Method for detecting vehicles left on highway section

Abstract

The invention provides a method for detecting vehicles retained in a highway section, and relates to the technical field of data processing, and the method comprises the steps: obtaining multi-source vehicle passing data in a target space-time scene, and enabling the multi-source vehicle passing data to be fused with each other through a vehicle identifier, and obtaining vehicle track data; constructing a candidate road network topology according to the vehicle trajectory data, and identifying abnormal road sections and abnormal events in the candidate road network topology; optimizing the candidate road network topology according to an identification result to obtain a local road network topology; selecting a target detection time window according to the to-be-detected time, and determining candidate remaining vehicles, missing detection vehicles and accident remaining vehicles of the to-be-detected road section in the target detection time window; and according to the candidate reserved vehicles, the missed vehicles and the accident retained vehicles, determining final reserved vehicles. By adopting the method for detecting the retained vehicles on the highway section, the road network topology can be dynamically updated, the retained vehicles are detected under the updated road network topology, and the accuracy of detecting the retained vehicles on the highway section is improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and more specifically, to a method for detecting vehicles remaining on highway sections. Background Technology

[0002] With the continuous growth of people's travel demand, the need for highway construction is also increasing. Highways, with their unique advantages of being enclosed and fast, greatly satisfy people's travel needs and play a vital role in promoting national economic development. However, the workload and difficulty of identifying abnormally loitering vehicles on highways are gradually increasing. In existing technologies, vehicle loitering detection methods typically involve capturing vehicle images with monitoring equipment, extracting license plate information from the images, and then determining the vehicle's loitering time through license plate matching. Finally, vehicles whose loitering time exceeds a preset threshold are identified as abnormally loitering vehicles.

[0003] Existing vehicle retention detection methods assume that the road network connectivity is fixed. When dynamic events such as road closures for construction or accidents occur, they rely solely on the passage time at fixed entrance and exit points to determine whether vehicles are stuck. This makes it difficult to accurately detect vehicle retention and fails to meet the dynamic requirements of complex road networks. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide a method for detecting vehicles remaining on highway sections, so as to overcome at least one of the above-mentioned defects.

[0005] In a first aspect, embodiments of this application provide a method for detecting vehicles remaining on highway sections, including: Acquire multi-source vehicle traffic data in the target spatiotemporal scenario, and use vehicle identifiers to fuse the multi-source vehicle traffic data to obtain vehicle trajectory data. The target spatiotemporal scenario is determined based on the road segment to be detected and the time to be detected. Candidate road network topologies are constructed based on vehicle trajectory data. Abnormal road segments and events in the candidate road network topologies are identified. The candidate road network topologies are optimized based on the identification results to obtain local road network topologies. Identify candidate vehicles, missed vehicles, and accident-related vehicles on the road segments to be detected under the local road network topology; The final list of vehicles to be retained is determined based on the candidate vehicles to be retained, the vehicles that were missed during inspection, and the vehicles that were stranded due to accidents.

[0006] Optionally, the multi-source vehicle passage data includes license plate recognition data, gantry transaction data, and entrance / exit flow data. The step of integrating the multi-source vehicle passage data using vehicle identifiers to obtain vehicle trajectory data includes: using the passage record identifier as the primary association key and the gantry identifier as the secondary association key, linking the license plate recognition data, gantry transaction data, and entrance / exit flow data together to generate vehicle trajectory data.

[0007] Optionally, the method further includes: performing a two-way confidence assessment on the vehicle trajectory data; optimizing the vehicle trajectory data based on the two-way confidence assessment results to obtain optimized vehicle trajectory data, and constructing a candidate road network topology based on the optimized vehicle trajectory data.

[0008] Optionally, the two-way confidence assessment results include a first confidence assessment result and a second confidence assessment result. The steps for performing a two-way confidence assessment on vehicle trajectory data include: assessing the confidence of vehicle trajectory data from three dimensions—image quality, lighting conditions, and temporal continuity—to obtain a first confidence assessment result; and assessing the confidence of vehicle trajectory data from three dimensions—signal strength, transaction chain continuity, and time synchronization—to obtain a second confidence assessment result.

[0009] Optionally, the method further includes: performing kernel density estimation on vehicle trajectory data to determine the baseline travel time of the road segment to be detected in the local time period corresponding to the time to be detected; and determining the target detection time window based on the time to be detected and the baseline travel time.

[0010] Optionally, the candidate road network topology includes multiple candidate road segments. The steps of identifying abnormal road segments and abnormal events in the candidate road network topology and optimizing the candidate road network topology based on the identification results to obtain a local road network topology include: judging the authenticity of each candidate road segment from the dimensions of spatiotemporal reliability and node traffic requirements, and optimizing the candidate road network topology based on the judgment results to obtain an optimized road network topology; identifying abnormal events occurring in the optimized road network topology; and dynamically adjusting the connectivity of road segments in the optimized road network topology based on the abnormal event identification results to obtain a local road network topology.

[0011] Optionally, the step of judging the authenticity of each candidate road segment from the dimensions of spatiotemporal reliability and node traffic requirements includes: determining the spatiotemporal weighted support and node traffic distribution entropy of the candidate road segment; judging the authenticity of the candidate road segment based on the spatiotemporal weighted support and node traffic distribution entropy to eliminate noisy road segments.

[0012] Optionally, the abnormal event identification results include construction closures and accident delays. The road segment optimization topology includes the road segment type for each road segment, which includes ordinary road segments and toll station segments. The steps to dynamically adjust the connectivity of road segments in the road network optimization topology based on the abnormal event identification results to obtain the local road network topology include: determining the road segment type of the target road segment where the abnormal event occurred; if the road segment type is an ordinary road segment, interrupting the connectivity of the target road segment when the duration of the abnormal event meets a preset duration condition; if the road segment type is an toll station segment, updating the connectivity of the target road segment based on the traffic flow data of the target road segment.

[0013] Optionally, the steps for determining candidate retained vehicles, missed detection vehicles, and accident-stuck vehicles for the road segment to be detected under the local road network topology include: selecting vehicles from those passing through the road segment to be detected that did not simultaneously pass through the starting traffic monitoring node and the ending traffic monitoring node within the target detection time window to obtain candidate retained vehicles; using a bidirectional verification strategy between upstream and downstream under the local road network topology to determine missed detection vehicles for the road segment to be detected, wherein the bidirectional verification strategy between upstream and downstream is a strategy for handling missed detection based on the upstream vehicle fusion data of the starting traffic monitoring node and the downstream vehicle fusion data of the ending traffic monitoring node; determining the upstream and downstream traffic monitoring nodes for the road segment to be detected under the local road network topology, and determining candidate accident-stuck vehicles based on the monitoring data of the upstream and downstream traffic monitoring nodes; assessing the staging risk of candidate accident-stuck vehicles, and determining accident-stuck vehicles based on the staging risk assessment results.

[0014] Optionally, the step of constructing a candidate road network topology based on vehicle trajectory data includes: extracting traffic monitoring nodes with connectivity from the vehicle trajectory data; connecting two adjacent traffic monitoring nodes with connectivity to obtain candidate road segments; and constructing a candidate road network topology based on multiple candidate road segments that are connected to each other.

[0015] The embodiments of this application bring the following beneficial effects: This application provides a method for detecting vehicles remaining on highway sections. It can construct candidate road network topologies and dynamically update the candidate road network topologies when abnormal road sections and abnormal events are identified. The updated local road network topologies are used to detect vehicles remaining on highway sections. Even in complex road networks and situations with changes in road network structure, it can accurately screen out vehicles remaining on highway sections. Compared with existing methods for detecting vehicles remaining on highway sections, this method solves the problem of low detection accuracy of vehicles remaining on highways under complex road network conditions.

[0016] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart of the method for detecting vehicles remaining on highway sections provided in an embodiment of this application is shown; Figure 2 A flowchart illustrating the steps for determining vehicle trajectory data provided in an embodiment of this application is shown; Figure 3 A flowchart illustrating the steps for obtaining the local road network topology provided in an embodiment of this application is shown; Figure 4 A flowchart illustrating the detection steps for different types of retained vehicles provided in embodiments of this application is shown; Figure 5 A schematic diagram of the local road network topology provided in an embodiment of this application is shown; Figure 6 This paper shows a schematic diagram of the structure of the highway section vehicle detection device provided in an embodiment of this application; Figure 7 A schematic diagram of the structure of the electronic device provided in the embodiments of this application is shown. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. Based on the embodiments of this application, every other embodiment obtained by those skilled in the art without inventive effort falls within the scope of protection of this application.

[0020] To facilitate understanding of this embodiment, the following description uses the highway section vehicle detection method provided in this application embodiment applied to a server as an example to illustrate the above exemplary steps provided in this application embodiment.

[0021] Please see Figure 1 , Figure 1This is a flowchart illustrating a method for detecting vehicles remaining on a highway section, as provided in an embodiment of this application. Figure 1 As shown in the embodiments of this application, the method for detecting vehicles remaining on highway sections includes: Step S101: Obtain multi-source vehicle traffic data in the target spatiotemporal scene, and use vehicle identifiers to fuse the multi-source vehicle traffic data to obtain vehicle trajectory data.

[0022] Multi-source vehicle traffic data refers to vehicle traffic data from different sources in the context of highways. Multi-source vehicle traffic data includes license plate recognition data, gantry transaction data, and entrance / exit flow data.

[0023] The license plate recognition data refers to LPR (License Plate Recognition) data, which includes, but is not limited to: vehicle identifier (vehicle_plate), gantry identifier (gantry_id), transit time (trans_time), and lane position. For example, the vehicle identifier can be the license plate number, the gantry identifier can be the gantry number, and the transit time can refer to the time the vehicle passed through the traffic monitoring node.

[0024] Gantry transaction data can refer to ETC (Electronic Toll Collection) data, which includes, but is not limited to: pass_id, vehicle_plate, gantry_id, and transaction time. The pass_id refers to a unique code assigned to a vehicle for a single complete passage, and the transaction time is the timestamp indicating when the ETC system completes a complete billing / deduction operation.

[0025] Entry and exit flow data can refer to the entry and exit data of toll stations in highway toll collection or road network monitoring scenarios. Entry and exit flow data includes, but is not limited to: passage record identifier (pass_id), vehicle identifier (vehicle_plate), entry and exit time, and station number.

[0026] The following reference Figure 2 This section will introduce the process of determining vehicle trajectory data.

[0027] Figure 2 A flowchart illustrating the steps for determining vehicle trajectory data provided in an embodiment of this application is shown, as follows: Figure 2 As shown, the steps for determining vehicle trajectory data include: Step S1011: Determine the target spatiotemporal scene.

[0028] When acquiring multi-source vehicle traffic data, it is necessary to first determine the target spatiotemporal scenario in order to filter out multi-source vehicle traffic data under the target spatiotemporal scenario from different data sources.

[0029] The target spatiotemporal scene is determined based on the road segment to be detected and the time to be detected. The target spatiotemporal scene can refer to the spatiotemporal scene jointly corresponding to the associated monitoring node of the road segment to be detected on the highway and the local time period corresponding to the time to be detected.

[0030] The associated monitoring nodes include multiple traffic monitoring nodes that are within a preset range of distance from the road segment to be monitored. Traffic monitoring nodes can refer to a collection of hardware devices and logical units deployed in key locations of the road network in an intelligent transportation system, which have the functions of vehicle information collection, status monitoring, and data uploading. They are the basic units that constitute the road network monitoring network and are used to obtain vehicle traffic status in real time, support traffic management, toll settlement, safety warnings, and other services. As an example, a traffic monitoring node can be a gantry or a toll station.

[0031] For example: The road segment to be detected can refer to the road segment that starts from the starting traffic monitoring node and ends at the ending traffic monitoring node. The road segment to be detected includes the starting traffic monitoring node and the ending traffic monitoring node. The associated monitoring node can refer to the traffic monitoring node that is less than 10 kilometers away from the starting traffic monitoring node of the road segment to be detected and the traffic monitoring node that is less than 10 kilometers away from the ending traffic monitoring node of the road segment to be detected. The associated monitoring node includes the starting traffic monitoring node and the ending traffic monitoring node of the road segment to be detected.

[0032] The local time period includes the time range of each day within a preset number of days. The time range can refer to the time interval between the time to be detected and the time before and after the preset time. For example, if it is desired to detect which vehicles are present in road segment A at 7:00 AM, then the road segment to be detected is road segment A, and the time to be detected is 7:00 AM. Since vehicles cannot pass through the two gantries before and after the road segment to be detected at the same time, the time range for obtaining data can be expanded to two hours before and after the time to be detected, that is, the time range is from 5:00 AM to 9:00 AM. Assuming the preset number of days is 7 days, the local time period is the time interval from 5:00 AM to 9:00 AM every day within the most recent week.

[0033] Step S1012: Obtain multi-source vehicle traffic data in the target spatiotemporal scenario.

[0034] After determining the target spatiotemporal scenario, multi-source vehicle traffic data under the target spatiotemporal scenario can be selected from historical multi-source vehicle traffic data for analysis to determine which vehicles are retained on the road segment to be detected during the detection time.

[0035] Meanwhile, preprocess the multi-source vehicle passing data to improve the detection accuracy of the retained vehicles. The preprocessing includes invalid data cleaning and initial travel time calculation.

[0036] When cleaning invalid data, a dual cleaning mechanism of spatio-temporal continuity verification and minimum time difference matching can be adopted to clean invalid data.

[0037] For example: for passing data with invalid license plates such as "Mo A00000", abnormal timestamps, and over-limit positions, based on the adjacent gantry trajectory continuity model, spatio-temporal interpolation can be used to complete the invalid data to implement the spatio-temporal continuity verification mechanism. At the same time, calculate the timestamp difference between the LPR data and the ETC data within the same spatio-temporal range, retain the optimal matching record (such as the record with the smallest timestamp difference), and regard the data with a timestamp difference greater than the set time difference threshold as misidentified data and eliminate the misidentified data to implement the minimum time matching mechanism.

[0038] When calculating the initial travel time, the multi-source vehicle passing data can be sorted in ascending order of time based on the passing record identifier and the vehicle identifier. Then, the time difference between adjacent passing monitoring nodes is determined as the initial travel time of the road section, and a basic data set including the passing record identifier, the vehicle identifier, the gantry identifier, the passing time, and the initial travel time is formed.

[0039] Step S1013, use the vehicle identifier to fuse the multi-source vehicle passing data to obtain vehicle trajectory data.

[0040] For example: taking the passing record identifier pass_id as the main association key and the gantry identifier gantry_id as the auxiliary association key, fuse the basic data sets corresponding to the license plate recognition data, the gantry transaction data, and the entrance and exit flow data to generate vehicle trajectory data. The vehicle trajectory data includes the vehicle trajectory data of each vehicle, and the vehicle trajectory data of each vehicle includes the vehicle trajectory data of multiple road sections.

[0041] For example, the vehicle trajectory data of a single vehicle on a certain road segment includes the following: pass_id, vehicle_plate, start_gantry_id, end_gantry_id, start_trans_time, end_trans_time, and travel_time. Among them, start_gantry_id represents the identifier of the starting passage monitoring node in the road segment, end_gantry_id represents the identifier of the ending passage monitoring node in the road segment. When a vehicle passes through the road segment, it enters the road segment through the starting passage monitoring node and leaves the road segment through the ending passage monitoring node. start_trans_time represents the passage time of the vehicle passing through the starting passage monitoring node, end_trans_time represents the passage time of the vehicle passing through the ending passage monitoring node, and travel_time represents the initial travel time of the vehicle passing through the road segment.

[0042] If the start_gantry_id in the vehicle trajectory data of the next road segment is the same as the end_gantry_id in the vehicle trajectory data of the previous road segment, it indicates that the vehicle has passed through these two road segments consecutively. The vehicle trajectory data of multiple road segments that the vehicle has passed through consecutively can be combined to obtain the complete vehicle trajectory data.

[0043] In one embodiment, when generating vehicle trajectory data, there may be a situation where the pass_id cannot be determined. Therefore, the pass_id can be filled in.

[0044] For example: when multiple pass_ids correspond to the same vehicle identifier, the most recent valid pass_id is selected as the passage record identifier for that vehicle's trajectory data. When a unique pass_id corresponds to the vehicle trajectory data, that pass_id is used directly as the passage record identifier. When no pass_id exists, the vehicle identifier is used as a temporary passage record identifier based on the entrance / exit flow data.

[0045] Step S1014: Perform a two-way confidence assessment on the vehicle trajectory data.

[0046] Two-way confidence assessment refers to assessing the confidence of vehicle trajectory data from two directions. The results of two-way confidence assessment include the first confidence assessment result and the second confidence assessment result.

[0047] On the one hand, the confidence level of LPR data in vehicle trajectory data can be evaluated from three dimensions: image quality, lighting conditions, and temporal continuity, to obtain the first confidence level evaluation result.

[0048] For example: The result of the first confidence level assessment is denoted as: , ,in, Indicates image quality score. Indicates the illumination condition factor. The time continuity index, α, β, and γ, is determined through training using gradient descent. Image quality scoring is determined using image recognition algorithms, assessing image sharpness, contrast, and resolution. Illumination condition factors are determined based on time and weather conditions (e.g., nighttime, rainy days). The time continuity index is calculated based on the deviation between actual and expected travel time; the formula for calculating the time continuity index is: ,in, τ represents the attenuation coefficient of the time deviation, which is used to quantify the acceptable degree of deviation between the actual travel time and the expected travel time.

[0049] On the other hand, the confidence level of ETC data in vehicle trajectory data can be assessed from three dimensions: signal strength, transaction chain continuity, and time synchronization, to obtain a second confidence level assessment result.

[0050] For example, the result of the second confidence level assessment is denoted as: , ,in, Indicates signal strength index, Indicates the continuity of the transaction chain. Indicating time synchronization, δ, ε, and ζ are determined through grid search optimization. The formula for calculating the signal strength index is: RSSI (Received Signal Strength Indicator) represents the received signal strength indicator, and MAX_RSSI (Maximum Received Signal Strength Indicator) represents the maximum received signal strength indicator. Transaction chain continuity can be characterized by the proportion of continuous gantry transactions, and the formula for calculating the proportion of continuous gantry transactions is as follows: , Indicates the number of consecutive gantry transactions. This represents the total number of gantry transactions; the formula for calculating time synchronization is: , This represents the transaction timestamp of the ETC device. This indicates the timestamp of the license plate recognition device's capture. This represents the maximum permissible time difference threshold.

[0051] Step S1015: Optimize the vehicle trajectory data based on the two-way confidence evaluation results to obtain optimized vehicle trajectory data, and construct candidate road network topology based on the optimized vehicle trajectory data.

[0052] Specifically, this data optimization process optimizes each individual data point in the vehicle trajectory data.

[0053] For a single LPR data point in the vehicle trajectory data, determine whether the first confidence level assessment result is less than a first preset threshold (e.g., 0.6). If it is less than the first preset threshold, then the vehicle trajectory data is supplemented using the continuous transaction chain of the gantry transaction data. For example: in If the reliability of a single LPR data point is deemed insufficient, a spatiotemporal search range can be constructed centered on the gantry location of the low-confidence LPR data point. Within this search range, the occurrence records of all ETC gantries corresponding to the target vehicle in the target spatiotemporal scenario are retrieved. If the target vehicle exhibits a continuous transaction chain, the LPR data point is deemed valid; otherwise, if the vehicle does not appear in any ETC records, the LPR data point is deemed invalid and discarded.

[0054] For a single ETC data point in the vehicle trajectory data, determine whether the second confidence level assessment result is less than a second preset threshold (e.g., 0.7). If it is less than the second preset threshold, then correct the vehicle trajectory data based on the frequency of occurrence of adjacent gantries in the license plate recognition data. For example: in If the reliability of a single ETC data entry is deemed insufficient, a spatiotemporal search range can be constructed centered on the gantry location and passage time of the low-confidence ETC data entry (only querying ETC data for the same vehicle on the same day in the database, excluding historical ETC data). Within this search range, the target vehicle corresponding to this ETC data entry is retrieved, and LPR records are found at all adjacent gantries in the target spatiotemporal scenario. If the target vehicle shows a continuous trajectory at multiple adjacent gantries, the ETC data entry is considered valid; otherwise, if the vehicle does not appear in any LPR records, the ETC data entry is considered invalid and is removed.

[0055] In one embodiment, a reference travel time for the road segment to be detected can be determined using a kernel density estimation method, so that the remaining vehicles can be detected based on the reference travel time.

[0056] Specifically, kernel density estimation is performed on vehicle trajectory data to determine the baseline travel time of the road segment to be detected in the local time period corresponding to the time to be detected. Then, the target detection time window is determined based on the time to be detected and the baseline travel time.

[0057] For example: vehicle trajectory data is input into a kernel density estimation function to obtain multiple estimated travel times. These estimated travel times are then sorted in ascending order, and the estimated travel time at a preset percentile is selected as the baseline travel time. Based on the baseline travel time and the time to be detected, the target detection time window is determined. The kernel density estimation function is as follows: ; in, Indicates the actual travel time. Indicates adaptive bandwidth. This represents the Gaussian kernel function. , , Indicates actual traffic volume. This represents the baseline flow rate.

[0058] Step S102: Construct candidate road network topology based on vehicle trajectory data, identify abnormal road segments and abnormal events in the candidate road network topology, optimize the candidate road network topology based on the identification results, and obtain local road network topology.

[0059] Specifically, traffic monitoring nodes that are connected to other traffic monitoring nodes can be extracted from vehicle trajectory data. For example, gantries, toll stations, and other traffic monitoring nodes that are connected to other traffic monitoring nodes can be extracted. Two adjacent traffic monitoring nodes that are connected can be connected together, such as gantries-gantries, toll stations-gantries, and gantries-toll stations, to obtain multiple candidate road segments. In this way, a candidate road network topology can be constructed based on the multiple candidate road segments that are connected to each other. The candidate road network topology includes multiple candidate road segments, and a candidate road network topology structure can be formed based on the candidate road network topology.

[0060] Abnormal road sections refer to road sections that do not meet the requirements for spatiotemporal reliability and node traffic flow. Abnormal events refer to events that affect the normal passage of vehicles, such as construction closures and traffic accidents.

[0061] The following reference Figure 3 This section will introduce the process of obtaining the local road network topology.

[0062] Figure 3 A flowchart illustrating the steps for obtaining the local road network topology provided in an embodiment of this application is shown, as follows: Figure 3 As shown, the steps for obtaining the local road network topology include: Step S1021: The authenticity of each candidate road segment is judged from the dimensions of spatiotemporal reliability and node traffic requirements, so as to optimize the candidate road network topology based on the judgment results and obtain the optimized road network topology.

[0063] The first step is to determine the spatiotemporal weighted support of each candidate road segment in the candidate road network topology. This spatiotemporal weighted support characterizes the spatiotemporal reliability of the candidate road segment. Assuming a candidate road segment includes a starting traffic monitoring node A and an ending traffic monitoring node B, the spatiotemporal weighted support of that candidate road segment is denoted as: The spatiotemporal weighted support of the candidate road segment can be determined using the following formula: ; in, This represents the number of consecutive passages from access monitoring node A (such as a gantry) to access monitoring node B. This indicates the total number of passes through monitoring node A. Indicates time consistency weight, Indicates the reachability weight of the control. , Indicates the standard deviation of travel time. Indicates the average travel time; , Indicates the actual distance. This represents the path distance; the actual distance and the path distance can be obtained from the database.

[0064] For each candidate road segment in the candidate road network topology, determine the node flow distribution entropy of that candidate road segment. The node flow distribution entropy characterizes the node flow requirements of the candidate road segment. The node flow distribution entropy is denoted as H(i), and can be determined using the following formula: ,in, This represents the probability of traffic transfer, i.e., the proportion of outflow traffic from upstream monitoring node i that flows to downstream monitoring node j.

[0065] The second step is to determine the authenticity of the candidate road segment based on the spatiotemporal weighted support and the node traffic distribution entropy, so as to eliminate noisy road segments and obtain the optimized road network topology.

[0066] Candidate road segments with spatiotemporal weighted support less than a preset support threshold are identified as non-real road segments with insufficient spatiotemporal reliability and are removed from the candidate road network topology.

[0067] If H(i) is less than the first distribution entropy threshold (e.g., 1), then the traffic transfer probability threshold is determined to be the first traffic transfer probability threshold (e.g., 0.1), and the road segments corresponding to the traffic monitoring nodes whose traffic transfer probability is less than the first traffic transfer probability threshold are determined to be non-real road segments and removed from the candidate road network topology; if H(i) is greater than the second distribution entropy threshold (e.g., 2.5), then the traffic transfer probability threshold is determined to be the second traffic transfer probability threshold (e.g., 0.01), and the road segments corresponding to the traffic monitoring nodes whose traffic transfer probability is less than the second traffic transfer probability threshold are determined to be non-real road segments and removed from the candidate road network topology.

[0068] In one embodiment, after verifying the authenticity of each candidate road segment, spatial verification is also required.

[0069] For example, vehicle trajectory data also includes lane markings. Combined with geographical information such as road network markers, number of lanes, and lane markings, accessibility checks are performed on each road segment. Road segments that fail the accessibility check are removed to obtain a static optimized road network topology. Then, a road segment type label is set for each road segment in the optimized road network topology. Road segment types include ordinary road segments, toll station / entrance / exit road segments, and parallel road segments.

[0070] Step S1022: Identify abnormal events occurring in the road network optimization topology.

[0071] For each road segment, the vehicle trajectory data corresponding to that road segment is sequentially input into a preset spatiotemporal graph convolutional network in chronological order to determine the abnormal events that occur in that road segment.

[0072] The spatiotemporal graph convolutional network is a three-layer spatiotemporal graph convolutional network. The spatiotemporal graph convolutional network outputs abnormal events through a 64-neuron ReLU activation layer. For example, if the traffic flow of a road segment drops by ≥50% and lasts for 10 minutes, the abnormal event of the road segment is determined to be construction closure; if the vehicles in the road segment deviate from their trajectory and their speed drops sharply, the abnormal event of the road segment is determined to be a traffic accident.

[0073] Step S1023: Based on the abnormal event identification results, dynamically adjust the connectivity of road segments in the road network optimization topology to obtain the local road network topology.

[0074] The results of abnormal event identification include construction closures and traffic accidents.

[0075] Based on the set road segment type labels, the road segment type of the target road segment where the abnormal event occurred is determined. If the road segment type is a normal road segment and the abnormal event identification result is construction closure, then if the duration of the abnormal event meets the first preset duration condition, the connection between the target road segment and other road segments is interrupted. For example, if the duration exceeds 5 hours, the traffic association of the target road segment is interrupted. If the road segment type is a normal road segment and the abnormal event identification result is a traffic accident, then if the duration of the abnormal event meets the second preset duration condition, the connection between the target road segment and other road segments is interrupted. For example, if the duration exceeds 1 hour, the traffic association of the target road segment is interrupted. If the road segment type is an entrance / exit toll station road segment, the traffic status of the target road segment is updated based on the traffic flow data of the target road segment. For example, if the traffic flow exceeds the set traffic flow threshold, the traffic association of the target road segment is interrupted.

[0076] Step S103: Determine the candidate remaining vehicles, missed detection vehicles, and accident-stuck vehicles for the road segment to be detected under the local road network topology.

[0077] The time for determining the retained vehicles, the time for detection, and the time for constructing the local road network topology may differ. For example, if a user detects vehicles retained at 10:00 AM at 2:00 PM, the determination time is 2:00 PM, and the detection time is 10:00 AM. Therefore, after determining the local road network topology, multi-source vehicle traffic data within the local road network topology can be retrieved from the database again, and step S101 can be executed on the re-retrieved multi-source vehicle traffic data within the local road network topology to obtain vehicle fusion data.

[0078] The following reference Figure 4 This section will introduce the inspection process for different types of vehicles that are retained.

[0079] Figure 4 The flowcharts illustrating the detection steps for different types of retained vehicles provided in embodiments of this application are shown, such as... Figure 4 As shown, the inspection steps for different types of retained vehicles include: Step S1031: Determine the dynamic time window.

[0080] Dynamic time windows include a start node time window and an end node time window.

[0081] The window range of the starting node time window is: [ , ],in, , Indicates the current speed. Indicates the free flow velocity. This indicates that it is based on adaptive bandwidth kernel density estimation.

[0082] The window range of the termination node time window is: [ ],in, , Indicates the current speed. Indicates the free flow velocity. This indicates that it is based on adaptive bandwidth kernel density estimation.

[0083] Step S1032: From the vehicles passing through the road section to be detected, select the vehicles that did not pass through the start traffic monitoring node and the end traffic monitoring node at the same time during the target detection time window, and obtain the candidate vehicles to be retained.

[0084] For the road segment to be detected, the vehicle fusion data within the starting node time window and the vehicle fusion data within the ending node time window are internally joined (i.e. correlated) to filter out candidate vehicles that have not reached the termination detection node of the road segment to be detected within the expected time.

[0085] Step S1033: Under the local road network topology, a two-way verification strategy between upstream and downstream is adopted to determine the missed vehicles in the road segment to be detected.

[0086] There may be vehicles that are missed in the road segment to be detected. For example, the starting traffic monitoring node of the road segment to be detected may detect the vehicle, but the ending traffic monitoring node may not detect the vehicle. However, the vehicle may appear in the downstream traffic monitoring node. This means that the vehicle may be in the road segment to be detected, but it was missed. Multiple backtracking is required to identify the missed vehicles.

[0087] The upstream and downstream two-way verification strategy is a strategy for handling missed detections based on the upstream vehicle fusion data of the starting traffic monitoring node and the downstream vehicle fusion data of the ending traffic monitoring node. The upstream vehicle fusion data can refer to the vehicle fusion data of the upstream traffic monitoring node of the starting traffic monitoring node, and the downstream vehicle fusion data can refer to the vehicle fusion data of the downstream traffic monitoring node of the starting traffic monitoring node.

[0088] For ordinary road sections, missed detection vehicles include vehicles missed at the destination and vehicles missed at the starting point. Vehicles missed at the destination can be those that were detected by the starting traffic monitoring node but not by the ending traffic monitoring node; vehicles missed at the starting point can be those that were detected by the ending traffic monitoring node but not by the starting traffic monitoring node.

[0089] When identifying vehicles that were missed at the destination, a first set of vehicles corresponding to the road segment to be detected is constructed under the local road network topology. The vehicle fusion data and vehicle travel time distribution corresponding to the vehicles in the first set of vehicles are input into the probability matching model. The probability matching model will determine the probability that each vehicle belongs to the missed vehicle based on the input data. When the probability of belonging to the missed vehicle meets the preset probability condition (such as the probability of belonging to the missed vehicle being greater than the set probability threshold), the vehicle Vi is determined to be the missed vehicle at the destination.

[0090] The first vehicle set can be a single-level vehicle set or a multi-level vehicle set. When the first vehicle set is a multi-level vehicle set, it is denoted as US. Here, the end-of-traffic monitoring node of the road segment to be monitored is referred to as the first target node, and the start-of-traffic monitoring node of the road segment to be monitored is referred to as the second target node. This represents the set of vehicles detected by the first-layer downstream nodes connected to the first target node. This represents the set of vehicles detected by the first target node. This represents the set of vehicles detected by upstream nodes in the first layer that are connected to the first target node; that is, the set of vehicles detected by upstream nodes in the same layer as the second target node that are connected to the first target node. This represents the set of vehicles detected by the second-layer upstream nodes connected to the first target node, and so on. This represents the set of vehicles detected by the k-th layer upstream node connected to the first target node.

[0091] The following reference Figure 5 Let's introduce the first group of vehicles.

[0092] Figure 5 A schematic diagram of the local road network topology provided in an embodiment of this application is shown, as follows: Figure 5 As shown, the road segment to be detected is the segment from node a1 to node b1. Node b1 is the first target node, and node a1 is the second target node. The first-layer downstream nodes connected to node b1 include nodes c1, c2, and c3. The first-layer upstream nodes connected to node b1 include nodes a1, a2, and a3. The second-layer upstream nodes connected to node b1 include nodes d1 and d2. Nodes b2 and c4 do not belong to the first vehicle set.

[0093] The calculation formula for the probability matching model is as follows: ,in, This represents the probability that vehicle Vi is the final vehicle that was missed during the detection process. This represents the probability of travel time. This represents the detection probability, that is, the probability that the traffic monitoring node successfully identifies and records vehicle traffic information. t represents the effective monitoring duration of the vehicle by the traffic monitoring node.

[0094] Similarly, a downstream vehicle set corresponding to the road segment to be detected is constructed to identify vehicles missed at the starting point. The method for determining vehicles missed at the starting point is symmetrical to that for vehicles missed at the ending point, and will not be elaborated here. The vehicle travel time is determined based on a kernel density estimation function.

[0095] For the entrance toll station section (without upstream gantries), when identifying missed vehicles at the starting point, the entrance toll station is designated as the first target node, and the exit toll station as the second target node. Parallel road segments are also identified, with the second target node serving as the end point of the road segment. Based on the vehicle fusion data of these parallel road segments, the probability of each vehicle passing through both the parallel road segments and the section to be detected is determined. Then, for each vehicle, the difference between its probability of passing through the parallel road segment and its probability of passing through the section to be detected is calculated. If the probability difference between a parallel road segment and the section to be detected exceeds a set threshold, the difference condition is met, indicating that the vehicle passed through that parallel road segment and is not a missed vehicle; vehicles meeting the difference condition can be removed. If no parallel road segment or section to be detected has a probability difference exceeding the set threshold, the difference condition is not met, indicating that the vehicle is a missed vehicle.

[0096] For exit toll station sections (without downstream gantries), when identifying missed vehicles at the destination, the exit toll station is designated as the first target node, the entrance toll station as the second target node, and first candidate nodes at the same level as the first target node are identified. None of these first candidate nodes are exit toll stations. Historical vehicle fusion data is input into a Bayesian inference model to determine the missed vehicles at the destination of the exit toll station section. For example, the Bayesian inference model determines a prior probability, and based on this prior probability, the probability of a vehicle leaving from the first target node is determined. If this probability is greater than a set probability threshold, the vehicle is identified as a missed vehicle at the destination. The prior probability includes the first proportion of the number of times a vehicle enters the second target node and exits from the first target node out of the total number of times, and the second proportion of the number of times a vehicle enters the second target node and exits from each of the first candidate nodes out of the total number of times. The total number of times can refer to the number of times a vehicle enters the second target node and exits from the first target node and all first candidate nodes.

[0097] Step S1034: Under the local road network topology, determine the upstream and downstream traffic monitoring nodes of the road segment to be detected, and determine the candidate accident stranded vehicles based on the monitoring data of the upstream and downstream traffic monitoring nodes.

[0098] Specifically, the vehicle differences between the vehicle fusion data of the upstream traffic monitoring node and the vehicle fusion data of the downstream traffic monitoring node of the road segment to be detected are calculated to obtain candidate accident-stuck vehicles. The upstream traffic monitoring node can refer to the traffic monitoring node located upstream of the starting traffic monitoring node of the road segment to be detected, and the downstream traffic monitoring node can refer to the traffic monitoring node located downstream of the ending traffic monitoring node of the road segment to be detected.

[0099] Step S1035: Assess the detention risk of candidate accident-detained vehicles to determine accident-detained vehicles based on the detention risk assessment results.

[0100] The vehicle fusion data of candidate accident vehicles are input into the survival analysis model, and the accident-stuck vehicles are identified based on the output of the survival analysis model.

[0101] For example: the survival analysis model is ,in, Represents the benchmark risk function. (like , , () represents covariates, which include, but are not limited to, traffic flow, speed, and weather.

[0102] Step S104: Determine the final vehicles to be retained based on the candidate vehicles to be retained, the vehicles that were missed during inspection, and the vehicles that were stranded due to accidents.

[0103] The candidate vehicles to be retained, the vehicles that were missed during inspection, and the vehicles that were stranded due to accidents are combined together, and duplicate vehicles are removed to determine the final vehicles to be retained. The retention information of the final vehicles to be retained is recorded, including the license plate, the start time of the stranding, the location of the stranding, and the type of stranding (such as ordinary stranding or accident stranding).

[0104] The method for detecting vehicles remaining on highway sections provided in this application can construct candidate road network topologies and dynamically update the candidate road network topologies when abnormal road sections and abnormal events are identified. The updated local road network topologies are used to detect vehicles remaining on highway sections. Even in the case of complex road networks and changes in road network structure, vehicles remaining can be accurately screened out, solving the problem of low detection accuracy of vehicles remaining under complex highway road network conditions.

[0105] Based on the same inventive concept, this application also provides a highway vehicle detection device corresponding to the highway vehicle detection method. Since the principle of the device in this application is similar to the highway vehicle detection method described above, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0106] Please see Figure 6 , Figure 6 This is a schematic diagram of a vehicle detection device for highway sections provided in an embodiment of this application. Figure 6 As shown, the highway section vehicle detection device 200 includes: Data acquisition module 201 is used to acquire multi-source vehicle traffic data in the target spatiotemporal scene, and use vehicle identifiers to associate the multi-source vehicle traffic data to obtain vehicle trajectory data. The target spatiotemporal scene is determined based on the road segment to be detected and the time to be detected. The topology construction module 202 is used to construct candidate road network topologies based on vehicle trajectory data, identify abnormal road segments and abnormal events in the candidate road network topologies, optimize the candidate road network topologies based on the identification results, and obtain local road network topologies. The first determining module 203 is used to determine the candidate remaining vehicles, missed vehicles, and accident-stuck vehicles on the road segment to be detected under the local road network topology. The second determining module 204 is used to determine the final retained vehicles based on the candidate retained vehicles, the missed inspection vehicles, and the accident-stuck vehicles.

[0107] Please see Figure 7 , Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 7 As shown, the electronic device 300 includes a processor 310, a memory 320, and a bus 330.

[0108] The memory 320 stores machine-readable instructions executable by the processor 310. When the electronic device 300 is running, the processor 310 and the memory 320 communicate via the bus 330. When the machine-readable instructions are executed by the processor 310, they can perform the operations described above. Figure 1 The specific implementation of the steps of the highway section vehicle detection method in the method embodiment shown can be found in the method embodiment, and will not be repeated here.

[0109] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can perform the above-described actions. Figure 1 The specific implementation of the steps of the highway section vehicle detection method in the method embodiment shown can be found in the method embodiment, and will not be repeated here.

[0110] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0111] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the shown or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0112] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0113] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0114] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0115] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The scope of protection of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for detecting a remaining vehicle on a highway section, characterized by, The method comprises the following steps: Obtain multi-source vehicle passing data in a target space-time scene, fuse the multi-source vehicle passing data with each other by using vehicle identification, and obtain vehicle trajectory data, wherein the target space-time scene is determined based on a to-be-detected road section and a to-be-detected time; Construct a candidate road network topology according to the vehicle trajectory data, identify abnormal road sections and abnormal events in the candidate road network topology, optimize the candidate road network topology according to the identification result, and obtain a local road network topology; Determine candidate remaining vehicles, missed vehicles and accident stranded vehicles of the to-be-detected road section in the local road network topology; Determine final remaining vehicles according to the candidate remaining vehicles, the missed vehicles and the accident stranded vehicles.

2. The method of claim 1, wherein, The multi-source vehicle passing data comprises license plate recognition data, gantry transaction data and entrance and exit flow data, and the step of fusing the multi-source vehicle passing data with each other by using vehicle identification to obtain vehicle trajectory data comprises the following steps: Associate the license plate recognition data, the gantry transaction data and the entrance and exit flow data together by taking a passing record identification as a main association key and taking a gantry identification as an auxiliary association key, and generate vehicle trajectory data.

3. The method of claim 2, wherein, The method further comprises the following steps: Perform bidirectional confidence evaluation on the vehicle trajectory data; Optimize the vehicle trajectory data according to the bidirectional confidence evaluation result, and obtain optimized vehicle trajectory data, so as to construct a candidate road network topology according to the optimized vehicle trajectory data.

4. The method of claim 3, wherein, The bidirectional confidence evaluation result comprises a first confidence evaluation result and a second confidence evaluation result, and the step of performing bidirectional confidence evaluation on the vehicle trajectory data comprises the following steps: Perform confidence evaluation on the vehicle trajectory data from three dimensions of image quality, illumination condition and time continuity, and obtain a first confidence evaluation result; Perform confidence evaluation on the vehicle trajectory data from three dimensions of signal strength, transaction chain continuity and time synchronization, and obtain a second confidence evaluation result.

5. The method of claim 1, wherein, The method further comprises the following steps: Perform kernel density estimation on the vehicle trajectory data, determine a reference travel time of the to-be-detected road section in a local time period corresponding to the to-be-detected time, and determine a target detection time window according to the to-be-detected time and the reference travel time. The candidate road network topology comprises a plurality of candidate road sections, and the step of identifying abnormal road sections and abnormal events in the candidate road network topology, and optimizing the candidate road network topology according to the identification result to obtain a local road network topology comprises the following steps:

6. The method of claim 1, wherein, Judge the authenticity of each candidate road section from the dimensions of space-time reliability and node flow requirement, optimize the candidate road network topology according to the judgment result, and obtain a road network optimization topology; Identify abnormal events occurring in the road network optimization topology; Dynamically adjust the connectivity of road sections in the road network optimization topology according to the abnormal event identification result, and obtain a local road network topology. The step of judging the authenticity of each candidate road section from the dimensions of space-time reliability and node flow requirement comprises the following steps:

7. The method of claim 6, wherein, Determine the space-time weighted support degree and the node flow distribution entropy of the candidate road section; ​ According to the spatio-temporal weighted support and the node traffic distribution entropy, the authenticity of the candidate road section is judged to eliminate noise road sections.

8. The method of claim 6, wherein, The abnormal event identification result includes construction closure and accident detention, the road section optimization topology includes a road section type of each road section, the road section type includes a general road section and an entrance and exit toll station road section, and the step of dynamically adjusting the connectivity of the road section in the road network optimization topology according to the abnormal event identification result to obtain the local road network topology includes: Determining a road section type of a target road section where the abnormal event occurs; If the road section type is a general road section, the connectivity of the target road section is interrupted when the duration of the abnormal event meets a preset time length condition; If the road section type is an entrance and exit toll station road section, the connectivity of the target road section is updated according to traffic flow data of the target road section.

9. The method of claim 5, wherein, The step of determining the candidate remaining vehicle, the missed vehicle and the accident detention vehicle of the to-be-detected road section under the local road network topology includes: From the vehicles passing through the to-be-detected road section, vehicles that do not pass through the starting and ending traffic monitoring nodes at the same time in the target detection time window are screened to obtain the candidate remaining vehicle; The missed vehicle of the to-be-detected road section is determined by using an upstream and downstream bidirectional verification strategy under the local road network topology, and the upstream and downstream bidirectional verification strategy is a strategy of performing missed detection processing according to upstream vehicle fusion data of the starting traffic monitoring node and downstream vehicle fusion data of the ending traffic monitoring node; The upstream and downstream traffic monitoring nodes of the to-be-detected road section are determined under the local road network topology, and the candidate accident detention vehicle is determined according to the monitoring data of the upstream and downstream traffic monitoring nodes; The detention risk of the candidate accident detention vehicle is evaluated to determine the accident detention vehicle according to the detention risk evaluation result.

10. The method of claim 1, wherein, The step of constructing the candidate road network topology according to the vehicle trajectory data includes: Extracting traffic monitoring nodes with a connectivity relationship from the vehicle trajectory data; Connecting two adjacent traffic monitoring nodes with a connectivity relationship to obtain a candidate road section, and constructing a candidate road network topology according to a plurality of candidate road sections connected to each other.

Images