Vehicle ramp intelligent control method and system based on multi-modal data fusion

Abstract

The invention provides a vehicle ramp intelligent control method and system based on multi-modal data fusion, and relates to the technical field of intelligent traffic control. Through the steps of multi-modal data acquisition, state portrait generation, dynamic adjustment, multi-stage judgment and the like, the problems of insufficient data perception and control lag in the prior art are solved, and the method has the advantages of improving ramp control precision and adaptability, reducing control lag and improving road network operation efficiency.

Description

Technical Field

[0001] This application relates to the field of intelligent traffic control technology, and in particular to a vehicle ramp intelligent control method and system based on multimodal data fusion. Background Technology

[0002] With the continuous expansion of urban transportation networks, highway ramps, as core hubs connecting main roads and auxiliary roads, have a decisive impact on the overall traffic capacity of the regional road network. Currently, mainstream ramp control systems generally employ single data acquisition methods, such as relying solely on geomagnetic induction coils or fixed-point cameras for monitoring. This type of technical architecture has significant limitations. At the data perception level, traditional detection equipment struggles to simultaneously acquire multi-dimensional information such as vehicle trajectory dynamics, driver behavior characteristics, and environmental interactions, leading to discrepancies between the generated traffic state descriptions and the actual scene. At the control logic level, existing systems mostly operate based on preset thresholds or historical statistical models. When encountering non-steady-state traffic flows such as holiday traffic surges, severe weather conditions, or sudden accidents, they cannot dynamically adjust strategy parameters, causing control commands to lag behind actual road condition changes. At the system coordination level, although vehicle-to-everything (V2X) technology has achieved basic vehicle-to-vehicle communication, it has failed to effectively integrate multimodal inputs from meteorological monitoring data, high-precision positioning information, and roadside sensing units. Data from different sources conflict due to differences in spatiotemporal references; for example, the mismatch in time synchronization between instantaneous traffic flow captured by cameras and radar ranging data reduces the reliability of the fusion results. The aforementioned defects directly lead to practical problems such as ramp queuing overflow and contradictions between control strategies and local environments. In particular, under the differentiated demand scenarios of peak and off-peak hours, the fixed rule system is difficult to achieve accurate adaptation, which seriously restricts the efficiency of road network operation.

[0003] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0004] This application provides a vehicle ramp intelligent control method and system based on multimodal data fusion. It has the advantages of improving ramp control accuracy and adaptability, reducing control lag, and enhancing road network operating efficiency.

[0005] Firstly, the intelligent vehicle ramp control method based on multimodal data fusion provided in this application adopts the following technical solution: A vehicle ramp intelligent control method based on multimodal data fusion includes: Real-time traffic data is acquired through a multimodal perception module integrated into the vehicle. The multimodal perception module includes a camera unit for video recognition, a radar unit for ranging, a GPS unit for positioning, and a lidar unit for traffic flow monitoring. A ramp traffic status profile is generated by combining the real-time traffic data with a data fusion algorithm. The ramp traffic status profile includes: real-time traffic flow label, average vehicle speed label, vehicle density label, and accident risk label. The specific information of the ramp traffic status profile is dynamically adjusted based on the local traffic feature database, which includes geographic information system data, historical traffic flow patterns, time period characteristics, and environmental factor labels. The adjusted ramp traffic status profile is matched and analyzed with traffic rules in the preset database to generate control decision results; Add multi-level judgment steps to the matching analysis process, including confidence assessment, risk priority ranking, and feedback verification; Based on the risk level and control strategy in the control decision results, corresponding ramp adjustment instructions are generated and executed. The ramp adjustment instructions include adjusting the traffic light cycle, speed limit sign display, and sending cooperative passage prompts to vehicles.

[0006] Optionally, the multimodal perception module has a built-in data verification subunit; the step of acquiring real-time traffic data through the vehicle-integrated multimodal perception module includes: The image clarity and frame rate stability of the camera unit are verified in parallel by the data verification subunit, and the ranging accuracy of the radar unit and the point cloud density of the lidar unit are verified simultaneously. When the image clarity is detected to be consistently below the first preset threshold and the radar ranging error is consistently above the second preset threshold, the redundant data compensation mode is activated. In this mode, the data from the lidar unit is used as the main value, and the high-frequency output of the camera unit is turned off to reduce power consumption. When the GPS signal strength is detected to be continuously lower than the third preset threshold and radar data is available, switch to radar master control mode. In this mode, the distance data calculated by the radar unit is used as the main value of the location information. During the data fusion process, a Kalman filter is used to perform spatiotemporal alignment and noise filtering on multi-source data, generating a fused data stream with timestamps and confidence scores, which serves as the input for the subsequent generation of ramp traffic status profiles.

[0007] Optionally, the step of dynamically adjusting the specific information of the ramp traffic state profile based on the local traffic feature database includes: Extract spatiotemporal features from real-time traffic data, including current timestamp, geographic location coordinates, and weather conditions, and perform similarity matching with historical patterns in the local feature database; Local label weights are calculated using a machine learning model. These local labels include peak hour markers, accident-prone area markers, and seasonal traffic flow change factors. The priority of the profile labels is then adjusted based on these weights. Based on the deviation between real-time data and local characteristics, the confidence thresholds of traffic flow labels and accident risk labels are dynamically adjusted to make the profile output adapt to localized scenarios. Generate an adjusted traffic status profile of the ramps and add a localization adaptation coefficient as a decision-making reference.

[0008] Optionally, the process of adding multi-level judgment steps to achieve reasonableness includes: Level 1 judgment: Confidence assessment of the traffic status profile of the ramp. When the confidence of the profile label is lower than the preset safety threshold, a data re-collection or manual review mechanism is triggered. Second-level judgment: Based on the risk priority ranking algorithm, calculate the expected impact value of different control strategies, and prioritize the strategy with the highest risk balance and efficiency. The third level of judgment: introduce a feedback verification loop to compare historical decision results with actual traffic flow changes, and optimize the matching rules through reinforcement learning models to reduce the misjudgment rate; The final control decision result must be verified by all judgment steps; otherwise, it will fall back to the default safe mode.

[0009] Optionally, the step of generating control decision results further includes: The state machine based on threshold determination is used to classify the traffic status of the ramps and output status labels, which include: smooth traffic, slow traffic, congestion, and accident labels. By using a spatiotemporal pattern mining algorithm, the deviation is calculated by comparing real-time traffic flow with historical peak patterns. The deviation is corrected by combining event data from the local feature database with weighted data.

[0010] Optionally, the traffic rule base in the preset database supports dynamic updates, the method of which includes: Record profile data, control strategies and effect feedback of all traffic events, use machine learning algorithms to analyze historical data, and automatically adjust the threshold and / or confidence level for traffic flow determination; New potential risk patterns are identified to generate new rule recommendations, which are then incorporated into the rule base after confirmation by the traffic management center.

[0011] Optionally, generating and executing corresponding ramp adjustment instructions based on the risk level and control strategy in the control decision results includes: For low-risk slow-moving conditions, a yellow warning signal is sent to the ramp traffic lights, and the green light cycle is extended. For medium-risk congestion conditions, the variable speed limit sign will be triggered to display a speed reduction prompt and a coordinated deceleration message will be broadcast to surrounding vehicles. In the event of a high-risk accident, the highest level alarm is sent to the traffic management platform, an emergency lane opening instruction is automatically generated, and the roadside broadcast system is triggered to broadcast traffic guidance information.

[0012] Secondly, this application provides a vehicle ramp intelligent control system based on multimodal data fusion, comprising: The data acquisition module is used to acquire real-time traffic data through the multimodal perception module integrated in the vehicle. The multimodal perception module includes a camera unit for video recognition, a radar unit for ranging, a GPS unit for positioning, and a lidar unit for traffic flow monitoring. The profile generation module is used to generate a ramp traffic status profile by combining the real-time traffic data with a data fusion algorithm. The ramp traffic status profile includes: real-time traffic flow label, average vehicle speed label, vehicle density label, and accident risk label. The data adjustment module is used to dynamically adjust the specific information of the ramp traffic status profile based on the local traffic feature database, which includes geographic information system data, historical traffic flow patterns, time period features, and environmental factor labels. The decision generation module is used to match and analyze the adjusted ramp traffic status profile with traffic rules in the preset database to generate control decision results; The judgment module is used to add multi-level judgment steps during the matching analysis process, including confidence assessment, risk priority ranking, and feedback verification. The instruction generation module is used to generate and execute corresponding ramp adjustment instructions based on the risk level and control strategy in the control decision results. The ramp adjustment instructions include adjusting the traffic light cycle, speed limit sign display, and sending cooperative passage prompts to vehicles.

[0013] Thirdly, this application provides a computer device, the device comprising: a memory and a processor, wherein the processor, when executing computer instructions stored in the memory, performs the method described above.

[0014] Fourthly, this application provides a computer-readable storage medium including instructions that, when executed on a computer, cause the computer to perform the method described above.

[0015] In summary, this application solves the problems of insufficient data perception and control lag in the prior art by acquiring multimodal data, generating state profiles, dynamically adjusting, and making multi-level judgments. It has the advantages of improving the accuracy and adaptability of ramp control, reducing control lag, and improving the efficiency of road network operation. Attached Figure Description

[0016] Figure 1This is a schematic diagram of the computer device structure of the hardware operating environment involved in the embodiments of this application; Figure 2 This is a flowchart illustrating the first embodiment of the intelligent vehicle ramp control method based on multimodal data fusion in this application; Figure 3 This is a structural block diagram of the first embodiment of the intelligent vehicle ramp control system based on multimodal data fusion in this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0018] Reference Figure 1 , Figure 1 This is a schematic diagram of the computer device structure of the hardware operating environment involved in the embodiments of this application.

[0019] like Figure 1 As shown, the computer device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wireless-Fidelity (Wi-Fi) interface). The memory 1005 may be high-speed random access memory (RAM) or stable non-volatile memory (NVM), such as a disk drive. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.

[0020] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on the computer device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0021] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a network communication module, a user interface module, and a vehicle ramp intelligent control program based on multimodal data fusion.

[0022] exist Figure 1 In the computer device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in this application can be set in the computer device. The computer device calls the vehicle ramp intelligent control program based on multimodal data fusion stored in the memory 1005 through the processor 1001, and executes the vehicle ramp intelligent control method based on multimodal data fusion provided in the embodiment of this application.

[0023] This application provides a vehicle ramp intelligent control method based on multimodal data fusion, referring to... Figure 2 , Figure 2 This is a flowchart illustrating the first embodiment of the intelligent vehicle ramp control method based on multimodal data fusion according to this application.

[0024] Traditional ramp control methods primarily rely on single data sources or fixed rules, which suffer from limitations such as insufficient data coverage, poor dynamic adaptability, and lack of coordinated control when dealing with continuously increasing urban traffic flow. These limitations lead to control lag, rigid strategies, and conflicts between multiple data sources, making it difficult to respond promptly to sudden congestion, adapt to dynamic traffic demands, and resulting in low reliability of fusion results. Consequently, these factors affect ramp efficiency and the overall performance of the regional road network.

[0025] In this embodiment, the intelligent vehicle ramp control method based on multimodal data fusion includes the following steps: Step S10: Obtain real-time traffic data through the multimodal perception module integrated in the vehicle. The multimodal perception module includes a camera unit for video recognition, a radar unit for ranging, a GPS unit for positioning, and a lidar unit for traffic flow monitoring.

[0026] Step S20: Generate a ramp traffic status profile by combining the real-time traffic data with a data fusion algorithm. The ramp traffic status profile includes: real-time traffic flow label, average vehicle speed label, vehicle density label, and accident risk label.

[0027] Step S30: Dynamically adjust the specific information of the ramp traffic status profile based on the local traffic feature database, wherein the local traffic feature database includes geographic information system data, historical traffic flow patterns, time period characteristics and environmental factor labels.

[0028] Step S40: Match the adjusted ramp traffic status profile with the traffic rules in the preset database to generate control decision results.

[0029] Step S50: Add multi-level judgment steps to the matching analysis process, including confidence assessment, risk priority ranking, and feedback verification.

[0030] Step S60: Generate and execute corresponding ramp adjustment instructions based on the risk level and control strategy in the control decision results. The ramp adjustment instructions include adjusting the traffic light cycle, speed limit sign display, and sending cooperative passage prompts to vehicles.

[0031] For ease of understanding, the following explains some key terms in this embodiment: A multimodal perception module refers to a hardware collection that integrates multiple different types of sensors to acquire multi-dimensional information. In this embodiment, the module is configured on a vehicle to perceive the vehicle's surrounding environment and traffic conditions. It may include a camera unit, a radar unit, a GPS unit, and a lidar unit to respectively realize functions such as video recognition, ranging, positioning, and traffic flow monitoring.

[0032] Real-time traffic data refers to the raw information stream reflecting traffic conditions collected at the current moment through multimodal sensing modules. This data may include, but is not limited to, vehicle images, distance information, location coordinates, and point cloud data of the surrounding environment. Its characteristics are immediacy and dynamism, and it can reflect the instantaneous changes in traffic flow.

[0033] Data fusion algorithms refer to methods for integrating, processing, and analyzing information from different sensors or data sources. These algorithms aim to eliminate data redundancy, compensate for missing data, and improve data accuracy and reliability, thereby extracting comprehensive and accurate traffic status information from multi-source heterogeneous data.

[0034] A ramp traffic status profile refers to a description of the current traffic conditions in a ramp area, generated by processing real-time traffic data through data fusion algorithms. This profile is presented in the form of tags, such as real-time traffic flow tags, average speed tags, vehicle density tags, and accident risk tags, to quantify and visualize the ramp's operational status.

[0035] A local traffic feature database is a database that stores historical and static traffic-related information for a specific geographic area. This database may contain geographic information system data, historical traffic flow patterns, time-period characteristics, and environmental factor labels, providing localized contextual knowledge for the analysis and adjustment of real-time traffic conditions.

[0036] Traffic rules refer to a set of pre-set or dynamically updated criteria used to regulate traffic behavior and control traffic flow. These rules may include speed limits, traffic light timing logic, lane management regulations, and emergency response plans, and serve as the basis for generating control decisions.

[0037] Control decision results refer to the instructions or suggestions output by the system regarding how to manage and regulate ramp traffic after analyzing the ramp traffic state profile and matching it with traffic rules. These results typically include specific control strategies and corresponding risk level assessments, guiding the generation of subsequent ramp regulation instructions.

[0038] Multi-level decision-making steps refer to multiple verification stages introduced during the generation of control decision outcomes to ensure the rationality and security of the decisions. These stages may include confidence assessment of data or profiles, risk prioritization of different control strategies, and feedback verification of historical decision effects to improve the robustness of the decisions.

[0039] Ramp adjustment instructions are operational commands sent to ramp infrastructure or vehicles based on control decisions. These instructions aim to change the operational status of ramps, such as adjusting traffic light cycles, displaying speed limit signs, or sending coordinated traffic prompts to vehicles, in order to optimize traffic flow, alleviate congestion, and respond to emergencies.

[0040] This embodiment provides a vehicle ramp intelligent control method based on multimodal data fusion.

[0041] First, real-time traffic data is acquired through a multimodal perception module integrated into the vehicle. This multimodal perception module can consist of multiple independent sensors. For example, it can be configured with a camera unit to capture video images, a radar unit to measure vehicle distance and speed, a GPS unit to provide vehicle location information, and a lidar unit to generate point cloud data of the surrounding environment to monitor traffic flow. These sensors independently collect their respective types of data and transmit the raw data directly to the processing unit.

[0042] Secondly, a traffic status profile of the ramp is generated by combining the real-time traffic data using a data fusion algorithm. In one implementation, weighted averaging or logical judgment rules can be used to integrate data from different sensors. For example, the number of vehicles identified by cameras, the average speed measured by radar, the vehicle location distribution provided by GPS, and the traffic density monitored by lidar can be aggregated. Then, based on preset thresholds or rules, real-time traffic flow labels, average vehicle speed labels, vehicle density labels, and accident risk labels are calculated and generated.

[0043] Furthermore, the specific information of the ramp traffic status profile is dynamically adjusted based on a local traffic feature database. This database can be a collection of geographic information system data for a specific area, historical traffic flow patterns, time-period characteristics, and environmental factor labels. During the adjustment process, the real-time generated ramp traffic status profile can be compared with the corresponding historical data in this feature database, and the profile labels can be adjusted according to preset correction factors to ensure it conforms to the actual local conditions.

[0044] Subsequently, the adjusted ramp traffic status profile is matched and analyzed against traffic rules in a pre-set database to generate a control decision. This pre-set database can store a series of traffic rules, such as "When traffic volume exceeds X and average speed is below Y, speed limits should be implemented." The system receives the adjusted ramp traffic status profile and compares it with the traffic rules in the database. When a label in the profile meets the triggering condition of a certain traffic rule, the corresponding control strategy is selected as the control decision.

[0045] In this matching analysis process, multiple judgment steps are added to ensure the rationality of control decisions. For example, a confidence assessment step can be set up to check the validity of the generated profile labels, such as checking whether the data is complete or contains anomalies. Next, risk prioritization can be performed, for example, prioritizing strategies involving accident risk over strategies that only involve congestion. Finally, feedback verification can be performed, such as manual inspection or comparison based on historical data, to confirm the rationality of the selected strategies.

[0046] Finally, based on the risk level and control strategy in the control decision result, the system generates and executes corresponding ramp adjustment instructions. For example, if the control decision result indicates "congestion, speed limit recommended," the system will generate an instruction requiring the speed limit signs on the ramp to display a specific speed value. If the decision result indicates "slow traffic, traffic light adjustment recommended," the system will generate an instruction requiring the traffic light controller to extend the green light cycle. In addition, instructions to send cooperative traffic prompts to surrounding vehicles can be generated, such as broadcasting deceleration or lane-changing suggestions via roadside units.

[0047] This embodiment acquires multi-dimensional real-time traffic data by integrating a multimodal perception module, solving the problem of insufficient data coverage from traditional single data sources. Through data fusion algorithms and dynamic adjustments based on a local traffic feature database, it can generate a comprehensive and realistic profile of ramp traffic conditions, overcoming the limitations of poor dynamic adaptability and rigid strategies in existing methods. Furthermore, the introduction of multi-level judgment steps ensures the rationality of control decisions, avoiding control lags caused by data conflicts or misjudgments, thereby improving ramp traffic efficiency and the overall operational performance of the regional road network.

[0048] It should be noted that real-time traffic data is acquired through a multimodal perception module integrated into the vehicle. This multimodal perception module includes a camera unit for video recognition, a radar unit for ranging, a GPS unit for positioning, and a lidar unit for traffic flow monitoring. However, in practical applications, the individual sensor units of the multimodal perception module may be affected by environmental factors, their own malfunctions, or signal interference, leading to unstable data quality or partial data loss. If unverified or unprocessed raw data is directly used for fusion, noise and errors may be introduced, thereby affecting the accuracy of the ramp traffic condition profile and the rationality of subsequent control decisions.

[0049] To address this, this embodiment further proposes a built-in data verification subunit for the multimodal perception module. The steps of acquiring real-time traffic data through the vehicle-integrated multimodal perception module include: parallel verification of the image clarity and frame rate stability of the camera unit via the data verification subunit, and simultaneous verification of the ranging accuracy of the radar unit and the point cloud density of the lidar unit; when the image clarity is continuously lower than a first preset threshold and the radar ranging error is continuously higher than a second preset threshold, a redundant data compensation mode is activated. In this mode, the data from the lidar unit is used as the primary value, and the high-frequency output of the camera unit is turned off to reduce power consumption; when the GPS signal strength is continuously lower than a third preset threshold and radar data is available, the system switches to radar master control mode. In this mode, the distance data calculated by the radar unit is used as the primary value of the location information; during the data fusion process, a Kalman filter is used to perform spatiotemporal alignment and noise filtering on the multi-source data, generating a fused data stream with timestamps and confidence scores, which serves as input for subsequently generating a ramp traffic state profile.

[0050] Specifically, the data verification subunit is an independent processing unit integrated within the multimodal sensing module. Its main function is to perform real-time quality assessment on the raw data from the camera unit, radar unit, GPS unit, and lidar unit. This subunit can be implemented using a dedicated microcontroller or FPGA to ensure parallel processing capabilities and low latency. Its role is to perform preliminary screening and verification of the validity and reliability of the data from each sensor before it enters the main fusion algorithm, thereby avoiding the negative impact of low-quality data on subsequent decision-making processes.

[0051] Parallel verification of image sharpness and frame rate stability of the camera unit, along with simultaneous verification of ranging accuracy of the radar unit and point cloud density of the lidar unit, refers to the data verification subunit simultaneously performing quality checks on data from different types of sensors. For the camera unit, image sharpness can be evaluated by calculating indicators such as image gradient, edge strength, or frequency domain energy, while frame rate stability is determined by monitoring the time interval between consecutive frames. For the radar unit, ranging accuracy can be verified by comparing with a known reference target or through an internal self-checking mechanism. For the lidar unit, point cloud density can be evaluated by statistically analyzing the number of points per unit volume or area to ensure sufficient environmental detail. These verification processes are performed synchronously to ensure the quality status of data from each sensor acquired within the same time window.

[0052] When image clarity consistently falls below a first preset threshold and radar ranging error consistently exceeds a second preset threshold, a redundant data compensation mode is activated. This typically occurs during severe weather (such as heavy fog or heavy rain) or when some sensors malfunction. The first and second preset thresholds are pre-set based on the actual application scenario and the system's data quality requirements. In this mode, the system prioritizes the data from the LiDAR unit, using it as the primary environmental perception input, as LiDAR typically exhibits good robustness in severe weather. Simultaneously, disabling the high-frequency output of the camera unit reduces its power consumption, extends device lifespan, and decreases the overall computational burden on the system.

[0053] When the GPS signal strength is detected to be consistently below the third preset threshold and radar data is available, the system switches to radar master control mode. This typically occurs in tunnels, areas with many tall buildings, or when GPS signals are interfered with. The third preset threshold is also preset based on the system's positioning accuracy requirements. In this mode, since GPS positioning is unreliable, the system uses distance data calculated by the radar unit to estimate the vehicle's relative position and motion state, using this as the primary value of the position information. By continuously measuring the distance and relative speed with surrounding obstacles, the radar can provide relatively accurate local positioning information, thus maintaining the vehicle's positioning capability when GPS signals are missing.

[0054] During data fusion, a Kalman filter is employed to perform spatiotemporal alignment and noise filtering on multi-source data. The Kalman filter is a highly efficient recursive filter suitable for processing noisy dynamic system data. It can weight and fuse data from different sensors (such as cameras, radar, lidar, and GPS) based on the system's dynamic and measurement models, achieving precise spatiotemporal alignment and effectively filtering out measurement and system noise. After processing by the Kalman filter, the generated data stream not only contains fused environmental information but also includes timestamps and confidence scores. The timestamps ensure the temporal sequence of the data, while the confidence scores quantify the reliability of the fusion results, providing high-quality and reliable input for subsequent generation of ramp traffic state profiles.

[0055] By introducing a data verification subunit to perform parallel verification of the raw data from the multimodal sensing module, sensor data quality issues can be detected and identified in a timely manner, preventing the direct use of low-quality data for subsequent processing. When a degradation in the data quality of a specific sensor is detected, the system can intelligently activate a redundant data compensation mode or switch to radar master control mode, ensuring reliable data input even when some sensors are limited or malfunctioning. This significantly enhances the system's adaptability and robustness to complex environments. Furthermore, employing a Kalman filter for spatiotemporal alignment and noise filtering of multi-source data effectively integrates data from different sensors, generating a high-quality, high-confidence fused data stream. This provides a solid foundation for generating accurate ramp traffic state profiles, making the control decisions generated based on these profiles more reasonable and reliable, effectively improving the overall performance and safety of the ramp intelligent control system.

[0056] It should be noted that real-time traffic data is acquired through the vehicle's integrated multimodal perception module to generate a ramp traffic state profile, which is then matched and analyzed with traffic rules in a pre-set database to generate control decision results. However, if the ramp traffic state profile is not sufficiently refined to incorporate local traffic patterns and environmental factors, the generated profile may deviate from the actual situation, thus affecting the accuracy and adaptability of the control decisions.

[0057] To address this, this embodiment further proposes a method for dynamically adjusting the specific information of the ramp traffic status profile based on a local traffic feature database. This method includes: extracting spatiotemporal features from real-time traffic data, including the current timestamp, geographic coordinates, and weather conditions, and performing similarity matching with historical patterns in the local feature database; calculating local label weights using a machine learning model, where local labels include peak hour markers, accident-prone area markers, and seasonal traffic flow change factors, and adjusting the priority of profile labels based on these weights; dynamically correcting the confidence thresholds of traffic flow labels and accident risk labels based on the deviation between real-time data and local features, enabling the profile output to adapt to localized scenarios; generating an adjusted ramp traffic status profile and attaching a localization adaptation coefficient as a decision-making reference.

[0058] Specifically, when extracting spatiotemporal features from real-time traffic data, the system obtains precise timestamp information, such as year, month, day, hour, minute, and second, as well as precise geographical coordinates of vehicles or ramp areas, such as latitude and longitude. Simultaneously, it acquires real-time weather condition data, such as temperature, humidity, rainfall, and visibility. These spatiotemporal features are then matched against historical patterns stored in the local traffic feature database. Historical patterns can be sets of traffic flow data indexed by time period (e.g., weekday morning rush hour, weekend off-peak), geographical region (e.g., specific ramp entrances and exits), and weather type (e.g., sunny, rainy, foggy). Similarity matching can be performed by calculating the distance between the current spatiotemporal features and the feature vectors of historical patterns (e.g., Euclidean distance, cosine similarity) or by using classification algorithms to identify the historical traffic pattern closest to the current situation.

[0059] When calculating local label weights, the system utilizes a pre-trained machine learning model. This model takes local labels (such as peak-hour markers, accident-prone area markers, and seasonal traffic flow variation factors) as input and outputs the weights of these labels' influence on various labels in the ramp traffic condition profile (such as real-time traffic flow labels, average speed labels, vehicle density labels, and accident risk labels). For example, during peak hours, the weights of traffic flow and vehicle density labels may be increased, while in accident-prone areas, the weight of the accident risk label will increase significantly. The machine learning model can be a support vector machine, random forest, or neural network, which determines the weights by learning the correlation between local labels and traffic profile labels in historical data. These weights are then used to adjust the priority of profile labels, ensuring that the most critical traffic information receives priority attention and processing in specific localized scenarios.

[0060] When dynamically adjusting the confidence thresholds for traffic flow and accident risk labels, the system first calculates the deviation between real-time data and local features. For example, if the real-time traffic flow significantly exceeds the historical average traffic flow under corresponding spatiotemporal conditions in the local feature library, the deviation is high. Based on this deviation, the system dynamically adjusts the confidence thresholds for traffic flow and accident risk labels. For instance, when the deviation is high, the system may increase the confidence level required to trigger the "congestion" or "high risk" label to avoid misjudgments due to abnormal fluctuations; or, under certain specific abnormal conditions, the system may lower the threshold to more sensitively capture potential risks. This dynamic adjustment mechanism enables the output of the ramp traffic status profile to better adapt to localized scenarios, improving the accuracy and robustness of the profile.

[0061] Ultimately, the system generates a traffic status profile of the ramps after the aforementioned adjustments. This profile includes not only real-time traffic flow labels, average speed labels, vehicle density labels, and accident risk labels, but also a localized adaptation coefficient. This coefficient is a quantitative indicator that reflects the profile's adaptability to local traffic characteristics and the strength of its correction. It serves as an important reference in the subsequent control decision-making process, guiding the formulation of control strategies.

[0062] By extracting spatiotemporal features from real-time traffic data and matching them with historical patterns in a local feature database, this embodiment can accurately identify the background information of the current traffic situation, laying the foundation for subsequent refined adjustments. Furthermore, by calculating local label weights and adjusting the priority of profile labels using a machine learning model, the ramp traffic status profile can be intelligently weighted according to local traffic patterns (such as peak hours and accident-prone areas), ensuring the prominent display of key information. In addition, the confidence thresholds for traffic flow labels and accident risk labels are dynamically adjusted based on the deviation between real-time data and local features, effectively avoiding misjudgments or omissions caused by fixed thresholds, making the profile output more localized and adaptable. Finally, the generated ramp traffic status profile with localized adaptation coefficients provides a more accurate and reliable reference for subsequent control decisions, significantly improving the accuracy and response speed of intelligent ramp control, thereby optimizing traffic flow management efficiency.

[0063] This embodiment proposes a vehicle ramp intelligent control method based on multimodal data fusion. It acquires real-time traffic data through a multimodal perception module integrated into the vehicle, and combines this data fusion algorithm to generate a ramp traffic state profile. The method then dynamically adjusts the approach based on a local traffic feature database and performs matching analysis with traffic rules in a pre-set database to generate control decisions. However, in practical applications, relying solely on preliminary matching analysis may lead to uncertainties or potential risks in the generated control decisions, such as data quality fluctuations, inappropriate strategy selection, or insufficient system adaptability to dynamic environments, thereby affecting the safety and efficiency of ramp control.

[0064] To address this, this embodiment further proposes adding multi-level judgment steps during the matching analysis process to ensure the rationality of control decisions. This process includes: a first-level judgment, which assesses the confidence level of the ramp traffic state profile; when the confidence level of the profile label falls below a preset safety threshold, a data re-collection or manual review mechanism is triggered; a second-level judgment, based on a risk priority ranking algorithm, calculates the expected impact value of different control strategies and prioritizes the strategy with the highest risk balance and efficiency; a third-level judgment, which introduces a feedback verification loop, comparing historical decision results with actual traffic flow changes, and optimizing the matching rules through a reinforcement learning model to reduce the misjudgment rate; the final generated control decision result must pass the verification of all judgment steps; otherwise, it reverts to the default safety mode.

[0065] Specifically, in the first-level judgment, the reliability or accuracy of each label in the ramp traffic status profile (such as real-time traffic flow label, average vehicle speed label, vehicle density label, and accident risk label) is quantitatively evaluated. This can be achieved in various ways, such as calculating based on the redundancy and consistency of data from each sensor in the multimodal perception module, the accuracy indicators of the sensors themselves, and the confidence interval or probability distribution output by the data fusion algorithm. The preset safety threshold refers to a lower limit of confidence set by the system in advance. This threshold can be set according to the requirements of the actual application scenario for the safety and reliability of ramp control. When the confidence of the profile label is lower than this threshold, the system will actively request the multimodal perception module to reacquire or supplement real-time traffic data in order to improve the confidence of the profile with more or more accurate data. This is data re-acquisition. If the confidence still cannot be improved, the current ramp traffic status profile and related raw data are submitted to a human operator for review and judgment. This is a manual review mechanism to ensure that safe and effective control can still be made under conditions of high uncertainty.

[0066] In the second-level judgment, a risk priority ranking algorithm is employed. This algorithm quantifies and compares the potential risks and benefits of different control strategies. It comprehensively considers multiple factors, such as the potential for traffic congestion, accident probability, changes in traffic efficiency, and environmental impact, assigning weights to each risk and benefit factor and calculating a comprehensive score. The expected impact value refers to the predicted quantitative value of the positive or negative effects that a specific control strategy may have on the ramp traffic system. These values ​​can be predicted using traffic flow simulation models, historical data analysis, or expert systems. The system selects a strategy that effectively reduces potential risks (such as accident risk and severe congestion risk) while maximizing traffic efficiency (such as increasing traffic volume and reducing travel time)—that is, a strategy with a risk balance and the highest efficiency. This typically involves a multi-objective optimization problem.

[0067] In the third-level judgment, a feedback verification loop is introduced. This is a continuously running mechanism used to evaluate the actual effectiveness of the control decisions made by the system in the past. It judges the effectiveness and accuracy of the decisions by comparing the ramp adjustment instructions executed by the system (i.e., historical decision results) with the actual traffic flow changes (such as traffic volume, speed, and congestion) after the instructions are executed. Reinforcement learning is a machine learning paradigm that learns optimal behavioral strategies through interaction with the environment. In this embodiment, the reinforcement learning model takes the ramp traffic state profile as input, the control decision result as the output action, and the actual traffic flow changes as a reward or penalty signal. Through continuous trial and error and learning, the model optimizes the matching rules, that is, it automatically adjusts and improves the logic and parameters for matching the ramp traffic state profile with traffic rules, thereby reducing the misjudgment rate and mitigating the negative impact caused by decision-making errors.

[0068] Ultimately, the verification of all judgment steps emphasizes the sequential nature and necessity of the multi-level judgment mechanism. Only when the confidence level of the ramp traffic state profile meets the requirements, the selected control strategy achieves an optimal balance between risk and efficiency, and the decision is consistent with historical feedback verification results or optimized by a reinforcement learning model, is the control decision considered valid. Otherwise, the system will revert to the default safety mode, a pre-set, conservative system operating state. This includes actions such as setting the ramp traffic lights to red, displaying minimum speed limit signs, prohibiting vehicles from entering the ramp, or sending emergency avoidance warnings to all relevant vehicles, to maximize traffic safety and avoid risks arising from uncertain decisions.

[0069] By introducing a multi-level judgment process, this embodiment effectively addresses the issues of insufficient reliability, suboptimal decision-making, and poor adaptability that may exist in generating control decision results. The first-level judgment, through confidence assessment and data re-collection or manual review, ensures the accuracy of decision data from the source. The second-level judgment, based on a risk priority ranking algorithm, intelligently selects the strategy with the highest risk balance and efficiency, guaranteeing the scientific and optimal nature of the decision. The third-level judgment introduces a feedback verification loop and a reinforcement learning model, enabling the system to continuously learn and optimize matching rules, significantly reducing the misjudgment rate and improving adaptability to dynamic traffic environments. Finally, all decision results must undergo rigorous verification; otherwise, they revert to the default safe mode, thereby ensuring the highest level of safety for ramp traffic and significantly improving the overall robustness, safety, and intelligence level of the ramp intelligent control system.

[0070] It should be noted that although a traffic status profile of the ramps was generated through multimodal data fusion and dynamically adjusted based on the local traffic feature database, how to effectively transform this profile information into refined and operable control decision results, especially how to accurately identify and classify the real-time traffic status of the ramps and quantify its deviation from historical patterns to ensure the accuracy and adaptability of the decisions, still needs further clarification.

[0071] In response, this embodiment further proposes that the steps for generating control decision results include: classifying the ramp traffic status using a state machine based on threshold determination and outputting status labels, including labels for smooth traffic, slow traffic, congestion, and accidents; calculating the deviation by comparing real-time traffic flow with historical peak patterns using a spatiotemporal pattern mining algorithm; and weighting and correcting the deviation by combining event data in the local feature database.

[0072] Specifically, to achieve accurate classification of ramp traffic conditions, this embodiment employs a threshold-based state machine. This state machine is a finite state automaton model that compares real-time traffic data (such as real-time traffic flow labels, average speed labels, and vehicle density labels) with multiple preset thresholds, thereby classifying complex traffic conditions into discrete and clear state labels. For example, when real-time traffic flow is below a certain threshold and average speed is above another threshold, the system can classify it as a "smooth" state; when traffic flow increases significantly while speed decreases noticeably, it may enter a "congested" state. The triggering of an "accident label" may be based on abnormal events detected by the multimodal perception module, such as abnormal vehicle stagnation, signs of collision, or abrupt traffic flow interruption. This state machine mechanism ensures the determinism and consistency of ramp traffic condition classification, providing a clear basis for subsequent decision-making.

[0073] Building upon this foundation, to gain a deeper understanding of the anomalies in real-time traffic flow, this embodiment introduces a spatiotemporal pattern mining algorithm. This algorithm aims to analyze the differences between real-time traffic flow data (including information such as traffic volume and speed in time and space) and historical peak patterns stored in a local traffic feature database. Historical peak patterns typically contain typical traffic behavior characteristics of a ramp within a specific time period (such as morning and evening rush hours). The spatiotemporal pattern mining algorithm can be implemented using various techniques, such as using the Dynamic Time Warping (DTW) algorithm to compare the similarity between real-time traffic flow time series and historical pattern time series, or using statistical methods to calculate the degree of deviation between real-time data points and the mean or distribution of historical patterns. In this way, the system can quantify the "deviation" between the current traffic conditions and normal peak-hour patterns, thereby identifying whether it belongs to regular peak congestion or abnormal traffic conditions requiring special attention.

[0074] Furthermore, to make the calculated deviation more meaningful and valuable for decision-making, this embodiment incorporates event data from a local feature database to perform a weighted correction on the deviation. The local feature database may contain information on various known events, such as large-scale events (concerts, sporting events), road construction, and peak holiday travel periods. These events often have predictable impacts on traffic flow. When the system detects a high deviation, it queries the local feature database. If it finds that the current deviation is associated with a known event, it can adjust the deviation based on the expected impact of that event. For example, if the deviation indicates severe congestion, but the local feature database shows a large event is taking place nearby, the system may lower the deviation, considering it predictable, thus avoiding overreaction. Conversely, if the deviation is high and there is no known event to explain it, it may be raised to highlight its anomaly. This weighted correction mechanism enables the decision-making system to distinguish between "expected deviations" and "unexpected anomalies," thereby improving the accuracy and adaptability of decision-making.

[0075] Through the above technical solution, this embodiment can clearly classify complex ramp traffic status profiles into discrete state labels such as smooth flow, slow traffic, congestion, and accidents, greatly improving the understanding of real-time traffic conditions and the intuitiveness of decision-making. Simultaneously, the application of spatiotemporal pattern mining algorithms enables the system to quantify the deviation between real-time traffic flow and historical peak patterns, thereby gaining a deeper understanding of the degree and nature of traffic anomalies. Furthermore, by combining event data from the local feature database to perform weighted correction on the deviation, misjudgments of predictable traffic changes are effectively avoided, allowing control decisions to more accurately reflect actual needs and exhibit stronger adaptability to emergencies. This makes the subsequent control decision generation process more refined and intelligent, ensuring the rationality and effectiveness of ramp adjustment commands, thereby improving the overall performance of the ramp intelligent control system.

[0076] It should be noted that control decisions are generated by matching the adjusted ramp traffic state profile with traffic rules in a pre-set database. However, ramp traffic conditions are complex and ever-changing. If the traffic rule database is statically preset, it will be difficult to adapt to constantly changing traffic flow patterns, emergencies, and new types of risks. This may lead to a decline in the accuracy and effectiveness of control decisions over time, thereby affecting the long-term performance of the ramp intelligent control system.

[0077] In response, this embodiment further proposes that the traffic rule base in the preset database supports dynamic updates. The method includes: recording the profile data, control strategies and effect feedback of all traffic events; using machine learning algorithms to analyze historical data; automatically adjusting the threshold and / or confidence level of traffic flow determination; mining potential new risk patterns to generate new rule suggestions, which are then incorporated into the rule base after confirmation by the traffic management center.

[0078] The traffic rule base is a core component of the ramp intelligent control system, storing the logic, thresholds, and parameters used to determine traffic conditions and generate control strategies. Supporting dynamic updates means that this rule base is not static but can be continuously revised and optimized based on actual operational conditions and data feedback to ensure it remains synchronized with the latest traffic environment and management needs. This dynamism is key to improving the system's adaptability and intelligence. The system continuously records a profile of the ramp traffic state at the time of each traffic event, including real-time traffic flow labels, average speed labels, vehicle density labels, accident risk labels, and the control strategies adopted by the system for that event, such as adjusting traffic light cycles, displaying speed limit signs, and providing coordinated traffic prompts. It further records the actual effects of these control strategies, such as whether traffic flow has returned to normal, whether congestion has been alleviated, and whether accidents have been effectively handled. These records constitute a valuable historical dataset, providing a foundation for subsequent analysis and learning.

[0079] Machine learning algorithms, such as classification, regression, clustering, or reinforcement learning algorithms, are applied to perform deep analysis of recorded historical data. By learning the complex relationships between historical traffic profile data, control strategies, and their feedback effects, algorithms can identify key factors affecting the accuracy of traffic flow assessments and automatically optimize or adjust thresholds used to distinguish different traffic flow states (such as smooth flow, slow flow, and congestion), as well as confidence assessment parameters for various profile labels (such as accident risk labels). For example, in specific weather conditions or time periods, the algorithm may find that the original congestion assessment threshold is no longer applicable and make fine-tuning to better reflect the actual situation. In addition to adjusting the parameters of existing rules, machine learning algorithms can also proactively discover new, previously undefined, or evolving traffic risk patterns through pattern recognition and anomaly detection of historical data. For example, the algorithm may identify that a certain combination of traffic flows is prone to triggering cascading congestion on specific road sections and under specific conditions, which is not directly covered by existing rules. Once such patterns are identified, the system can automatically generate corresponding rule suggestions, such as proposing new control strategies or early warning mechanisms. To ensure the safety, compliance, and effectiveness of newly generated rule recommendations, these recommendations undergo manual review and verification before being formally incorporated into the traffic rule base. Typically, this is assessed by a professional traffic management center or relevant department, who combines practical management experience, policies and regulations, and simulation test results to make the final approval or revision of the new rule recommendations, thus ensuring that the rule base updates are both intelligent and reliable.

[0080] Through the above technical solution, the intelligent ramp control method of this embodiment can overcome the limitations of traditional static traffic rule bases. By continuously recording traffic event profile data, control strategies, and effect feedback, and using machine learning algorithms to conduct in-depth analysis of this historical data, the system can automatically learn and adapt to the ever-changing traffic environment. Specifically, the machine learning algorithm can intelligently adjust the threshold and / or confidence level of traffic flow determination, making the classification and risk assessment of ramp traffic status profiles more accurate, thereby improving the accuracy of control decisions. In addition, this solution can proactively discover potential new risk patterns and generate new rule suggestions, enabling the system to respond promptly to emerging traffic challenges rather than passively waiting for manual intervention. Finally, with the confirmation of the traffic management center, these optimized and added rules are incorporated into the rule base, ensuring that the intelligent ramp control system can maintain a highly efficient, safe, and adaptive operating state in the long term, significantly improving the intelligence level of ramp traffic management and its ability to cope with complex traffic conditions.

[0081] It should be noted that multimodal data fusion and multi-level judgment can generate control decision results for ramp traffic conditions. However, when translating these abstract control decision results into specific, executable ramp adjustment instructions, the lack of a refined risk level correspondence strategy may lead to inaccurate execution of the instructions, failing to adequately address complex traffic conditions under different risk levels, thereby affecting the efficiency and safety of ramp traffic control.

[0082] To address this, this embodiment further proposes a method for generating and executing corresponding ramp adjustment instructions based on the risk level and control strategy in the control decision results. Specifically, it includes: for low-risk slow traffic conditions, sending a yellow warning signal to the ramp traffic lights and adjusting the green light cycle to be extended; for medium-risk congestion conditions, triggering variable speed limit signs to display speed reduction prompts and broadcasting coordinated deceleration messages to surrounding vehicles; for high-risk accident conditions, sending the highest level alarm to the traffic management platform, automatically generating emergency lane opening instructions, and triggering the roadside broadcast system to play traffic guidance information.

[0083] The low-risk slow-moving state refers to a situation where ramp traffic flow has slowed down to some extent, but has not yet reached the level of severe congestion or accidents, and there is still room for effective traffic management through adjustments. This state is usually determined by a combination of indicators such as real-time traffic flow and average speed in the ramp traffic status profile. In this state, a yellow warning signal is sent to the ramp traffic lights to remind drivers of changes in traffic conditions ahead, to slow down, and to be more vigilant. This is usually achieved by communicating with the traffic signal control system and sending specific signal control commands. Simultaneously, extending the green light cycle means increasing the green light time for vehicles entering the main road from the ramp. This operation involves sending commands to the traffic light controller to dynamically adjust the signal timing scheme, aiming to maximize the traffic flow efficiency of ramp vehicles without affecting main road traffic and effectively alleviate vehicle congestion within the ramps.

[0084] Medium-risk congestion refers to a situation where there is significant vehicle backlog and a marked decrease in vehicle speed on ramps, posing a risk of further deterioration into severe congestion or triggering accidents. This state is typically triggered when indicators such as vehicle density and average speed in the ramp traffic profile reach preset thresholds. In this state, triggering variable speed limit signs to display speed reduction prompts involves using roadside variable information signs or variable speed limit signs to display suggested or mandatory speeds below the conventional speed limit to drivers. This helps smooth traffic flow, reduce speed differences between vehicles, and thus lower the risk of rear-end collisions. Simultaneously, broadcasting coordinated deceleration messages to surrounding vehicles involves using vehicle-to-everything (V2X) communication technology to send digital messages containing deceleration suggestions or information about upcoming congestion to connected vehicles on the ramp and in the surrounding area. Vehicles receiving these messages can adjust their speed accordingly, and even implement coordinated deceleration in autonomous driving mode, further improving the overall stability of traffic flow.

[0085] A high-risk accident status refers to an accident that has occurred on the ramp or an extremely high risk of such an accident, requiring immediate emergency intervention. This status is typically triggered by the activation of an accident risk tag in the ramp traffic status profile or by the direct detection of an abnormal event by a multimodal perception module. In this status, sending the highest-level alert to the traffic management platform means the system automatically generates and sends an emergency alert containing detailed information such as the accident location, type, and severity to the traffic management center. This ensures that traffic management departments can obtain information immediately and initiate appropriate emergency response procedures. Automatically generating emergency lane opening instructions means the system automatically plans and sends instructions to relevant traffic facilities (such as variable lane signs and roadblocks) based on the accident location and traffic conditions to open or designate a dedicated lane for emergency vehicles. This is crucial for ensuring that rescue vehicles can quickly reach the scene. Linking with the roadside broadcast system to play traffic guidance information means the system activates roadside broadcast equipment installed on and around the ramp to play preset or real-time generated traffic guidance information, such as informing drivers of detour routes, accident details, or precautions, guiding drivers to pass in an orderly manner and avoiding secondary accidents.

[0086] Through the above technical solution, this embodiment can transform abstract control decision results into specific and targeted ramp adjustment instructions. For low-risk slow-moving conditions, adjusting traffic light cycles and warning signals can effectively guide traffic flow and prevent congestion from escalating. For medium-risk congestion conditions, variable speed limits and coordinated deceleration messages can smooth traffic flow and reduce accident risk. For high-risk accident conditions, a multi-party coordinated emergency response can be quickly initiated to minimize the impact of accidents and ensure access for rescue operations. This refined, hierarchical control strategy significantly improves the response speed, accuracy, and overall safety of ramp traffic control, ensuring that the most appropriate intervention measures can be taken under different traffic conditions, thereby optimizing ramp traffic efficiency and emergency response capabilities.

[0087] The following example will provide a more detailed explanation of the above technical solution: Imagine a highway ramp entrance in a city, where vehicles are preparing to merge into the main road. This ramp frequently experiences congestion during peak hours and occasional minor collisions. This solution aims to address issues in existing ramp control methods, such as insufficient data coverage, poor dynamic adaptability, lack of collaborative control, control lag, rigid strategies, and conflicts between multi-source data.

[0088] First, as a vehicle approaches the ramp, its integrated multimodal perception module activates, acquiring real-time traffic data. This module includes a camera unit for video recognition, a radar unit for ranging, a GPS unit for positioning, and a lidar unit for traffic flow monitoring. For example, the camera unit continuously captures video streams of the ramp area, the radar unit precisely measures the distance and speed of vehicles ahead, the GPS unit provides accurate vehicle location information, and the lidar unit generates high-density point cloud data to identify vehicle type, trajectory, and lane occupancy. Compared to traditional single data sources (such as geomagnetic coils), this multimodal perception method can comprehensively capture the multidimensional traffic characteristics of the ramp, effectively solving the problem of insufficient data coverage.

[0089] During data acquisition, the data verification subunit built into the multimodal perception module verifies the image clarity and frame rate stability of the camera unit in parallel, and simultaneously verifies the ranging accuracy of the radar unit and the point cloud density of the lidar unit. For example, when the image clarity is consistently below a first preset threshold and the radar ranging error is consistently above a second preset threshold, the system activates a redundant data compensation mode. In this mode, the lidar unit's data is used as the primary value, and the high-frequency output of the camera unit is turned off to reduce power consumption. When the GPS signal strength is consistently below a third preset threshold and radar data is available, the system switches to radar master control mode. In this mode, the distance data calculated by the radar unit is used as the primary value of the location information. During data fusion, the system uses a Kalman filter to perform spatiotemporal alignment and noise filtering on multi-source data, generating a fused data stream with timestamps and confidence scores, which serves as input for subsequently generating a ramp traffic status profile. This data verification and fusion mechanism effectively solves the problem of multi-source data conflicts and improves the reliability of the fusion results.

[0090] Next, a traffic status profile of the ramp is generated by combining real-time traffic data with a data fusion algorithm. This profile includes real-time traffic flow labels, average vehicle speed labels, vehicle density labels, and accident risk labels. For example, the system calculates the number of vehicles passing through the current ramp per minute, the average speed of all vehicles, and the number of vehicles per unit area based on the fused data, and assesses potential accident risks by combining abnormal behavior patterns (such as sudden braking and frequent lane changes).

[0091] Subsequently, the system dynamically adjusts the specific information of the ramp traffic status profile based on a local traffic feature database. This database includes geographic information system data, historical traffic flow patterns, time-period features, and environmental factor labels. For example, the system extracts the current timestamp, geographic coordinates, and weather conditions from real-time traffic data and performs similarity matching with historical patterns in the local feature database. A machine learning model calculates local label weights, which include peak-hour markers, accident-prone area markers, and seasonal traffic flow change factors. The priority of profile labels is adjusted based on these weights. For instance, during weekday morning rush hours, the system assigns higher weights to traffic flow and vehicle density labels, and makes corrections based on historical peak-hour traffic flow patterns. If the current weather is rainy or snowy, the weight of the accident risk label is increased. Based on the deviation between real-time data and local features, the system dynamically adjusts the confidence thresholds of traffic flow and accident risk labels, enabling the profile output to adapt to localized scenarios, generating an adjusted ramp traffic status profile, and adding a localization adaptation coefficient as a decision-making reference. This dynamic adjustment mechanism overcomes the shortcomings of existing technologies, such as rigid strategies and a lack of adaptive adjustment capabilities to real-time traffic conditions.

[0092] Then, the adjusted ramp traffic status profile is matched and analyzed with traffic rules in a pre-set database to generate control decision results. For example, if the profile shows high traffic volume, high vehicle density, and average speed below normal levels, the system will match it with the "moderate congestion" rule defined in the database. This pre-set database of traffic rules supports dynamic updates. By recording profile data, control strategies, and effect feedback for all traffic events, it uses machine learning algorithms to analyze historical data, automatically adjusts the threshold and / or confidence level for traffic volume determination, and identifies potential new risk patterns to generate new rule suggestions, which are then incorporated into the rule database after confirmation by the traffic management center.

[0093] During the matching analysis process, the system adds multi-level judgment steps to ensure the rationality of control decisions.

[0094] Level 1 Judgment: A confidence level assessment is performed on the ramp traffic status profile. When the confidence level of the profile label falls below a preset safety threshold—for example, if there is a significant discrepancy between LiDAR data and radar data—the system will trigger a data re-collection or manual review mechanism to avoid misjudgments caused by data uncertainty.

[0095] The second level of judgment involves calculating the expected impact of different control strategies based on a risk priority ranking algorithm, and prioritizing the strategy with the highest risk balance and efficiency. For example, for a potential congestion situation, the system will evaluate the risks and benefits of two strategies: "extending the green light period" and "limiting the number of vehicles entering," and select the optimal solution.

[0096] The third level of judgment involves introducing a feedback verification loop. Historical decision results are compared with actual traffic flow changes, and matching rules are optimized through a reinforcement learning model to reduce the false positive rate. For example, if a decision to extend the green light period fails to effectively alleviate congestion, the system will learn and adjust its decision logic for similar situations in the future. The final control decision must pass verification through all judgment steps; otherwise, it reverts to the default safe mode. This multi-level judgment mechanism effectively solves the problem of control lag, ensuring the accuracy and reliability of decisions.

[0097] Finally, based on the risk level and control strategy in the control decision results, corresponding ramp adjustment instructions are generated and executed. These ramp adjustment instructions include adjusting traffic light cycles, speed limit sign displays, and sending coordinated passage prompts to vehicles.

[0098] For example, using a threshold-based state machine, the system classifies ramp traffic conditions and outputs state labels, including: smooth traffic, slow traffic, congestion, and accident labels. A spatiotemporal pattern mining algorithm compares real-time traffic flow with historical peak patterns to calculate the deviation, and then uses event data from a local feature database to perform a weighted correction on the deviation.

[0099] For low-risk slow-moving situations, the system will send a yellow warning signal to the ramp traffic lights and adjust the green light cycle to lengthen in order to smooth traffic flow.

[0100] In the case of medium-risk congestion, the system will trigger a variable speed limit sign to display a speed reduction prompt, such as adjusting the speed limit from 80 km / h to 60 km / h, and broadcast a coordinated deceleration message to surrounding vehicles to remind drivers to slow down in advance and avoid sudden braking.

[0101] In the event of a high-risk accident, the system sends the highest-level alert to the traffic management platform, automatically generates an emergency lane opening instruction, and triggers the roadside broadcast system to broadcast traffic guidance information, directing vehicles to detour or pass through in an orderly manner. This vehicle-road cooperative optimization method effectively integrates multimodal perception data, realizes multi-vehicle cooperative merging control, improves ramp traffic efficiency, and solves the problem of lack of cooperative control in existing technologies.

[0102] Furthermore, this application also proposes a computer-readable storage medium storing a program for intelligent vehicle ramp control based on multimodal data fusion. When the program for intelligent vehicle ramp control based on multimodal data fusion is executed by a processor, it implements the steps of the method for intelligent vehicle ramp control based on multimodal data fusion as described above.

[0103] Reference Figure 3 , Figure 3 This is a structural block diagram of the first embodiment of the intelligent vehicle ramp control system based on multimodal data fusion of this application.

[0104] like Figure 3 As shown in the embodiments of this application, the intelligent vehicle ramp control system based on multimodal data fusion includes: The data acquisition module 10 is used to acquire real-time traffic data through the multimodal perception module integrated in the vehicle. The multimodal perception module includes a camera unit for video recognition, a radar unit for ranging, a GPS unit for positioning, and a lidar unit for traffic flow monitoring. The profile generation module 20 is used to generate a ramp traffic status profile by combining the real-time traffic data with a data fusion algorithm. The ramp traffic status profile includes: real-time traffic flow label, average vehicle speed label, vehicle density label, and accident risk label. Data adjustment module 30 is used to dynamically adjust the specific information of the ramp traffic status profile based on the local traffic feature database, wherein the local traffic feature database includes geographic information system data, historical traffic flow patterns, time period features and environmental factor labels; The decision generation module 40 is used to match and analyze the adjusted ramp traffic state profile with the traffic rules in the preset database to generate control decision results. The judgment module 50 is used to add multi-level judgment steps during the matching analysis process, including confidence assessment, risk priority ranking and feedback verification; The instruction generation module 60 is used to generate and execute corresponding ramp adjustment instructions based on the risk level and control strategy in the control decision results. The ramp adjustment instructions include adjusting the traffic light cycle, speed limit sign display, and sending cooperative passage prompts to vehicles.

[0105] It should be understood that the above are merely illustrative examples and do not constitute any limitation on the technical solution of this application. In specific applications, those skilled in the art can make settings as needed, and this application does not impose any restrictions on this.

[0106] This embodiment solves the problems of insufficient data perception and control lag in the prior art by acquiring multimodal data, generating state profiles, dynamic adjustment, and multi-level judgment. It has the advantages of improving the accuracy and adaptability of ramp control, reducing control lag, and improving the efficiency of road network operation.

[0107] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of this application. In practical applications, those skilled in the art can select some or all of it to achieve the purpose of this embodiment according to actual needs, and no restrictions are imposed here.

[0108] In addition, for technical details not described in detail in this embodiment, please refer to the method for intelligent control of vehicle ramps based on multimodal data fusion provided in any embodiment of this application, which will not be repeated here.

[0109] Furthermore, it should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0110] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0111] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory (ROM) / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods of the various embodiments of this application. The above are only preferred embodiments of this application and do not limit the patent scope of this application. All equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A vehicle ramp intelligent control method based on multimodal data fusion, characterized in that, include: Real-time traffic data is acquired through a multimodal perception module integrated into the vehicle. The multimodal perception module includes a camera unit for video recognition, a radar unit for ranging, a GPS unit for positioning, and a lidar unit for traffic flow monitoring. A ramp traffic status profile is generated by combining the real-time traffic data with a data fusion algorithm. The ramp traffic status profile includes: real-time traffic flow label, average vehicle speed label, vehicle density label, and accident risk label. The specific information of the ramp traffic status profile is dynamically adjusted based on the local traffic feature database, which includes geographic information system data, historical traffic flow patterns, time period characteristics, and environmental factor labels. The adjusted ramp traffic status profile is matched and analyzed with traffic rules in the preset database to generate control decision results; Add multi-level judgment steps to the matching analysis process, including confidence assessment, risk priority ranking, and feedback verification; Based on the risk level and control strategy in the control decision results, corresponding ramp adjustment instructions are generated and executed. The ramp adjustment instructions include adjusting the traffic light cycle, speed limit sign display, and sending cooperative passage prompts to vehicles.

2. The method according to claim 1, characterized in that, The multimodal perception module has a built-in data verification subunit; the step of acquiring real-time traffic data through the vehicle-integrated multimodal perception module includes: The image clarity and frame rate stability of the camera unit are verified in parallel by the data verification subunit, and the ranging accuracy of the radar unit and the point cloud density of the lidar unit are verified simultaneously. When the image clarity is detected to be consistently below the first preset threshold and the radar ranging error is consistently above the second preset threshold, the redundant data compensation mode is activated. In this mode, the data from the lidar unit is used as the main value, and the high-frequency output of the camera unit is turned off to reduce power consumption. When the GPS signal strength is detected to be continuously lower than the third preset threshold and radar data is available, switch to radar master control mode. In this mode, the distance data calculated by the radar unit is used as the main value of the location information. During the data fusion process, a Kalman filter is used to perform spatiotemporal alignment and noise filtering on multi-source data, generating a fused data stream with timestamps and confidence scores, which serves as the input for the subsequent generation of ramp traffic status profiles.

3. The method according to claim 1, characterized in that, The steps for dynamically adjusting the specific information of the ramp traffic status profile based on the local traffic feature database include: Extract spatiotemporal features from real-time traffic data, including current timestamp, geographic location coordinates, and weather conditions, and perform similarity matching with historical patterns in the local feature database; Local label weights are calculated using a machine learning model. These local labels include peak hour markers, accident-prone area markers, and seasonal traffic flow change factors. The priority of the profile labels is then adjusted based on these weights. Based on the deviation between real-time data and local characteristics, the confidence thresholds of traffic flow labels and accident risk labels are dynamically adjusted to make the profile output adapt to localized scenarios. Generate an adjusted traffic status profile of the ramps and add a localization adaptation coefficient as a decision-making reference.

4. The method according to claim 1, characterized in that, The process of adding multi-level judgment steps to achieve reasonableness includes: Level 1 judgment: Confidence assessment of the traffic status profile of the ramp. When the confidence of the profile label is lower than the preset safety threshold, a data re-collection or manual review mechanism is triggered. Second-level judgment: Based on the risk priority ranking algorithm, calculate the expected impact value of different control strategies, and prioritize the strategy with the highest risk balance and efficiency. The third level of judgment: introduce a feedback verification loop to compare historical decision results with actual traffic flow changes, and optimize the matching rules through reinforcement learning models to reduce the misjudgment rate; The final control decision result must be verified by all judgment steps; otherwise, it will fall back to the default safe mode.

5. The method according to claim 1, characterized in that, The step of generating control decision results also includes: The state machine based on threshold determination is used to classify the traffic status of the ramps and output status labels, which include: smooth traffic, slow traffic, congestion, and accident labels. By using a spatiotemporal pattern mining algorithm, the deviation is calculated by comparing real-time traffic flow with historical peak patterns. The deviation is corrected by combining event data from the local feature database with weighted data.

6. The method according to claim 1, characterized in that, The traffic rule base in the preset database supports dynamic updates, and the methods include: Record profile data, control strategies and effect feedback of all traffic events, use machine learning algorithms to analyze historical data, and automatically adjust the threshold and / or confidence level for traffic flow determination; New potential risk patterns are identified to generate new rule recommendations, which are then incorporated into the rule base after confirmation by the traffic management center.

7. The method according to claim 1, characterized in that, The step of generating and executing corresponding ramp adjustment instructions based on the risk level and control strategy in the control decision results includes: For low-risk slow-moving conditions, a yellow warning signal is sent to the ramp traffic lights, and the green light cycle is extended. For medium-risk congestion conditions, the variable speed limit sign will be triggered to display a speed reduction prompt and a coordinated deceleration message will be broadcast to surrounding vehicles. In the event of a high-risk accident, the highest level alarm is sent to the traffic management platform, an emergency lane opening instruction is automatically generated, and the roadside broadcast system is triggered to broadcast traffic guidance information.

8. A vehicle ramp intelligent control system based on multimodal data fusion, characterized in that, include: The data acquisition module is used to acquire real-time traffic data through the multimodal perception module integrated in the vehicle. The multimodal perception module includes a camera unit for video recognition, a radar unit for ranging, a GPS unit for positioning, and a lidar unit for traffic flow monitoring. The profile generation module is used to generate a ramp traffic status profile by combining the real-time traffic data with a data fusion algorithm. The ramp traffic status profile includes: real-time traffic flow label, average vehicle speed label, vehicle density label, and accident risk label. The data adjustment module is used to dynamically adjust the specific information of the ramp traffic status profile based on the local traffic feature database, which includes geographic information system data, historical traffic flow patterns, time period features, and environmental factor labels. The decision generation module is used to match and analyze the adjusted ramp traffic status profile with traffic rules in the preset database to generate control decision results; The judgment module is used to add multi-level judgment steps during the matching analysis process, including confidence assessment, risk priority ranking, and feedback verification. The instruction generation module is used to generate and execute corresponding ramp adjustment instructions based on the risk level and control strategy in the control decision results. The ramp adjustment instructions include adjusting the traffic light cycle, speed limit sign display, and sending cooperative passage prompts to vehicles.

9. A computer device, characterized in that, The device includes a memory and a processor, wherein when the processor executes computer instructions stored in the memory, it performs the vehicle ramp intelligent control method based on multimodal data fusion as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The method includes instructions that, when executed on a computer, cause the computer to perform the vehicle ramp intelligent control method based on multimodal data fusion as described in any one of claims 1 to 7.

Images