Lane level management and control method and device based on tunnel vehicle continuing algorithm

Abstract

The invention is suitable for the technical field of intelligent traffic, and provides a lane level management and control method and device based on a tunnel vehicle continuing algorithm. In the embodiment of the invention, the intelligent traffic platform firstly obtains a plurality of section acquisition data in real time through the front-end sensing device installed in the target tunnel, then performs feature transmission on each vehicle in the section acquisition data, determines the operation data of each vehicle, and transmits the operation data to the target tunnel. The method comprises the following steps: acquiring running data of vehicles, determining collision risks and lane changing probabilities of the vehicles and traffic states of the lanes according to the running data to obtain control strategies corresponding to the lanes, and finally controlling control equipment in a tunnel according to the control strategies to realize respective indication of the vehicles in different lanes. The corresponding lane level management and control strategy is generated according to the real-time traffic flow, the timeliness and effectiveness of early warning are improved, and therefore the accident occurrence probability is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent transportation, and in particular to a lane-level management and control method and device based on a tunnel vehicle connection algorithm. BACKGROUND

[0002] With the development of intelligent transportation, the safety management of tunnels, which are bottlenecks and accident-prone areas of road traffic, is particularly important. At present, tunnel management and control mainly relies on video monitoring, signal lights, variable information signs and other facilities to guide speed limiting, lane closure and other macro-level guidance. However, such methods have significant limitations: first, their perception ability is often fragmented, making it difficult to accurately and continuously track the complete driving trajectory, real-time speed and vehicle distance of vehicles in the tunnel, resulting in an inability to accurately assess transient and dynamic collision risks; second, traditional management and control strategies have a coarse granularity, usually taking the entire tunnel or large sections of the lane as a unit, and are unable to implement differentiated lane-level fine guidance for specific vehicles facing microscopic risks (such as frequent lane changes and too close following) on specific lanes, resulting in a high probability of accidents in tunnels. SUMMARY

[0003] In view of this, the present application provides a lane-level management and control method and device based on a tunnel vehicle connection algorithm to generate lane-level management and control strategies according to actual traffic flow changes in tunnels, thereby improving the timeliness and effectiveness of early warnings and reducing the probability of accidents.

[0004] The first aspect of the present application provides a lane-level management and control method based on a tunnel vehicle connection algorithm, applied to an intelligent transportation platform, the method comprising: acquiring a plurality of cross-section collection data through a front-end perception device in a target tunnel, and performing feature transmission on each vehicle in the cross-section collection data to determine the running data of each vehicle, wherein the cross-section collection data is used to represent the static characteristics of the vehicle, and the running data is used to represent the dynamic characteristics of the vehicle during driving; determining the collision risk, lane change probability of each vehicle and the traffic state of each lane through the running data, and then fusing the collision risk, lane change probability and traffic state to obtain multi-dimensional fusion data in the target tunnel; determining the management and control strategy of each lane in the target tunnel through the multi-dimensional fusion data, and indicating the vehicles in the target tunnel according to the management and control strategy.

[0005] Optionally, the cross-section collection data includes license plate information, and the feature transmission on each vehicle in the cross-section collection data to determine the running data of each vehicle comprises: For any target vehicle, license plate matching is performed on the section collection data according to the license plate information of the target vehicle to obtain target data of the target vehicle in the section collection data; The fusion frame data of the target vehicle is determined through the target data, and the running data of the target vehicle is determined according to the fusion frame data.

[0006] Optionally, the section collection data includes feature information of the vehicle, and the feature transmission is performed on each vehicle in the section collection data to determine the running data of each vehicle, including: For any target vehicle, feature matching is performed on the section collection data according to the feature information of the target vehicle to obtain target data of the target vehicle in the section collection data; The fusion frame data of the target vehicle is determined through the target data, and the running data of the target vehicle is determined according to the fusion frame data.

[0007] Optionally, the section collection data further includes feature information of the vehicle, and the feature transmission is performed on each vehicle in the section collection data to determine the running data of each vehicle, including: For any target vehicle, when the license plate information of the target vehicle cannot be determined, feature matching is performed on the section collection data according to the feature information of the target vehicle to obtain target data of the target vehicle in the section collection data; The fusion frame data of the target vehicle is determined through the target data, and the running data of the target vehicle is determined according to the fusion frame data.

[0008] Optionally, the determination of the collision risk, the lane changing probability and the traffic state of each lane through the running data includes: Collision prediction data is determined through the running data, and the collision risk level of each vehicle is determined through the collision prediction data, so as to determine the corresponding collision risk, wherein the collision prediction data includes the vehicle distance, the vehicle distance change rate and the relative speed of each vehicle; Lane changing prediction data is determined through the running data, and the lane changing probability of each vehicle is determined through the lane changing prediction data, wherein the lane changing prediction data includes the lateral position change trend and the vehicle distance; Lane state data is determined through the running data, and the traffic state of each lane is determined through the lane state data, wherein the lane state data includes the vehicle distance, the vehicle density and the average vehicle speed.

[0009] The second aspect of the application provides a lane-level control device based on a tunnel vehicle connection algorithm, applied to an intelligent traffic platform, the device comprising: running data determination unit, configured to acquire a plurality of section collection data by a front-end sensing device in a target tunnel, and perform feature transmission on each vehicle in the section collection data to determine running data of each vehicle, wherein the section collection data is used to represent static features of the vehicle, and the running data is used to represent dynamic features of the vehicle in driving; data fusion unit, configured to determine a collision risk, a lane changing probability and a traffic state of each lane in the target tunnel by the running data, and fuse the collision risk, the lane changing probability and the traffic state to obtain multi-dimensional fusion data in the target tunnel; strategy determination unit, configured to determine a management and control strategy of each lane in the target tunnel by the multi-dimensional fusion data, and instruct vehicles in the target tunnel according to the management and control strategy.

[0010] Optionally, the section collection data in the running data determination unit comprises license plate information, and the feature transmission on each vehicle in the section collection data to determine the running data of each vehicle comprises: for any target vehicle, performing license plate matching operation on each section collection data according to license plate information of the target vehicle to obtain target data of the target vehicle in each section collection data; determining fusion frame data of the target vehicle by the target data, and determining running data of the target vehicle according to the fusion frame data.

[0011] Optionally, the section collection data in the running data determination unit comprises feature information of the vehicle, and the feature transmission on each vehicle in the section collection data to determine the running data of each vehicle comprises: for any target vehicle, performing feature matching operation on each section collection data according to feature information of the target vehicle to obtain target data of the target vehicle in each section collection data; determining fusion frame data of the target vehicle by the target data, and determining running data of the target vehicle according to the fusion frame data.

[0012] Optionally, the section collection data in the running data determination unit further comprises feature information of the vehicle, and the feature transmission on each vehicle in the section collection data to determine the running data of each vehicle comprises: for any target vehicle, when license plate information of the target vehicle cannot be determined, performing feature matching operation on each section collection data according to feature information of the target vehicle to obtain target data of the target vehicle in each section collection data; The fusion frame data of the target vehicle is determined according to the target data, and the operation data of the target vehicle is determined according to the fusion frame data.

[0013] Optionally, the determining, by the data fusion unit, the collision risk of each vehicle, the lane changing probability of each vehicle, and the traffic state of each lane according to the operation data comprises: The collision prediction data is determined according to the operation data, and the collision risk level of each vehicle is determined according to the collision prediction data, so as to determine the corresponding collision risk, wherein the collision prediction data comprises the distance between vehicles, the distance change rate and the relative speed of each vehicle; The lane changing prediction data is determined according to the operation data, and the lane changing probability of each vehicle is determined according to the lane changing prediction data, wherein the lane changing prediction data comprises the lateral position change trend and the distance between vehicles; The lane state data is determined according to the operation data, and the traffic state of each lane is determined according to the lane state data, wherein the lane state data comprises the distance between vehicles, the vehicle density and the average vehicle speed.

[0014] In the embodiments provided in the present application, for a target tunnel, a plurality of section collection data are acquired in real time by a front-end sensing device installed in the tunnel, feature transmission is performed on each vehicle in the section collection data, operation data of each vehicle is determined, collision risk of each vehicle, lane changing probability of each vehicle, and traffic state of each lane are determined according to the operation data, so as to obtain a corresponding lane-level control strategy, and finally the control equipment in the tunnel is controlled according to the control strategy, so as to realize the indication of vehicles in different lanes. This realizes the generation of a corresponding lane-level control strategy according to real-time traffic flow, improves the timeliness and effectiveness of early warning, and reduces the probability of accidents. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 The method flowchart provided in the embodiments of the present application is provided; Figure 2 The device structure diagram provided in the embodiments of the present application is provided; Figure 3 The internal structure diagram of the computer device provided in the embodiments of the present application is provided. DETAILED DESCRIPTION

[0016] The exemplary embodiments will be described in detail herein below with reference to the drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present disclosure. Instead, they are merely examples of apparatuses and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0017] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting. As used in this application and the appended claims, the singular forms "a," "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will also be understood that the term "and / or," as used herein, refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0018] It will be understood that, although the terms first, second, third, etc. can be used herein to describe various information, these terms are not intended to denote a temporal or chronological order. Rather, these terms are used only as distinguishable identifiers, for example, to identify different types of information. For example, a first information can also be termed a second information, and, similarly, a second information can also be termed a first information, without departing from the scope of the application. Depending on the context, the word "if' as used herein can be interpreted to mean "when" or "in response to determining" or "in response to a determination."

[0019] The application provides a lane-level management and control method and device based on a tunnel vehicle connection algorithm to reduce the probability of accidents in the tunnel.

[0020] The technical solutions of the application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes can not be described in detail in some embodiments.

[0021] As shown in the flowchart of the lane-level management and control method based on a tunnel vehicle connection algorithm provided by the application, the flowchart can include the following steps: Figure 1 Step S101, acquiring a plurality of cross-section collection data through a front-end perception device in a target tunnel, and performing feature transmission on each vehicle in the cross-section collection data to determine the running data of each vehicle.

[0022] In this embodiment, the cross-section collection data is used to represent the static features of the vehicle, such as license plate number, vehicle color, vehicle model, vehicle shape, size, etc., for identifying the same vehicle in different cross-section collection data. The running data is used to represent the dynamic features of the vehicle in driving, such as vehicle speed, vehicle distance (including adjacent vehicle distance, front and rear vehicle distance, and distance from the adjacent lane, etc.), vehicle distance change rate, vehicle density, etc., for judging the current road conditions.

[0023] The front-end perception device can be a high-definition camera, and a high-definition camera can be installed at a fixed distance, such as 500 meters, to collect a plurality of cross-section collection data. After collection, the data is pre-processed, such as cleaning, completion and deduplication, etc. After removing invalid data, the cross-section collection data is sorted in time sequence.​

[0024] The process of determining the running data through feature transfer is as follows: When the vehicle passes through the first section, the system creates an independent ID and trajectory file for the vehicle according to its features, and then transfers and associates the identity ID of the vehicle through a real-time feature matching algorithm at the downstream section, thereby seamlessly integrating the passing time, speed, position and other information of the vehicle at each section, and finally constructing the complete running data of each vehicle in the whole tunnel range, and temporarily storing the running data in the computing memory, preparing for the subsequent data fusion and extraction link.

[0025] In another embodiment, the feature transfer is performed on each vehicle in the section collection data to determine the running data of each vehicle, including: For any target vehicle, license plate matching is performed on the section collection data according to the license plate information of the target vehicle to obtain target data of the target vehicle in the section collection data; The fusion frame data of the target vehicle is determined through the target data, and the running data of the target vehicle is determined according to the fusion frame data.

[0026] In this embodiment, the license plate number is taken as the identification mark, and the license plate recognition method can adopt the extraction logic based on contour detection + character segmentation. First, the Canny edge detection is used to locate the license plate area, and then the projection method is used to segment the characters, and the output is unified as the UTF-8 encoding format of “provincial abbreviation (2 characters) + letter / number (5-6 characters)”, and the invalid characters are automatically replaced with “*”. Then, the license plate full match operation is performed, that is, the license plate number is completely matched to identify the same vehicle, so as to obtain the target data of the same vehicle in the section collection data, such as vehicle image, spatial coordinates, time stamp, etc. Then, the sections are associated and fused under the unified time and space reference to generate a “fusion frame data” that can continuously describe the complete tunnel journey of the vehicle; finally, based on the fusion data, the displacement and time difference of the vehicle between different sections are calculated, and the key running data such as interval average speed, travel time and acceleration can be accurately analyzed. The specific method of identifying and determining the target data of the same vehicle in the section collection data is described in detail below.

[0027] In another embodiment, the feature transfer is performed on each vehicle in the section collection data to determine the running data of each vehicle, including: For any target vehicle, feature matching is performed on the section collection data according to the feature information of the target vehicle to obtain target data of the target vehicle in the section collection data; The fusion frame data of the target vehicle is determined according to the target data, and the operation data of the target vehicle is determined according to the fusion frame data.

[0028] The embodiment takes the characteristic information of the vehicle, such as the vehicle model, the vehicle body color, and the vehicle size, as the identification mark, and the specific identification method is as follows: 1. Vehicle head feature: key feature points are extracted by SIFT (Scale-Invariant Feature Transform) algorithm to generate 128-dimensional feature vectors, and the vectors are normalized to the interval [0, 1] to ensure that the feature vectors of different vehicles can be compared.

[0029] 2. Vehicle brand: based on a pre-trained ResNet-18 convolutional neural network, input the vehicle head cropped image (size 224x224), output the brand probability distribution (such as "BMW: 0.92, Mercedes-Benz: 0.05") through the fully connected layer, take the brand with probability ≥0.8 as the standardized result, otherwise mark it as "unknown".

[0030] 3. Vehicle color: convert the RGB image to HSV color domain, count the H channel (hue) distribution, match the preset standard color threshold (such as black: H 0-180, S 0-255, V 0-40; red: H 0-10 or 160-180, S 40-255, V 40-255), output the unique standard color (black / white / red / blue / silver gray / other).

[0031] For the collected characteristic information, it can also be optimized. For example, a sliding window filter can be used to smooth the collected instantaneous speed and eliminate interference. The output unit is unified as "km / h", and the first decimal place is retained. If the deviation of the collected speed is >20km / h for 3 consecutive times (the tunnel speed limit is usually ≤80km / h), it is marked as "abnormal speed" and does not participate in matching.

[0032] The abnormal values in the characteristic information can also be detected and then deleted, and the specific method is as follows: License plate: character length ≠7 (small car) or ≠8 (large car), contains 3 or more identical characters (such as "JingAAAAAA"), directly excluded; Vehicle head feature: SIFT feature point number <30 (judged as image blur), excluded; Speed: >100km / h (exceeds the normal tunnel speed limit) or <5km / h (judged as congestion and stagnation), excluded; Timestamp: the collection time difference of the same vehicle in the same section is >1s (judged as repeated collection), and the first collected data is retained.

[0033] Then, the feature matching operation is performed on the feature information of the vehicle to obtain target data of the same vehicle in the data collected at each section, and then the sections are associated and fused under a unified space-time reference to generate a "fusion frame data" capable of continuously describing the complete tunnel journey of the vehicle. Finally, based on the fusion data, the displacement and time difference of the vehicle between different sections are calculated to accurately analyze the interval average speed, travel time, acceleration and other key operation data of the vehicle.

[0034] In another embodiment, the feature transfer of each vehicle in the section data is performed to determine the operation data of each vehicle, including: For any target vehicle, when the license plate information of the target vehicle cannot be determined, the feature matching operation is performed on the feature information of the target vehicle from the data collected at each section to obtain target data of the target vehicle in the data collected at each section. The fusion frame data of the target vehicle is determined through the target data, and the operation data of the target vehicle is determined according to the fusion frame data.

[0035] In this embodiment, the license plate number is first used as an identification mark, and the feature information recognition matching is performed again for the vehicle whose license plate cannot be recognized, so as to further improve the accuracy of vehicle identification. Further, if the vehicle cannot be identified through the feature information, a mixed matching can also be performed in combination with multiple identification dimensions. For example, the vehicle brand can be identified based on a pre-trained ResNet-18 convolutional neural network, and a cropped image (size 224x224) of the vehicle head is input. The brand probability distribution (such as "BMW: 0.92, Mercedes-Benz: 0.05") is output through a fully connected layer. The brand with a probability ≥0.8 is taken as a standardized result, otherwise it is marked as "unknown".

[0036] In the above several embodiments, the accurate matching of the vehicle identity is realized through "feature weight distribution + voting screening + score determination", and the specific weight distribution is set as shown in the following table.

[0037] Feature Type Weight (points) Weight Basis (technical level) License Plate Information 40 The strongest anti-interference ability (even if partially obscured, it can still be restored by character segmentation), the highest matching priority Vehicle Head Feature 25 Core supplementary feature when license plate is invalid (completely obscured, dirty) Vehicle Brand 20 Pre-trained CNN model recognition accuracy ≥ 88% (tunnel low light scene), can quickly narrow down the matching range (such as first screening "BMW" brand vehicles) Vehicle Color 10 HSV color domain matching accuracy ≥ 95%, but similar colors (such as dark blue / light blue) are easily confused, anti-interference ability is weak, weight is low Vehicle Speed Data 5 Tunnel vehicle speed fluctuation is small (within ±8km / h, accounting for 90%), only as an auxiliary verification feature, not alone to determine the matching result Then, 1 independent voting unit is set for each feature dimension. If the matching threshold is met, 1 vote is counted, otherwise 0 votes are counted. The matching threshold is set as shown in the following table: Feature Type Matching Threshold (voting standard) Technical Implementation Details License Plate Information Complete matching (characters are completely consistent) or partial matching (edit distance ≤1, such as "Jing A12345" and "Jing A12346") Use edit distance algorithm to calculate character difference, edit distance = replacement / insertion / deletion operation times Vehicle Head Feature SIFT feature vector similarity ≥ 0.75 (cosine similarity) Calculate feature vector similarity through OpenCV's FlannBasedMatcher, threshold 0.75 is the optimal value (missed matching rate <3%) Vehicle Brand Identical identification results (such as "BMW 3 Series") Brand identification results contain vehicle model subdivision (pre-trained model output subdivision vehicle model), ensure matching accuracy Vehicle Color Standard color consistent or similar color (such as dark blue and light blue, HSV color domain H channel deviation ≤10°) Calculate the H channel difference between the two colors, ≤10° is considered similar color Vehicle Speed Data Deviation from the speed of the last section ≤10km / h Speed deviation comparison Then, the voting result screening is performed: if the total number of votes ≥3 (that is, at least 3 of the 5 features meet the matching threshold), it is determined as a "preliminary matching candidate set" and enters the score link; if the number of votes <3, the candidate vehicle is directly excluded.

[0038] Finally, the score is determined based on the weighted sum of feature weights, with the formula: Total Score = Σ (Feature Weight x Feature Score Coefficient), where the "Feature Score Coefficient" is set according to the matching degree (perfect match = 1, partial match = 0.25-0.5, no match = 0), and the specific coefficients are shown in the following table: Feature Type Complete matching coefficient Partial matching coefficient Non-matching coefficient License Plate Information 1 (40 points) 0.25 (10 points) 0 (0 points) Vehicle Head Feature 1 (25 points) 0.4 (10 points) 0 (0 points) Vehicle Brand 1 (20 points) 0 (0 points) 0 (0 points) Vehicle Color 1 (10 points) 0.5 (5 points) 0 (0 points) Vehicle Speed Data 1 (5 points) 0 (0 points) 0 (0 points) If the total score is ≥ 70 points (i.e., the matching degree is ≥ 70%), it is confirmed that the current section vehicle identity is consistent with the previous section transmitted identity; if the score is 60-69 points, it enters "secondary verification" (re-extract the vehicle head features, increase the number of feature points to more than 50 for matching); if the score is < 60 points, exclude the candidate and re-search the matching candidate set (if there is no candidate, mark it as a "newly entered vehicle" and assign a temporary identity ID).

[0039] After identifying the vehicles in each section data, the continuous transmission process of the same vehicle between different sections is as follows: First, based on the physical model of "vehicle speed - distance - time", the time range of the vehicle arriving at the next section can be accurately determined, with the formula T = [t_prev + (d / (v_prev + Δv)), t_prev + (d / (v_prev - Δv))], where t_prev: the vehicle passing time stamp of the previous section (millisecond level), d: the actual distance between the two sections (calibrated by laser range finder, error ≤ 0.5 meters), v_prev: the instantaneous vehicle speed of the previous section (filtered value of the sliding window), Δv: vehicle speed fluctuation redundancy (take 5km / h, based on the fluctuation standard deviation statistics of 100,000 vehicle speed samples in the tunnel). Further, the above data can be corrected by the slope in the tunnel, if there is a slope in the tunnel (> 5°), the slope correction coefficient (3km / h reduction for uphill, 2km / h increase for downhill) is added to v_prev to avoid time window deviation.

[0040] Then the vehicle is connected, and after the next section device collects data, the vehicle data in the "time window" (timestamp within T range) is first screened. For vehicles within this range, "voting + scoring" matching is performed one by one (same as single-section logic). The vehicle with the highest score and ≥70 points is associated with the last section identity ID, the section_info (section number, passing time, and distance from the next section) is updated, and a new identity data packet is generated and passed to the next section. When there are multiple candidate vehicles within the time window, the IOU (intersection over union) of the bounding box of the vehicle in the last section and the bounding box of the candidate vehicle in the current section is calculated, and the IOU≥0.3 is preferentially matched. For candidate vehicles that meet the IOU standard, the total feature score is sorted in descending order, and the highest score is taken as the matching result.

[0041] Further, the present embodiment can also retrieve interrupted vehicles. First, the method for interrupt determination is: if the matching scores of two consecutive sections are <60 points, it is marked as "tracking interruption". For interrupted vehicles, "feature similarity≥0.6" vehicles are continuously searched within the next 3 sections, and if a vehicle with a score≥65 points is found, the identity association is restored; if it is not found after 3 sections, it is marked as "tracking termination" and the interruption position is recorded.

[0042] In step S102, the collision risk, lane changing probability of each vehicle, and the traffic state of each lane are determined through the running data, and the collision risk, lane changing probability, and traffic state are fused to obtain multi-dimensional fusion data in the target tunnel.

[0043] In the present embodiment, historical running data can be determined through historical section data collection in the target tunnel, and then the risk warning model, behavior prediction model, and traffic state evaluation model are trained respectively through the historical running data. The trained risk warning model outputs the collision risk of each vehicle; the trained behavior prediction model outputs the lane changing probability of each vehicle; and the trained traffic state evaluation model outputs the traffic state of each lane. The collision risk can be a probability value between low, medium, high, or 0 to 1, and the traffic state can be smooth, slow, congested, etc.

[0044] The collision risk level, lane changing probability of each vehicle, and traffic state of each lane are further spatio-temporally aligned and associatedly integrated through a preset data fusion algorithm, such as weighted fusion or rule-based reasoning engine. For example, a vehicle with high collision risk and high lane changing probability is superimposed with the state information of the "congested" lane it is in, thereby generating a multi-dimensional fusion data view that can reflect both individual vehicle risk and overall traffic flow status, ultimately providing a comprehensive and three-dimensional decision basis for tunnel management.

[0045] In another embodiment, the determining, by the operation data, the collision risk of each vehicle, the lane-changing probability of each vehicle, and the traffic state of each lane comprises: determining, by the operation data, collision prediction data, and determining, by the collision prediction data, the collision risk level of each vehicle, thereby determining the corresponding collision risk; determining, by the operation data, lane-changing prediction data, and determining, by the lane-changing prediction data, the lane-changing probability of each vehicle; determining, by the operation data, lane state data, and determining, by the lane state data, the traffic state of each lane, wherein the lane state data comprises vehicle distance, vehicle density, and average vehicle speed.

[0046] In this embodiment, for the collision risk, the vehicle distance, relative speed, and vehicle distance change rate between the front and rear adjacent vehicles in the same lane of the certain vehicle are calculated in real time by the operation data of the certain vehicle to form key collision prediction data. When the vehicle distance is lower than a safety threshold, the relative speed is too high, and the vehicle distance is rapidly decreasing, the system determines that the collision risk level is rising, and accordingly, the risk level of the vehicle is divided into different levels such as no risk, low risk, medium risk, and high risk, thereby accurately determining the real-time collision risk of each vehicle.

[0047] For the lane-changing probability, the lane-changing prediction data is generated by the operation data of the certain vehicle, especially the trend of the lateral position change (such as whether continuously approaching a certain side lane line) of the certain vehicle relative to the lane line and the real-time vehicle distance with the front and rear vehicles in the adjacent lane. When the certain vehicle is steadily moving laterally and the target adjacent lane has sufficient safe vehicle distance, the lane-changing probability of the certain vehicle is determined to be high. Conversely, if the lateral movement hesitates or the target lane distance is insufficient, the lane-changing probability is determined to be low, thereby realizing accurate quantification and prediction of the lane-changing intention of the vehicle.

[0048] For the lane state data, the average vehicle distance, vehicle density (number of vehicles per unit length), and average vehicle speed of each lane are calculated in real time based on the operation data of all vehicles in the lane to form lane state data representing the macro state of the lane. Subsequently, the system comprehensively judges these indicators according to the preset threshold rules or machine learning model: if the average vehicle speed is high, the vehicle density is low, and the vehicle distance is large, the lane is determined to be in a "smooth" state; if the average vehicle speed significantly decreases, the vehicle density increases, and the vehicle distance decreases, the lane is determined to be in a "congestion" state; if the vehicle density is extremely high, the average vehicle speed is close to zero, and the vehicle distance is extremely small, the lane is determined to be in a "blockage" state, thereby accurately distinguishing the real-time traffic state of each lane.

[0049] Step S103, determining a management and control strategy of each lane in the target tunnel through the multi-dimensional fused data, and indicating the vehicle in the target tunnel according to the management and control strategy.

[0050] In the embodiment, the tunnel interior can be grouped horizontally and vertically through the fine management and control of "partition, segmentation, and lane", and the corresponding management and control strategy is generated according to the grouped lanes. The specific process is as follows: For vertical grouping, linear accurate positioning can be based on "post number".

[0051] Precise coordinates: the precise "linear coordinates" are given to each event and each device by using the milepost number in the tunnel. This is the spatial reference of all control instructions.

[0052] Dynamic segmentation: the control strategy is no longer limited to fixed partition, but can be more flexible "dynamic segmentation" according to the post number. For example, in a 150-meter control area, the 50 meters from K100+50 to K101+00 can be accurately divided into the core control area according to the event influence range.

[0053] Cascade control: based on the post number sequence, the smooth transition of control measures is realized. For example, the content of the variable information sign can be changed step by step according to the post number (such as: K101 prompts "accident ahead", K100+500 prompts "left lane closed", and K100+200 prompts "please change to the right lane") upstream of the accident point, guiding the smooth transition of traffic flow.

[0054] For horizontal grouping, the fine management and control dimension based on "lane" is used. Specifically, it includes: Lane as the smallest control unit: each lane is regarded as an independent control resource, which can be opened, closed, speed limited, and induced independently.

[0055] Dynamic lane function: according to the characteristics of traffic flow and tides or sudden events, the lane function is dynamically changed. For example, one opposite lane is changed to a tidal lane during the morning peak period, or one small passenger car lane is temporarily changed to an emergency repair passage after a large truck accident.

[0056] Lane-level driving environment adjustment: the environment of different lanes is adjusted independently according to the state of the lanes. For example, the lighting brightness is reduced in the closed lane to save energy, and the independent strong smoke exhaust mode is started in the lane where the fire occurs.

[0057] After grouping and generating the corresponding strategy, each device in the tunnel is controlled to execute the instructions, and the instruction execution process is as follows: Traffic signal: execute the most direct lane opening / closing instruction. Adopt red / green / yellow (warning) three-color signal, clearly display on the lane, combined with sound and light alarm, force standardize driver behavior.

[0058] Variable message sign: Global induction: before the tunnel entrance, prompt the overall road condition and regulation information in the tunnel.

[0059] Zoning preview: at the starting point of each zone in the tunnel, preview the lane opening / closing state and speed limit value of the front zone.

[0060] Lane-level indication: dynamically display the arrow (straight, left turn, right turn), speed limit (60, 40, prohibited passage) of the lane on the lane, realize the most precise guidance.

[0061] Broadcast system: Zoning broadcast: realize "sound source zoning following", warning or guidance information is only played in the event-related zone and upstream zone, avoid interference to irrelevant areas.

[0062] Directional broadcast: use phased array speaker technology, can perform "sound beam" directional broadcast to specific lanes, for example, only to the queuing lane for pacification and prompt, do not affect the driving environment of the adjacent smooth lane.

[0063] Through the above execution units such as traffic signal, variable message sign, broadcast system, clear lane use, speed control and path guidance instructions can be issued to the driver, so as to realize precise traffic induction and safety control.

[0064] So far, the flowchart shown in Figure 1 is completed.

[0065] In the embodiment of the present application, the intelligent traffic platform first acquires a plurality of section collection data in real time through the front-end sensing device installed in the tunnel, then performs feature transmission on each vehicle in the section collection data, determines the running data of each vehicle, and then determines the collision risk, lane changing probability of each vehicle and the traffic state of each lane according to the running data, to obtain the corresponding control strategy of each lane, and finally controls the control equipment in the tunnel according to the control strategy, to realize the indication of vehicles in different lanes. This makes it possible to generate corresponding lane-level control strategy according to real-time traffic flow, improves the timeliness and effectiveness of early warning, and thus reduces the probability of accidents.

[0066] As shown in Figure 2 , the present application also provides a lane-level control device based on a tunnel vehicle connection algorithm, applied to an intelligent traffic platform, the device comprises: The running data determination unit 201 is configured to acquire a plurality of cross-section collection data by a front-end sensing device in the target tunnel, and perform feature transmission on each vehicle in the cross-section collection data to determine running data of each vehicle, wherein the cross-section collection data is used to represent static features of the vehicle, and the running data is used to represent dynamic features of the vehicle in driving; The data fusion unit 202 is configured to determine a collision risk, a lane changing probability and a traffic state of each lane in the target tunnel by the running data, and fuse the collision risk, the lane changing probability and the traffic state to obtain multi-dimensional fusion data in the target tunnel; The strategy determination unit 203 is configured to determine a management and control strategy of each lane in the target tunnel by the multi-dimensional fusion data, and instruct the vehicle in the target tunnel according to the management and control strategy.

[0067] In another embodiment, the cross-section collection data in the running data determination unit includes license plate information, and the feature transmission on each vehicle in the cross-section collection data to determine the running data of each vehicle includes: For any target vehicle, license plate matching is performed on each cross-section collection data according to the license plate information of the target vehicle to obtain target data of the target vehicle in each cross-section collection data; The fusion frame data of the target vehicle is determined by the target data, and the running data of the target vehicle is determined according to the fusion frame data.

[0068] In another embodiment, the cross-section collection data in the running data determination unit includes feature information of the vehicle, and the feature transmission on each vehicle in the cross-section collection data to determine the running data of each vehicle includes: For any target vehicle, feature matching is performed on each cross-section collection data according to the feature information of the target vehicle to obtain target data of the target vehicle in each cross-section collection data; The fusion frame data of the target vehicle is determined by the target data, and the running data of the target vehicle is determined according to the fusion frame data.

[0069] In another embodiment, the cross-section collection data in the running data determination unit further includes feature information of the vehicle, and the feature transmission on each vehicle in the cross-section collection data to determine the running data of each vehicle includes: For any target vehicle, when the license plate information of the target vehicle cannot be determined, feature matching is performed on each cross-section collection data according to the feature information of the target vehicle to obtain target data of the target vehicle in each cross-section collection data; The target vehicle's fused frame data is determined using the target data, and then the target vehicle's operating data is determined based on the fused frame data.

[0070] In another embodiment, the data fusion unit's determination of the collision risk, lane change probability, and traffic status of each lane using the operational data includes: The collision prediction data is determined by the operational data, and then the collision risk level of each vehicle is determined by the collision prediction data, thereby determining the corresponding collision risk. The collision prediction data includes the distance between each vehicle, the rate of change of distance between vehicles, and the relative speed. Lane change prediction data is determined by the operational data, and then the lane change probability of each vehicle is determined by the lane change prediction data. The lane change prediction data includes the lateral position change trend and vehicle distance. Lane status data is determined by the operational data, and then the traffic status of each lane is determined by the lane status data. The lane status data includes vehicle distance, vehicle density, and average vehicle speed.

[0071] The above embodiments of the present invention provide a lane-level control method based on a tunnel vehicle connection algorithm, and a lane-level control device based on the same method. Through the above method and device, a corresponding lane-level control strategy can be generated according to the real-time traffic flow, which improves the timeliness and effectiveness of early warning and reduces the probability of accidents.

[0072] This embodiment also discloses a computer device, such as... Figure 3 As shown, the computer device includes a processor and a memory, the memory storing at least one instruction, which is loaded and executed by the processor to implement any of the lane-level control methods based on the tunnel vehicle connection algorithm described above.

[0073] Furthermore, in the above-described implementation of the lane-level control device based on the tunnel vehicle connection algorithm, the logical division of each program module is merely illustrative. In actual applications, the above functions can be assigned to different program modules as needed, for example, for the sake of corresponding hardware configuration requirements or the convenience of software implementation. That is, the internal structure of the lane-level control device based on the tunnel vehicle connection algorithm can be divided into different program modules to complete all or part of the functions described above.

[0074] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A lane-level management method based on a tunnel vehicle connection algorithm, characterized in that, The method is applied to an intelligent traffic platform, and comprises the following steps: Obtaining a plurality of cross-section collection data through a front-end sensing device in a target tunnel, and performing feature transmission on each vehicle in the cross-section collection data to determine running data of each vehicle, wherein the cross-section collection data is used to represent static features of the vehicle, and the running data is used to represent dynamic features of the vehicle in driving; Determining collision risks, lane-changing probabilities and traffic states of each lane in the target tunnel through the running data, and fusing the collision risks, the lane-changing probabilities and the traffic states to obtain multi-dimensional fusion data in the target tunnel; Determining a control strategy of each lane in the target tunnel through the multi-dimensional fusion data, and indicating vehicles in the target tunnel according to the control strategy.

2. The method of claim 1, wherein, The cross-section collection data contains license plate information, and the feature transmission on each vehicle in the cross-section collection data to determine the running data of each vehicle comprises the following steps: For any target vehicle, performing license plate matching on each cross-section collection data according to license plate information of the target vehicle to obtain target data of the target vehicle in each cross-section collection data; Determining fusion frame data of the target vehicle through the target data, and determining running data of the target vehicle according to the fusion frame data.

3. The method of claim 1, wherein, The cross-section collection data contains feature information of the vehicle, and the feature transmission on each vehicle in the cross-section collection data to determine the running data of each vehicle comprises the following steps: For any target vehicle, performing feature matching on each cross-section collection data according to feature information of the target vehicle to obtain target data of the target vehicle in each cross-section collection data; Determining fusion frame data of the target vehicle through the target data, and determining running data of the target vehicle according to the fusion frame data.

4. The method of claim 2, wherein, The cross-section collection data contains feature information of the vehicle, and the feature transmission on each vehicle in the cross-section collection data to determine the running data of each vehicle comprises the following steps: For any target vehicle, when the license plate information of the target vehicle cannot be determined, performing feature matching on each cross-section collection data according to feature information of the target vehicle to obtain target data of the target vehicle in each cross-section collection data; Determining fusion frame data of the target vehicle through the target data, and determining running data of the target vehicle according to the fusion frame data.

5. The method of claim 1, wherein, The determination of the collision risks, the lane-changing probabilities and the traffic states of each lane through the running data comprises the following steps: Determining collision prediction data through the running data, and determining a collision risk level of each vehicle through the collision prediction data to determine a corresponding collision risk, wherein the collision prediction data comprises a vehicle distance, a vehicle distance change rate and a relative speed of each vehicle; Determining lane-changing prediction data through the running data, and determining a lane-changing probability of each vehicle through the lane-changing prediction data, wherein the lane-changing prediction data comprises a lateral position change trend and a vehicle distance; The lane state data is determined according to the operation data, and the traffic states of the lanes are determined according to the lane state data, wherein the lane state data comprises a distance between vehicles, a vehicle density and an average vehicle speed.

6. A lane-level management device based on a tunnel vehicle connection algorithm, characterized in that, The device is applied to an intelligent traffic platform, and comprises: An operation data determination unit is configured to acquire a plurality of cross-section collection data by a front-end sensing device in a target tunnel, and perform feature transmission on each vehicle in the cross-section collection data to determine operation data of each vehicle, wherein the cross-section collection data is used to represent static features of vehicles, and the operation data is used to represent dynamic features of vehicles in driving; A data fusion unit is configured to determine collision risks of each vehicle, lane changing probabilities and traffic states of each lane according to the operation data, and fuse the collision risks, the lane changing probabilities and the traffic states to obtain multi-dimensional fusion data in the target tunnel; A strategy determination unit is configured to determine a management and control strategy of each lane in the target tunnel according to the multi-dimensional fusion data, and instruct vehicles in the target tunnel according to the management and control strategy.

7. The apparatus of claim 6, wherein, The cross-section collection data in the operation data determination unit comprises license plate information, and the feature transmission on each vehicle in the cross-section collection data to determine operation data of each vehicle comprises: For any target vehicle, license plate matching is performed on each cross-section collection data according to license plate information of the target vehicle to obtain target data of the target vehicle in each cross-section collection data; Fusion frame data of the target vehicle is determined according to the target data, and operation data of the target vehicle is determined according to the fusion frame data.

8. The apparatus of claim 6, wherein, The cross-section collection data in the operation data determination unit comprises feature information of vehicles, and the feature transmission on each vehicle in the cross-section collection data to determine operation data of each vehicle comprises: For any target vehicle, feature matching is performed on each cross-section collection data according to feature information of the target vehicle to obtain target data of the target vehicle in each cross-section collection data; Fusion frame data of the target vehicle is determined according to the target data, and operation data of the target vehicle is determined according to the fusion frame data.

9. The apparatus of claim 7, wherein, The cross-section collection data in the operation data determination unit further comprises feature information of vehicles, and the feature transmission on each vehicle in the cross-section collection data to determine operation data of each vehicle comprises: For any target vehicle, when license plate information of the target vehicle cannot be determined, feature matching is performed on each cross-section collection data according to feature information of the target vehicle to obtain target data of the target vehicle in each cross-section collection data; Fusion frame data of the target vehicle is determined according to the target data, and operation data of the target vehicle is determined according to the fusion frame data.

10. The apparatus of claim 6, wherein, The determination of the collision risks of each vehicle, the lane changing probabilities and the traffic states of each lane according to the operation data in the data fusion unit comprises: The collision prediction data is determined according to the operation data, and the collision risk level of each vehicle is determined according to the collision prediction data, so as to determine the corresponding collision risk, wherein the collision prediction data comprises the distance between vehicles, the distance change rate and the relative speed of each vehicle; The lane change prediction data is determined according to the operation data, and the lane change probability of each vehicle is determined according to the lane change prediction data, wherein the lane change prediction data comprises the lateral position change trend and the distance between vehicles; The lane state data is determined according to the operation data, and the traffic state of each lane is determined according to the lane state data, wherein the lane state data comprises the distance between vehicles, the vehicle density and the average vehicle speed.

Images