Precision perception and decision-making integrated traffic control method, system and equipment

Abstract

The invention provides a precision perception and decision-making integrated traffic control method, system and device, and relates to the technical field of traffic control, and the method comprises the steps: collecting high-precision traffic flow information and pedestrian and non-motor vehicle information of a preset lane range, and constructing a target traffic signal lamp preset range scene model; collecting a high-precision lane image set; vehicle type identification and vehicle distance identification are carried out to generate a road congestion coefficient; positioning the type and the position of the sudden traffic event; performing influence range prediction to obtain a traffic event prediction influence range; and performing traffic signal lamp control decision analysis to obtain a target control parameter, and performing a control decision of a target traffic signal lamp. The technical problems that an existing traffic control system is usually set as timing period signal lamp control, traffic demand differences in different time periods and different road conditions are neglected, flexible adjustment cannot be carried out when sudden traffic accidents occur, and congestion or traffic accidents are caused are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic control, and particularly to a traffic control method, system and device for integrated precision perception and decision-making. Background Art

[0002] Most existing traffic signal control systems use fixed green and red light cycles for traffic flow management. However, this does not take into account real-time traffic flow, vehicle speed, traffic density and other factors. Especially during peak hours or when special events occur, the fixed cycle may lead to unbalanced traffic flow, and problems such as vehicle congestion or low traffic efficiency may occur. Moreover, currently, traffic management systems usually separate traffic information collection from signal control decision-making. The information collection part uses traditional sensors or cameras to collect data, while the decision-making part is usually controlled by a preset traffic signal cycle. This separation of information and decision-making cannot optimize the signal control strategy in real time, resulting in information lag and low decision-making efficiency. Summary of the Invention

[0003] The present application provides a traffic control method, system and device for integrated precision perception and decision-making, aiming to solve the technical problem that existing traffic control systems are usually set with timed-cycle signal control, ignoring the traffic demand differences in different time periods and different road conditions, and being unable to make flexible adjustments in case of sudden traffic events, resulting in congestion or traffic accidents.

[0004] In the first aspect disclosed in the present application, a traffic control method for integrated precision perception and decision-making is provided. The method includes: collecting high-precision traffic flow information, pedestrian and non-motor vehicle information within a preset lane range ending at a target traffic signal, and constructing a scene model for the preset range of the target traffic signal; connecting a high-precision traffic image acquisition device to collect a set of high-precision lane images within the preset lane range; based on the set of high-precision lane images, performing vehicle type recognition and vehicle distance recognition, and generating a road congestion coefficient according to the recognition results; when receiving sudden traffic event information, locating the type and location of the sudden traffic event; based on the type and location of the sudden traffic event, predicting the influence range to obtain the predicted influence range of the traffic event; according to the predicted influence range of the traffic event, using the road congestion coefficient as a decision reference coefficient, performing traffic signal control decision analysis in the scene model for the preset range of the target traffic signal to obtain target control parameters, and making a control decision for the target traffic signal based on the target control parameters.

[0005] The second aspect disclosed in this application provides a traffic control system integrating precision perception and decision-making. The system is used for the above-mentioned traffic control method integrating precision perception and decision-making, and the system includes: a scenario model construction module, configured to collect high-precision traffic flow information, pedestrian and non-motor vehicle information within a preset lane range with a target traffic signal as the end point, and construct a scenario model for the preset range of the target traffic signal; a lane image acquisition module, configured to connect to a high-precision traffic image acquisition device and acquire a set of high-precision lane images within the preset lane range; a vehicle distance recognition module, configured to perform vehicle type recognition and vehicle distance recognition based on the set of high-precision lane images, and generate a road congestion coefficient according to the recognition results; a traffic event positioning module, configured to, when receiving sudden traffic event information, locate the type and location of the sudden traffic event; an influence range prediction module, configured to perform influence range prediction based on the type of the sudden traffic event and the location of the sudden traffic event to obtain a predicted influence range of the traffic event; a control decision-making module, configured to perform traffic signal control decision analysis in the scenario model for the preset range of the target traffic signal according to the predicted influence range of the traffic event, using the road congestion coefficient as a decision reference coefficient, to obtain target control parameters, and perform control decision-making for the target traffic signal based on the target control parameters.

[0006] The third aspect disclosed in this application provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the traffic control method integrating precision perception and decision-making in the first aspect.

[0007] One or more technical solutions provided in this application have at least the following beneficial effects:

[0008] By collecting traffic flow, pedestrian, and non-motor vehicle information within the preset lane range covered by the target traffic signal, this high-precision collection can ensure a real-time reflection of the true situation of traffic flow. Based on the collected traffic information, a scene model of the preset range in the area where the target traffic signal is located is constructed. This model contains multi-dimensional information such as traffic flow, pedestrians, and non-motor vehicles, which helps to comprehensively understand the road traffic conditions and provides a scenario-based reference for control decisions. The image set of the target lane is obtained through a high-precision traffic image acquisition device. By performing vehicle type recognition and vehicle distance recognition on the image set, the vehicle types on the lane and the distances between adjacent vehicles can be accurately extracted, and a congestion coefficient of the road can be generated, providing a data basis for subsequent traffic signal control decisions. When receiving information about a sudden traffic event, the event type and location are located, so as to make a response in the first time when the incident occurs. Based on the type and location of the event, the impact range of the event on the surrounding traffic is predicted. This prediction ability can timely adjust the traffic flow control strategy when a sudden event occurs, avoiding the further expansion of traffic congestion. By combining the analysis of the road congestion coefficient and the predicted impact range of traffic events, more accurate traffic signal control decisions are made. This decision-making method can adjust the signal cycle in real time, optimize the traffic flow, reduce unnecessary vehicle detention and congestion, and reduce the traffic pressure during peak hours and traffic problems caused by sudden events.

[0009] The above description is only an overview of the technical solution of this application. In order to be able to understand the technical means of this application more clearly, it can be implemented in accordance with the content of the description. And in order to make the above and other purposes, features, and advantages of this application more obvious and understandable, the specific embodiments of this application are specifically given below. Brief Description of the Drawings

[0010] Figure 1 It is a schematic flowchart of a precision perception and decision-making integrated traffic control method provided by an embodiment of this application.

[0011] Figure 2 It is a schematic structural diagram of a precision perception and decision-making integrated traffic control system provided by an embodiment of this application.

[0012] Figure 3 It is a schematic structural diagram of an exemplary computer device provided by an embodiment of this application.

[0013] Description of the reference numerals: Scene model construction module 10, lane image acquisition module 20, vehicle distance recognition module 30, traffic event location module 40, impact range prediction module 50, control decision module 60, bus 300, receiver 301, processor 302, transmitter 303, memory 304, bus interface 305. Detailed Embodiments

[0014] Embodiments of the present application provide a traffic control method, system and device for integrated precision perception and decision-making, which solve the technical problem that existing traffic control systems are usually set to control traffic lights at fixed time intervals, ignoring the traffic demand differences in different time periods and different road conditions, and being unable to make flexible adjustments in case of sudden traffic events, resulting in congestion or traffic accidents.

[0015] After introducing the basic principle of the present application, various non-limiting embodiments of the present application will be specifically introduced below in conjunction with the accompanying drawings of the specification. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0016] Embodiment 1, as Figure 1 shown, embodiments of the present application provide a traffic control method for integrated precision perception and decision-making, and the method includes:

[0017] Collect high-precision traffic flow information, pedestrian and non-motor vehicle information within a preset lane range ending at a target traffic light, and construct a scene model for the preset range of the target traffic light.

[0018] Determine the target traffic light for which data is to be collected and its preset lane range, including selecting a research area, defining lane ranges such as left-turn lanes, straight lanes, right-turn lanes, etc., and the time period of the scene, such as peak hours and off-peak hours. Use high-precision sensors such as lidar, ultrasonic sensors or high-definition cameras to obtain traffic flow data on the lane. The flow information includes vehicle speed, lane occupancy rate, number of vehicles, etc.; collect the flow of pedestrians and non-motor vehicles within the collection area. A vision recognition system can be used to monitor the positions, movement trajectories and numbers of pedestrians and non-motor vehicles, and identify information such as the speeds, flow densities and residence times of these targets.

[0019] Utilize the collected data to construct a high-precision traffic flow model through data fusion and analysis. Traditional statistical models can be used to process traffic flow, pedestrian and non-motor vehicle data to generate a dynamic scene model. This model should be able to present the changes of the target traffic light under different traffic conditions, including traffic flow, pedestrian flow, etc. during peak hours, off-peak hours and different weather conditions.

[0020] Connect a high-precision traffic image acquisition device to collect and obtain a set of high-precision lane images for the preset lane range.

[0021] Install and calibrate high-precision traffic image acquisition devices, such as high-definition cameras, panoramic cameras, etc., to ensure that the devices can cover the target lane range. Pay attention to the image quality under different time periods, weather conditions, etc. when collecting images. Capture traffic images within the lane range in real time, and perform image processing and enhancement on the collected high-precision images, such as denoising, contrast adjustment, image stabilization, etc., to ensure the accuracy of subsequent analysis. Through image acquisition and processing, obtain a set of high-precision lane images.

[0022] Based on the set of high-precision lane images, perform vehicle type recognition and vehicle distance recognition, and generate a road congestion coefficient according to the recognition results.

[0023] Use image recognition technology to process lane images and identify different vehicle types in the images. Common vehicle type recognition methods include object detection based on convolutional neural networks. The model can be based on a pre-trained vehicle classification model or can be customized through a training set to adapt to vehicle type classification in a specific environment, accurately identifying different vehicle types in the lane, such as small cars, large cars, trucks, etc.

[0024] Vehicle distance recognition is to calculate the distance between adjacent vehicles through the spatial information in the image. According to the recognized vehicle type information and vehicle distance data, combined with traffic flow and road lane occupancy, calculate the road congestion coefficient. This coefficient reflects the vehicle density and traffic flow on the road. Through the calculated congestion coefficient, the traffic conditions at different time periods and sections can be displayed in real time, including congestion, smooth or slow states.

[0025] When receiving information about a sudden traffic event, locate the type and location of the sudden traffic event.

[0026] Information about sudden traffic events usually comes from multiple data sources, such as traffic signal control systems, traffic monitoring cameras, sensor networks, emergency response systems, or other traffic management platforms. Event information includes types of emergencies such as car accidents, traffic jams, road construction, weather changes, etc., and is usually transmitted to the system in the form of alarms, signals, or direct data transmission.

[0027] According to the received event information, first identify the type of the event. By analyzing changes in traffic flow and abnormal behaviors appearing in the images, such as vehicle collisions, the appearance of obstacles, etc., automatically judge the type of sudden event. For example, a car accident event may be manifested as a sudden traffic standstill and vehicles deviating from the normal driving trajectory, while a traffic jam is manifested as a long-term high traffic flow density. Then, according to the position of the traffic camera and the real-time image data collected, locate the occurrence location of the sudden event. For example, by calculating the coverage area of the traffic camera and combining the time point when the event occurred, determine the accurate location where the event occurred, such as a specific section, lane, or intersection.

[0028] Based on the types and locations of the sudden traffic events, predict the affected range to obtain the predicted affected range of the traffic events.

[0029] According to the types and occurrence locations of the sudden traffic events, analyze the characteristics of the events and their potential impacts on the surrounding traffic flow. For example, a traffic accident may cause the closure of one or more lanes, while road construction may reduce the traffic capacity but not necessarily lead to a complete blockage. Utilize the records of historical traffic events to analyze the affected ranges of past similar events. This can be achieved through regression analysis or machine learning methods on the data of historical events to predict the affected range of the event output, including the roads, lanes, and areas with traffic flow changes affected. Further, visualization tools can be used to display the predicted affected range, showing which road sections will be significantly affected to assist decision-makers in making decisions.

[0030] Based on the predicted affected range of the traffic events, using the road congestion coefficient as the decision reference coefficient, conduct traffic signal control decision analysis in the preset range scenario model of the target traffic signal to obtain the target control parameters, and make the control decision of the target traffic signal based on the target control parameters.

[0031] Based on the predicted affected range of the traffic events, conduct a comprehensive analysis in combination with the road congestion coefficient. It is predicted that the traffic flow within the affected range will be directly affected by the sudden event, so the congestion coefficient will change significantly. For the affected area, calculate the congestion conditions of each lane and each signal intersection. If the traffic flow at a certain intersection increases significantly, such as due to an accident or road closure causing vehicles to detour, the congestion coefficient of this area will increase. Combine the preset range scenario model of the target traffic signal to analyze the current traffic conditions. According to the affected range of the traffic event, the predicted congestion coefficient, and the current traffic conditions, conduct traffic signal control decision analysis with the goal of optimizing the traffic flow, minimizing congestion, and improving the road traffic efficiency. Exemplarily, according to the change of the congestion coefficient, adjust the green light and red light durations of the signal lights. For example, if the traffic flow of a certain lane surges, the green light duration of this lane can be automatically extended, or traffic flow balance can be achieved through dynamic signal control; within the event affected range, guide the traffic flow to the unaffected roads or lanes through signal control to relieve congestion; if the event occurs in an area where emergency vehicles need to pass, provide a priority passage for rescue vehicles, police cars, etc., which can be achieved through a rapid signal switching control strategy.

[0032] According to the above analysis results, the optimized target control parameters are calculated, including the adjusted signal cycle, green light duration, red light duration, etc., and they are transmitted to the target traffic signal control system for the control decision of the target traffic signal. Through these steps, the control strategy of the traffic signal can be flexibly adjusted according to the predicted influence range of real-time traffic events and the road congestion situation, so as to improve the traffic flow capacity, relieve road congestion, and ensure traffic order and safety.

[0033] Furthermore, before constructing the target traffic signal preset range scenario model, the method includes:

[0034] Locate the adjacent traffic signals of the target traffic signal; obtain the traffic facility information between the target traffic signal and the adjacent traffic signals, where the traffic facility information includes road signs, speed limit signs, zebra crossings, non-motor vehicle lanes, and bus lanes; based on the traffic facility information, collect the sample traffic road data set; perform lane boundary accuracy perception on the sample traffic road data set to obtain the preset lane range.

[0035] Using map and geographic information system technology, locate the adjacent traffic signals of the target traffic signal. Adjacent traffic signals usually refer to other signals that are adjacent to the target traffic signal in the geographical space. These signals are usually located around the intersection where the target signal is located or at adjacent intersections, constituting the key nodes of the traffic flow.

[0036] Collect all traffic facility information between the target traffic signal and its adjacent traffic signals. These information include but are not limited to road signs, speed limit signs, zebra crossings, non-motor vehicle lanes, and bus lanes. Among them, road signs such as stop signs, turning indication signs, and no-entry signs; speed limit signs are the identification information of road speed limits, which affect the driving speed of vehicles.

[0037] The sample traffic road data set refers to a data set formed by collecting traffic flow, driving speed, traffic density, etc. of the roads near the target traffic signal and its associated signals. This data set includes all data related to traffic facility information. Specifically, combining traffic facility information such as road signs, speed limit signs, and zebra crossings, analyze how they affect traffic flow. For example, speed limit signs will affect the driving speed of vehicles, and zebra crossings will affect the parking and driving processes of vehicles. Integrate the collected data to construct the sample traffic road data set. The data set includes not only lane flow information but also multi-faceted data such as traffic facility information, road conditions information, and weather conditions. These information helps to further analyze the relationship between traffic flow and signal control.

[0038] By using high-precision image recognition technology, analyze the image data in the sample traffic road dataset to identify the boundaries of lanes. This process can be based on computer vision technology, such as convolutional neural networks, for automatic detection and positioning of lane lines. Detect the accuracy of lane boundaries through algorithms and calibrate according to road conditions and facility characteristics. For example, special processing is required for more complex intersections or areas blocked by obstacles to ensure the accuracy of lane boundaries. Utilize the collected traffic facility information, such as road signs and markings, to further optimize lane boundary perception. By comparing with traffic facilities, the recognition accuracy of lane boundaries can be improved and errors reduced. Based on the accuracy perception of lane boundaries, calculate the specific range of the lane, including the width, length, and changes in the boundary lines of the lane. This information will be used as the preset lane range to assist the traffic signal control system in making accurate decisions.

[0039] Furthermore, the method for constructing the preset range scene model of the target traffic signal includes:

[0040] Locate the target traffic signal to generate the end coordinates; based on the end coordinates, extend forward with the preset lane range to obtain the lane extension coordinates; construct the scene acquisition range based on the lane extension coordinates; according to the scene acquisition range, collect the high-precision traffic flow information, pedestrian and non-motor vehicle information, and construct the preset range scene model of the target traffic signal.

[0041] By using a geographic information system or a positioning system, accurately locate the position of the target traffic signal and determine the end position of the section or intersection where the target signal is located. These end coordinates represent the exact position of the road where the target traffic signal is located and will be used as the reference point for subsequent analysis and acquisition.

[0042] The preset lane range defines the traffic road area within a certain distance starting from the target traffic signal. These lane ranges take into account the number of lanes, lane width, and traffic flow requirements at intersections. According to the end coordinates, extend in the road direction. During the extension process, a straight or curved path model can be adopted to ensure that the extended lane area can cover the traffic range controlled by the signal. This process can be combined with road design standards to ensure the accuracy of the lane extension coordinates. Based on the end coordinates of the target traffic signal, combined with the road geometry and extension rules, calculate the lane extension coordinates. These coordinates represent the road position after extending a certain range forward from the target signal, forming the extended range of the lane.

[0043] The scene collection range refers to the actual collection area determined based on the lane extension coordinates. Usually, it is the lane range extending forward from the target traffic signal, combined with a certain width to form the collection area. This range will determine the geographical area that needs to be concerned when collecting traffic flow, pedestrian, and non-motor vehicle information. This range can be dynamically adjusted according to traffic management needs.

[0044] Within the constructed scene collection range, traffic flow information is collected by using devices such as high-precision cameras, traffic sensors (such as radar, lidar), and ground sensors. The data that needs to be focused on collecting are the traffic flow, vehicle speed, lane occupancy, etc. on each lane. In addition, information such as the type, quantity, and driving trajectory of vehicles in the traffic flow is obtained through image recognition technology and sensor detection. Besides motor vehicle information, cameras and sensors are used to accurately monitor the flow, movement trajectory, and staying situation of pedestrians crossing zebra crossings, non-motor vehicle lanes, etc., to help analyze the behavior patterns of different traffic participants.

[0045] Based on the collected traffic flow information, pedestrian, and non-motor vehicle data, combined with the geometric information of the lane and the configuration of traffic facilities, etc., a scene model of the area where the target traffic signal is located is constructed. The scene model includes information such as the control parameters of the target traffic signal, traffic flow, pedestrian flow, non-motor vehicle flow, etc., and can reflect the real-time traffic condition. The model can be optimized through data processing, simulation prediction, and machine learning algorithms to ensure its application effect in actual traffic management.

[0046] Furthermore, for vehicle type recognition and vehicle distance recognition based on the high-precision lane image set, and generating a road congestion coefficient according to the recognition results, the method includes:

[0047] Based on the high-precision lane image set, extract the first lane image sequence within a preset time window; input the first lane image sequence into a pre-constructed vehicle type recognizer for vehicle image segmentation and vehicle type recognition to obtain multiple recognized vehicle type information, where the vehicle type recognizer includes an image segmentation channel and a vehicle type recognition channel; according to the multiple recognized vehicle type information, mark the vehicles in the first lane image sequence, calculate the time difference between adjacent vehicles passing through a preset position based on the vehicle marking result to obtain multiple vehicle distance recognition information; perform road congestion analysis according to the multiple vehicle distance recognition information to obtain the road congestion coefficient.

[0048] The preset time window is a continuous time period during which image data is selected from the high-precision lane image set for analysis. For example, the time window can be set from several seconds to several minutes, depending on the requirements of traffic flow analysis. The first lane image sequence extracted is for consecutive image frames within the target lane, and each frame in the sequence represents the traffic state of the lane at that moment, helping the system to perform dynamic analysis of the traffic conditions.

[0049] Image segmentation refers to separating each object in the image from the background. First, the model identifies the vehicle area through the image segmentation channel, separating the vehicle area in the lane image from other areas. After vehicle image segmentation, the segmented vehicle image area is processed through the vehicle type recognition channel. The vehicle type recognition channel classifies the vehicle image through a pre-trained model to identify the type of the vehicle, and finally obtains multiple recognized vehicle type information, each piece of information including the type, position, and appearance time of the vehicle.

[0050] Based on the multiple recognized vehicle type information obtained, the position of each vehicle in the image sequence is marked. These marks include not only the position of the vehicle, but also information such as the appearance order, speed, and driving direction of each vehicle on the time axis. After vehicle marking is completed, the vehicle distance is calculated according to the time difference between two adjacent vehicles passing through a preset position. The preset position can be a fixed point in the image or a reference position on the physical road section. By calculating the time difference between adjacent vehicles passing through this position, the relative vehicle speed and vehicle distance between vehicles are deduced. The smaller the time difference, the closer the vehicle distance, and vice versa. According to the time difference between adjacent vehicles, multiple vehicle distance recognition information is obtained, each piece of vehicle distance recognition information including the specific distance and time interval between two adjacent vehicles.

[0051] Based on multiple vehicle distance recognition information, road congestion analysis is carried out. Generally, a smaller vehicle distance and a shorter time interval between adjacent vehicles mean a larger traffic flow, which may lead to road congestion. Congestion analysis can be carried out by calculating the average value, standard deviation of the vehicle distance or analyzing the traffic flow density. If the vehicle distance remains small and multiple vehicle distance recognition information indicates that the vehicles are very close, it can be judged that there may be congestion on this section. Through the above analysis, a road congestion coefficient is obtained, which represents the degree of road congestion. When the coefficient is high, it indicates that the traffic flow on the road is too large and serious congestion may occur; when the coefficient is low, it indicates that the road traffic condition is good.

[0052] Furthermore, to construct the image segmentation channel, the method includes:

[0053] Obtain the historical lane image set of the high-precision traffic image acquisition device; segment and identify the vehicle images in the historical lane image set to obtain the historical image segmentation result set; based on semantic segmentation, construct an image segmentation model, and use the historical lane image set and the historical image segmentation result set to train the image segmentation channel until convergence to obtain the image segmentation channel.

[0054] Through the high-precision traffic image acquisition device, obtain the historical lane image set, which is collected by fixed cameras on the lane during a certain period in the past and contains vehicle image information under different times and different traffic conditions.

[0055] When processing the historical lane image set, first separate the vehicles in the image from the background through an image segmentation algorithm, specifically classify each pixel in the lane image as a vehicle or the background, so as to obtain the accurate contour of the vehicle. After segmentation and identification in each frame of the historical lane image, a segmentation result set is generated, and each segmentation result contains image data marked with the vehicle position and classification information.

[0056] Based on the semantic segmentation method, construct an image segmentation model. The goal of semantic segmentation is to classify each pixel in the image into the corresponding category, which is a vehicle or the background here. The input of the model is the historical lane image set, and the output is the pixel-level classification of each image. The semantic segmentation model extracts image features and reconstructs the segmentation result through an encoder-decoder structure. Use the historical lane image set and the historical image segmentation result set to train the segmentation model. During the training process, the model will use the image and the corresponding segmentation label for supervised learning to optimize the model parameters to achieve better segmentation accuracy. The training process continues until the loss function of the model reaches a lower value and the segmentation result tends to be stable, indicating that the model has converged. At this time, the model can accurately segment the vehicles in the image and has good performance. The trained model is the final image segmentation channel, which can process new high-precision lane images and automatically perform vehicle segmentation and identification to generate accurate segmentation results.

[0057] Furthermore, to construct the vehicle type recognition channel, the method includes:

[0058] Identify and label the vehicle types in the historical image segmentation result set to obtain the vehicle type recognition result set; based on the deep convolutional network, construct a vehicle type recognition model, and use the historical image segmentation result set and the vehicle type recognition result set to train the vehicle type recognition model until convergence to obtain the vehicle type recognition channel.

[0059] Vehicle type recognition refers to further classifying each vehicle on an image where vehicle segmentation has already been performed, such as identifying types like sedans, SUVs, trucks, electric vehicles, etc., to obtain a set of vehicle type recognition results.

[0060] A vehicle type recognition model is constructed based on a deep convolutional network. The purpose of this model is to perform fine-grained classification on the images of each vehicle, accurately identify different vehicle types. The input data is the vehicle images after image segmentation processing. Each image represents a vehicle area that has been separated from the background, and the image only contains the features of the vehicle part. The training objective of the model is to enable the model to output vehicle type labels consistent with the historical vehicle type recognition result set according to the input images. During the training process, the model gradually improves the recognition accuracy of vehicle types by repeatedly learning the historical image segmentation results and vehicle type recognition results. As the training progresses, the model continuously adjusts the weight parameters of convolutional layers, fully connected layers, etc. Eventually, it achieves high-precision recognition of unknown vehicle images. The convergence of the model is reflected in the training error and validation error tending to be stable, the loss function value decreasing, and the recognition accuracy rate output by the model reaching the expected level. After training is completed, the model is the final vehicle type recognition channel, and this model can input the segmented vehicle images and quickly identify and calibrate the vehicle types.

[0061] Furthermore, for the method of performing road congestion analysis based on the multiple vehicle distance recognition information to obtain the road congestion coefficient, it further includes:

[0062] Construct a vehicle type congestion comparison table, where the vehicle type congestion comparison table includes the mapping relationships between multiple preset vehicle types and multiple preset traffic diversion obstruction coefficients; traverse the vehicle type congestion comparison table according to the multiple recognized vehicle type information to obtain multiple traffic diversion obstruction coefficients; calculate the average value of the multiple traffic diversion obstruction coefficients, and correct the road congestion coefficient according to the calculation result.

[0063] Construct a vehicle type congestion comparison table. The vehicle type congestion comparison table is a mapping table that records the relationship between different types of vehicles and the degree of traffic congestion. Each preset vehicle type, such as sedans, trucks, large SUVs, etc., is associated with a corresponding preset traffic diversion obstruction coefficient. The traffic diversion obstruction coefficient is used to quantify the impact of different types of vehicles on traffic flow. For example, large trucks occupy more road space and have a slower speed, so they will have a greater obstructive effect on traffic flow, and the corresponding traffic diversion obstruction coefficient will be higher.

[0064] Search for the corresponding traffic diversion obstruction coefficients in the vehicle type congestion comparison table according to the multiple recognized vehicle type information. Specifically, traverse the vehicle type congestion comparison table according to the type of each vehicle. For each vehicle, find the obstruction coefficient corresponding to its type in the comparison table to obtain multiple traffic diversion obstruction coefficients, and each coefficient corresponds to the impact of a vehicle type on traffic flow.

[0065] Calculate the average value of the obtained multiple traffic flow obstruction coefficients. The average value reflects the overall impact degree of different vehicle types on the traffic flow within a specific area or time period. This average value is used to describe the overall congestion situation of the traffic flow because different types of vehicles have different congestion impacts, and the average value calculation can comprehensively reflect these impacts. Use the calculated average value of the traffic flow obstruction coefficients to correct the overall congestion coefficient of the road. The original road congestion coefficient is calculated based on factors such as vehicle flow, vehicle speed, and vehicle distance. However, by introducing the average value of the traffic flow obstruction coefficients, a more refined correction of the vehicle type impact in the traffic flow can be carried out. In this way, the traffic flow obstruction caused by different vehicle types can be more accurately reflected, and thus the congestion degree of the road can be more precisely evaluated.

[0066] Furthermore, the method further includes:

[0067] Evaluate the control effect of the traffic signal based on the target control parameters at a preset time node; when the evaluation result does not meet the preset effect, optimize and adjust the target control parameters based on the evaluation result.

[0068] The preset time node refers to a predefined specific time point, usually during some key periods, such as peak hours, off-peak hours, after an accident, etc., or an evaluation after a specific event. The selection of these time nodes usually depends on factors such as the change pattern of traffic flow, the frequency of traffic accidents, and the daily traffic peak hours. For example, the evaluation can be carried out at the whole hour of each hour, at the start or end of the traffic peak period, or according to the dynamic changes of real-time traffic monitoring.

[0069] At each preset time node, use the current target control parameters to perform target traffic signal control and evaluate the control effect. The evaluation indicators include traffic flow, road congestion, traffic efficiency, vehicle speed, and vehicle distance. During the evaluation process, generate an evaluation report by integrating the above indicators to show the actual effect of the traffic signal control decision.

[0070] If the evaluation result fails to achieve the expected effect, optimize and adjust the target control parameters according to the evaluation result. The optimization goal is to improve traffic flow according to the actual traffic situation, avoid congestion, and improve traffic efficiency. For example, if the signal duration of a certain lane is unreasonable, the green light cycle can be appropriately extended, or the red light cycle can be shortened to improve the traffic capacity; if the traffic flow at some intersections is large, more passing time can be preferentially provided for these intersections or lanes.

[0071] Through these steps, the control system of the traffic signal can be continuously optimized to cope with different traffic flow changes and actual traffic demands, maximize the traffic efficiency of the road, and reduce the occurrence of congestion and accidents.

[0072] In summary, the precision perception and decision-making integrated traffic control method provided by the embodiments of the present application has the following technical effects:

[0073] By collecting traffic flow, pedestrian, and non-motor vehicle information within the preset lane range covered by the target traffic signal, this high-precision collection can ensure that the real situation of traffic flow is reflected in real time. Based on the collected traffic information, a preset range scene model of the area where the target traffic signal is located is constructed. This model contains multi-dimensional information such as traffic flow, pedestrians, and non-motor vehicles, which helps to comprehensively understand the road traffic conditions and provides a scenario-based reference for control decisions; by using a high-precision traffic image acquisition device to obtain an image set of the target lane, and through vehicle type recognition and vehicle distance recognition of the image set, the vehicle types on the lane and the distance between adjacent vehicles can be accurately extracted, generating a congestion coefficient of the road, which provides a data basis for subsequent traffic signal control decisions; when receiving sudden traffic event information, the event type and location are located, so as to make a response in the first time when the event occurs. Based on the type and location of the event, the influence range of the event on the surrounding traffic is predicted. This prediction ability can timely adjust the traffic flow control strategy when a sudden event occurs, avoiding the further expansion of traffic congestion; by combining the analysis of the road congestion coefficient and the predicted influence range of traffic events, a more accurate traffic signal control decision is made. This decision-making method can adjust the signal cycle in real time, optimize the traffic flow, reduce unnecessary vehicle detention and congestion, and reduce the traffic pressure during peak hours and traffic problems caused by sudden events.

[0074] Embodiment 2, based on the same inventive concept as the precision perception and decision-making integrated traffic control method in the foregoing embodiment, as Figure 2 shown, the embodiments of the present application provide a precision perception and decision-making integrated traffic control system, and the system includes:

[0075] The scene model construction module 10 is used to collect high-precision traffic flow information, pedestrian and non-motor vehicle information within a preset lane range with the target traffic signal as the end point, and construct a scene model for the preset range of the target traffic signal; the lane image acquisition module 20 is used to connect to a high-precision traffic image acquisition device and acquire a set of high-precision lane images of the preset lane range; the vehicle distance recognition module 30 is used to perform vehicle type recognition and vehicle distance recognition based on the set of high-precision lane images, and generate a road congestion coefficient according to the recognition results; the traffic event positioning module 40 is used to, when receiving sudden traffic event information, locate the type and location of the sudden traffic event; the influence range prediction module 50 is used to perform influence range prediction based on the type of the sudden traffic event and the location of the sudden traffic event to obtain the predicted influence range of the traffic event; the control decision-making module 60 is used to perform traffic signal control decision analysis in the scene model of the preset range of the target traffic signal according to the predicted influence range of the traffic event, with the road congestion coefficient as the decision reference coefficient, to obtain target control parameters, and perform control decision-making on the target traffic signal based on the target control parameters.

[0076] Furthermore, the system further includes a preset lane range acquisition module to perform the following operation steps:

[0077] Locate the adjacent traffic signals of the target traffic signal; obtain the traffic facility information between the target traffic signal and the adjacent traffic signals, where the traffic facility information includes road signs, speed limit signs, zebra crossings, non-motor vehicle lanes, and bus lanes; based on the traffic facility information, collect a sample traffic road data set; perform lane boundary accuracy perception on the sample traffic road data set to obtain the preset lane range.

[0078] Furthermore, the system further includes a preset range scene model construction module to perform the following operation steps:

[0079] Locate the target traffic signal and generate end point coordinates; based on the end point coordinates, extend forward with the preset lane range to obtain lane extension coordinates; construct a scene acquisition range based on the lane extension coordinates; according to the scene acquisition range, collect the high-precision traffic flow information, pedestrian and non-motor vehicle information, and construct the scene model for the preset range of the target traffic signal.

[0080] Furthermore, the system further includes a road congestion coefficient acquisition module to perform the following operation steps:

[0081] Based on the set of high-precision lane images, extract the first lane image sequence within a preset time window; input the first lane image sequence into a pre-constructed vehicle type recognizer for vehicle image segmentation and vehicle type recognition to obtain multiple recognized vehicle type information, where the vehicle type recognizer includes an image segmentation channel and a vehicle type recognition channel; according to the multiple recognized vehicle type information, mark the vehicles in the first lane image sequence, calculate the time difference between adjacent vehicles passing through a preset position based on the vehicle marking result to obtain multiple vehicle distance recognition information; perform road congestion analysis based on the multiple vehicle distance recognition information to obtain the road congestion coefficient.

[0082] Furthermore, the system further includes an image segmentation channel construction module to perform the following operating steps:

[0083] Obtain the historical lane image set of the high-precision traffic image acquisition device; segment and identify the vehicle images in the historical lane image set to obtain a set of historical image segmentation results; based on semantic segmentation, construct an image segmentation model, and use the historical lane image set and the set of historical image segmentation results to train the image segmentation channel until convergence to obtain the image segmentation channel.

[0084] Furthermore, the system further includes a vehicle type recognition channel construction module to perform the following operating steps:

[0085] Identify and mark the vehicle types in the set of historical image segmentation results to obtain a set of vehicle type recognition results; based on a deep convolutional network, construct a vehicle type recognition model, and use the set of historical image segmentation results and the set of vehicle type recognition results to train the vehicle type recognition model until convergence to obtain the vehicle type recognition channel.

[0086] Furthermore, the system further includes a correction module to perform the following operating steps:

[0087] Construct a vehicle type congestion comparison table, where the vehicle type congestion comparison table includes the mapping relationships between multiple preset vehicle types and multiple preset traffic guidance obstruction coefficients; traverse the vehicle type congestion comparison table according to the multiple recognized vehicle type information to obtain multiple traffic guidance obstruction coefficients; calculate the average value of the multiple traffic guidance obstruction coefficients, and correct the road congestion coefficient according to the calculation result.

[0088] Furthermore, the system further includes an optimization and adjustment module to perform the following operating steps:

[0089] Evaluate the traffic signal control decision effect based on the target control parameters at a preset time node; when the evaluation result does not meet the preset effect, perform optimization and adjustment of the target control parameters based on the evaluation result.

[0090] Through the foregoing detailed description of a traffic control method integrating precision perception and decision-making, those skilled in the art can clearly understand a traffic control system integrating precision perception and decision-making in this embodiment. Since it corresponds to the method disclosed in the embodiment, the description is relatively simple. For related parts, reference may be made to the description in the method section.

[0091] Embodiment 3, as Figure 3 shown, is a schematic structural diagram of an exemplary computer device of the present application. In Figure 3 , the bus architecture is represented by bus 300. Bus 300 may include any number of interconnected buses and bridges. Bus 300 connects various circuits including one or more processors represented by processor 302 and a memory represented by memory 304 together. Bus 300 may also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art. Therefore, further description thereof will not be provided herein. Bus interface 305 provides an interface between bus 300 and receiver 301 and transmitter 303. Receiver 301 and transmitter 303 may be the same element, i.e., a transceiver, providing a unit for communicating with various other devices over a transmission medium. Processor 302 is responsible for managing bus 300 and general processing, while memory 304 may be used to store data used by processor 302 when performing operations.

[0092] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0093] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A traffic control method integrating precision perception and decision making, characterized in that: The method comprises: Collect high-precision traffic flow information, pedestrian and non-motor vehicle information in a preset lane range with the target traffic light as the end point, and build a scene model of the preset range of the target traffic light; Connecting to a high-precision traffic image acquisition device to acquire a set of high-precision lane images of the preset lane range; Based on the high-precision lane image set, vehicle type recognition and vehicle distance recognition are performed, and a road congestion coefficient is generated according to the recognition results; When receiving traffic emergency information, locate the type and location of the traffic emergency; Based on the type of the sudden traffic event and the location of the sudden traffic event, predict the impact range to obtain the predicted impact range of the traffic event; According to the predicted impact range of the traffic event and taking the road congestion coefficient as a decision reference coefficient, a traffic light control decision analysis is performed in a preset range scenario model of the target traffic light to obtain target control parameters, and a control decision of the target traffic light is performed based on the target control parameters.

2. The method for integrated precision perception and decision-making traffic control according to claim 1, characterized in that: Before constructing the target traffic light preset range scene model, the method includes: Locating adjacent traffic lights of the target traffic light; Acquire traffic facility information between the target traffic light and the adjacent traffic light, wherein the traffic facility information includes road signs, speed limit signs, zebra crossings, non-motorized vehicle lanes, and bus lanes; Based on the traffic facility information, collecting a sample traffic road data set; Lane boundary accuracy perception is performed on the sample traffic road data set to obtain the preset lane range.

3. The method for integrated precision perception and decision-making traffic control according to claim 1, characterized in that: The method for constructing a preset range scene model of a target traffic light comprises: Locating the target traffic light and generating the destination coordinates; Based on the end point coordinates, extending forward within the preset lane range to obtain lane extension coordinates; Constructing a scene acquisition range based on the lane extension coordinates; According to the scene collection range, the high-precision traffic flow information, pedestrian and non-motor vehicle information are collected to construct a preset range scene model of the target traffic light.

4. The method for integrated precision perception and decision-making traffic control according to claim 1, characterized in that: The method of performing vehicle type recognition and vehicle distance recognition based on the high-precision lane image set and generating a road congestion coefficient according to the recognition result includes: Based on the high-precision lane image set, extracting a first lane image sequence within a preset time window; Inputting the first lane image sequence into a pre-built vehicle type identifier to perform vehicle image segmentation and vehicle type identification to obtain a plurality of identified vehicle type information, wherein the vehicle type identifier includes an image segmentation channel and a vehicle type identification channel; According to the multiple identification vehicle type information, the first lane image sequence is marked with vehicles, and according to the vehicle marking result, the time difference between adjacent vehicles passing through the preset position is calculated to obtain multiple vehicle distance identification information; A road congestion analysis is performed according to the plurality of vehicle distance recognition information to obtain the road congestion coefficient.

5. The method for integrated precision perception and decision-making traffic control according to claim 4, characterized in that: Constructing the image segmentation channel, the method includes: Acquire a historical lane image collection from a high-precision traffic image acquisition device; Segmenting and marking vehicle images in the historical lane image set to obtain a historical image segmentation result set; Based on semantic segmentation, an image segmentation model is constructed, and the image segmentation channel is trained using the historical lane image set and the historical image segmentation result set until convergence, so as to obtain the image segmentation channel.

6. A precision perception and decision-making integrated traffic control method as claimed in claim 5, characterized in that: The vehicle type identification channel is constructed, and the method includes: Identifying and labeling the vehicle types in the historical image segmentation result set to obtain a vehicle type recognition result set; Based on a deep convolutional network, a vehicle type recognition model is constructed, and the vehicle type recognition model is trained using the historical image segmentation result set and the vehicle type recognition result set until convergence, thereby obtaining the vehicle type recognition channel.

7. The method for integrated precision perception and decision-making traffic control according to claim 4, characterized in that: The method further comprises: performing a road congestion analysis according to the plurality of vehicle distance recognition information to obtain the road congestion coefficient; Constructing a vehicle type congestion comparison table, wherein the vehicle type congestion comparison table includes a mapping relationship between a plurality of preset vehicle types and a plurality of preset diversion obstruction coefficients; According to the plurality of identified vehicle type information, the vehicle type congestion comparison table is traversed to obtain a plurality of diversion obstruction coefficients; The average of the plurality of traffic congestion obstacle coefficients is calculated, and the road congestion coefficient is corrected according to the calculation result.

8. The method for integrated precision perception and decision-making traffic control according to claim 1, characterized in that: The method further comprises: Conducting a traffic light control decision effect evaluation based on the target control parameters at a preset time node; When the evaluation result does not meet the preset effect, the target control parameter is optimized and adjusted based on the evaluation result.

9. A precision perception and decision-making integrated traffic control system, characterized in that: A system for implementing a precision perception and decision-making integrated traffic control method as described in any one of claims 1 to 8, comprising: A scene model building module is used to collect high-precision traffic flow information, pedestrian and non-motor vehicle information within a preset lane range with a target traffic light as the end point, and build a scene model of the preset range of the target traffic light; A lane image acquisition module, used to connect to a high-precision traffic image acquisition device to acquire a set of high-precision lane images within the preset lane range; A vehicle distance recognition module, used to perform vehicle type recognition and vehicle distance recognition based on the high-precision lane image set, and generate a road congestion coefficient according to the recognition result; A traffic incident locating module, used to locate the type and location of a traffic incident when receiving information about a traffic incident; An impact range prediction module, used to predict the impact range based on the type of the sudden traffic event and the location of the sudden traffic event, and obtain the predicted impact range of the traffic event; A control decision module is used to predict the impact range of the traffic event and use the road congestion coefficient as a decision reference coefficient to perform traffic light control decision analysis in a preset range scenario model of the target traffic light, obtain target control parameters, and make a control decision for the target traffic light based on the target control parameters.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: The processor executes the computer program to implement the steps of a precision perception and decision-making integrated traffic control method according to any one of claims 1 to 8.

Images