Toll station traffic flow intelligent monitoring and scheduling system and method

Abstract

The invention provides a toll station traffic flow intelligent monitoring system and method. The toll station traffic flow intelligent monitoring system comprises a real-time monitoring module, a data processing and analyzing module, a navigation guiding module and an intervention processing module. The real-time monitoring module collects vehicle and traffic flow information and transmits the information to the monitoring center, the data processing and analyzing module preprocesses data, establishes a prediction model, triggers an early warning signal and transmits the early warning signal to the navigation guiding and intervention processing module, and the navigation guiding module receives a prediction result and the early warning signal. The optimal guiding route is calculated and transmitted to the vehicle navigation system, and the intervention processing module monitors the driving condition of the vehicle, carries out marking and laser guiding on the vehicle which does not comply with guiding, projects a warning mark to the vehicle which violates the rule, and feeds back information. According to the scheme, through cooperation of various technologies, the goals of improving the toll station passing efficiency and reducing the operation cost are achieved, intelligent traffic is promoted to be converted from artificial experience to data algorithm driving, and a replicable intelligent scheduling solution is provided.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent transportation, and particularly relates to a toll station vehicle flow intelligent monitoring and dispatching system and method. Background Art

[0002] The toll station vehicle flow intelligent monitoring and dispatching system is the core infrastructure in the field of intelligent transportation. By integrating sensor technology, intelligent algorithms and execution devices, it realizes the real-time monitoring, flow prediction, route planning and dynamic guidance of the vehicle flow in the expressway toll station area, aiming to improve the utilization rate of lane resources, alleviate congestion and reduce the cost of manual intervention.

[0003] Traditional toll station vehicle flow management systems mainly rely on single-point monitoring devices (such as fixed cameras, static traffic signs) and simple rule engines, with the following characteristics: Fragmented data collection: Only local vehicle flow data is obtained through the toll station entrance camera or inductive loop, lacking the collaborative monitoring of the macroscopic vehicle flow before entering the station and the microscopic behavior at the entrance; Weak decision-making ability: Flow prediction relies on a single time series model, without integrating spatio-temporal correlation features, and the prediction error generally exceeds 15%; Route planning mostly adopts static rules, and static route planning cannot respond to dynamic congestion in real time; Coarse guidance method: Lane guidance relies on fixed LED screens or manual command, lacking precise dynamic intervention for vehicles, with a high rate of illegal lane changes, resulting in frequent vehicle flow weaving conflicts at the entrance square; Lack of closed-loop control: The system has no self-optimization mechanism, resulting in the need for manual adjustment of parameters regularly, with a lag in response to sudden congestion, often leading to the coexistence of over-limit queue lengths, idle and overloaded channel resources.

[0004] With the continuous growth of the daily average flow of expressways, the daily average of some hub toll stations reaches more than 100,000 vehicles. The problems of "extensive monitoring, lagging decision-making, and inefficient intervention" of traditional systems have led to an increase in congestion duration and an increase in manual dispatching costs, and it has been difficult to meet the development needs of "efficient, low-carbon, and unmanned" intelligent transportation. There is an urgent need to develop a new generation of systems with three-dimensional perception, dynamic guidance, intelligent algorithm integration and closed-loop self-optimization. Summary of the Invention

[0005] Aiming at the industry pain points of fragmented data collection, extensive dispatching strategies and passive emergency responses in traditional toll stations, the present invention provides a toll station vehicle flow intelligent monitoring and dispatching system and method, realizing a paradigm shift from manual experience dominance to data algorithm drive, and providing a replicable, low-cost, highly reliable and easily expandable intelligent dispatching solution for the field of intelligent transportation.

[0006] The solution of the present invention to solve its technical problems is: to provide a toll station vehicle flow intelligent monitoring system, including a real-time monitoring module, a data processing and analysis module, a navigation guidance module and an intervention processing module.

[0007] Real-time Monitoring Module: It collects vehicle position, speed, vehicle type, vehicle distance, and traffic flow information of each toll lane in real time, and transmits the information to the monitoring center through a wired or wireless communication network.

[0008] Data Processing and Analysis Module: The monitoring center first preprocesses the traffic flow data obtained from real-time monitoring through the data processing and analysis module to remove noise and error data to ensure the accuracy of data analysis. Then, it uses big data analysis technology combined with historical data to establish a traffic flow prediction model, trains and optimizes the model using deep learning algorithms, predicts the future traffic flow of each toll lane, and compares the traffic flow prediction results with the traffic capacity threshold of each toll lane. If the predicted traffic flow exceeds the traffic capacity, an intelligent warning signal is triggered, and the warning signal is transmitted to the navigation guidance module and the intervention processing module.

[0009] Navigation Guidance Module: After receiving the prediction results and warning signals transmitted by the data processing and analysis module, it combines the real-time position, speed, direction of the vehicle, and information of surrounding vehicles, and uses a path planning algorithm to calculate the best guidance route for each vehicle. The guidance route information is transmitted to the in-vehicle navigation system through the communication module, and the in-vehicle navigation system displays guiding arrows and text prompts on the screen to guide the driver to drive according to the recommended route.

[0010] Intervention Processing Module: It monitors whether the vehicle drives according to the navigation guidance through the real-time monitoring module and marks the vehicles that do not receive or comply with the guidance. For the marked vehicles, it projects guiding arrows and / or toll lane numbers to their operation console positions using a laser projection guiding device. The laser line marker adjusts the projection angle in real time according to the vehicle speed, distance, and direction to ensure that the guiding information is accurately projected without disturbing the driver's line of sight. When obvious traffic violations are detected in the vehicle, the laser line marker projects yellow or red warning signs, and at the same time, it feeds back the violation information of these vehicles to the traffic management department.

[0011] An intelligent monitoring and dispatching method for vehicle flow at toll stations includes the following steps: Step 1, data collection and intelligent prediction: Near-point monitoring devices collect real-time vehicle flow data at the toll station entrance and each channel, and transmit it to the monitoring center. Based on historical data and external factors, a traffic flow prediction model is constructed through big data analysis and deep learning algorithms to accurately predict the future vehicle flow of each channel. When the predicted value exceeds the channel capacity threshold, an intelligent warning is triggered, and a warning signal is sent to the navigation guidance module and the interference processing module simultaneously; Step 2, dynamic route planning and dispatching: The monitoring center integrates the real-time position, driving direction, speed, channel traffic status, and prediction data of vehicles, and uses route planning algorithms to generate the first navigation information before the vehicle enters the station, and synchronizes it to the navigation guidance module and the interference processing module in real time to achieve unified dispatching of the incoming vehicle flow; Step 3, in-vehicle navigation guidance: After receiving the warning signal, the navigation guidance module transmits the data of the first navigation information to each in-vehicle navigation system through the communication module to guide each vehicle to drive to the designated toll channel according to the navigation information; at the same time, the navigation guidance module sends the vehicle information of those that have not received the first navigation information to the monitoring center; Step 4, laser projection interference control: The monitoring center marks the vehicles that have not received the first navigation information or have not complied with the first navigation information based on the near-point real-time monitoring data and the feedback data of the navigation guidance module, and sends the marked vehicle information to the interference processing module. The interference processing module transmits the data of the first navigation information to the laser navigation system through the communication module. The laser navigation system controls each near-point laser marking instrument to project each first navigation information onto the instrument panel of the corresponding vehicle near the point in the form of laser projection. The near-point laser marking instrument is assembled on the front bracket of each toll channel.

[0012] The beneficial effects of the present invention:

[0013] 1. Three-dimensional perception improves prediction accuracy and congestion warning ability: Through the double-layer data collection system of far-point and near-point, integrating millimeter-wave radar trajectory tracking and dynamic calibration of meteorological data, a multi-dimensional input vector containing spatio-temporal characteristics is constructed, significantly reducing the vehicle flow prediction error and solving the problems of microscopic behavior oversight and macroscopic trend misjudgment caused by traditional single-camera monitoring.

[0014] 2. The dynamic laser guidance system realizes refined control of vehicle flow: While the navigation guidance module is guiding, an intervention processing module is additionally used. The far and near-point laser marking instruments involved in the intervention processing module use multi-motor linkage adjustment and extended Kalman filter trajectory prediction mechanism to track the vehicle position in real time, enabling the vehicle to complete the lane change decision in advance. The near-point anti-glare laser forced intervention shortens the average queuing length of the entrance square, breaking the problems of lag in traditional static sign guidance and imbalance of lane resources.

[0015] 3. Intelligent algorithm fusion improves decision-making efficiency and resource allocation rationality: The LSTM hybrid model enables millisecond-level data access and multi-cycle prediction. Combining with the dynamic threshold of the M / M / C queuing model significantly reduces the calculation error of channel throughput capacity. The improved ant colony algorithm shortens the path planning time, and the hierarchical guidance strategy enhances the utilization rate of ETC channels and reduces the queuing overflow rate of manual channels, solving the problems of insufficient global optimization and lagging real-time response of traditional heuristic algorithms.

[0016] 4. Through the cooperation of near and far devices, precise laser intervention, multi-modal interaction, and reinforcement learning self-optimization, the dual goals of improving toll station throughput efficiency and reducing operating costs are achieved. A paradigm shift from manual experience-driven to data algorithm-driven is realized, providing a replicable, low-cost, highly reliable, and easily expandable intelligent scheduling solution for the field of intelligent transportation. Brief Description of the Drawings

[0017] Figure 1 It is a block diagram of the system framework of the present invention;

[0018] Figure 2 It is a block diagram of the system module relationship of the present invention;

[0019] Figure 3 It is a comparison diagram of the distribution of near and far monitoring devices;

[0020] Figure 4 It is a block diagram of the intervention processing module system;

[0021] Figure 5 It is a schematic diagram of the structure of the laser marking instrument;

[0022] Figure 6 It is a schematic diagram of the projection effect of the laser marking instrument;

[0023] Figure 7 It is a flowchart of the method of the present invention;

[0024] Figure 8 It is a comparison diagram of the intervention timing of the intervention processing module;

[0025] Figure 9 It is a comparison diagram of the included angle between the vehicle head and the lane and the included angle of the pixel board.

[0026] Reference numerals in the figure: base 1; base 2; buffer block 3; fixing frame 4; main L frame 5; auxiliary L frame 6; fixing seat 7; left and right corner motors 8; rotating motor 9; upper and lower corner motors 10; laser marking instrument 11; pixel board 12; laser reference 13; high-definition camera 14; marking instrument fixing frame 15; camera fixing frame 16. Detailed Embodiments

[0027] The present invention will be further described below in conjunction with the drawings and embodiments.

[0028] Embodiment 1: An intelligent monitoring and dispatching system for traffic flow at toll stations. This system constructs a four-layer technical architecture of "perception - calculation - control - interaction". The system architecture is as shown in Figure 1 shown, and the relationship between system modules is as shown in Figure 2 shown. It realizes the collection of global data through a hardware device cluster, completes traffic flow prediction and path planning relying on an intelligent algorithm engine, and achieves precise guidance with the help of a multi-modal interaction terminal, ultimately forming a closed-loop management system for data collection, intelligent decision-making, and dynamic regulation.

[0029] Far - point monitoring device (sub - system): Deployed on the cross - beam bracket 1 - 2 kilometers away from the toll station entrance at the end of the highway. It uses a high - definition intelligent camera (20 million pixels, supporting license plate recognition and vehicle type classification), a millimeter - wave radar (128 channels, detecting vehicle speed, vehicle distance, and lane distribution), and a meteorological sensor (integrated with temperature - humidity, wind speed, and visibility detectors). It collects macroscopic data of traffic flow within a range of 500 meters, including vehicle queue length, average vehicle speed, vehicle type distribution, and driving trajectory, etc., to form an initial data pool of the pre - approaching traffic flow.

[0030] Near - point monitoring device (sub - system): Deployed at the front end of the toll station entrance square and each toll lane. It uses a panoramic fisheye camera (360° coverage, frame rate 60fps); a combination of an inductive loop (detecting the presence of vehicles) and an RFID reader (identifying vehicle types), and a license plate recognition all - in - one machine. It real - time collects microscopic data such as traffic flow density at the entrance square, queue length of each lane, and vehicle passing time, and constructs a channel - level real - time traffic flow model.

[0031] Far - point guiding device of the intervention processing module: A laser projection module (such as a laser marking instrument) deployed on the bracket at the end of the highway, evenly distributed at intervals of 50 meters; Technical parameters: 635nm red laser, projection distance 50 - 100 meters, supporting dynamic spot calibration, and display content, as shown in Figure 6 shown, in the form of arrows such as forward, left, right, deceleration, stop, and acceleration in a dot - combination form. Near - point guiding device of the intervention processing module: A laser projection module (such as a laser marking instrument) installed on the bracket 5 meters in front of the front end of each toll lane, with adjustable angles (pitch ±15°, rotation ±30°); Technical parameters: such as 532nm green laser or 635nm red laser, projection range covering the entire lane, supporting anti - strong - light interference; Display content: as shown in Figure 6 shown, in the form of arrows such as forward, left, right, deceleration, stop, and acceleration in a dot - combination form.

[0032] Structure of the intervention processing module: As shown in Figure 5As shown in the figure, an intervention processing module includes a buffer fixing mechanism, a main frame assembly, an auxiliary frame assembly, a fixed seat assembly, a laser line marker, a high-definition camera, and a controller. Among them, the buffer fixing mechanism includes a base 1, a base 2, buffer blocks 3, and a fixing frame 4. The base 1 is fixed to the road surface bracket through the fixing frame 4. A plurality of buffer blocks 3 are evenly installed between the base 2 and the base 1. A relief hole is provided at the center of the base 1. The left and right corner motors 8 are located in the relief hole, and the rear end of the left and right corner motors 8 is fixed to the base 2. Main frame assembly: It includes a main L-shaped frame 5 and a rotation motor 9. The rotating shaft of the left and right corner motors 8 is connected to the outer side of the upper arm of the main L-shaped frame 5. A rotation motor 9 is fixed to the inner side of the lower arm of the main L-shaped frame 5. The auxiliary frame assembly includes an auxiliary L-shaped frame 6 and an upper and lower corner motor 10. The rotating shaft of the rotation motor 9 is connected to the outer side of the rear arm of the auxiliary L-shaped frame 6. An upper and lower corner motor 10 is fixed to the inner side of the side arm of the auxiliary L-shaped frame 6. The fixed seat assembly includes a fixed seat 7, a line marker fixing frame 15, and a camera fixing frame 16. The rotating shaft of the upper and lower corner motors 10 is connected to the side wall of the fixed seat 7. A line marker fixing frame 15 and a camera fixing frame 16 are respectively installed on the front side of the fixed seat 7 through bolts. The laser line marker includes a main body part and a front projection end. The main body part is sleeved in the fixed seat 7, and the front projection end is sleeved in the center of the annular frame of the line marker fixing frame 15. The front projection end is an annular sleeve. The pixel board 12 is located at the center position inside the annular sleeve. Four laser reference devices 13 are evenly distributed around the inside of the annular sleeve. Before the controller controls the pixel board 12 to emit a projected cursor pattern, it first locates the vehicle instrument panel through the four laser reference devices 13 (the vehicle height can be detected by combining a millimeter-wave radar or visual recognition can be used for auxiliary positioning). The projection situation of the laser reference devices 13 is detected by the high-definition camera 14. When the projection emitted by the laser reference devices 13 is located at the center of the vehicle instrument panel, the controller then controls the pixel board 12 to project a cursor pattern onto the surface of the instrument panel. By detecting the projection position of the laser reference devices 13 through the high-definition camera 14, the controller respectively controls the left and right corner motors 8, the rotation motor 9, and the upper and lower corner motors 10 to rotate. When controlling the left and right corner motors 8 to rotate, the horizontal swing angle of the pixel board 12 can be changed. When controlling the upper and lower corner motors 10 to rotate, the up and down swing angle of the pixel board 12 can be changed. When controlling the rotation motor 9 to rotate, the rotation angle of the pixel board 12 can be controlled, so as to continuously correct the projection angle of the laser reference devices 13. In addition to analyzing the video information collected by the high-definition camera 14 to judge the vehicle driving state and steering situation, it is also possible to judge the precise movement state of the vehicle by judging the spacing of multiple projection points of the laser reference devices 13, and then control each motor to achieve fine adjustment with the vehicle movement. For example Figure 9As shown in the figure, for the four projection points A, B, C, and D of the laser reference 13, when the distance between AB and CD gradually decreases, it is determined that the vehicle is moving forward, and the upper and lower angle motors 10 are controlled to rotate downward. When the moving speed of AD relative to the vehicle is greater than that of BC, it is determined that the vehicle is moving rightward. When the moving speed of AD relative to the vehicle is less than that of BC, it is determined that the vehicle is moving leftward, and the left and right angle motors 8 are controlled to rotate leftward or rightward. At the same time, according to the degree of the above-mentioned distance expansion or contraction, the steering angle of the vehicle is calculated, and then the rotation motor 9 is synchronously controlled to rotate an approximate angle (rotating between -90 degrees, -0 degrees, and 90 degrees). The high-definition camera 14 includes a main body part and a front lens part. The main body part is sleeved in the fixed seat 7, and the front lens part is sleeved in the center of the annular frame of the marking instrument fixing frame 15.

[0033] Communication module: Adopt vehicle networking communication, such as dedicated short-range communication or cellular vehicle networking; for wide-area communication, adopt 5G private network and fiber optic ring network (for internal data transmission in toll stations). Use the national secret SM4 algorithm for encryption, and the communication link authentication period ≤ 100 ms.

[0034] Hardware configuration of the monitoring center core platform. The server cluster can adopt a 16-node distributed architecture. Each node is configured with a dual-way Intel Xeon Platinum 8380 processor, 512 GB DDR4 memory, and 20 TB NVMe storage. The computing power platform can adopt an NVIDIA DGXA100 cluster, supporting model parallel and data parallel training. The data middle platform of the software system can adopt a real-time data stream processing engine based on Flink, supporting millisecond-level data access and Hudi data lake storage, and can realize unified management of historical data and real-time data. The algorithm engine can adopt an integrated TensorFlow / PyTorch framework, supporting algorithms such as LSTM-GRU hybrid model, spatio-temporal graph convolutional network (ST-GCN), and Transformer time series prediction.

[0035] Traffic flow prediction system. Use a CART decision tree to perform feature engineering on historical data, and construct a multi-dimensional feature vector including time features (day of the week / time period / holiday), spatial features (traffic flow correlation degree of adjacent channels), and environmental features (weather / construction impact coefficient). The prediction model uses an improved LSTM model (with an attention mechanism added), and the prediction period supports 15 minutes / 30 minutes / 1 hour, with the mean absolute error ≤ 8%. Threshold model: Based on the M / M / C queuing theory, construct a channel capacity model, and the formula is as follows:

[0036]

[0037] Where: Ui is the channel processing rate, and Pi is the channel utilization rate. Path planning system, state modeling: Build a road network map of the toll station area with more than 100 nodes. The node attributes include the number of lanes, turning restrictions, and real-time travel time (based on floating car data). Optimization algorithm: Improved ant colony algorithm (ACA), introducing a dynamic pheromone update strategy. The objective function is: f = α·T + β·D + γ·C; (T is the estimated travel time, D is the driving distance, C is the channel congestion coefficient, and α / β / γ are dynamic weight parameters).

[0038] Navigation guidance module: The data interface accesses the prediction data of the monitoring center and the destination data of the in-vehicle system (obtained through the OBU on-vehicle unit) in real time. The scheduling strategy adopts a hierarchical guidance mechanism. The first-level guidance (lane pre-allocation) is pushed 2 kilometers away from the toll station, the second-level guidance (accurate lane number) is pushed 500 meters away, and dynamic calibration is triggered 100 meters away (to cope with sudden lane changes); The output format conforms to the navigation instructions of the NMEA-0183 protocol, including the recommended channel ID (1-8 bit encoding), estimated waiting time, and steering angle.

[0039] Intelligent marking system, multi-object tracking algorithm based on YOLOv8, supporting cross-camera trajectory association); Marking rules: Vehicles that have not received navigation information for more than 30 seconds, deviated from the recommended route by more than 50 meters, or continuously changed lanes ≥ 2 times trigger a yellow mark, and multiple violations (such as 2 or 3 times) trigger a red mark. The data output generates four-dimensional marking data including license plate number (encrypted), violation type, timestamp, and current location.

[0040] Laser guidance control: Coordinate calibration uses the Zhang's calibration method to calibrate the spatial coordinates of the laser marking instrument. Based on the extended Kalman filter to predict the vehicle's driving trajectory, and adjust the laser projection angle in real time (update frequency 20Hz); Angular position control: As Figure 9 shown, by analyzing the moving images of the corresponding vehicle collected by the high-definition camera 14, or by analyzing the change in the distance between the projection points in the picture by the high-definition camera 14, the angle of the corresponding vehicle going straight or turning is obtained, and then the rotation angle of the rotation motor 9 is controlled to be consistent to ensure that the position and direction of the projection image before and after are basically unchanged.

[0041] Display strategy: Priority is given to displaying a red warning border for marked vehicles, and a green guiding arrow for normal vehicles to achieve visual differential guidance. The display content is as Figure 6 shown, in the form of arrow shapes such as forward, left, right, deceleration, stop, and acceleration in a dot combination form.

[0042] In-vehicle navigation fusion system: Adopt a multi-source fusion method to integrate the first / second navigation information of this system, the real-time road conditions of third-party navigation, and the vehicle's own state.

[0043] Temporary Vehicle Database: This database records vehicle characteristics, including multi-dimensional feature vectors of license plate features, appearance features, and behavioral characteristics (lane change frequency / speed fluctuation). It uses the Hungarian algorithm to match vehicles across cameras, supporting simultaneous tracking of over 1,000 vehicles with a missed detection rate of ≤3%. The database uses a Redis cluster to store real-time data and HBase to store historical data (retention period: 90 days).

[0044] Through modular hardware design and innovative software algorithms, this system integrates data collection, intelligent computing, and precise control into a smart toll station dispatching solution. Suitable for highway toll stations with daily traffic volumes of 10,000 to 100,000 vehicles, it effectively improves traffic efficiency and reduces manual intervention costs. By synergizing near- and far-point devices, it achieves three-dimensional data collection combining macro-situational awareness and micro-behavior monitoring. A deep learning model incorporating spatiotemporal features reduces traffic flow prediction error to 7% and shortens route planning time to less than 200ms.

[0045] Example 2: Based on the system described in Example 1, the monitoring center uses a data processing and analysis module to perform intelligent monitoring and dispatching of traffic flow at a toll station, which includes the following steps.

[0046] Step 1: Global data collection and intelligent prediction. Near-point monitoring equipment collects real-time traffic data from toll station entrances and each channel and transmits it to the monitoring center. Based on historical data and external factors, a traffic flow prediction model is constructed through big data analysis and deep learning algorithms to accurately predict the future traffic flow of each channel. When the predicted value exceeds the channel capacity threshold, an intelligent warning is triggered, and a warning signal is sent to the navigation guidance module and the interference processing module simultaneously.

[0047] Real-time data collection from multiple sources: Near-point monitoring equipment collects near-point microscopic data using high-definition cameras, ground sensors, and RFID detectors deployed at toll station entrance plazas and in various lanes. This data, including entrance traffic density, queue lengths in each lane, vehicle transit time, and vehicle type, is transmitted to the monitoring center at a 20Hz frequency via a fiber optic ring network. Far-point monitoring equipment collects far-point macroscopic data using 20-megapixel smart cameras and 128-channel millimeter-wave radar deployed on beam-mounted brackets 1-2 kilometers from the end of the highway. This data collects queue lengths, vehicle type distribution, and driving trajectories of pre-entry vehicles in real time. It also accesses a meteorological API for real-time weather (precipitation, visibility, and temperature) and a traffic management platform for external data such as holidays and construction schedules, generating a far-point data stream updated at 10Hz.

[0048] The spatio-temporal data fusion processing monitoring center fuses data through the following process: sending the traffic flow information of the far point and the near point to the monitoring center simultaneously. The monitoring center establishes a temporary vehicle database and fuses the traffic flow information of the far point with that of the near point. Information closed-loop: The laser guidance result of the far point is transmitted back to the monitoring center in real time, and is associated with the license plate recognition data of the near point (the missed detection rate ≤ 2%) for trajectory association, establishing a two-layer control system of "pre-guidance at the far point - precise intervention at the near point". The fusion method is to achieve time synchronization based on the GPS timing module to calibrate the time stamps of the far / near point data, use the UTM projection coordinate system to unify the coordinates, and calibrate the camera view angle error through the Zhang's calibration method to achieve spatial calibration, constructing an input vector containing multi-dimensional features. The processing method for external factors is as follows: mapping visibility < 500 meters to a congestion impact coefficient of 1.5 and moderate rain to 1.2 through fuzzy logic; establishing one-hot encoding (weekday = 000, weekend = 001, Spring Festival = 010, etc.), obtaining the distance between the construction location and the toll station through GIS data, generating an attenuation factor (impact coefficient 0.8 when the distance < 1 km), analyzing the characteristics of each vehicle and the driving habits of each vehicle driver by comparing the front and back data (comparing indicators such as the driving trajectory deviation and speed change rate of the same vehicle at the far point and the near point), avoiding the problems of image error and data loss in the near point monitoring equipment, and at the same time solving the problem that the prediction effect deteriorates in a short time due to a large amount of data. Comparing the front and back data can unify the data of the front and back vehicles, improve the accuracy of analyzing the characteristics of each vehicle and the driver's behavior. At the same time, according to the results of analyzing the intervention timing of the intervention processing module obtained from the comparison of the front and back data: high intervention level and low intervention level, the analysis results as shown in Figure 8 can be adopted. When the far point and near point data judgments are inconsistent, 1 indicates non-compliance with the driving non-standard requirements, 0 indicates compliance with the driving standard requirements. Strong intervention: intervene when the result is 1 or 10 or 01; Non-strong intervention: intervene when the result is 1 or 10, or when the result is 1 or 01; If intervene: intervene when the result is 0. On the other hand, comparing the front and back data can avoid the problem of system analysis result deviation caused by the lack of necessary traffic data by mutually supplementing the front and back databases to improve the data integrity in the case of vehicle data loss or incomplete vehicle data.

[0049] Construction of Traffic Flow Prediction Model: An LSTM-Transformer hybrid model is adopted, and the modeling process is as follows: Normalize the 3-year historical data (including more than 100,000 congestion events), generate time series samples according to a 15-minute window. Embedding layer: Convert discrete features (such as holidays, construction marks) into multi-dimensional vectors. Encoder: 6-layer TransformerEncoder captures spatio-temporal dependencies (8 attention heads). Decoder: A bidirectional LSTM layer (256 units) processes the dynamic changes of the time series. Output layer: A fully connected layer outputs the predicted traffic flow values for each channel in the next 1 hour (step size 15 minutes). Training optimization: Use the AdamW optimizer, and the loss function is a combination of MAE + MSE. Conduct distributed training on the DGXA100 cluster (batch size 1024), with a training period of 72 hours, and finally MAE ≤ 7.5%.

[0050] Passing Capacity Threshold Comparison Mechanism: First, establish a channel passing capacity model based on the M / M / C queuing model: When p i > 2, C i = μ i ·(1 - p i ) / (p i ·In(1 - p i )); when 1 < p i < 2, Ci = μ i ·C·(1 - p i )” (C is the number of channels); where: The processing rate of manual channels μ i = 60 - 120 vehicles / hour, preferably 80 vehicles / hour, and for ETC channels, μ i = 300 vehicles / hour, preferably 240 vehicles / hour; The channel utilization rate p i = current traffic flow / design traffic flow (the design traffic flow is determined based on lane width and equipment performance). The monitoring center dynamically compares the predicted traffic flow with the threshold: When the predicted value exceeds the threshold by 80%, a yellow warning is triggered; when it exceeds 90%, an orange warning is triggered; when it exceeds 100%, a red warning is triggered. The warning information includes the channel number, the estimated congestion duration, and the recommended number of standby channels to be opened, etc.

[0051] Step 2: Dynamic Route Planning and Hierarchical Scheduling. The monitoring center integrates the real-time position, driving direction, speed, channel passing status, and prediction data of vehicles, uses a route planning algorithm to generate the first navigation information before the vehicle enters the station, and synchronizes it to the navigation guidance module and the interference processing module in real time to achieve the unified scheduling of the incoming traffic flow.

[0052] Multi-dimensional Data Fusion Modeling: The monitoring center integrates the following data to build a dynamic road network model: The real-time vehicle status obtains vehicle GPS location, driving direction, vehicle speed, and destination information through C-V2X, and the current queue length and average passing time obtained by near-point monitoring. Prediction information: The traffic index of each channel in the next 30 minutes (based on predicted traffic flow / capacity, 0-10 points, ≥8 points indicates congestion). A road network of the toll station area containing 120 nodes is constructed through a graph database, and the node attributes include the number of lanes, turning restrictions, and real-time passing time (based on the moving average of floating car data).

[0053] Implementation of Intelligent Route Planning Algorithm: The improved ant colony algorithm is used to calculate the first navigation information. The specific steps are as follows:

[0054] First, establish the objective function: f = α·T p +β·D+γ·C f ; where: T p is the predicted passing time (including queuing time, weight α = 0.6, automatically increased to 0.8 during peak hours); D is the driving distance (after standardization processing, weight β = 0.2); C f is the channel congestion coefficient (predicted traffic flow / capacity, weight γ = 0.2).

[0055] Then, the vehicle selects a lane according to the roulette wheel strategy. The transfer probability formula is:

[0056]

[0057] (τ is the pheromone concentration, η is the heuristic factor, dynamically adjusted based on the real-time congestion index).

[0058] Then, perform the planning output: Generate the first navigation information including the recommended channel ID (such as "ETC-3"), the estimated arrival time, and the lane change prompt (turn left 200 meters before the next exit). Dynamic traffic diversion: When the queue length of a certain channel exceeds 50 meters, the far-point system automatically increases the guiding weight in the corresponding direction of this channel (such as extending the display duration of the guiding arrow of the ETC-3 channel from 2 seconds to 4 seconds) to balance the traffic flow distribution of each channel; through the far-point pre-guiding system, realize the vehicle's early lane change decision before entering the toll station square, and cooperate with the near-point laser forced intervention to control the illegal lane change rate below 5%, and curb the traffic flow chaos problem in the entrance area from the source.

[0059] Step 3: Multi-modal Guidance Execution and Intervention. After receiving the warning signal, the navigation guidance module transmits the data of the first navigation information to each vehicle navigation system through the communication module to guide each vehicle to drive towards the designated toll channel according to the navigation information; at the same time, the navigation guidance module sends the vehicle information of the vehicles that have not received the first navigation information to the monitoring center.

[0060] Collaborative guidance and data sharing via the connected vehicle system: First, navigation information is transmitted by the navigation guidance module to the vehicle's head-mounted system via a dedicated 5G network or DSRC protocol. Voice notifications are supported, such as "Enter ETC Lane 2 500 meters ahead. The estimated waiting time is 1 minute." The AR-HUD displays lane guidance arrows superimposed with real-time congestion levels. Dashboard icons dynamically update the lane type (ETC / manual) and estimated transit time. Second, after generating and sharing destination data with the vehicle's head-mounted system, the navigation system uses the Dijkstra algorithm to generate exit routes, combining real-time traffic conditions (integrating speed data from the surrounding 3 kilometers of roads using the AutoNavi API) to avoid congested areas. For example, for vehicles heading downtown, the following message appears: "We recommend taking the auxiliary road after exiting the station. The main road is currently congested, with a 15-minute delay." For trucks, the following message appears: "We recommend detouring via the eastern logistics corridor to avoid areas with high concentrations of small vehicles."

[0061] Pre-guidance at near and far points and entrance order control: Conventional guidance projects a green dynamic arrow (such as "→ETC-2 channel") to vehicles receiving the first navigation information, and the arrow brightness is dynamically adjusted with vehicle speed (the flashing frequency is increased when the vehicle speed is >80km / h); early warning intervention triggers a red warning pattern for vehicles that do not receive the navigation signal (no response for more than 45 seconds), and simultaneously sends a strong reminder signal through the DSRC protocol; emergency control is to activate the "entry flow control mode" when the predicted congestion level at the entrance reaches red, project a red cross signal of "pause entry", and at the same time release the message "congestion at the toll station ahead, it is recommended to detour at the next intersection" through the variable information board.

[0062] Laser navigation precision intervention mechanism: Fusion of far-point millimeter-wave radar and YOLOv8 target detection data, through the extended Kalman filter to predict the vehicle's position in the next 3 seconds, including the angle α between the front of the vehicle and the lane (such as Figure 9 As shown), the laser projection angle of the laser marking instrument (corresponding to the center of the vehicle dashboard) and the rotation angle of the pixel board are dynamically adjusted (the rotation direction and rotation angle are basically consistent with the angle α of the front lane). Each set of laser devices covers 3 lanes and uses time-division multiplexing technology with a frequency of 20Hz to project different lane guidance information in different time periods to avoid interference from adjacent lane signals. The integrated light intensity sensor automatically adjusts the laser power to ensure projection clarity in rainy and foggy weather. The 635nm laser device (projection distance 50-100 meters) at the end of the highway for far-point advance guidance is displayed as follows: Figure 6As shown, in the form of forward, left, right, deceleration, stop, and acceleration arrows in a dot combination. The front-end 532nm laser or 635nm red laser device (anti-strong light interference) of the near-point forced guidance channel dynamically adjusts the display according to the marked data of the monitoring center: green arrows for normal vehicles, red flashing arrows for vehicles that have not received navigation, and warning information surrounded by a red border for vehicles that violate lane-changing regulations. The projection angle real-time tracks the vehicle trajectory through extended Kalman filtering (update frequency 20Hz), and the calibration error ≤ 5cm to ensure accurate projection of information to the designated area of the instrument panel.

[0063] Step 4: Outbound path dynamic planning and personalized guidance. The navigation guidance module simultaneously shares scheduling data with the in-vehicle navigation system. After data sharing, the navigation guidance module combines the vehicle's current location, destination, and the real-time traffic conditions of each toll channel, and uses path planning algorithms to calculate the best second navigation information for each vehicle after leaving the station. Then, it transmits the second guidance information to the in-vehicle navigation system through the communication module for display, guiding the driver to drive according to the recommended route. Specifically, after completing the inbound traffic flow scheduling, the navigation guidance module, based on the real-time shared in-vehicle data, including the vehicle's current location, destination POI, load type, ETC signing status, etc., combines the real-time operation data of the road network within 3 kilometers around the toll station, and obtains road speed, accident points, and temporary control information through the Gaode / Baidu Map API to construct an outbound dynamic path planning model.

[0064] Step 5: Closed-loop control and data optimization. The monitoring center marks the vehicles that have not received the first navigation information or have not complied with the first navigation information based on the near-point real-time monitoring data and the feedback data of the navigation guidance module, and sends the marked vehicle information to the interference processing module. The interference processing module transmits the data of the first navigation information to the laser navigation system through the communication module. The laser navigation system controls each near-point laser marking instrument to project each first navigation information onto the instrument panel of the corresponding vehicle at the near point in the form of laser projection. The near-point laser marking instrument is assembled on the front-end bracket of each toll channel.

[0065] Vehicle marking and trajectory association: Vehicles that have not received navigation information, that is, vehicles that have not responded to the vehicle networking signal for more than 30 seconds, trigger a yellow mark. Vehicles that have not complied with navigation, that is, vehicles that deviate from the recommended route by more than 50 meters or change lanes twice in a row without following the guidance, trigger a red mark. Trajectory association is realized based on the YOLOv8 + Hungarian algorithm for cross-camera tracking, extracting multi-dimensional features (license plate character encoding, vehicle type length-width ratio, and lane-changing frequency), and the missed detection rate ≤ 3% to achieve continuous identification of vehicle identities from far points to near points.

[0066] Far and near point data fusion optimization: Establish a temporary vehicle database. Through license plate recognition and feature matching, associate the vehicle IDs detected at the far point with the near point data, calculate driver behavior indicators, and use them to optimize the input features of the prediction model. For large vehicles missed by the near point detection, supplement the detection with far point radar data.

[0067] System self-optimization mechanism. The monitoring center collects the following KPIs every hour for model iteration: Guidance success rate = Number of vehicles successfully following the navigation / Total number of guided vehicles (target ≥ 92%), Queue length reduction rate = (Historical queue length - Current length) / Historical queue length (target ≥ 35%), Prediction error rate = |Actual traffic - Predicted traffic| / Actual traffic (target ≤ 8%). Dynamically adjust the path planning weights (α / β / γ) and prediction model parameters through reinforcement learning to form a closed loop of "data collection - decision execution - feedback optimization", and continuously improve the system adaptability. The panoramic view of the method flow is as Figure 7 shown.

[0068] The above method in this embodiment strictly follows the technical route of "data collection - intelligent decision - precise execution - closed-loop optimization". Through deep fusion of multi-source data, precise modeling with intelligent algorithms, and collaborative multi-modal guidance, it realizes the full-process intelligent scheduling of the toll station traffic flow, and significantly improves the traffic efficiency and control accuracy.

[0069] The above specific embodiments of the present invention are only used for exemplary illustration or explanation of the principles of the present invention, and do not constitute a limitation to the present invention. Therefore, any modifications, equivalent replacements, improvements, etc. made without departing from the spirit and scope of the present invention shall be included within the protection scope of the present invention.

Claims

1. An intelligent monitoring system for vehicle flow at a toll station, characterized in that, It includes the following modules: Real-time monitoring module: It collects vehicle position, speed, vehicle type, vehicle distance and traffic flow information of each toll lane in real time, and transmits it to the monitoring center through a wired or wireless communication network; Data processing and analysis module: The monitoring center first performs preprocessing operations on the traffic flow data obtained from real-time monitoring through the data processing and analysis module to remove noise and error data to ensure the accuracy of data analysis. Then, it uses big data analysis technology combined with historical data to establish a traffic flow prediction model, trains and optimizes the model using deep learning algorithms, predicts the future traffic flow of each toll lane, and compares the traffic flow prediction results with the traffic capacity threshold of each toll lane. If the predicted traffic flow exceeds the traffic capacity, an intelligent warning signal is triggered, and the warning signal is transmitted to the navigation guidance module and the intervention processing module; Navigation guidance module: After receiving the prediction results and warning signals transmitted by the data processing and analysis module, it combines the vehicle's real-time position, speed, direction and surrounding vehicle information, and uses a path planning algorithm to calculate the best guidance route for each vehicle, and transmits the guidance route information to the in-vehicle navigation system through the communication module. The in-vehicle navigation system displays a guidance arrow and text prompt on the screen to guide the driver to drive according to the recommended route; Intervention processing module: It monitors whether the vehicle is driving according to the navigation guidance through the real-time monitoring module, and marks the vehicles that do not receive or comply with the guidance; for the marked vehicles, it projects a guidance arrow and / or toll lane number to its operation console position using a laser projection guidance device, and the laser line marker adjusts the projection angle in real time according to the vehicle's driving speed, distance and direction to ensure that the guidance information is accurately projected without disturbing the driver's line of sight; when obvious traffic violations are detected in the vehicle, the laser line marker projects a yellow or red warning sign, and at the same time feedbacks the violation information of these vehicles to the traffic management department.

2. The intelligent toll station traffic monitoring system according to claim 1, characterized in that It also includes a far-end monitoring device and a far-end laser line marker, which are respectively installed on the crossbeam brackets at the end of the highway to monitor the traffic flow information before entering the toll station entrance in real time, and send the traffic flow information at the far end to the monitoring center. The interference processing module transmits the data of the first navigation information to the far-end laser navigation system through the communication module. The laser navigation system controls each far-end laser line marker to project each first navigation information onto the instrument panel of the corresponding vehicle at the far end in the form of laser projection to guide the vehicles at the far end in advance.

3. The intelligent traffic flow monitoring system for toll stations according to claim 1, wherein The intervention processing module includes a buffer fixing mechanism, a main frame assembly, an auxiliary frame assembly, a fixing seat assembly, a laser alignment instrument, a high-definition camera, and a controller. The main frame assembly is fixed to the buffer fixing mechanism. The main frame assembly includes a main L-shaped frame (5) and a rotating motor (9). The rotating shaft of the left and right corner motor (8) is connected to the outer side of the upper arm of the main L-shaped frame (5). A rotating motor (9) is fixed to the inner side of the lower arm of the main L-shaped frame (5). The auxiliary frame assembly includes an auxiliary L-shaped frame (6) and an upper and lower corner motor (10). The rotating shaft of the rotating motor (9) is connected to the outer side of the rear arm of the auxiliary L-shaped frame (6). An upper and lower corner motor (10) is fixed to the inner side of the side arm of the auxiliary L-shaped frame (6). The fixing seat assembly includes a fixing seat (7). The rotating shaft of the upper and lower corner motor (10) is connected to the side wall of the fixing seat (7). The laser alignment instrument includes a main body part and a front projection end. The main body part is sleeved in the fixing seat (7). The pixel board (12) is located at the center position inside the annular sleeve. Four laser reference devices (13) are evenly distributed around the inside of the annular sleeve. Before the controller controls the pixel board (12) to emit a projection cursor pattern, it first locates the vehicle instrument table through the four laser reference devices (13). The high-definition camera (14) includes a main body part and a front lens part. The main body part is sleeved in the fixing seat (7). The front lens part is sleeved in the center of the annular frame of the alignment instrument fixing frame (15).

4. The intelligent traffic flow monitoring system for toll stations according to claim 3, characterized in that, The buffer fixing mechanism includes a base (1), a base plate (2), buffer blocks (3), and a fixing frame (4). The base (1) is fixed to the road surface support through the fixing frame (4). A plurality of buffer blocks (3) are evenly installed between the base plate (2) and the base (1). A relief hole is provided at the center of the base (1). The left and right corner motor (8) is located in the relief hole and the rear end of the left and right corner motor (8) is fixed to the base plate (2).

5. The intelligent toll station traffic flow monitoring system according to claim 1, characterized in that The real-time monitoring module includes a high-definition camera, a millimeter-wave radar, and a ground loop detector.

6. An intelligent monitoring and scheduling method for the vehicle flow at a toll station, characterized in that, It includes the following steps: Step 1, data collection and intelligent prediction: The near-point monitoring device collects the traffic flow data at the toll station entrance and each channel in real time and transmits it to the monitoring center. Based on historical data and external factors, a traffic flow prediction model is constructed through big data analysis and deep learning algorithms to accurately predict the future traffic flow of each channel. When the predicted value exceeds the channel passing capacity threshold, an intelligent warning is triggered, and a warning signal is sent to the navigation guidance module and the interference processing module simultaneously. Step 2, dynamic path planning and scheduling: The monitoring center integrates the real-time position, driving direction, speed, channel passing status, and prediction data of the vehicle, and uses path planning algorithms to generate the first navigation information before the vehicle enters the station, and synchronizes it to the navigation guidance module and the interference processing module in real time to achieve unified scheduling of the incoming traffic flow. Step 3, in-vehicle navigation guidance: After receiving the warning signal, the navigation guidance module transmits the data of the first navigation information to each in-vehicle navigation system through the communication module, guiding each vehicle to drive to the designated toll channel according to the navigation information; at the same time, the navigation guidance module sends the vehicle information that has not received the first navigation information to the monitoring center. Step 4, Laser Projection Interference Control: Based on the near-point real-time monitoring data and the feedback data of the navigation guidance module, the monitoring center marks the vehicles that have not received the first navigation information or have not complied with the first navigation information after receiving it, and sends the marked vehicle information to the interference processing module. The interference processing module transmits the data of the first navigation information to the laser navigation system through the communication module. The laser navigation system controls each near-point laser marking instrument to project each first navigation information onto the instrument panel of the corresponding vehicle at the near point in the form of laser projection. The near-point laser marking instrument is assembled on the front bracket of each toll lane.

7. The intelligent toll station traffic flow monitoring system according to claim 1, wherein In Step 3, the navigation guidance module shares scheduling data with the in-vehicle navigation system at the same time. After the data sharing, the navigation guidance module combines the current position, destination of the vehicle, and the real-time traffic conditions of each toll lane, and uses the path planning algorithm to calculate the best second navigation information for each vehicle after leaving the station. Then, it transmits the second guidance information to the in-vehicle navigation system through the communication module for display, guiding the driver to drive according to the recommended route.

8. The intelligent toll station traffic flow monitoring system according to claim 1, characterized in that, It further includes a far-point monitoring device, which is assembled on the crossbeam bracket at the end of the highway and far from the toll station entrance. The far-point monitoring device is used to monitor the traffic flow information before entering the toll station entrance in real time, and sends the traffic flow information at the far point to the monitoring center. The monitoring center establishes a temporary vehicle database, including fusing the traffic flow information at the far point with the traffic flow information at the near point, and analyzing the characteristics of each vehicle and the driving habits of each vehicle driver through the comparison of the front and rear data.

9. The intelligent traffic flow monitoring system for toll stations according to claim 1, wherein, It further includes far-point laser marking instruments, which are respectively assembled on the crossbeam brackets at the end of the highway and far from the toll station entrance. The interference processing module transmits the data of the first navigation information to the far-point laser navigation system through the communication module. The laser navigation system controls each far-point laser marking instrument to project each first navigation information onto the instrument panel of the corresponding vehicle at the far point in the form of laser projection, so as to guide the far-point vehicles to decelerate, change direction and pause in advance.

Images