecting it with the urban traffic database, and integrating traffic flow status data from various checkpoints and roads to construct a global traffic status. Based on the global traffic status assessment, the checkpoint traffic pressure and efficiency are evaluated, and key vehicle control and traffic management are implemented in conjunction with the high-frequency abnormal behavior vehicles. A unique identifier is established for each vehicle and associated with its historical passage records. The passage characteristics, time period habits and historical behaviors of each vehicle are recorded to construct the vehicle behavior profile database.

2. The intelligent traffic checkpoint management method based on vehicle status perception as described in claim 1, characterized in that, By collecting multi-dimensional vehicle information through front-end sensing devices, an initial vehicle status information stream is obtained, including: Cameras are deployed to acquire vehicle image information, and license plate, vehicle type, and color are extracted based on the vehicle image information; Set up geomagnetic coils to collect vehicle speed, location, and transit time; Use a wireless identification device to read vehicle identification information; Collect environmental data at checkpoints, including visibility, lighting conditions, and meteorological factors; The license plate, vehicle type, color, vehicle speed, location, passage time, vehicle identification information, and environmental status data are encoded into the initial vehicle status information stream.

3. The intelligent traffic checkpoint management method based on vehicle status perception as described in claim 1, characterized in that, The initial vehicle state information stream is aligned and fused in time and space to construct a standardized vehicle state vector, including: The initial vehicle state information stream is synchronized using timestamp calibration and spatially registered to form a unified vehicle state vector. The unified vehicle state vector is tagged and stored in the cache to construct the standardized vehicle state vector.

4. A traffic checkpoint intelligent management device based on vehicle status perception, characterized in that, The steps for implementing the intelligent traffic checkpoint management method based on vehicle status perception as described in any one of claims 1 to 3 include: The initial vehicle status information stream acquisition module is used to collect multi-dimensional vehicle information through front-end sensing devices to obtain the initial vehicle status information stream; A standardized vehicle state vector construction module is used to perform temporal and spatial alignment and fusion of the initial vehicle state information stream to construct a standardized vehicle state vector. The high-frequency abnormal behavior vehicle acquisition module is used to perform preliminary behavior identification on the vehicle state based on the standardized vehicle state vector to obtain high-frequency abnormal behavior vehicles. The vehicle intent prediction label acquisition module is used to perform time-series modeling of short-term vehicle behavior through a deep learning model to identify and obtain vehicle intent prediction labels. The behavior deviation degree acquisition module is used to compare the intent prediction label with the vehicle behavior profile database to evaluate the behavior deviation detection and obtain the behavior deviation degree. The signal control scheme generation module is used to make risk judgments based on the behavioral deviation and to perform traffic strategy optimization to generate a signal control scheme.

Images