Cloning vehicle identification method and apparatus

Abstract

This application provides a method and apparatus for identifying cloned vehicles. Through an innovative spatiotemporal analysis model, it achieves accurate screening of unreachable vehicles via distance calculation and speed verification. A feature recognition system is constructed, combining image segmentation and tag extraction to establish a reliable mechanism for identifying cloned vehicles. Predictive deployment is introduced, using trajectory analysis and route prediction to ensure the accuracy of strikes. This method effectively addresses the shortcomings of traditional technologies in spatiotemporal analysis, feature recognition, and predictive deployment, providing technical support for the control of cloned vehicles.

Description

Technical Field

[0001] This application relates to the field of data processing, specifically to a method and apparatus for identifying cloned vehicles. Background Technology

[0002] Existing methods for identifying cloned vehicles have significant shortcomings. Traditional systems perform poorly in data acquisition and spatiotemporal analysis, failing to effectively integrate multi-source information and thus affecting identification accuracy.

[0003] Furthermore, existing technologies suffer from bottlenecks in feature extraction and similarity matching. Most systems lack robust detailed identification mechanisms and label comparison strategies, resulting in inaccurate identification of cloned vehicles.

[0004] Existing systems have technical shortcomings in predictive deployment. The lack of in-depth analysis of historical trajectories makes it difficult to achieve effective deployment through model prediction, thus impacting the effectiveness of strikes. Solving these problems is crucial for improving the efficiency of cloned vehicle identification. Summary of the Invention

[0005] To address the problems in existing technologies, this application provides a method and apparatus for identifying cloned vehicles, which can effectively solve the shortcomings of traditional technologies in spatiotemporal analysis, feature recognition, and predictive deployment, and provide technical support for the management of cloned vehicles.

[0006] To solve at least one of the above problems, this application provides the following technical solution: Firstly, this application provides a method for identifying cloned vehicles, including: The system collects vehicle traffic data from urban traffic checkpoint equipment and roadside radar sensors, extracts basic information and location information for each vehicle, reads distance information between adjacent checkpoints, calculates vehicle travel distance based on the semi-sine formula, extracts timestamps to calculate time differences, calculates travel speed based on the travel distance and time difference, constructs a road network parameter table containing distance thresholds and speed thresholds, compares the travel distance, time difference, and travel speed with the road network parameter table, and filters out candidate vehicles with cloned license plates that cannot be reached within a unit of time. Read the exterior image of the candidate vehicle with cloned license plate, extract the key parts of the vehicle based on the image segmentation model, input the key parts into the feature recognition model to extract the basic features of the vehicle, construct the vehicle feature vector, calculate the similarity between the feature vectors of different vehicles, mark the vehicles with similarity exceeding a preset threshold as cloned vehicle candidate pairs, input the images of the cloned vehicle candidate pairs with the same shooting angle into the label recognition model, extract the vehicle detail identification features, and generate the vehicle label feature set. The system retrieves violation record data of the cloned vehicle candidate pair from the traffic violation management system, extracts vehicle images from the processed violation records, inputs the vehicle images into a label recognition model to extract feature labels, compares the feature labels with the vehicle label feature set, identifies real vehicles and cloned vehicles based on label matching degree, inputs the historical trajectory data of the cloned vehicles into a time series prediction model, calculates the future driving route and necessary checkpoints, and generates a deployment command containing cloned vehicle feature information and driving prediction.

[0007] Furthermore, it also includes: reading vehicle capture records from the urban traffic checkpoint database, parsing vehicle passage information from roadside radar data stream, grouping the capture records and passage information by license plate number, extracting the license plate number, capture time, and checkpoint location information from each group of data, constructing a vehicle basic information table, reading the road network topology data from the electronic map, mapping the checkpoint location information to road network nodes, calculating the road connection relationship between adjacent checkpoints, and generating a checkpoint distance matrix; Substitute the latitude and longitude coordinates of adjacent checkpoints into the semi-sine formula to calculate the actual driving distance of the vehicle. Extract the timestamp information of adjacent capture points from the vehicle basic information table, convert the timestamp information into a time difference value, divide the driving distance by the time difference value to obtain the driving speed, and construct a parameter calculation table containing license plate number, driving distance, time difference value, and driving speed.

[0008] Furthermore, it also includes: reading road network topology data, classifying roads into expressways, urban expressways, and urban arterial roads according to functional levels, extracting speed limit standards for each type of road, setting speed thresholds based on the speed limit standards, reading the road segment lengths in the checkpoint distance matrix, setting distance thresholds based on the road segment lengths, writing the speed thresholds and distance thresholds into the road network parameter table according to road level, and constructing road traffic constraint rules; Read the vehicle driving records from the parameter calculation table, compare the driving distance with the distance threshold in the road network parameter table, compare the driving speed with the speed threshold in the road network parameter table, mark driving records with both distance and speed exceeding the threshold as abnormal data, group the abnormal data by license plate number, and generate a dataset of candidate vehicles with cloned license plates.

[0009] Furthermore, it also includes: extracting vehicle appearance images from the candidate vehicle dataset for counterfeit license plates, inputting the appearance images into the image enhancement module to perform illumination normalization and noise suppression, constructing an image segmentation network containing convolutional layers and deconvolutional layers, inputting the enhanced image into the image segmentation network, generating a vehicle region segmentation mask, segmenting the vehicle image into a front face region, a side region, and a rear region based on the segmentation mask, and scaling the segmented regions to a uniform size according to a preset ratio; The scaled segmented region is input into the feature recognition model, and vehicle appearance features are extracted based on the convolutional neural network. Local detail features are extracted based on the region attention mechanism. The appearance features and detail features are fused according to preset weights to construct a multi-dimensional vehicle feature matrix. Dimensionality reduction calculation is performed on the feature matrix to generate a fixed-dimensional vehicle feature vector.

[0010] Furthermore, it also includes: reading vehicle feature vectors, constructing a feature matching matrix, calculating the Euclidean distance and cosine similarity between feature vectors, weighting and combining the Euclidean distance and cosine similarity to obtain a comprehensive similarity, determining a similarity threshold based on cluster analysis, marking vehicle pairs with a comprehensive similarity higher than the threshold as clone vehicle candidate pairs, reading the image shooting parameters of the clone vehicle candidate pairs, dividing the images into front view, rear view, and side view based on camera angle information, and filtering image groups with consistent shooting angles; Images taken from the same angle are input into the label recognition model. The location and content of the annual inspection mark are extracted from the front view, the shape of the taillight assembly and the sub-brand logo are extracted from the rear view, and the location and shape of the body trim are extracted from the side view. The mark features, component features and trim features are combined into a label feature set, and a feature index table containing image number, shooting angle and label features is constructed.

[0011] Furthermore, it also includes: reading the license plate number of the cloned vehicle candidate pair, querying the traffic violation management system database, extracting violation records within a specified time range, filtering out unprocessed and appealed violation data, classifying the processed violation records into manual processing records and automatic processing records according to processing type, extracting violation images and processing times from the processed records, grouping the violation images according to shooting angle, and constructing a violation image dataset. The violation image is input into the label recognition model, vehicle identification features and decoration features are extracted, violation image feature labels are generated, the vehicle label feature set is read, the feature labels and the label feature set are used to calculate the matching degree score, the label similarity matrix is ​​constructed based on the matching degree score, the vehicle with the highest label similarity is marked as a real car, and the vehicle with the lower label similarity is marked as a cloned car.

[0012] Furthermore, it also includes: reading the label similarity matrix, setting the label matching threshold, identifying vehicles with a label matching degree higher than the threshold as real vehicles, identifying vehicles with a label matching degree lower than the threshold as cloned vehicles, extracting the historical passage records of the cloned vehicles from the vehicle trajectory database, sorting the passage records by timestamp to generate a trajectory sequence, constructing a trajectory prediction model based on a long short-term memory network, inputting the trajectory sequence into the prediction model to generate a driving route within a future time window, and determining the location of the necessary checkpoints based on the road network topology. Read the vehicle characteristic information of the cloned vehicle, combine the characteristic information with the predicted route to generate an analysis report, read the checkpoint deployment rules, write the location of the necessary checkpoint into the deployment command, set the deployment time and deployment area, construct a deployment command set including checkpoint number, deployment time period, vehicle characteristics, and warning level, and send the deployment command set to the traffic control platform.

[0013] Secondly, this application provides a cloned vehicle identification device, comprising: The vehicle feature acquisition module is used to collect vehicle passage data from urban traffic checkpoint equipment and roadside radar sensors, extract basic information and location information of each vehicle, read distance information between adjacent checkpoints, calculate vehicle travel distance based on the semi-sine formula, extract timestamps to calculate time difference, calculate travel speed based on the travel distance and time difference, construct a road network parameter table containing distance thresholds and speed thresholds, compare the travel distance, time difference, and travel speed with the road network parameter table, and filter out candidate vehicles with cloned license plates that cannot be reached within a unit of time. The cloned vehicle identification module is used to read the appearance image of the candidate vehicle with cloned license plates, extract the key parts of the vehicle based on the image segmentation model, input the key parts into the feature recognition model to extract the basic features of the vehicle, construct the vehicle feature vector, calculate the similarity between different vehicle feature vectors, mark the vehicles with similarity exceeding a preset threshold as cloned vehicle candidate pairs, input the images with the same shooting angle in the cloned vehicle candidate pairs into the label recognition model, extract the vehicle detail identification features, and generate the vehicle label feature set. The early warning and control module is used to retrieve the violation record data of the cloned vehicle candidate pair from the traffic violation management system, extract vehicle images from the processed violation records, input the vehicle images into the label recognition model to extract feature labels, compare the feature labels with the vehicle label feature set, identify real vehicles and cloned vehicles based on label matching degree, input the historical trajectory data of the cloned vehicles into the time series prediction model, calculate the future driving route and the necessary checkpoints, and generate a control instruction containing cloned vehicle feature information and driving prediction.

[0014] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the cloned vehicle identification method.

[0015] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the cloned vehicle identification method.

[0016] Fifthly, this application provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the cloned vehicle identification method.

[0017] As can be seen from the above technical solution, this application provides a method and device for identifying cloned vehicles. Through an innovative spatiotemporal analysis model, and by using distance calculation and speed verification, it achieves accurate screening of unreachable vehicles. A feature recognition system is constructed, combining image segmentation and tag extraction to establish a reliable cloned vehicle determination mechanism. Predictive deployment is introduced, using trajectory analysis and route prediction to ensure the accuracy of the crackdown. This method effectively solves the shortcomings of traditional technologies in spatiotemporal analysis, feature recognition, and predictive deployment, providing technical support for the control of cloned vehicles. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the cloned vehicle identification method in the embodiments of this application; Figure 2 This is a structural diagram of the cloned vehicle identification device in the embodiments of this application; Figure 3 This is a schematic diagram of the structure of the electronic device in the embodiments of this application.

[0020] Figure label: Electronic device 9600, central processing unit 9100, memory 9140, communication module 9110, input unit 9120, audio processor 9130, display 9160, power supply 9170, buffer memory 9141, application / function storage unit 9142, data storage unit 9143, driver storage unit 9144, antenna 9111, speaker 9131, microphone 9132. Detailed Implementation

[0021] 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, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] The acquisition, storage, use, and processing of data in this application comply with relevant laws and regulations.

[0023] In view of the problems existing in the prior art, this application provides a method and device for identifying cloned vehicles. Through the innovative design of a spatiotemporal analysis model, and by using distance calculation and speed verification, it achieves accurate screening of unreachable vehicles. A feature recognition system is constructed, combining image segmentation and tag extraction to establish a reliable mechanism for identifying cloned vehicles. Predictive deployment is introduced, using trajectory analysis and route prediction to ensure the accuracy of strikes. This method effectively solves the shortcomings of traditional technologies in spatiotemporal analysis, feature recognition, and predictive deployment, providing technical support for the control of cloned vehicles.

[0024] To effectively address the shortcomings of traditional technologies in spatiotemporal analysis, feature recognition, and predictive deployment, and to provide technical support for the management of cloned vehicles, this application provides an embodiment of a cloned vehicle identification method, see [link to embodiment]. Figure 1 The cloned vehicle identification method specifically includes the following: Step S101: Collect vehicle passage data from urban traffic checkpoint equipment and roadside radar sensors, extract basic information and location information of each vehicle, read the distance information between adjacent checkpoints, calculate the vehicle travel distance based on the semi-sine formula, extract timestamps to calculate time differences, calculate travel speed based on the travel distance and time difference, construct a road network parameter table containing distance thresholds and speed thresholds, compare the travel distance, time difference, and travel speed with the road network parameter table, and filter out candidate vehicles with cloned license plates that cannot be reached within a unit of time. First, vehicle passage data from urban traffic checkpoints and roadside radar sensors are integrated and time-aligned to create continuous records for the same vehicle based on license plate number. Coordinate system one is then implemented for data sources from different devices, mapping the latitude and longitude of the checkpoints to node locations under a unified geographic benchmark. Simultaneously, the capture timestamp and travel direction fields are extracted to form raw trajectory segments that can be advanced over time. To reduce the interference of anomalies on subsequent calculations, records with missing timestamps or significant coordinate jumps are removed, while the reason for removal is retained for later review.

[0025] Based on the original trajectory segments, basic information and location information for each vehicle are extracted, establishing a basic information set with the license plate number as the key. Fields include the initial vehicle appearance image, checkpoint node number, and capture time. To calculate the distance between adjacent checkpoints, the connection relationships between nodes in the road network topology are read. For each pair of adjacent capture points, the corresponding geographical coordinates are queried according to the node number and recorded as adjacent point pairs. Adjacent point pairs maintain a one-to-one correspondence with the capture time, serving as input for the distance and time difference.

[0026] The vehicle's travel distance is calculated based on the adjacent point pairs, and the corresponding time difference is extracted simultaneously. The distance is calculated using a spherical approximation based on the semi-versus principle, and the result is stored in meters. The time difference is obtained by subtracting the previous moment from the next moment, and the unit is recorded in seconds. To avoid excessive zero-distance records caused by short-term jitter, entries with a time difference of zero are marked as invalid and transferred to the exception list. This step outputs a distance-time difference aligned sequence, which will serve as the sole input for speed estimation.

[0027] The driving speed is calculated based on the distance-time difference aligned sequence and then written back and bound to the license plate number. The speed is obtained by dividing the distance by the time difference, stored in a uniform unit, and the generation time window is recorded. Entries with abnormally high speeds are assigned a source weight to prompt subsequent comparison stages to reduce their impact. This speed sequence, along with the distance and time difference values, serves as the data input for subsequent threshold comparisons.

[0028] Based on the data input obtained from the previous processing, the initial elements of the road network parameters are read, and a parameter set including road function level, distance threshold, and speed threshold is established. The parameter set is derived from the road network topology and speed limit standards, and a level mapping is established according to highways, urban expressways, and urban arterial roads. The passable length between adjacent nodes is registered as a level-related distance threshold. To achieve joint judgment, a scoring function is introduced, which is only called when a unified evaluation is required. Q = u1·Jd + u2·Jv u3·Jc.

[0029] In the formula, Q is the joint accessibility score; Jd is the distance relative exceedance, which represents the normalized result of the difference between the driving distance and the distance threshold of the corresponding level; Jv is the speed relative exceedance, which represents the normalized result of the difference between the driving speed and the speed threshold of the corresponding level; Jc is the connection complexity index, which represents the normalized result of the minimum number of road transitions between adjacent points; u1, u2, and u3 are non-negative weights, whose values ​​are configured according to the road level and remain unchanged within the same decision window.

[0030] After the parameter set is determined, the travel distance, time difference, and travel speed are compared one by one with the corresponding distance and speed thresholds to obtain primary anomaly markers. Simultaneously, the joint reachability score Q is calculated, and the initially marked entries are further filtered, retaining records with scores exceeding a set range, and recording the corresponding road level and connection complexity. The results of this filtering are written into the anomaly sequence list for subsequent vehicle-to-vehicle aggregation.

[0031] Based on the list of abnormal sequences, vehicles are aggregated by license plate number to form a set of vehicles to be judged. During aggregation, the time order and checkpoint node sequence are maintained, and the proportion of abnormal entries for each vehicle in the current window is calculated. To avoid the influence of single-shot sensing errors on the final result, vehicles with an abnormality ratio below a set lower limit are only retained as observation items and are not included in the candidate output for this cycle; for vehicles reaching the lower limit, cloned license plate candidate labels are generated, along with the source window and road level distribution, for downstream image verification.

[0032] Based on the candidate tags for cloned license plates, a list of unreachable cloned vehicles within a unit of time is output and associated with the vehicle appearance image index. This list serves as the input for subsequent image segmentation and feature extraction. After reading this list, the image processing unit directly extracts the corresponding appearance image by license plate number and time window and performs key part identification. Simultaneously, the list of abnormal sequences and the parameter set are used in subsequent tag matching and trajectory prediction to trace back the road segments and nodes that caused the anomalies, serving as boundary condition inputs for path reachability constraints. Through this data chain, the output of step S101 is continuously used in the next stage for image comparison and authenticity determination.

[0033] Step S102: Read the appearance image of the candidate vehicle with cloned license plate, extract the key parts of the vehicle based on the image segmentation model, input the key parts into the feature recognition model to extract the basic features of the vehicle, construct the vehicle feature vector, calculate the similarity between the feature vectors of different vehicles, mark the vehicles with similarity exceeding the preset threshold as cloned vehicle candidate pairs, input the images with the same shooting angle in the cloned vehicle candidate pairs into the label recognition model, extract the vehicle detail identification features, and generate the vehicle label feature set. First, the license plate number and time window index are read from the list of candidate vehicles for cloned license plates output in step S101 to locate the corresponding appearance image source. Illumination normalization and noise suppression are performed on each image, and the focal length and distortion are corrected according to the shooting parameters to ensure that the boundaries of the subsequent areas are clearly distinguishable. The corrected images are recorded with the shooting angle and checkpoint number to establish a reference relationship with abnormal road sections on the trajectory side.

[0034] Based on the preprocessed image, an image segmentation model is invoked to output segmentation masks for the front, side, and tail regions. The model uses convolutional features as its backbone, guides boundary transitions with region attention, and applies weakened weights to reflective license plates and shadow areas to reduce the impact of missegmentation. The three segmented regions are scaled to a uniform size, and region type labels are retained as input batches for subsequent feature extraction.

[0035] Based on the segmented regions, the data are fed into the appearance branch and detail branch of the feature recognition model, respectively. The appearance branch extracts texture and contour direction, while the detail branch focuses on local features such as the grille, headlight contours, and trim seams. The two types of features are aligned along the channel dimension and then weighted and fused to output a fixed-length vector as the vehicle feature vector. Each vector includes region type and shooting angle labels to ensure that similarity calculations are performed under comparable conditions.

[0036] Pair matching relationships are established on the vehicle feature vector set, and the comprehensive similarity is calculated for vectors across license plates within the same time window. To avoid bias caused by angle differences, only vector pairs with consistent shooting angles and matching region types are compared. Vehicle pairs with a comprehensive similarity higher than a preset threshold are recorded as clone vehicle candidate pairs, and the image number and time index that generated the candidate relationship are registered for use by the tagging side.

[0037] Based on the candidate cloned vehicles, original images with consistent shooting angles are selected and fed into the label recognition model. This model uses a component-level detector to identify the location and content of the annual inspection sticker, the shape of the sub-brand tail badge characters, the geometric arrangement of the taillight components, and the relative positions of the body trim pieces. The recognition results are output in key-value pair format, along with confidence labels and occlusion markers, facilitating the downstream filtering of low-confidence items.

[0038] Consistency cleaning is performed on the tag recognition output to remove tags with inconsistent angles and low confidence. Multiple images of the same vehicle taken from the same angle are then merged using a voting process to form stable tag entries. During merging, the source image number and shooting time are retained to support subsequent comparisons with traffic violation records. The result of this cleaning and merging process serves as the initial draft of the tag features.

[0039] Based on the initial draft of the label features obtained from the previous processing, a vehicle label feature set is constructed. This feature set uses the license plate number and shooting angle as indexes, and the entries include four categories of fields: annual inspection sticker status, sub-brand tail badge category, taillight component style, and decorative part location. It also records references to abnormal road sections from step S101, so as to link trajectory-side information during authenticity determination. The feature set synchronously outputs a version number to ensure consistency across different cycles.

[0040] After the vehicle tag feature set is determined, the corresponding cloned vehicle candidate pairs are backfilled, and the number of comparable tag fields and missing information for each pair of vehicles at various angles are marked. This backfilling result is used for the next step of tag matching and real vehicle recognition, and at the same time provides a list of target fields and shooting angle conditions for step S103 to extract tags from the violation image, avoiding repeated parsing and unnecessary cross-angle comparisons.

[0041] Step S103: Retrieve the violation record data of the cloned vehicle candidate pair from the traffic violation management system, extract the vehicle images from the processed violation records, input the vehicle images into the label recognition model to extract feature labels, compare the feature labels with the vehicle label feature set, identify the real vehicle and the cloned vehicle based on the label matching degree, input the historical trajectory data of the cloned vehicle into the time series prediction model, calculate the future driving route and the necessary checkpoints, and generate a deployment instruction containing cloned vehicle feature information and driving prediction.

[0042] First, the candidate cloned vehicles and their shooting angle fields, filled in in step S102, are read. A query is then initiated with the traffic violation management system based on the license plate number and time window. The returned violation records are grouped by processing status, retaining only processed entries, and the violation image, processing time, and shooting angle are extracted. To ensure consistency, the angle of the violation image is mapped to the angle index of the vehicle tag feature set; samples that cannot be mapped are marked as not participating in this round of matching.

[0043] The label recognition model is invoked on the processed violation image to output four types of feature labels: annual inspection sticker, sub-brand tail badge, taillight assembly style, and decorative component location. Occlusion and blur marks are recorded simultaneously. These feature labels are aligned with the vehicle label feature set generated in step S102 according to license plate and angle, forming a one-to-one matching entry. To reduce mismatches, fields with true occlusion or insufficient confidence are removed, and the reasons for removal are recorded along with the entry number for subsequent verification.

[0044] The label matching degree is calculated based on the comparison entries, and the determination of whether a vehicle is a genuine or cloned vehicle is output. The matching degree is obtained by weighting the fields according to their consistency ratio and weight. The weights are assigned configuration values ​​related to the distinguishability, with the annual inspection mark and sub-brand tail badge having higher priority than the position of the decorative parts. For multiple violation images of the same candidate pair from the same angle, a voting strategy is used. If a vehicle meets the threshold condition on most entries, it is marked as a genuine vehicle, and the other vehicle is marked as a cloned vehicle. The determination result is written back to the candidate pair index as an input switch for subsequent trajectory prediction.

[0045] After the cloned vehicle is identified, its historical travel records are extracted from the vehicle trajectory database, and a trajectory sequence is constructed in chronological order. The sequence is organized using triplets of checkpoint nodes, passage time, and travel direction, while simultaneously reading the road network topology to establish node connectivity. Missing short segments are filled using interpolation, but without violating road connectivity constraints to prevent unreachable jumps. The cleaned sequence is then delivered to the time series prediction model as input samples.

[0046] The time-series prediction model, constructed using a Long Short-Term Memory (LSTM) network based on the trajectory sequence, outputs the driving route within the future time window. The model uses the sequence window as input unit, learns the temporal preference and directional continuity of checkpoint transfers, and predicts a set of candidate nodes sorted by time, providing an arrival time range for each node. Considering the connectivity of the road network topology, nodes that do not meet the connectivity conditions are eliminated, while waiting markers are retained at boundary conditions for configuration during control periods.

[0047] Based on the candidate node set, mandatory checkpoints are determined. The mandatory checkpoint criterion relies on the uniqueness of historical high-frequency passage statistics and the current connected path. When a node is irreplaceable in the feasible path and appears frequently in the past, it is marked as a mandatory checkpoint. The mandatory checkpoint results and the predicted route together constitute a driving prediction summary, which is then bound to the vehicle label feature set produced in step S102 to form a searchable judgment entry.

[0048] Based on the analysis criteria, a control command is generated. The command includes the cloned vehicle's license plate number, key tag fields, predicted route node sequence, required checkpoint numbers, and suggested control time period, while also recording the road network version and model version. This command is written into the traffic control platform interface as the input for downstream control execution; simultaneously, the required checkpoints and arrival time range are provided for subsequent control feedback comparison, forming a closed-loop link from identification to early warning.

[0049] As described above, the cloned vehicle identification method provided in this application can accurately screen unreachable vehicles through innovative spatiotemporal analysis model design and distance calculation and speed verification. It constructs a feature recognition system, combining image segmentation and label extraction to establish a reliable cloned vehicle determination mechanism. Predictive deployment is introduced, using trajectory analysis and route prediction to ensure the accuracy of the crackdown. This method effectively solves the shortcomings of traditional technologies in spatiotemporal analysis, feature recognition, and predictive deployment, providing technical support for cloned vehicle control.

[0050] In one embodiment of the cloned vehicle identification method of this application, it may further include the following: Step S201: Read vehicle capture records from the urban traffic checkpoint database, parse vehicle passage information from roadside radar data stream, group the capture records and passage information by license plate number, extract the license plate number, capture time, and checkpoint location information from each group of data, construct a vehicle basic information table, read the road network topology data from the electronic map, map the checkpoint location information to road network nodes, calculate the road connection relationship between adjacent checkpoints, and generate a checkpoint distance matrix; Step S202: Substitute the latitude and longitude coordinates of adjacent checkpoints into the semi-versus formula to calculate the actual driving distance of the vehicle. Extract the timestamp information of adjacent capture points from the vehicle basic information table, convert the timestamp information into a time difference value, divide the driving distance by the time difference value to obtain the driving speed, and construct a parameter calculation table containing license plate number, driving distance, time difference value, and driving speed.

[0051] First, the system connects to the city's traffic checkpoint database and roadside radar data stream, establishing a joint index based on license plate numbers and aligning it by time. For captured images, license plate numbers, capture times, and checkpoint latitude and longitude are extracted. For radar traffic information, license plate numbers, passage times, and trajectory points are extracted, and data is merged and deduplicated based on temporal and spatial proximity. During the merging process, records with missing times or abnormal coordinates are marked with reasons and temporarily stored in an anomaly set to avoid entering subsequent mapping stages.

[0052] Based on the composite index, core fields for each vehicle group are extracted to generate a basic information set with the license plate number as the key. These fields include the capture time sequence, checkpoint latitude and longitude sequence, and source marker. Subsequently, road network topology data from the electronic map is read to establish a mapping relationship between checkpoint coordinates and road network nodes. The mapping uses the nearest reachable node principle and verifies whether the road direction matches the capture direction; inconsistent samples are written back to the anomaly set. After mapping, a node sequence sorted by time is obtained, which serves as input for calculating the connectivity relationships.

[0053] The edge set in the road network topology is queried based on the node sequence to identify the road connection relationship between adjacent checkpoints. For adjacent nodes with multiple feasible edges, the shortest connectivity length and the number of candidate edges are recorded to reflect the route ambiguity. The shortest connectivity lengths of all adjacent node pairs are aggregated into a matrix form by license plate number, with the matrix index being the time sequence and node pair number, forming a checkpoint distance matrix. The road network version number is also saved to ensure subsequent traceability.

[0054] After the checkpoint distance matrix is ​​generated, the latitude and longitude coordinates and corresponding capture times of adjacent checkpoints are read, and the distance and time difference calculation process begins. The distance is approximated using the semi-sine principle, and the result is recorded in meters; the time difference is obtained by subtracting the previous time from the next time, and the unit is seconds. Entries with zero or negative time differences are marked as abnormal and are not included in the speed calculation. At the same time, the associated node pair number is recorded for verification by the upstream data side.

[0055] Based on the aforementioned distance and time difference, the driving speed between adjacent capture points is calculated and linked to the license plate number for write-back. The speed is obtained by dividing the distance by the time difference, with the unit uniformly set to meters per second, and includes the generation of a time window and node pair number. Entries with abnormally high speeds and a large number of corresponding candidate edges are additionally marked with ambiguity alerts to prompt a reduction in confidence during subsequent threshold comparison stages.

[0056] Once the speed results are available, parameter calculation entries are organized to construct a parameter set containing license plate number, adjacent node pair number, travel distance, time difference, and travel speed. This set maintains index consistency with the previous checkpoint distance matrix for easy cross-table lookups; simultaneously, the road network version and source tag are retained for each entry to ensure no confusion between different data batches.

[0057] The parameter calculation table is output based on the parameter set and used as the direct input for subsequent threshold comparison. During the threshold comparison stage, distance and speed thresholds are read based on the road function level. The parameter calculation table is searched line by line to generate anomaly markers. Simultaneously, this table is also read by the image side in step S102 to limit the time window and checkpoint locations for extracting the appearance image, reducing unnecessary image resolution.

[0058] Finally, the parameter calculation table and the checkpoint distance matrix together serve as the upstream inputs for steps S101 and S103: the former provides the three elements of distance, time, and speed required for accessibility determination, while the latter provides a location basis for retrieving and matching violation records through node pair numbers and time windows. Through the above organization and write-back relationship, data maintains a consistent index and version closure across different stages.

[0059] In one embodiment of the cloned vehicle identification method of this application, it may further include the following: Step S301: Read the road network topology data, classify the roads into expressways, urban expressways and urban arterial roads according to their functional levels, extract the speed limit standards for each type of road, set speed thresholds based on the speed limit standards, read the road segment lengths in the checkpoint distance matrix, set distance thresholds based on the road segment lengths, write the speed thresholds and distance thresholds into the road network parameter table according to the road level, and construct road traffic constraint rules. Step S302: Read the vehicle driving records in the parameter calculation table, compare the driving distance with the distance threshold in the road network parameter table, compare the driving speed with the speed threshold in the road network parameter table, mark driving records with distances exceeding the thresholds and speeds exceeding the thresholds as abnormal data, group the abnormal data by license plate number, and generate a dataset of candidate vehicles with cloned license plates.

[0060] First, after reading the road network topology data from the electronic map, the edge set is layered according to road function level to clarify the mapping relationship between expressways, urban expressways, and urban arterial roads. Speed ​​limit standards and direction attributes are extracted for each layer and registered as the source of speed boundaries. Then, combined with the aligned adjacent node pairs in the checkpoint distance matrix, the corresponding road segment length and candidate edge count are read to initially form a draft threshold configuration based on road level, for subsequent rule implementation.

[0061] Based on the aforementioned threshold configuration draft, consistency checks are performed on the speed and distance thresholds within each level. The checks are based on the distribution patterns of historical traffic samples at the same level, eliminating the influence of extreme values ​​to determine the speed and distance threshold ranges for that level. These ranges are then cross-checked with road segment directionality; samples that do not meet directional consistency requirements are transferred to an anomaly list. This results in a threshold mapping indexed by level, used for traffic constraint generation.

[0062] Based on the threshold mapping, road traffic constraint rules are constructed. These rules include two types of criteria: one based on speed and distance boundaries for a single road segment, and the other based on connection complexity indicators related to the number of candidate edges. For adjacent node pairs with excessively high connection complexity, a conservative threshold condition is added to reduce the probability of misjudgment. This rule set shares node pair numbers with the checkpoint distance matrix, ensuring that subsequent comparisons can locate each node individually.

[0063] After the rule set is determined, the vehicle driving records in the parameter calculation table are read, and the node numbers are aligned with the license plate numbers. For each record, the driving distance is compared with the distance threshold of the corresponding level, and the driving speed is compared with the speed threshold of the corresponding level, resulting in two Boolean results with a connection complexity warning flag. Entries with zero or negative time differences are retained as exceptions and are not included in this round of judgment.

[0064] Based on the aforementioned comparison results, records exhibiting both distance and speed exceeding the threshold were filtered out and marked as abnormal data. Their node pair numbers, road classifications, and warning labels were then recorded. To reduce sporadic errors, records were aggregated within a time window by license plate number, the proportion of abnormal entries was statistically analyzed, and the location indices of first and consecutive occurrences were retained as clues for determining trajectory continuity.

[0065] Based on the aggregation results, a candidate vehicle dataset for license plate cloning is generated. This dataset is indexed by license plate number, and each entry includes a list of abnormal records, the distribution of road levels involved, and the distribution of connection complexity. It also establishes a reference with the abnormal sequence list from step S101, forming cross-stage evidence association. The dataset synchronously outputs a retrieval key for the image side, pointing to the time window and checkpoint location of the appearance image to be extracted, supporting the segmentation and feature extraction entry point in step S102.

[0066] Finally, the road traffic constraint rules and the dataset of candidate vehicles with cloned license plates are written into the reading interface of subsequent steps: the former is called as an accessibility boundary in the subsequent authenticity determination, and the latter is called as a joint retrieval index for images and violations. Through this organization, steps S301 and S302 solidify the comparison process of topology, threshold, and driving records into a traceable determination link.

[0067] In one embodiment of the cloned vehicle identification method of this application, it may further include the following: Step S401: Extract vehicle exterior images from the candidate vehicle dataset for cloned license plates, input the exterior images into the image enhancement module to perform illumination normalization and noise suppression, construct an image segmentation network containing convolutional and deconvolutional layers, input the enhanced image into the image segmentation network to generate a vehicle region segmentation mask, segment the vehicle image into a front face region, a side region, and a rear region based on the segmentation mask, and scale the segmented regions to a uniform size according to a preset ratio; Step S402: Input the scaled segmented region into the feature recognition model, extract vehicle appearance features based on convolutional neural network, extract local detail features based on region attention mechanism, fuse the appearance features and detail features according to preset weights to construct a vehicle multidimensional feature matrix, perform dimensionality reduction calculation on the feature matrix to generate a fixed-dimensional vehicle feature vector.

[0068] First, license plate numbers and corresponding time windows are read from the candidate vehicle dataset for cloned license plates generated in step S302 to locate the source of the appearance images and complete batch retrieval. Each image enters a preprocessing stage, where the image enhancement module is called to perform illumination normalization and noise suppression, and distortion correction and resolution standardization are performed based on the shooting parameters. This stage also records the shooting angle and checkpoint number, which serve as index fields for subsequent region segmentation and cross-angle comparison, preventing samples from different angles from being mixed in the same batch.

[0069] An image segmentation network containing convolutional and deconvolutional layers is constructed based on the enhanced image, employing a dual-path structure for foreground suppression and edge refinement. The enhanced image is fed into the segmentation network, outputting a vehicle region segmentation mask, with confidence bars appended to the mask boundaries to mark uncertain regions. Based on the mask, the vehicle image is segmented into three sub-images: front, side, and rear regions. Region confidence bars guide subsequent cropping, preserving the geometric integrity of component boundaries as much as possible.

[0070] Based on the three types of sub-images, each type is scaled to a uniform size according to a preset ratio, while maintaining a stable pixel aspect ratio. During the scaling process, the region type, shooting angle, and checkpoint number are written into the sub-image metadata to ensure that vehicles within the same batch can be directly binned according to angle. A batch index is established for the scaled sub-images as the input queue for the feature recognition model, while retaining the reference to the node pair number in step S302, facilitating subsequent tracing of the association between abnormal road sections and image sources.

[0071] Once the input queue is available, each sub-image is fed into two branches of the feature recognition model: the appearance branch extracts overall texture, contour, and shape information based on a convolutional neural network, while the detail branch focuses on local features such as grille bars, light group contours, decorative seams, and the edges of annual inspection signs under the guidance of a region attention mechanism. Each branch outputs a fixed-length representation, which is then aligned along the channel dimension and fused according to preset weights to obtain a fused representation of a single region.

[0072] According to the fusion representation, the representations of the three types of regions are concatenated according to the vehicle and the shooting angle to form a multi-dimensional feature matrix for the vehicle. Each column in the matrix corresponds to the source of the region, and each row corresponds to the channel dimension. The metadata retains the region type and angle labels to ensure that subsequent similarity comparisons are only performed within comparable scenes. To reduce the computational burden caused by redundant channels, dimensionality reduction is performed on the feature matrix to retain the components that contribute significantly to the vehicle differentiation task.

[0073] In the dimensionality reduction process, the dimensionality reduction mapping is kept fixed within the current batch, and the mapping version and training source are recorded to prevent drift introduced across batches. The dimensionality reduction result is output as a fixed-dimensional vector, which is strongly bound to the license plate number, shooting angle, and regional participation to form the vehicle feature vector. For samples with significant confidence band coverage, a quality label is added for weight adjustment in subsequent similarity calculation stages.

[0074] Based on the vehicle feature vector, a pairwise retrieval index is established according to the needs of license plate cross-matching. This index only participates in subsequent matching if the shooting angle is consistent and all three types of regions are present or the proportion of comparable fields meets the requirements. This index structure refers back to the abnormal entries and time windows in step S302, facilitating the simultaneous examination of the unreachable context in the road network when highly similar pairs are discovered, thus forming a consistent cross-domain evidence chain.

[0075] Finally, the vehicle feature vector is output to the similarity calculation and cloned vehicle candidate pair determination input in step S102; the segmentation mask and sub-image metadata are synchronously retained for direct reuse in step S102 when extracting detail identification features from images with consistent shooting angles in the label recognition model. Through the above organization, steps S401 and S402 connect image enhancement, region segmentation, feature extraction, and dimensionality reduction compression into a traceable processing link, providing stable input for subsequent similarity measurement and label-level comparison.

[0076] In one embodiment of the cloned vehicle identification method of this application, it may further include the following: Step S501: Read the vehicle feature vector, construct the feature matching matrix, calculate the Euclidean distance and cosine similarity between the feature vectors, weight the Euclidean distance and cosine similarity to obtain the comprehensive similarity, determine the similarity threshold based on cluster analysis, mark the vehicle pairs with comprehensive similarity higher than the threshold as clone vehicle candidate pairs, read the image shooting parameters of the clone vehicle candidate pairs, divide the images into front view, rear view and side view based on camera angle information, and filter the image groups with consistent shooting angles; Step S502: Input images with consistent shooting angles into the label recognition model, extract the location and content of the annual inspection label from the front view, extract the taillight component shape and sub-brand logo from the rear view, extract the location and shape of the body trim from the side view, combine the label features, component features, and trim features into a label feature set, and construct a feature index table containing image number, shooting angle, and label features.

[0077] First, the vehicle feature vectors output in step S402 are read. Batch indexes are established based on license plate number and shooting angle, and paired retrieval sets are constructed under the same shooting angle. Two types of metrics are calculated for each retrieval set: Euclidean distance to measure the magnitude difference between vectors, and cosine similarity to measure directional consistency. Both metrics are normalized using the quantiles within the batch to avoid extreme samples dominating the combination results, while retaining the regional participation and quality label of each pair of samples as the basis for subsequent weighting.

[0078] Based on the measurement results, a feature matching matrix is ​​generated. The Euclidean distance and cosine similarity of each vehicle pair are weighted and combined to obtain a comprehensive similarity. The combination weights are adjusted according to region completeness and quality labels, and the dependence on directional measurement is reduced when regions are missing. To obtain a robust decision boundary, cluster analysis is used to separate high-similarity and low-similarity groups from the comprehensive similarity, and a threshold is determined at the gap between the groups to output a preliminary set of high-similarity vehicle pairs, while recording the corresponding shooting angle index.

[0079] Based on the set of highly similar vehicle pairs, their image capture parameters were read back and divided into front view, rear view, and side view according to camera angle information. Figure 3 The process involves grouping images and performing a consistency check within each group. Samples with angle drift or missing metadata are removed, and a representative frame closest to the standard angle is retained for each group. This filtering process yields image groups with consistent shooting angles, which, along with the overall similarity score, are written back to the candidate pair index for use by the tagging side.

[0080] The label recognition model is invoked on the image group with consistent angles, and feature extraction is performed according to the angle. For the front view, the position and content of the annual inspection sticker are detected, and the layout, relative coordinates, and character confidence are output. For the rear view, the geometry of the taillight assembly, the color of the lampshade, and the character structure of the sub-brand's taillight logo are analyzed. For the side view, the relative position and shape outline of the body trim pieces are located. Each recognition result is accompanied by occlusion and blur markers to remove unreliable fields during subsequent merging.

[0081] Based on the recognition results, field-level cleaning is performed, removing entries with true occlusion or insufficient confidence from the current round of judgment, and merging multiple frames of the same vehicle at the same angle through voting. The merging strategy prioritizes retaining entries with stable layout and consistent character structure, while also preserving the source image number and shooting time to ensure traceability later. The cleaned and merged output forms three subsets in the angle dimension, providing structured input for feature assembly.

[0082] Based on the three subsets, the annual inspection label features, taillight assembly and sub-brand logo features, and decorative part position and shape features are combined into a tag feature set according to a preset field order. To avoid cross-angle mismatch, the tag feature set retains angle labels and vehicle logos, and establishes a mapping relationship with the area quality mark in step S402 to ensure that low-quality areas are weakened in subsequent comparisons. This tag feature set is also associated with a comprehensive similarity score to provide prior clues in the authenticity determination stage.

[0083] Based on the aforementioned tag feature set, a feature index table is constructed, containing image ID, shooting angle, and tag features. The index table uses license plate number and candidate pair ID as the primary key, and records the feature vector batch from step S402 and the road level reference from step S301 in the extended key, facilitating cross-domain joint retrieval. The index table serves as the direct input for subsequent step S103, used for comparing violation image tags and identifying real and cloned vehicles. It also supports rapid backtracking and verification of the original image and feature vectors when conflicts are detected.

[0084] In one embodiment of the cloned vehicle identification method of this application, it may further include the following: Step S601: Read the license plate number of the cloned vehicle candidate pair, query the traffic violation management system database, extract the violation records within the specified time range, filter the unprocessed and appealed violation data, divide the processed violation records into manual processing records and automatic processing records according to the processing type, extract the violation images and processing time from the processed records, group the violation images according to the shooting angle, and construct a violation image dataset. Step S602: Input the violation image into the label recognition model, extract vehicle identification features and decoration features, generate violation image feature labels, read the vehicle label feature set, calculate the matching degree score between the feature labels and the label feature set, construct a label similarity matrix based on the matching degree score, mark the vehicle with the highest label similarity as a real vehicle, and mark the vehicle with the lower label similarity as a cloned vehicle.

[0085] First, the candidate pair numbers and license plate numbers from the feature index table output in step S502 are read. After creating a search list by license plate, a query is initiated to the traffic violation management system. The returned results are limited to a specified time range, and the processing status, processing type, and violation image of each record are extracted. Unprocessed records and records in the appeal stage are directly removed, and the reason for removal and timestamp are recorded to avoid interference in subsequent comparison stages.

[0086] Based on the processed records, the samples are divided into manually processed records and automatically processed records according to the processing type. The violation image and processing time fields are extracted for each type of sample. The extracted violation images are then read for their shooting angle and device number metadata. Based on the angle label, the samples are categorized into front view, rear view, and side view. Figure 3 Groups. Samples with missing angles are marked as needing to be supplemented and removed from the current batch. The three groups of samples, along with the candidate pair numbers, together form the violation image dataset, which serves as the input for label recognition.

[0087] Based on the aforementioned dataset of traffic violation images, a label recognition model is invoked group by group to output vehicle identification features and decorative features. The relative position and text content of the annual inspection sticker are extracted from the front view; the geometric style of the taillight assembly and the character structure of the sub-brand's taillight logo are analyzed from the rear view; and the relative coordinates and shape contours of the decorative parts are extracted from the side view. Each result is simultaneously written to an occlusion marker and a sharpness marker for subsequent weight control. This step produces feature labels for the traffic violation images, which are strongly bound to the image number and shooting angle.

[0088] Based on the violation image feature tags, the vehicle tag feature set formed in step S502 is read, and the entries on both sides are aligned according to the license plate number and the shooting angle. To avoid low-quality samples affecting the matching, fields with true occlusion or insufficient clarity are first removed, and then tags from multiple frames at the same angle are merged into a single entry by voting. After alignment and cleaning, the matching degree calculation stage begins, and the similarity score at the tag level is output for each entry, and the number of participating fields is recorded.

[0089] Based on the aforementioned matching output, a label similarity matrix is ​​constructed, and true / false pairs are determined within the candidate pair dimension. To ensure consistency in the determination, a weighted scoring method is introduced, which is only explained in this section and will be used in the next section: P = r1·Mf + r2·Mb r3·Md.

[0090] In the formula, P is the comprehensive score for tag matching; Mf is the consistency ratio of the identification field, which comes from the matching result of the annual inspection label and the sub-brand tail label; Mb is the consistency ratio of the component style, which comes from the matching result of the taillight assembly and the decorative parts; Md is the inconsistency field ratio, which is used to penalize conflict items; r1, r2, and r3 are non-negative weights, which are configured according to the shooting angle and the discernibility of the field, and remain unchanged within the current candidate pair.

[0091] After the score P is calculated, the P values ​​of the two vehicles within each candidate pair are compared. The vehicle with the higher P value is marked as the genuine vehicle, and the other is marked as a cloned vehicle. In cases where scores are close and the number of fields is insufficient, the vehicle is marked as requiring review, and a list of angles and fields to be supplemented is written back for subsequent batches of samples. The judgment result, along with the candidate pair number, shooting angle, and list of fields used, is recorded to form a traceable authenticity label.

[0092] Based on the authenticity labeling, the license plate numbers of vehicles marked as cloned are extracted, and the trajectory records and node pair indices established in steps S101 and S201 are read and organized into a time-ordered passage sequence. This sequence, along with the connectivity of the road network topology, is used as input to the trajectory-side temporal prediction model for subsequent route and checkpoint inference. Simultaneously, the label similarity matrix, along with the scoring weights, is saved as an assessment attachment, facilitating its distribution along with the cloned vehicle feature information during the deployment phase.

[0093] In one embodiment of the cloned vehicle identification method of this application, it may further include the following: Step S701: Read the label similarity matrix, set the label matching threshold, identify vehicles with a label matching degree higher than the threshold as real vehicles, identify vehicles with a label matching degree lower than the threshold as cloned vehicles, extract the historical passage records of the cloned vehicles from the vehicle trajectory database, sort the passage records by timestamp to generate a trajectory sequence, construct a trajectory prediction model based on a long short-term memory network, input the trajectory sequence into the prediction model to generate the driving route within the future time window, and determine the location of the necessary checkpoints based on the road network topology. Step S702: Read the vehicle feature information of the cloned vehicle, combine the feature information with the predicted route to generate an analysis report, read the checkpoint deployment rules, write the location of the necessary checkpoint into the deployment instruction, set the deployment time and deployment area, construct a deployment instruction set including checkpoint number, deployment time period, vehicle features, and warning level, and send the deployment instruction set to the traffic control platform.

[0094] First, the label similarity matrix formed in step S602 is read, and each candidate pair is retrieved by its number, with a label matching threshold set. The matching degree of each vehicle under each shooting angle is summarized to obtain an angle-weighted overall matching degree entry, and the number of participating fields and missing information are recorded in each entry. Vehicles with an overall matching degree higher than the threshold are marked as genuine vehicles, and vehicles with a matching degree lower than the threshold are marked as cloned vehicles. For critical samples, a checklist and a list of angles that need to be supplemented are registered to avoid misjudgment and entry into the prediction chain.

[0095] Based on the cloned vehicle tag set, a vehicle trajectory database is connected to extract the historical passage records of the corresponding license plates. The passage records are sorted by timestamp to generate trajectory sequences, which include three pieces of information: checkpoint nodes, passage time, and direction of travel. These sequences are then mapped to the node connectivity relationships in the road network topology. Short missing segments are partially filled in with time and node information without violating connectivity constraints. Records with reversed time sequences are written back as an anomaly and removed to ensure that the input samples can be used for time series modeling.

[0096] Based on the trajectory sequence, a trajectory prediction model based on a Long Short-Term Memory (LSTM) network is constructed, hereinafter referred to as the trajectory prediction model. This model uses a fixed-length time window as the input unit. The input dimension includes the embedded representation of node numbers and the numerical encoding of adjacent time intervals. The output is the sequence of candidate nodes within the future time window and their corresponding arrival time ranges. To avoid unreasonable jumps, both model training and inference are constrained by a connectivity mask, which is derived from the current road network topology version.

[0097] Based on the candidate node sequence output by the model, the locations of necessary checkpoints are determined by combining the road network topology. The determination of necessity uses two types of evidence: first, the intersection nodes of all feasible paths; and second, nodes that are high-frequency in historical trajectories and have no alternative detour in the current connected graph. For cases with unclear boundaries, multiple alternatives are retained with confidence labels for selection by the control rules when setting time periods. This step generates a list of future driving routes and necessary checkpoints, establishing a one-to-one correspondence with cloned vehicle markers.

[0098] Based on the aforementioned list of future driving routes and necessary checkpoints, the vehicle characteristic information of the cloned vehicle is read, including license plate number, annual inspection sticker status, sub-brand tail logo category, and taillight component style, as well as the confidence label field retained in step S602. This characteristic information is combined with the predicted route to generate an analysis report containing a timeline, node sequence, and key labels. The report references the road network version and model version, and indicates the prediction generation time and validity period for unified management on the platform side.

[0099] Based on the aforementioned analysis report, the checkpoint deployment rules are retrieved. These rules include time period lengths, warning level settings, and multi-point linkage strategies. The locations of essential checkpoints are written into the deployment instructions, and deployment time windows are set according to the arrival time range. Deployment areas are then divided according to the order of the routes. For cases with multiple alternative essential points, a hierarchical configuration is adopted, prioritizing the deployment of high-confidence checkpoints while setting secondary windows for low-confidence checkpoints to reduce resource consumption.

[0100] Based on the deployment configuration, a deployment instruction set is constructed, including checkpoint number, deployment time period, vehicle characteristics, and warning level. The instruction set establishes a primary key based on the license plate number and records the referenced tag similarity matrix number and trajectory prediction batch number to ensure traceability of subsequent feedback. To adapt to the platform's distribution interface, the instruction set is encoded into batch messages, with each message including its effective and expiration times to avoid cross-time period interference.

[0101] Once the instruction set is prepared, it is sent to the traffic control platform and linked to the platform's feedback channel. The deployment results returned by the platform include whether a match was achieved, the actual capture time, and the checkpoint number, serving as the data source for subsequent model evaluation and threshold adjustment. A foreign key relationship is established between the hit record and the aforementioned analysis report to verify the accuracy of the mandatory checkpoint determination and drive the next cycle's update of the tag matching threshold and time window length.

[0102] To effectively address the shortcomings of traditional technologies in spatiotemporal analysis, feature recognition, and predictive deployment, and to provide technical support for the management of cloned vehicles, this application provides an embodiment of a cloned vehicle identification device for implementing all or part of the aforementioned cloned vehicle identification method. See [link to embodiment]. Figure 2 The cloned vehicle identification device specifically includes the following components: The vehicle feature acquisition module 10 is used to collect vehicle passage data from urban traffic checkpoint equipment and roadside radar sensors, extract basic information and location information of each vehicle, read distance information between adjacent checkpoints, calculate vehicle travel distance based on the semi-sine formula, extract timestamps to calculate time difference, calculate travel speed based on the travel distance and time difference, construct a road network parameter table containing distance thresholds and speed thresholds, compare the travel distance, time difference, and travel speed with the road network parameter table, and filter out candidate vehicles with cloned license plates that cannot be reached within a unit of time. The cloned vehicle identification module 20 is used to read the appearance image of the candidate vehicle with cloned license plates, extract the key parts of the vehicle based on the image segmentation model, input the key parts into the feature recognition model to extract the basic features of the vehicle, construct the vehicle feature vector, calculate the similarity between different vehicle feature vectors, mark vehicles with similarity exceeding a preset threshold as cloned vehicle candidate pairs, input the images with the same shooting angle in the cloned vehicle candidate pairs into the label recognition model, extract the vehicle detail identification features, and generate a vehicle label feature set. The early warning and control module 30 is used to retrieve the violation record data of the cloned vehicle candidate pair from the traffic violation management system, extract vehicle images from the processed violation records, input the vehicle images into the label recognition model to extract feature labels, compare the feature labels with the vehicle label feature set, identify real vehicles and cloned vehicles based on label matching degree, input the historical trajectory data of the cloned vehicles into the time series prediction model, calculate the future driving route and the necessary checkpoints, and generate a control instruction containing cloned vehicle feature information and driving prediction.

[0103] As described above, the cloned vehicle identification device provided in this application can accurately screen unreachable vehicles through an innovative spatiotemporal analysis model, using distance calculation and speed verification. It constructs a feature recognition system, combining image segmentation and tag extraction to establish a reliable cloned vehicle determination mechanism. Predictive deployment is introduced, using trajectory analysis and route prediction to ensure the accuracy of the crackdown. This method effectively addresses the shortcomings of traditional technologies in spatiotemporal analysis, feature recognition, and predictive deployment, providing technical support for cloned vehicle control.

[0104] From a hardware perspective, in order to effectively address the shortcomings of traditional technologies in spatiotemporal analysis, feature recognition, and predictive deployment, and to provide technical support for the management of cloned vehicles, this application provides an embodiment of an electronic device for implementing all or part of the cloned vehicle identification method. The electronic device specifically includes the following components: The system comprises a processor, memory, a communications interface, and a bus; wherein the processor, memory, and communications interface communicate with each other via the bus; the communications interface is used to transmit information between the cloned vehicle identification device and core business systems, user terminals, and related databases and other related devices; the logic controller can be a desktop computer, tablet computer, or mobile terminal, etc., and this embodiment is not limited to these. In this embodiment, the logic controller can be implemented with reference to the embodiments of the cloned vehicle identification method and the cloned vehicle identification device in the embodiments, the content of which is incorporated herein, and repeated details will not be described again.

[0105] It is understood that the user terminal may include smartphones, tablet computers, network set-top boxes, portable computers, desktop computers, personal digital assistants (PDAs), in-vehicle devices, smart wearable devices, etc. Among these, the smart wearable devices may include smart glasses, smartwatches, smart bracelets, etc.

[0106] In practical applications, the cloned vehicle identification method can be partially executed on the electronic device side as described above, or all operations can be completed in the client device. The choice can be made based on the processing power of the client device and the limitations of the user's usage scenario. This application does not impose any limitations on this. If all operations are completed in the client device, the client device may further include a processor.

[0107] The aforementioned client device may have a communication module (i.e., a communication unit) that can communicate with a remote server to achieve data transmission with the server. The server may include a server on the task scheduling center side; in other implementation scenarios, it may also include a server on an intermediate platform, such as a server on a third-party server platform that has a communication link with the task scheduling center server. The server may include a single computer device, a server cluster consisting of multiple servers, or a distributed server structure.

[0108] Figure 3 This is a schematic block diagram illustrating the system configuration of the electronic device 9600 according to an embodiment of this application. Figure 3 As shown, the electronic device 9600 may include a central processing unit 9100 and a memory 9140; the memory 9140 is coupled to the central processing unit 9100. It is worth noting that... Figure 3 This is an example; other types of structures can also be used to supplement or replace this structure to achieve telecommunications functions or other functions.

[0109] In one embodiment, the cloned vehicle identification method function can be integrated into the central processing unit 9100. The central processing unit 9100 can be configured to perform the following control: Step S101: Collect vehicle passage data from urban traffic checkpoint equipment and roadside radar sensors, extract basic information and location information of each vehicle, read the distance information between adjacent checkpoints, calculate the vehicle travel distance based on the semi-sine formula, extract timestamps to calculate time differences, calculate travel speed based on the travel distance and time difference, construct a road network parameter table containing distance thresholds and speed thresholds, compare the travel distance, time difference, and travel speed with the road network parameter table, and filter out candidate vehicles with cloned license plates that cannot be reached within a unit of time. Step S102: Read the appearance image of the candidate vehicle with cloned license plate, extract the key parts of the vehicle based on the image segmentation model, input the key parts into the feature recognition model to extract the basic features of the vehicle, construct the vehicle feature vector, calculate the similarity between the feature vectors of different vehicles, mark the vehicles with similarity exceeding the preset threshold as cloned vehicle candidate pairs, input the images with the same shooting angle in the cloned vehicle candidate pairs into the label recognition model, extract the vehicle detail identification features, and generate the vehicle label feature set. Step S103: Retrieve the violation record data of the cloned vehicle candidate pair from the traffic violation management system, extract the vehicle images from the processed violation records, input the vehicle images into the label recognition model to extract feature labels, compare the feature labels with the vehicle label feature set, identify the real vehicle and the cloned vehicle based on the label matching degree, input the historical trajectory data of the cloned vehicle into the time series prediction model, calculate the future driving route and the necessary checkpoints, and generate a deployment instruction containing cloned vehicle feature information and driving prediction.

[0110] As described above, the electronic device provided in this application, through an innovative spatiotemporal analysis model, achieves accurate screening of unreachable vehicles via distance calculation and speed verification. It constructs a feature recognition system, combining image segmentation and tag extraction to establish a reliable mechanism for identifying cloned vehicles. Predictive deployment is introduced, using trajectory analysis and route prediction to ensure the accuracy of strikes. This method effectively addresses the shortcomings of traditional technologies in spatiotemporal analysis, feature recognition, and predictive deployment, providing technical support for the control of cloned vehicles.

[0111] In another embodiment, the cloned vehicle identification device can be configured separately from the central processing unit 9100. For example, the cloned vehicle identification device can be configured as a chip connected to the central processing unit 9100, and the cloned vehicle identification method function can be implemented through the control of the central processing unit.

[0112] like Figure 3 As shown, the electronic device 9600 may further include: a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It is worth noting that the electronic device 9600 does not necessarily need to include these components. Figure 3 All components shown; in addition, the electronic device 9600 may also include Figure 3 For components not shown, please refer to existing technologies.

[0113] like Figure 3 As shown, the central processing unit 9100, sometimes also referred to as a controller or operating control, may include a microprocessor or other processor device and / or logic device, which receives inputs and controls the operation of various components of the electronic device 9600.

[0114] The memory 9140 may be, for example, one or more of a cache, flash memory, hard drive, removable media, volatile memory, non-volatile memory, or other suitable devices. It may store the aforementioned failure-related information, and also store a program for executing that information. The central processing unit 9100 may execute the program stored in the memory 9140 to perform information storage or processing, etc.

[0115] Input unit 9120 provides input to central processing unit 9100. Input unit 9120 may be, for example, a keypad or touch input device. Power supply 9170 provides power to electronic device 9600. Display 9160 displays images and text. Display may be, for example, an LCD display, but is not limited thereto.

[0116] The memory 9140 can be a solid-state memory, such as a read-only memory (ROM), random access memory (RAM), a SIM card, etc. It can also be a memory that retains information even when power is off, can be selectively erased, and contains more data; examples of this type of memory are sometimes referred to as EPROMs. The memory 9140 can also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes referred to as a buffer). The memory 9140 may include an application / function storage unit 9142 for storing application programs and function programs or processes for executing the operation of the electronic device 9600 via the central processing unit 9100.

[0117] The memory 9140 may also include a data storage unit 9143 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 9144 of the memory 9140 may include various drivers for the electronic device for communication functions and / or for performing other functions of the electronic device (such as messaging applications, address book applications, etc.).

[0118] The communication module 9110 is a transmitter / receiver that sends and receives signals via the antenna 9111. The communication module 9110 (transmitter / receiver) is coupled to the central processing unit 9100 to provide input signals and receive output signals, which is the same as in a conventional mobile communication terminal.

[0119] Based on different communication technologies, multiple communication modules 9110 can be configured in the same electronic device, such as cellular network modules, Bluetooth modules, and / or wireless LAN modules. The communication module 9110 (transmitter / receiver) is also coupled to a speaker 9131 and a microphone 9132 via an audio processor 9130 to provide audio output via the speaker 9131 and receive audio input from the microphone 9132, thereby realizing typical telecommunications functions. The audio processor 9130 may include any suitable buffer, decoder, amplifier, etc. Additionally, the audio processor 9130 is coupled to a central processing unit 9100, enabling on-device recording via the microphone 9132 and on-device playback of stored audio via the speaker 9131.

[0120] Embodiments of this application also provide a computer-readable storage medium capable of implementing all steps of the cloned vehicle identification method with a server or client as the execution subject in the above embodiments. The computer-readable storage medium stores a computer program that, when executed by a processor, implements all steps of the cloned vehicle identification method with a server or client as the execution subject in the above embodiments. For example, when the processor executes the computer program, it implements the following steps: Step S101: Collect vehicle passage data from urban traffic checkpoint equipment and roadside radar sensors, extract basic information and location information of each vehicle, read the distance information between adjacent checkpoints, calculate the vehicle travel distance based on the semi-sine formula, extract timestamps to calculate time differences, calculate travel speed based on the travel distance and time difference, construct a road network parameter table containing distance thresholds and speed thresholds, compare the travel distance, time difference, and travel speed with the road network parameter table, and filter out candidate vehicles with cloned license plates that cannot be reached within a unit of time. Step S102: Read the appearance image of the candidate vehicle with cloned license plate, extract the key parts of the vehicle based on the image segmentation model, input the key parts into the feature recognition model to extract the basic features of the vehicle, construct the vehicle feature vector, calculate the similarity between the feature vectors of different vehicles, mark the vehicles with similarity exceeding the preset threshold as cloned vehicle candidate pairs, input the images with the same shooting angle in the cloned vehicle candidate pairs into the label recognition model, extract the vehicle detail identification features, and generate the vehicle label feature set. Step S103: Retrieve the violation record data of the cloned vehicle candidate pair from the traffic violation management system, extract the vehicle images from the processed violation records, input the vehicle images into the label recognition model to extract feature labels, compare the feature labels with the vehicle label feature set, identify the real vehicle and the cloned vehicle based on the label matching degree, input the historical trajectory data of the cloned vehicle into the time series prediction model, calculate the future driving route and the necessary checkpoints, and generate a deployment instruction containing cloned vehicle feature information and driving prediction.

[0121] As described above, the computer-readable storage medium provided in this application, through an innovative spatiotemporal analysis model, achieves accurate screening of unreachable vehicles via distance calculation and speed verification. It constructs a feature recognition system, combining image segmentation and tag extraction to establish a reliable mechanism for identifying cloned vehicles. Predictive deployment is introduced, using trajectory analysis and route prediction to ensure the accuracy of strikes. This method effectively addresses the shortcomings of traditional technologies in spatiotemporal analysis, feature recognition, and predictive deployment, providing technical support for the control of cloned vehicles.

[0122] Embodiments of this application also provide a computer program product capable of implementing all steps of the cloned vehicle identification method with the execution subject being a server or client in the above embodiments. When this computer program / instruction is executed by a processor, it implements the steps of the cloned vehicle identification method. For example, the computer program / instruction implements the following steps: Step S101: Collect vehicle passage data from urban traffic checkpoint equipment and roadside radar sensors, extract basic information and location information of each vehicle, read the distance information between adjacent checkpoints, calculate the vehicle travel distance based on the semi-sine formula, extract timestamps to calculate time differences, calculate travel speed based on the travel distance and time difference, construct a road network parameter table containing distance thresholds and speed thresholds, compare the travel distance, time difference, and travel speed with the road network parameter table, and filter out candidate vehicles with cloned license plates that cannot be reached within a unit of time. Step S102: Read the appearance image of the candidate vehicle with cloned license plate, extract the key parts of the vehicle based on the image segmentation model, input the key parts into the feature recognition model to extract the basic features of the vehicle, construct the vehicle feature vector, calculate the similarity between the feature vectors of different vehicles, mark the vehicles with similarity exceeding the preset threshold as cloned vehicle candidate pairs, input the images with the same shooting angle in the cloned vehicle candidate pairs into the label recognition model, extract the vehicle detail identification features, and generate the vehicle label feature set. Step S103: Retrieve the violation record data of the cloned vehicle candidate pair from the traffic violation management system, extract the vehicle images from the processed violation records, input the vehicle images into the label recognition model to extract feature labels, compare the feature labels with the vehicle label feature set, identify the real vehicle and the cloned vehicle based on the label matching degree, input the historical trajectory data of the cloned vehicle into the time series prediction model, calculate the future driving route and the necessary checkpoints, and generate a deployment instruction containing cloned vehicle feature information and driving prediction.

[0123] As described above, the computer program product provided in this application, through the innovative design of a spatiotemporal analysis model and the use of distance calculation and speed verification, achieves accurate screening of unreachable vehicles. It constructs a feature recognition system, combining image segmentation and tag extraction to establish a reliable mechanism for identifying cloned vehicles. Predictive deployment is introduced, using trajectory analysis and route prediction to ensure the accuracy of strikes. This method effectively addresses the shortcomings of traditional technologies in spatiotemporal analysis, feature recognition, and predictive deployment, providing technical support for the control of cloned vehicles.

[0124] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0125] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0126] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0127] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0128] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A method for identifying cloned vehicles, characterized in that, The method includes: The system collects vehicle traffic data from urban traffic checkpoint equipment and roadside radar sensors, extracts basic information and location information for each vehicle, reads distance information between adjacent checkpoints, calculates vehicle travel distance based on the semi-sine formula, extracts timestamps to calculate time differences, calculates travel speed based on the travel distance and time difference, constructs a road network parameter table containing distance thresholds and speed thresholds, compares the travel distance, time difference, and travel speed with the road network parameter table, and filters out candidate vehicles with cloned license plates that cannot be reached within a unit of time. Read the exterior image of the candidate vehicle with cloned license plate, extract the key parts of the vehicle based on the image segmentation model, input the key parts into the feature recognition model to extract the basic features of the vehicle, construct the vehicle feature vector, calculate the similarity between the feature vectors of different vehicles, mark the vehicles with similarity exceeding a preset threshold as cloned vehicle candidate pairs, input the images of the cloned vehicle candidate pairs with the same shooting angle into the label recognition model, extract the vehicle detail identification features, and generate the vehicle label feature set. The system retrieves violation record data of the cloned vehicle candidate pair from the traffic violation management system, extracts vehicle images from the processed violation records, inputs the vehicle images into a label recognition model to extract feature labels, compares the feature labels with the vehicle label feature set, identifies real vehicles and cloned vehicles based on label matching degree, inputs the historical trajectory data of the cloned vehicles into a time series prediction model, calculates the future driving route and necessary checkpoints, and generates a deployment command containing cloned vehicle feature information and driving prediction.

2. The cloned vehicle identification method according to claim 1, characterized in that, The process involves collecting vehicle traffic data from urban traffic checkpoint equipment and roadside radar sensors, extracting basic and location information for each vehicle, reading distance information between adjacent checkpoints, calculating vehicle travel distance based on the semi-versus formula, extracting timestamps to calculate time differences, and calculating travel speed based on the travel distance and time difference. This includes: Vehicle capture records are read from the urban traffic checkpoint database, vehicle passage information is parsed from the roadside radar data stream, the capture records and passage information are grouped by license plate number, the license plate number, capture time and checkpoint location information in each group of data are extracted, a basic vehicle information table is constructed, the road network topology data of the electronic map is read, the checkpoint location information is mapped to the road network nodes, the road connection relationship between adjacent checkpoints is calculated, and a checkpoint distance matrix is ​​generated. Substitute the latitude and longitude coordinates of adjacent checkpoints into the semi-sine formula to calculate the actual driving distance of the vehicle. Extract the timestamp information of adjacent capture points from the vehicle basic information table, convert the timestamp information into a time difference value, divide the driving distance by the time difference value to obtain the driving speed, and construct a parameter calculation table containing license plate number, driving distance, time difference value, and driving speed.

3. The cloned vehicle identification method according to claim 1, characterized in that, The process involves constructing a road network parameter table that includes distance and speed thresholds. The travel distance, time difference, and travel speed are compared with the road network parameter table to filter out candidate vehicles with cloned license plates that are unreachable within a unit of time. These include: Read the road network topology data, classify the roads into highways, urban expressways, and urban arterial roads according to their functional levels, extract the speed limit standards for each type of road, set speed thresholds based on the speed limit standards, read the road segment lengths in the checkpoint distance matrix, set distance thresholds based on the road segment lengths, and write the speed thresholds and distance thresholds into the road network parameter table according to the road level to construct road traffic constraint rules. Read the vehicle driving records from the parameter calculation table, compare the driving distance with the distance threshold in the road network parameter table, compare the driving speed with the speed threshold in the road network parameter table, mark driving records with both distance and speed exceeding the threshold as abnormal data, group the abnormal data by license plate number, and generate a dataset of candidate vehicles with cloned license plates.

4. The cloned vehicle identification method according to claim 1, characterized in that, The process involves reading the exterior image of the candidate vehicle with the counterfeit license plate, extracting key parts of the vehicle based on an image segmentation model, inputting these key parts into a feature recognition model to extract basic vehicle features, and constructing a vehicle feature vector, including: Vehicle exterior images are extracted from the candidate vehicle dataset for counterfeit license plates. The exterior images are input into the image enhancement module to perform illumination normalization and noise suppression. An image segmentation network containing convolutional and deconvolutional layers is constructed. The enhanced images are input into the image segmentation network to generate a vehicle region segmentation mask. Based on the segmentation mask, the vehicle image is segmented into a front face region, a side region, and a rear region. The segmented regions are scaled to a uniform size according to a preset ratio. The scaled segmented region is input into the feature recognition model, and vehicle appearance features are extracted based on the convolutional neural network. Local detail features are extracted based on the region attention mechanism. The appearance features and detail features are fused according to preset weights to construct a multi-dimensional vehicle feature matrix. Dimensionality reduction calculation is performed on the feature matrix to generate a fixed-dimensional vehicle feature vector.

5. The cloned vehicle identification method according to claim 1, characterized in that, The process involves calculating the similarity between feature vectors of different vehicles, marking vehicles with similarity exceeding a preset threshold as clone vehicle candidate pairs, inputting images with consistent shooting angles from the clone vehicle candidate pairs into a label recognition model, extracting vehicle detail identification features, and generating a vehicle label feature set, including: Read vehicle feature vectors, construct feature matching matrices, calculate Euclidean distance and cosine similarity between feature vectors, weight and combine the Euclidean distance and cosine similarity to obtain a comprehensive similarity, determine a similarity threshold based on cluster analysis, mark vehicle pairs with comprehensive similarity higher than the threshold as clone vehicle candidate pairs, read the image shooting parameters of the clone vehicle candidate pairs, divide the images into front view, rear view and side view based on camera angle information, and filter image groups with consistent shooting angles; Images taken from the same angle are input into the label recognition model. The location and content of the annual inspection mark are extracted from the front view, the shape of the taillight assembly and the sub-brand logo are extracted from the rear view, and the location and shape of the body trim are extracted from the side view. The mark features, component features and trim features are combined into a label feature set, and a feature index table containing image number, shooting angle and label features is constructed.

6. The cloned vehicle identification method according to claim 1, characterized in that, The process of retrieving violation record data of the cloned vehicle candidate pair from the traffic violation management system, extracting vehicle images from processed violation records, inputting the vehicle images into a label recognition model to extract feature labels, and comparing the feature labels with the vehicle label feature set includes: Read the license plate number of the cloned vehicle candidate pair, query the traffic violation management system database, extract the violation records within a specified time range, filter out the unprocessed and appealed violation data, divide the processed violation records into manual processing records and automatic processing records according to the processing type, extract the violation images and processing time from the processed records, group the violation images according to the shooting angle, and construct a violation image dataset. The violation image is input into the label recognition model, vehicle identification features and decoration features are extracted, violation image feature labels are generated, the vehicle label feature set is read, the feature labels and the label feature set are used to calculate the matching degree score, the label similarity matrix is ​​constructed based on the matching degree score, the vehicle with the highest label similarity is marked as a real car, and the vehicle with the lower label similarity is marked as a cloned car.

7. The cloned vehicle identification method according to claim 1, characterized in that, The method for identifying real and cloned vehicles based on label matching involves inputting the historical trajectory data of the cloned vehicles into a time-series prediction model to calculate future driving routes and necessary checkpoints, and generating deployment instructions that include cloned vehicle feature information and driving predictions, including: Read the label similarity matrix, set the label matching threshold, identify vehicles with a label matching degree higher than the threshold as real vehicles, and identify vehicles with a label matching degree lower than the threshold as cloned vehicles. Extract the historical passage records of the cloned vehicles from the vehicle trajectory database, sort the passage records by timestamp to generate a trajectory sequence, construct a trajectory prediction model based on a long short-term memory network, input the trajectory sequence into the prediction model to generate the driving route within the future time window, and determine the location of the necessary checkpoints based on the road network topology. Read the vehicle characteristic information of the cloned vehicle, combine the characteristic information with the predicted route to generate an analysis report, read the checkpoint deployment rules, write the location of the necessary checkpoint into the deployment command, set the deployment time and deployment area, construct a deployment command set including checkpoint number, deployment time period, vehicle characteristics, and warning level, and send the deployment command set to the traffic control platform.

8. A cloned vehicle identification device, characterized in that, The device includes: The vehicle feature acquisition module is used to collect vehicle passage data from urban traffic checkpoint equipment and roadside radar sensors, extract basic information and location information of each vehicle, read distance information between adjacent checkpoints, calculate vehicle travel distance based on the semi-sine formula, extract timestamps to calculate time difference, calculate travel speed based on the travel distance and time difference, construct a road network parameter table containing distance thresholds and speed thresholds, compare the travel distance, time difference, and travel speed with the road network parameter table, and filter out candidate vehicles with cloned license plates that cannot be reached within a unit of time. The cloned vehicle identification module is used to read the appearance image of the candidate vehicle with cloned license plates, extract the key parts of the vehicle based on the image segmentation model, input the key parts into the feature recognition model to extract the basic features of the vehicle, construct the vehicle feature vector, calculate the similarity between different vehicle feature vectors, mark the vehicles with similarity exceeding a preset threshold as cloned vehicle candidate pairs, input the images with the same shooting angle in the cloned vehicle candidate pairs into the label recognition model, extract the vehicle detail identification features, and generate the vehicle label feature set. The early warning and control module is used to retrieve the violation record data of the cloned vehicle candidate pair from the traffic violation management system, extract vehicle images from the processed violation records, input the vehicle images into the label recognition model to extract feature labels, compare the feature labels with the vehicle label feature set, identify real vehicles and cloned vehicles based on label matching degree, input the historical trajectory data of the cloned vehicles into the time series prediction model, calculate the future driving route and the necessary checkpoints, and generate a control instruction containing cloned vehicle feature information and driving prediction.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the cloned vehicle identification method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the cloned vehicle identification method according to any one of claims 1 to 7.

Images