Crossing intelligent verification and risk blocking method and device based on multi-modal feature correlation and dynamic threshold determination

Abstract

The embodiment of the application provides a crossing intelligent verification and risk blocking method and device based on multi-modal feature association and dynamic threshold determination, which realizes effective protection of information through format verification and security encryption. A verification mechanism is constructed, combined with multi-modal analysis and risk assessment, to establish a reliable determination strategy. The introduction of traffic control ensures the accuracy of verification through record comparison and alarm pushing. The method effectively solves the shortcomings of traditional technology in data processing, multi-modal verification and traffic control, etc., and provides technical support for crossing verification.

Description

Technical Field

[0001] This application relates to the field of data processing, specifically to a method and apparatus for intelligent verification and risk prevention at road crossings based on multimodal feature association and dynamic threshold determination. Background Technology

[0002] Existing methods for verifying road crossings have significant shortcomings. Traditional systems perform poorly in data collection and standardization, failing to effectively integrate multi-dimensional information and impacting the foundation of verification.

[0003] Furthermore, existing technologies face bottlenecks in multimodal verification and risk assessment. Most systems lack robust feature matching mechanisms and contraband analysis strategies, resulting in insufficient verification rigor.

[0004] The existing system has technical shortcomings in traffic control. It lacks in-depth analysis of vehicle passage data, making it difficult to efficiently handle anomalies through comparison mechanisms, thus affecting verification effectiveness. Solving these problems is crucial for improving the verification capabilities at level crossings. Summary of the Invention

[0005] To address the problems in existing technologies, this application provides a method and apparatus for intelligent verification and risk prevention at level crossings based on multimodal feature association and dynamic threshold determination. This method effectively solves the shortcomings of traditional technologies in data processing, multimodal verification, and passage control, and provides technical support for level crossing verification.

[0006] To solve at least one of the above problems, this application provides the following technical solution: Firstly, this application provides a method for intelligent verification and risk prevention at road crossings based on multimodal feature association and dynamic threshold determination, including: A raw dataset is generated by collecting multidimensional data from intelligent level crossing equipment. The multidimensional data includes license plate images, RFID card numbers, vehicle roof images, vehicle under-vehicle images, face images, and equipment identifiers. The raw dataset is then subjected to format and integrity checks to generate standardized data packets. The standardized data packets are then subjected to hierarchical encryption to generate a safety dataset. Based on the safety dataset, a feature index table is established according to vehicle identifiers. The feature index table is then used to construct a state mapping model according to lane numbers. Based on the state mapping model, multimodal verification of passing vehicles is performed to generate a verification dataset. The license plate images in the verification dataset are subjected to dual recognition to generate license plate feature vectors. The license plate feature vectors are associated and matched with RFID card numbers to generate a vehicle-license mapping table. Based on a preset model library, the roof and undercarriage images in the verification dataset are analyzed for prohibited items to generate a risk assessment table. A verification state matrix is ​​constructed based on the risk assessment table. The verification state matrix is ​​then judged according to dynamic thresholds to generate a verification result set. Based on the verification result set, the control gate equipment generates a passage instruction. When the passage instruction is to allow passage, the gate equipment is opened, and the rear of the passing vehicle is captured to generate a vehicle data packet. The vehicle data packet is correlated and compared with the verification status matrix to generate a verification record. When the verification record shows a mismatch, an alarm message is generated and pushed to the management platform. The verification record and the alarm message are encrypted and stored to generate an evidence chain database.

[0007] Furthermore, it also includes: acquiring multiple data streams from intelligent checkpoint equipment based on a distributed acquisition mechanism to generate a data collection set; dividing the multiple data streams into time-series segments according to a preset acquisition cycle to generate segmented data groups; extracting and encoding fields from the segmented data groups according to a preset data template to generate device data packets; and organizing the license plate image, RFID card number, vehicle roof image, vehicle under image, face image, and device identifier in the device data packets to generate an acquisition record table. The data collection record table is validated to generate an original dataset. The data items in the data collection record table are validated according to the field integrity rules to generate a validation result set. Based on the validation result set, the data completion interface is called to obtain missing field information and generate a completion data package. The validated data is reorganized according to the preset field specifications to generate an original dataset containing license plate images, RFID card numbers, vehicle roof images, vehicle under images, face images and device identifiers.

[0008] Furthermore, it also includes: constructing a hierarchical encryption module to generate an encryption strategy table based on the data security level; dividing standardized data packets into data security levels according to preset classification rules to generate hierarchical data packets; allocating keys to sensitive fields in the hierarchical data packets to generate an encryption parameter set; encrypting the hierarchical data packets based on the encryption parameter set to generate a security dataset; and establishing a data mapping table for the security dataset according to vehicle identifiers and timestamps. Based on the data mapping table, a feature index system is constructed to generate a state mapping model. The vehicle identification information in the data mapping table is serialized to generate a vehicle feature sequence. The vehicle feature sequence is grouped and mapped according to lane number to generate an index relationship table. Based on the index relationship table, a lane-level feature mapping is established to generate a state mapping model. The state mapping model is used for subsequent vehicle verification analysis.

[0009] Furthermore, it also includes: constructing a multimodal verifier based on a state mapping model to generate a verification process table; collecting traffic vehicle data from multiple sources according to preset collection rules to generate a real-time data stream; performing time-series alignment processing on the real-time data stream to generate a synchronous data packet; and extracting features from the synchronous data packet according to the verification process table to generate a verification dataset. The verification dataset is used to generate a vehicle-license mapping table by performing identity feature recognition. The license plate images in the verification dataset are recognized by high-position and low-position cameras respectively to generate dual license plate features. The dual license plate features are then used to perform consistency verification to generate a license plate feature vector. Based on the license plate feature vector and the RFID card number, a spatiotemporal correlation calculation is performed to generate the vehicle-license mapping table.

[0010] Furthermore, it also includes: constructing a prohibited items identifier generation analysis rule table based on a preset model library, performing image preprocessing on the roof and undercarriage images in the verification dataset according to preset resolution rules to generate a standard image set, performing target detection on the standard image set based on a deep learning prohibited items identification model to generate a detection feature set, and calculating a risk probability score based on the detection feature set to generate a risk assessment table; Based on the risk assessment table, a verification result set is generated by determining the verification status. The risk scores in the risk assessment table are quantified according to preset weight rules to generate a verification status matrix. The verification status matrix is ​​then dynamically analyzed based on an adaptive threshold model to generate a threshold determination result. Based on the threshold determination result, the vehicle traffic status is classified and labeled to generate a verification result set.

[0011] Furthermore, it also includes: constructing a control instruction table for the access controller based on the verification result set, mapping the verification result set to a state according to a preset access rule to generate an access state package, performing permission verification on the access state package to generate an authorization result set, generating access instructions according to the authorization result set and a preset control strategy, and driving the barrier gate device to perform opening and closing actions based on the access instructions; Based on the vehicle-triggered lane-breaking mechanism to generate vehicle data packets, the status information of the barrier gate equipment and vehicle passage information are collected in real time to generate a status data stream. The rear images of the vehicles in the status data stream are captured and processed to generate a rear feature set. The license plate information and timestamp are extracted from the rear feature set to generate a vehicle data packet.

[0012] Furthermore, it also includes: establishing a comparison and verification device based on the vehicle data packet to generate a matching rule table, extracting the license plate information and time information in the vehicle data packet according to a preset mapping rule to generate a vehicle feature set, performing spatiotemporal correlation calculation on the vehicle feature set and the verification state matrix to generate a comparison result set, and determining and generating a verification record according to the comparison result set and a preset threshold condition. Anomaly analysis is performed on the verification records to generate an evidence chain database. Based on preset alarm rules, mismatch analysis is performed on the verification records to generate an alarm trigger table. The abnormal information in the alarm trigger table is processed in a hierarchical manner to generate an alarm information package. The verification records and the alarm information package are distributed and encrypted for storage to generate an evidence chain database.

[0013] Secondly, this application provides a smart checkpoint verification and risk prevention device based on multimodal feature association and dynamic threshold determination, comprising: The data acquisition module is used to collect multidimensional data based on the intelligent level crossing equipment to generate a raw dataset. The multidimensional data includes license plate images, RFID card numbers, vehicle roof images, vehicle under images, face images, and equipment identifiers. The module performs format and integrity checks on the raw dataset to generate a standardized data packet. The standardized data packet is then subjected to hierarchical encryption to generate a safety dataset. Based on the safety dataset, a feature index table is established according to vehicle identifiers. The feature index table is then used to construct a state mapping model according to lane numbers. The feature recognition module is used to perform multimodal verification of passing vehicles based on the state mapping model to generate a verification dataset, perform dual recognition on the license plate images in the verification dataset to generate license plate feature vectors, associate and match the license plate feature vectors with RFID card numbers to generate a vehicle-license mapping table, perform contraband analysis on the vehicle roof and vehicle bottom images in the verification dataset based on a preset model library to generate a risk assessment table, construct a verification state matrix based on the risk assessment table, and generate a verification result set by judging the verification state matrix according to a dynamic threshold. The intelligent verification and risk blocking module for road crossings based on multimodal feature association and dynamic threshold determination is used to control the road crossing equipment to generate passage instructions according to the verification result set. When the passage instruction is to allow passage, the gate equipment is opened, and the rear of the passing vehicle is captured to generate a vehicle data packet. The vehicle data packet is correlated and compared with the verification status matrix to generate a verification record. When the verification record shows a mismatch, an alarm message is generated and pushed to the management platform. The verification record and the alarm message are encrypted and stored to generate an evidence chain database.

[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. When the processor executes the program, it implements the steps of the intelligent checkpoint verification and risk blocking method based on multimodal feature association and dynamic threshold determination.

[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 intelligent checkpoint verification and risk blocking method based on multimodal feature association and dynamic threshold determination.

[0016] Fifthly, this application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the intelligent checkpoint verification and risk blocking method based on multimodal feature association and dynamic threshold determination.

[0017] As can be seen from the above technical solution, this application provides a method and device for intelligent verification and risk blocking at level crossings based on multimodal feature association and dynamic threshold determination. Effective information protection is achieved through format verification and security encryption. A verification mechanism is constructed, combining multimodal analysis and risk assessment to establish a reliable determination strategy. Access control is introduced, ensuring verification accuracy through record comparison and alarm push notifications. This method effectively solves the shortcomings of traditional technologies in data processing, multimodal verification, and access control, providing technical support for intelligent verification and risk blocking at level crossings based on multimodal feature association and dynamic threshold determination. 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 intelligent checkpoint verification and risk blocking method based on multimodal feature association and dynamic threshold determination in the embodiments of this application. Figure 2 This is a structural diagram of the intelligent checkpoint verification and risk blocking device based on multimodal feature association and dynamic threshold determination 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 intelligent verification and risk blocking at level crossings based on multimodal feature association and dynamic threshold determination. Effective information protection is achieved through format verification and security encryption. A verification mechanism is constructed, combining multimodal analysis and risk assessment to establish a reliable determination strategy. Access control is introduced, ensuring the accuracy of verification through record comparison and alarm push. This method effectively solves the shortcomings of traditional technologies in data processing, multimodal verification, and access control, providing technical support for intelligent verification and risk blocking at level crossings based on multimodal feature association and dynamic threshold determination.

[0024] To effectively address the shortcomings of traditional technologies in data processing, multimodal verification, and passage control, and to provide technical support for level crossing verification, this application provides an embodiment of an intelligent level crossing verification and risk prevention method based on multimodal feature association and dynamic threshold determination. See [link to relevant documentation]. Figure 1 The intelligent checkpoint verification and risk blocking method based on multimodal feature association and dynamic threshold determination specifically includes the following: Step S101: Generate a raw dataset by collecting multidimensional data from the intelligent toll gate device. The multidimensional data includes license plate images, RFID card numbers, vehicle roof images, vehicle under-vehicle images, face images, and device identifiers. Perform format and integrity checks on the raw dataset to generate a standardized data packet. Perform hierarchical encryption processing on the standardized data packet to generate a safety dataset. Based on the safety dataset, establish a feature index table according to vehicle identifiers. Construct a state mapping model based on the feature index table according to lane numbers. First, six types of data generated by the intelligent toll gate equipment are collected: license plate images, RFID card numbers, vehicle roof images, vehicle under-vehicle images, facial images, and equipment identifiers. These are aligned according to a unified timeline to generate records aggregated by collection time. For each record, the number, source device, and timestamp are extracted based on preset field templates, and missing items are recorded to form candidate entries for the original dataset. Size and encoding checks are performed on image fields, and character set and length checks are performed on RFID card numbers and equipment identifiers. Non-compliant items are marked for subsequent backfilling.

[0025] Based on the candidate entries, integrity checks are performed, and the missing field backfilling interface is invoked to re-acquire images or reread RFID card numbers. Image backfilling uses the trajectory of the same vehicle in the same lane within the nearest time neighborhood as the search scope, while RFID backfilling is based on the reader cache in the same lane within the same time window. Entries that pass the check are rearranged in field order and standardized in naming, packaged into standardized data packets, and entries that fail the check are removed from the current process after being annotated with the reason.

[0026] The standardized data packets are sent to a hierarchical encryption module. Based on the sensitivity level of the fields, three categories are identified: vehicle identity elements, image elements, and device elements, with key lengths and encryption strengths selected for each category. To prevent cross-lane leakage, temporary keys are generated only within the same lane and bound to a timestamp interval. The encrypted output and the minimum necessary index entries of the plaintext are stored separately to form a secure dataset, where the plaintext retains only the hash index used for retrieval and necessary time positioning information.

[0027] A feature index table, grouped by vehicle identifier, is established based on the aforementioned security dataset. The vehicle identifier is determined by the spatiotemporal consistency criteria of the license plate character sequence and the RFID card number; when either is missing, the device identifier and the collection queue order are used as temporary markers, which are then written back after subsequent completion. Index entries include timestamps, lane numbers, and reference pointers to the security dataset, ensuring that subsequent queries can locate the target encrypted block without decrypting the image content.

[0028] Based on the feature index table generated in the previous processing, the system is binned by lane number to construct the skeleton of the state mapping model. Within each lane bin, index entries are arranged in ascending order by time, and adjacency relationships are established to represent continuous observations of the same vehicle within a short period. To reduce the probability of false merging across lanes, the model introduces lane consistency constraints when constructing adjacency relationships, establishing strong connections only when observations fall within the same lane.

[0029] Therefore, a one-time index score is defined to determine whether an entry is merged into the same vehicle trajectory, denoted by the formula: S = a1·U + a2·V-a3·W.

[0030] In the formula, S is the index score, U is the license plate character similarity, V is the RFID card number matching indication, and W is the time difference penalty term; a1, a2, and a3 are weights, and their values ​​are set to non-negative according to the lane configuration strategy. U takes a higher value when the license plate dual recognition is consistent and a lower value when inconsistent; V takes a higher value when the RFID is consistent, a medium value when missing, and a lower value when inconsistent; W increases with the time interval between adjacent entries. Only when S exceeds the lane-level threshold condition are the entries merged into the same trajectory node. This score result is written as a field into the state mapping model for subsequent reading and judgment in the verification stage.

[0031] In the state mapping model, each vehicle trajectory node retains a read-only pointer to the safety dataset, an adjacency index to previous and subsequent observations, and lane number and time range label. The image content remains encrypted and is only decrypted into a short-term buffer in memory via an authorized interface when subsequent dual license plate recognition or contraband identification is required; it is not stored on disk.

[0032] Next, a mapping relationship is established between the model index and the lane policy table configured on the central side, registering the identification order, time window length, and temporary number replacement rules used for that lane. This mapping is written into the policy field of the index table, enabling subsequent multimodal verification to select the image decryption order and RFID matching window accordingly, avoiding cross-policy misuse.

[0033] Then, a consistency self-check is performed on the generated state mapping model. The self-check reads the lane number, time range, and index score of adjacent nodes. If a strong cross-lane connection or a connection with a score below the threshold is found, it is immediately disconnected and reverted to a weak connection. After the self-check is completed, the model version number is output and the index snapshot of the current version is frozen to ensure determinism in subsequent reads.

[0034] Finally, the state mapping model and feature index table are exposed as input interfaces for subsequent steps. When reading the model, multimodal verification first determines the decryption order based on the policy domain, and then accesses the secure dataset within the authorized range through the index pointer. The dual license plate recognition and RFID association calculation directly references the aforementioned index score S as a priori, narrowing the matching search space, and after verification, writes the result back to the same trajectory node for subsequent risk assessment and access control.

[0035] Step S102: Based on the state mapping model, perform multimodal verification on passing vehicles to generate a verification dataset. Perform dual recognition on the license plate images in the verification dataset to generate license plate feature vectors. Associate and match the license plate feature vectors with RFID card numbers to generate a vehicle-license mapping table. Based on a preset model library, perform contraband analysis on the roof and undercarriage images in the verification dataset to generate a risk assessment table. Construct a verification state matrix based on the risk assessment table. Generate a verification result set by judging the verification state matrix according to dynamic thresholds. First, the trajectory nodes and policy domains in the aforementioned state mapping model are read. Within the authorized range, the minimum necessary data is decrypted according to the node's index pointer, generating a candidate packet containing license plate images, high- and low-position camera annotations, RFID card number placeholders, vehicle roof images, vehicle under-vehicle images, timestamps, and lane numbers. Based on the policy domain's acquisition order and time window length, frame-level aggregation of the same vehicle within the same lane is completed, forming data segments for verification. The source node number is registered for write-back.

[0036] Based on the candidate packets, the timeline is refined and aligned. License plate images from high- and low-position cameras are sorted by trigger time, and duplicate frames from the same camera are filtered out, retaining the pair with higher clarity as dual identification inputs. Simultaneously, RFID readings within the same time window are linked to this pair of images to establish a one-to-many labeling relationship. Camera pose and trigger mode labels are added to the roof and undercarriage images for subsequent viewpoint selection in contraband identification. After the above processing, the dataset is packaged into a verification dataset, with references to the original trajectory nodes included in each data segment.

[0037] The verification dataset is fed into the license plate dual recognition subprocess. The license plate recognizer is invoked on both the high-resolution and low-resolution images, outputting character sequences and confidence scores. Consistency is then checked using character-level alignment and occlusion tolerance rules; if inconsistencies are found, both sequences are retained and marked as conflicting. The character sequences are then encoded into a license plate feature vector, which includes character embedding, layout encoding, and color encoding, with a uniform length for subsequent association. This license plate feature vector, along with the timestamp and lane number from the data segment, is written into a temporary index to maintain a traceable relationship with the aforementioned trajectory nodes.

[0038] Vehicle-license association matching is performed based on the temporary index. RFID readings are selected according to the time range of trajectory nodes to construct a candidate set of matching RFID tags, with lane number as a hard constraint. If multiple RFID candidates exist, time proximity and acquisition intensity are used as the first sorting key, and the consistency status of dual license plate recognition is used as the second sorting key to obtain a one-to-one pairing result. After pairing is completed, a vehicle-license mapping table is output. Each record in the table contains a license plate feature vector, RFID card number, timestamp, and pairing confidence flag, which can be used for subsequent risk and status determination.

[0039] Next, the prohibited items recognition model library was called on the images of the vehicle roof and undercarriage in the verification dataset to perform viewpoint classification and target detection, respectively. Viewpoint classification was used to filter out frames that did not meet the imaging conditions, and target detection output suspicious region boxes and category scores. To reduce false detections, the suspicious categories of the vehicle roof and undercarriage were summarized in parallel within the same time window, and a risk assessment table for each segment was output. The table included elements such as category-level scores, region counts, and frame coverage ratios, and recorded the image source and backtracking pointer.

[0040] A verification status matrix is ​​generated based on the vehicle-license mapping table and risk assessment table. Rows in the matrix correspond to data segments, and columns are divided into two groups: identity verification items and risk verification items. Identity verification items are filled with license plate consistency markers, vehicle-license pairing confidence markers, and lane consistency markers, respectively; risk verification items are filled with roof risk scores, under-vehicle risk scores, and joint risk markers. To maintain consistency with upstream data, matrix cells only store standardized values ​​and enumerated states, and do not directly store images or plaintext identifiers.

[0041] Therefore, a comprehensive verification score is defined to facilitate subsequent threshold determination, denoted as the formula: R = b1·H + b2·J-b3·K.

[0042] In the formula, R is the comprehensive score; H is the identity consistency component, derived from the combined value of license plate dual recognition consistency and vehicle-document pairing confidence; J is the safety perspective component, derived from the weighted result of the inverse mapping of the vehicle roof and vehicle under category scores and the frame coverage ratio; K is the time consistency penalty, reflecting the maximum time difference between the image frame and the RFID reading within the window; b1, b2, and b3 are non-negative weights, given by the lane policy domain and remaining unchanged within the current session. The above R is written as a derived column of the matrix for the threshold model to read.

[0043] Dynamic threshold determination is applied to the verification state matrix. The threshold is initialized within a range given by the lane policy domain and adaptively fine-tuned with reference to the R distribution over a recent period. The determination logic first checks whether the hard constraints of the identity verification item are met, then reads R and the joint risk label to jointly determine the verification result. The result is marked as pass or fail, and carries a reason code. All determinations record parameter snapshots to ensure that experimental conditions can be reproduced in subsequent traceability.

[0044] The verification result set is then written back to the corresponding trajectory node, updating the node's identity verification status, joint risk field, and judgment timestamp, while retaining a read-only pointer to the security dataset. After the write-back is complete, the temporary decryption buffer is released, and necessary indexes are retained for downstream queries. This verification result set serves as input for the next step, being read by the access control logic to generate access status packets and compare them with subsequent rear-end capture images.

[0045] Step S103: Based on the verification result set, control the level crossing device to generate a passage instruction. When the passage instruction is to allow passage, open the gate device, capture the rear of the passing vehicle to generate a vehicle data packet, associate and compare the vehicle data packet with the verification status matrix to generate a verification record, generate an alarm message when the verification record shows a mismatch and push it to the management platform, and encrypt and store the verification record and the alarm message to generate an evidence chain database.

[0046] First, the index relationship between the aforementioned verification result set and trajectory nodes is read. Based on the hard constraints and comprehensive scoring fields in the policy domain, a passage status package is generated. The passage status package includes trajectory number, lane number, identity verification conclusion, joint risk marker, and reason code. Permission verification is applied to this data to confirm that the lane is in a controllable state and the device activation marker is true. Subsequently, it is mapped to execution parameters according to the policy domain to form passage instruction candidates.

[0047] Based on the candidate passage instructions, device-level control instructions are generated. These control instructions are categorized into three types according to device identification: barrier gates, access control gates, and indicator displays, and are bound to a one-time control token and a timestamp. When the instruction is to allow passage, an opening action is sent to the barrier gate device, and the control session number and valid time window are registered for subsequent closed-loop association of vehicle passage. If the instruction is to deny passage, only a reason code is pushed to the display device; the barrier gate is not triggered.

[0048] The release session triggers a lane-tripping mechanism, which collects the original frame captured from the rear of the vehicle and the device status, generating a candidate vehicle to pass that includes the rear image, capture time, gate status, and lane number. Subsequently, the captured frame is processed for character region localization and clarity filtering, the license plate characters at the rear of the vehicle are extracted, and the image source is registered, forming a vehicle passage data packet. Simultaneously, references to the corresponding trajectory nodes and control sessions are retained.

[0049] Based on the vehicle data packets, an association comparison is performed. The comparison targets are the identity entries within the verification state matrix and trajectory nodes, using same lane and same control session as hard constraints, and time proximity as the primary sorting key. If the license plate characters of the rear vehicle map to the same character sequence as the feature vector of the previous vehicle, and the time difference falls within the session window, a matching verification record is generated; otherwise, a mismatch is marked and a difference field is attached, including the character difference position and time offset.

[0050] To reduce false alarms, a comparison score is constructed to record the verification reliability, denoted by the formula: T = c1·M + c2·N-c3·P.

[0051] In the formula, T is the comparison score, M is the consistency measure of the characters of the license plate at the rear of the vehicle and the license plate at the front, N is the piecewise function of the session time fit, and P is the device status fluctuation penalty item; c1, c2, and c3 are non-negative weights, and their values ​​are set according to the lane policy domain. T is written into the verification record and used as the retrieval key for subsequent alarms and evidence storage.

[0052] If there is a mismatch in the verification record or if T is below the policy threshold, an alarm information packet is generated. The alarm information packet includes lane number, trajectory number, cause code, difference summary, and minimum necessary image pointer. The alarm is pushed to the management platform via a message channel, and the handling status is registered on the edge, awaiting manual confirmation or policy callback.

[0053] The verification records and alarm information packets are then subjected to hierarchical encryption and stored, generating evidence chain database entries. The encryption process follows the key management strategy of the secure dataset, encrypting and storing the identity field, image pointer, and control session number separately, and writing timestamps and hash verification values ​​to ensure consistency during subsequent retrieval. Each evidence chain entry returns a read-only index for querying by trajectory nodes and the management platform.

[0054] Finally, the execution result of the passage command and the vehicle comparison conclusion are written back to the closed-loop domain of the trajectory node. The closed-loop domain records the command type, execution time, comparison score T, and final conclusion, serving as the input for downstream passage statistics and re-verification. At the end of this session, the temporary decryption buffer is cleared, and only the necessary indexes are retained for subsequent traceability.

[0055] As described above, the intelligent checkpoint verification and risk blocking method based on multimodal feature association and dynamic threshold determination provided in this application can effectively protect information through format verification and security encryption. A verification mechanism is constructed, combining multimodal analysis and risk assessment to establish a reliable determination strategy. Access control is introduced, ensuring verification accuracy through record comparison and alarm push notifications. This method effectively solves the shortcomings of traditional technologies in data processing, multimodal verification, and access control, providing technical support for intelligent checkpoint verification and risk blocking based on multimodal feature association and dynamic threshold determination.

[0056] In one embodiment of the intelligent checkpoint verification and risk blocking method based on multimodal feature association and dynamic threshold determination in this application, the method may further include the following: Step S201: Based on the distributed acquisition mechanism, multiple data streams are acquired from the intelligent checkpoint device to generate a data acquisition set. The multiple data streams are then segmented into segmented data groups according to a preset acquisition cycle. Fields are extracted and encoded from the segmented data groups according to a preset data template to generate a device data package. The license plate image, RFID card number, vehicle roof image, vehicle under image, face image, and device identifier in the device data package are organized to generate an acquisition record table. Step S202: Perform data verification on the collection record table to generate an original dataset. Verify the validity of the data items in the collection record table according to the field integrity rules to generate a verification result set. Based on the verification result set, call the data completion interface to obtain missing field information and generate a completion data package. Reassemble the data that has passed the verification according to the preset field specifications to generate an original dataset containing license plate images, RFID card numbers, vehicle roof images, vehicle under images, face images, and device identifiers.

[0057] First, multiple data streams from various smart toll gate devices are accessed, including high-position and low-position license plate camera channels, RFID readers, roof-mounted cameras, under-vehicle cameras, and facial recognition panels. Device identifiers and heartbeat status are simultaneously read and recorded as data collection sets. Based on a preset collection cycle, continuous data within the same lane is divided into segmented data groups along the time axis. Each group covers a stable sampling window, and the start and end times and trigger sources are recorded during segmentation to ensure cross-device event alignment.

[0058] Based on the segmented data groups, field extraction and encoding are performed according to the data template. For image fields, image hash, width, height, and color space annotations are written; for RFID readings, card number, read strength, and reader number are written; for face fields, face bounding box coordinates and sharpness labels are written. Device identifiers are standardized as a combination of device type and device number. After encoding, a device data package is generated, with each field in the package including a timestamp and lane number for easy subsequent aggregation and tracing.

[0059] The device data packets are organized into a collection record table. Data from the same lane within the same time window is grouped into a single record, which includes dual-channel license plate image bits, RFID card number placeholders, roof image bits, under-vehicle image bits, face image bits, and a device identifier list. Missing bits are marked with placeholders and the reason for the absence is added, such as not triggered, collection failure, or device disabled, while the original pointer is retained for possible re-collection.

[0060] Data validation is performed on the data collection record table to generate candidates for the original dataset. Validation includes field completeness, timestamp monotonicity, and device identifier validity: image bits must meet minimum resolution and size requirements, RFID card numbers must meet character set and length requirements, and face bits must have a face bounding box and quality label. Validation results are written to the validation result set, and non-compliant items are marked with either a completion or rejection flag.

[0061] The data completion interface is invoked based on the verification result set. Image completion uses a nearby time window within the same lane to retrieve alternative frames; RFID completion uses the reader cache and session queue for re-examination; and face completion uses the results of subsequent captures within the same window. Successfully completed fields are backfilled into the corresponding records, along with the source, time, and reason for the backfill operation, for future traceability. Fields that cannot be completed remain placeholders, and a missing label is retained in the record header.

[0062] Then, data reorganization is performed on the verified and backfilled records. Field naming and order are standardized, internal diagnostic fields are removed, and only six core elements are retained: license plate image, RFID card number, roof image, under-vehicle image, face image, and device identifier. Timestamps and lane numbers are also retained as search keys. The reorganized record set is the original dataset, and it is consistent with the timeline strategy established in the previous step S101 to ensure that it can directly enter the standardization and hierarchical encryption process.

[0063] After the original dataset is generated, the index keys of the records are written into the state mapping entry table, pointing to the corresponding lane and time window. Subsequent steps will use this entry table as the starting point to complete standardization, encryption, and feature index construction, ensuring that the data link from collection to verification remains coherent and traceable.

[0064] In one embodiment of the intelligent checkpoint verification and risk blocking method based on multimodal feature association and dynamic threshold determination in this application, the method may further include the following: Step S301: Construct a hierarchical encryption module based on the data security level to generate an encryption strategy table, divide standardized data packets into data security levels according to preset classification rules to generate hierarchical data packets, allocate keys to sensitive fields in the hierarchical data packets to generate an encryption parameter set, encrypt the hierarchical data packets based on the encryption parameter set to generate a security dataset, and establish a data mapping table for the security dataset according to vehicle identifier and timestamp. Step S302: Based on the data mapping table, construct a feature index system to generate a state mapping model. Serialize the vehicle identification information in the data mapping table to generate a vehicle feature sequence. Group and map the vehicle feature sequence according to lane number to generate an index relationship table. Establish a lane-level feature mapping based on the index relationship table to generate a state mapping model. Use the state mapping model for subsequent vehicle verification analysis.

[0065] First, the standardized data packets and entry indexes output from the preceding steps are read. Based on the data security level defined for the scenario, the fields are divided into three groups: identity elements, image elements, and device elements, and an encryption policy table is generated. The policy table records the key type, rotation period, and access domain constraints for each group. Then, hierarchical slicing is performed on the standardized data packets to form hierarchical data packets, and lane numbers and timestamps are written to the slice headers for subsequent location tracking.

[0066] Based on the hierarchical data packets, key allocation is performed to generate an encryption parameter set. Identity elements use session domain keys, image elements use lane domain keys, and device elements use device domain keys. Each key is bound to an expiration date and revocation pointer. Encryption is performed only on sensitive fields, and necessary digests are retrieved while maintaining the plaintext hash. After processing, a secure dataset is output, simultaneously recording encrypted metadata and the minimum necessary index to ensure location and authorization checks can be completed without decryption.

[0067] The safety dataset is mapped using a data mapping table based on vehicle identifiers and timestamps. Vehicle identifiers are derived from a consistent combination of license plate characters and RFID tags; in cases of missing identifiers, a temporary identifier is used and a write-back flag is set. Each entry in the mapping table is saved to a read-only pointer in the safety dataset, along with its lane number and time range, for subsequent construction of a searchable trajectory view.

[0068] The feature index construction process is initiated based on the data mapping table. First, the vehicle identification information is serialized, encoding the character sequence, color label, and RFID hash into a vehicle feature sequence of uniform length; simultaneously, the timestamp sequence is retained as the timeline for subsequent adjacency determination. This serialization result establishes a one-to-one correspondence with the read-only pointers in the mapping table, ensuring direct access from the index to the encrypted block.

[0069] Based on the aforementioned vehicle feature sequences, a lane number grouping and mapping are performed to generate an index relationship table. Each lane maintains an independent time-ordered list, and an adjacency index is established between adjacent entries to represent continuous observations within a short period. If an entry is generated from a temporary marker, a write-back flag is used to prompt subsequent verification stages to update the index after completion, avoiding permanent dangling.

[0070] To address this, a state mapping model is constructed on the lane-level index. Model nodes correspond to grouped sequence entries, and edges represent temporal adjacency and consistency within the same lane. Nodes are stored in a read-only pointer to the safety dataset along with a policy domain summary, while edges store the time interval and source camera location information. The model does not carry plaintext images; instead, it decrypts them into memory only via pointers within the authorized time period, preventing cross-domain leaks.

[0071] Next, a consistency check is performed on the state mapping model. The check includes lane consistency, time monotonicity, and pointer availability. Edges that do not meet the constraints are immediately downgraded to weak connections, and a revision record is registered in the model metadata. After the check passes, a version number is generated and the version is frozen to support stable downstream reading.

[0072] Finally, the state mapping model is mounted as the entry point for verification analysis. Subsequent multimodal verification, when reading the model, uses an indexed relation table to determine the decryption order and time window, and then accesses the secure dataset via a read-only pointer to complete identification and matching; the verification results are written back to the corresponding nodes, forming a traceable closed-loop trajectory.

[0073] In one embodiment of the intelligent checkpoint verification and risk blocking method based on multimodal feature association and dynamic threshold determination in this application, the method may further include the following: Step S401: Based on the state mapping model, construct a multimodal verifier to generate a verification process table, collect multiple data streams of passing vehicles according to preset collection rules to generate a real-time data stream, perform time-series alignment processing on the real-time data stream to generate a synchronous data packet, and extract features from the synchronous data packet according to the verification process table to generate a verification dataset. Step S402: Perform identity feature recognition on the verification dataset to generate a vehicle-license mapping table. Recognize the license plate images in the verification dataset using a high-position camera and a low-position camera to generate dual license plate features. Perform consistency verification on the dual license plate features to generate a license plate feature vector. Perform spatiotemporal correlation calculation on the license plate feature vector and the RFID card number to generate a vehicle-license mapping table.

[0074] First, the lane-level index and policy domain from the aforementioned state mapping model are read, a multimodal verifier is instantiated, and a verification process table is generated accordingly. The verification process table lists the collection order, effective time window, and mandatory inspection items by lane number. Among them, dual-channel license plate recognition, RFID reading, roof capture, and under-vehicle capture are marked as parallel triggering within the same session. Subsequently, after the vehicle enters the trigger line, multi-channel data collection is initiated according to the process table, forming a real-time data stream with session number and timestamp.

[0075] Timing alignment is performed based on the real-time data stream. First, a relative time axis is established with the trigger line time as the zero point. Then, frames and readings are mapped onto a unified axis according to the event times of the camera and the reader. A sharpness-first retention strategy is used for duplicate frames from the same device. For multiple RFID readings within the same window, a two-stage deduplication process based on intensity and time proximity is used to output synchronization data packets. Synchronization data packets are aggregated by session number, along with lane number and source pointer, facilitating subsequent write-back.

[0076] The synchronization data packet is sent to the feature extraction stage. For license plate images, character regions, layout encoding, and color components are extracted; for RFID readings, card numbers and reading intensity are extracted; for images of the vehicle roof and undercarriage, viewpoint labels and candidate target regions are extracted. All features are written with fixed field names to form a verification dataset, and the reference relationship with the state mapping model nodes is recorded to ensure that the verification results can fall back to the same trajectory node.

[0077] Identity feature recognition is performed based on the verification dataset. The license plate recognizer is called on both the high-resolution and low-resolution license plate images to obtain two sets of character sequences and confidence scores. Character-level alignment and occlusion tolerance processing are then performed. If the two sets of sequences are completely identical, they are marked as identical; if there are some differences, the location of the differences is recorded and the results are entered into a cross-correction branch, retaining both sets of results for subsequent association judgment. The recognition product is then converted into dual license plate features, including three components: character embedding, layout encoding, and color encoding.

[0078] Based on the aforementioned dual license plate features, a consistency check is performed to generate a license plate feature vector. The consistency check reads the character distance and layout consistency flags from the two channels, outputs a unified vector and a consistency flag, and writes the flag into the identity field within the session. This license plate feature vector, along with the timestamp and lane number from the verification dataset, is then stored in a temporary index as input for vehicle-license association.

[0079] The vehicle-license association is calculated based on the temporary index. First, RFID candidates are collected within the session time window, using lane number as a hard constraint. When there are more than one candidate, they are first sorted by temporal proximity, then the read strength is considered, and the consistency flag from the previous segment is used as a weighted sorting factor to obtain the optimal pairing. After pairing, a vehicle-license mapping table is written, recording the license plate feature vector, RFID card number, timestamp, lane number, and pairing confidence flag. The mapping table entries maintain references to nodes in the state mapping model.

[0080] To improve pairing stability, a relationship check is recorded in the verification process table. The check verifies the consistency between the time interval between two consecutive pairings read from adjacent sessions and the lane. If a short-term switch across sessions or a cross-lane reading is detected, the confidence flag of this pairing is downgraded and a reason code is added. The downgrade operation does not directly negate the pairing, but rather serves as an input condition for downstream risk assessment.

[0081] Next, the vehicle license mapping table is written back to the corresponding trajectory node. The written-back content includes the unified license plate feature vector, consistency flag, RFID card number, and pairing confidence flag, and updates the node's identity verification status to "verified". To avoid unnecessary plaintext exposure, the node only saves a read-only pointer and standardized flag value, and does not save the decrypted image content.

[0082] Finally, the verification dataset and vehicle permit mapping table are used as inputs for subsequent risk assessment steps. When reading the input, the prohibited items identification process selects available roof and undercarriage frames based on the imaging conditions in the verification process table, and then writes the identification results along with paired confidence tags into the same session domain, providing directly readable structured input for the subsequent construction of the verification state matrix and dynamic threshold determination.

[0083] In one embodiment of the intelligent checkpoint verification and risk blocking method based on multimodal feature association and dynamic threshold determination in this application, the method may further include the following: Step S501: Construct a prohibited items identifier generation analysis rule table based on a preset model library, preprocess the roof and undercarriage images in the verification dataset according to preset resolution rules to generate a standard image set, perform target detection on the standard image set based on a deep learning prohibited items identification model to generate a detection feature set, and calculate the risk probability score based on the detection feature set to generate a risk assessment table. Step S502: Based on the risk assessment table, determine the verification status and generate a verification result set. Quantify the risk scores in the risk assessment table according to preset weight rules to generate a verification status matrix. Perform dynamic threshold analysis on the verification status matrix based on an adaptive threshold model to generate threshold determination results. Classify and label the vehicle passage status according to the threshold determination results to generate a verification result set.

[0084] First, the roof and undercarriage images from the aforementioned verification dataset are read. Based on the lane policy domain, the imaging angle and minimum imaging requirements are determined, and a contraband detector is instantiated. Simultaneously, an analysis rule table is generated. The analysis rule table specifies the image resolution, aspect ratio, denoising method, and cropping boundaries, and configures illumination compensation and reflection suppression switches for the roof and undercarriage respectively. Subsequently, size normalization, color space standardization, and distortion correction are performed on the original frames, outputting a standard image set. The source session number and trajectory node reference are registered for each frame.

[0085] Based on the standard image set, a deep learning contraband detection model is invoked for target detection. The model, named the Contraband Detection Network, takes a single-frame standard image as input and outputs candidate bounding boxes, class labels, and confidence scores. To reduce false triggers, low-confidence candidates are first soft-screened according to class thresholds, and then temporal consistency is checked within the same session: if candidates of the same class in adjacent frames differ significantly in position and scale, they are downweighted rather than directly removed. The results of the check are organized into a detection feature set, including a class score sequence, candidate bounding box trajectories, and frame coverage ratios, while maintaining a reverse pointer to the original frames.

[0086] The detected feature set is fed into the risk calculation unit and summarized at two levels: category level and session level. Category-level statistics consider the stability of the same category across multiple frames, while session-level statistics integrate the synchronous occurrence of the vehicle roof and undercarriage, and label the temporal overlap. Based on these two levels of statistics, a risk probability score is generated for each session's data segment, and compiled into a risk assessment table. The risk assessment table records the category components, frame coverage ratio, temporal overlap, and source pointer, ensuring that key quantities can be read without tracing back to plaintext images during subsequent judgments.

[0087] Based on the risk assessment form, the verification status determination process is initiated. First, the risk score is aligned with the pairing confidence and lane consistency of the preceding identity domain within the session to form a computable input set. Then, multiple indicators are quantified according to preset weighting rules to construct the row vector of the verification status matrix. Identity-related indicators and risk-related indicators are written in separate columns, and the session number and trajectory node index that generated the row vector are recorded to maintain a closed loop of read and write operations.

[0088] Therefore, a state quantization function is defined to form a matrix derived column, denoted as the formula: Y = d1·G + d2·H-d3·J.

[0089] In the formula, Y is the state quantization output; G is the category-level risk component, derived from the combination of the category score and frame coverage ratio of the detection feature set; H is the session-level consistency component, derived from the weighted result of the time overlap between the vehicle roof and the vehicle floor and the stability of the candidate box trajectory; J is the identity association penalty, reflecting the superposition of identity pairing confidence downgrade, cross-session discontinuity, and cross-lane prompts; d1, d2, and d3 are non-negative weights, configured within the lane policy domain and remaining unchanged in the current version. This Y is written into a derived column of the verification state matrix for subsequent threshold analysis.

[0090] An adaptive threshold model is applied to the verification state matrix for dynamic threshold analysis. The model reads the Y distribution and joint risk markers of the same lane over a recent period to determine the decision threshold conditions for the current session, prioritizing the checking of whether the identity hard constraint is met. If the identity hard constraint is met, a threshold judgment result is given based on the combination of Y and category-level risk components; if not, it is directly marked as failing with a reason code. The threshold analysis process records parameter snapshots and version numbers throughout to ensure replayability.

[0091] A verification result set is generated based on the aforementioned threshold determination results. Each result includes a session number, trajectory node, identity verification status, risk conclusion, and reason code, and is stored in bidirectional pointers to the state matrix row vector and the risk assessment table for easy downstream querying. For entries marked as passed, a "can be driven and released" flag is written and the determination time is recorded; for entries marked as failed, a "requires manual review" flag is retained for subsequent processing.

[0092] Next, the verification result set is written back to the corresponding trajectory node, the synchronization status of the node's identity field and risk field is updated, and the read / write snapshot of this session is frozen. After the write-back is completed, the temporary video memory and decryption cache are released, and only the standardized indicator values ​​and read-only pointers are retained to avoid plaintext residue.

[0093] Finally, the verification result set is exposed as the input interface for the next step. The passage control logic reads the session number and "driveable release" flag from this interface, generates a passage status packet, and forms a device-level control command accordingly. At the same time, the rear-end capture comparison will reference the pointers of the status matrix and risk assessment table output in this segment to complete the vehicle passage closed loop and subsequent evidence chain storage.

[0094] In one embodiment of the intelligent checkpoint verification and risk blocking method based on multimodal feature association and dynamic threshold determination in this application, the method may further include the following: Step S601: Based on the verification result set, construct the access controller to generate a control instruction table, map the verification result set to the state according to the preset access rules to generate an access state package, perform permission verification on the access state package to generate an authorization result set, generate access instructions according to the authorization result set and the preset control strategy, and drive the barrier gate device to perform opening and closing actions based on the access instructions. Step S602: Based on the vehicle-triggered lane-breaking mechanism, a vehicle data packet is generated. The status information of the barrier gate equipment and the vehicle passage information are collected in real time to generate a status data stream. The rear image of the vehicle in the status data stream is captured and processed to generate a rear feature set. The license plate information and timestamp are extracted from the rear feature set to generate a vehicle data packet.

[0095] First, the session number, trajectory node, identity verification status, and risk conclusion of the aforementioned verification result set are read. A passage controller is instantiated, and passage rules and device constraints are loaded according to the lane policy domain to generate a control instruction table. The control instruction table clearly defines the mapping relationship between three states: release, prohibition, and manual review, as well as the execution order and timeout handling of the barrier, indicator display, and alarm buzzer. Then, the verification result set is aggregated by session number, and state mapping is performed according to the control instruction table to form a passage status package containing trajectory number, lane number, drivable release flag, and reason code.

[0096] Based on the passage status packet, permission verification is performed. Verification includes checking the lane controllability flag, the barrier gate device activation flag, the on-duty operator authorization, and whether the current time period policy allows passage. If verification passes, an authorization result set is written, and a one-time control token and valid time window are generated; if verification fails, the rejection reason is filled back, and the system is restricted to a prompt-only display mode. Subsequently, a passage instruction is generated based on the authorization result set and control policy. The instruction includes the action to be executed, the token, the valid time window, and the retry limit, and is prepared to be sent to the barrier gate device.

[0097] The passage command is sent to the barrier gate and associated equipment, driving the switch to operate. When the passage command takes effect, the command issuance time and equipment feedback time are recorded to form an execution snapshot, and session-level timing is started for subsequent vehicle passage closed loop. When the prohibition command takes effect, only the display panel and alarm buzzer status are updated, without triggering the barrier gate mechanism. The execution snapshot and the session number of the passage status packet are kept consistent for subsequent association.

[0098] Upon receiving the aforementioned clearance command, the barrier mechanism is activated and triggered when a vehicle crosses the line. This initiates the collection of the barrier gate's status and passage sensor data, generating a status data stream containing a session number, the current location of the barrier gate, the vehicle's location trigger, and a timestamp. This status data stream continues until the vehicle leaves the monitoring area or the session times out, during which time it is written into the device feedback and trigger event sequence according to the sampling period.

[0099] Based on the aforementioned state data stream, rear-end capture processing is performed. Upon triggering a line crossing, the capture camera initiates a short burst of shots, performing sharpness evaluation and motion blur detection on consecutive frames, retaining one to three frames as candidates. Character regions are located in the candidate frames, and brightness and contrast are normalized. The license plate character sequence, color markers, and capture time are extracted from the rear of the vehicle, and these are aggregated into a rear-end feature set. To ensure backtracking capability, feature set entries are saved to references of corresponding trajectory nodes and session numbers.

[0100] The license plate information and timestamp are extracted from the vehicle rear feature set, and combined with the lane number and gate status to encapsulate them into a vehicle passage data packet. The vehicle passage data packet includes the vehicle rear character sequence, capture time, session number, lane number, gate opening / closing status, and snapshot pointer. If multiple candidate frames are generated from continuous shooting, the correspondence between the main frame and the auxiliary frames is recorded in the packet to facilitate character consistency verification during downstream comparison.

[0101] Next, the vehicle passage data packet and control execution snapshot are registered together in the session loop table, forming the output interface of this step. The loop table uses the session number as the primary key and links the passage status packet, authorization result set, passage command and vehicle passage data packet together to ensure that subsequent verification and comparison can read the required context from a single entry point, and can complete the determination of identity consistency and time window consistency without decrypting the image content again.

[0102] Finally, the authorization result set and vehicle passage data generated in this step are written back to the passage domain of the trajectory node. The passage domain records the instruction type, token, effective time window, and the main frame pointer for rear vehicle capture, and synchronously updates the node's device feedback summary. This write-back provides a stable reference for subsequent steps to perform correlation comparison based on the verification status matrix and generate verification records, completing the closed link from the release decision to vehicle passage data collection.

[0103] In one embodiment of the intelligent checkpoint verification and risk blocking method based on multimodal feature association and dynamic threshold determination in this application, the method may further include the following: Step S701: Based on the vehicle data packet, establish a comparison and verification device to generate a matching rule table. Extract the license plate information and time information in the vehicle data packet according to the preset mapping rules to generate a vehicle feature set. Perform spatiotemporal correlation calculation on the vehicle feature set and the verification state matrix to generate a comparison result set. Based on the comparison result set, determine and generate a verification record according to the preset threshold conditions. Step S702: Perform anomaly analysis on the verification records to generate an evidence chain database; perform mismatch analysis on the verification records based on preset alarm rules to generate an alarm trigger table; perform hierarchical processing on the anomaly information in the alarm trigger table to generate an alarm information package; and perform distributed encrypted storage on the verification records and the alarm information package to generate an evidence chain database.

[0104] First, the aforementioned vehicle data packets and control execution snapshots are read, and a comparison verifier is instantiated to generate a matching rule table. The matching rule table specifies the lane number, allowed time window, character consistency requirements, and device feedback constraints for the same session. Then, the vehicle tail character sequence, capture time, and session number are extracted from the vehicle data packets. Combined with the lane number and the gate opening / closing status, field standardization is completed to form input entries for comparison.

[0105] Feature extraction is performed based on the input entries to generate a vehicle passing feature set. Feature extraction includes character embedding representation, color labeling, and frame sharpness grading, and records the capture time and session number as time anchors. If multiple candidate frames exist, the primary frame is selected based on a joint sorting of sharpness and motion stability; secondary frames are only used for conflict verification. Each record in the feature set maintains a reference to the trajectory node and the row vector of the verification state matrix, ensuring a closed loop for subsequent comparisons and write-back.

[0106] The vehicle passage feature set and verification state matrix are fed into a spatiotemporal correlation calculation. The correlation uses session number consistency and lane number consistency as hard constraints, sorts them by temporal proximity within an allowed time window, and then uses character consistency as the primary criterion. For sessions with dual license plate recognition conflict markers, a tolerance boundary is introduced in the character distance calculation, and the identity consistency marker is read as a secondary criterion. After the calculation is completed, a comparison result set is output, including fields such as matching markers, time offset, and character difference positions.

[0107] Based on the comparison result set, a threshold determination is performed to generate a verification record. The determination logic first checks whether the hard constraints are met, and then reads whether the character consistency value and time offset fall within the threshold conditions of the rule table; if not, a mismatch is marked and a reason code is recorded. To quantify the comparison reliability, a comparison score is calculated and written into the verification record, while bidirectional pointers to the vehicle owner frame and the row vector of the verification status matrix are retained for subsequent review.

[0108] To reduce false positives, an adjacency review is triggered after the verification record is generated. The review retrieves the most recent passage record from the nearest sessions in the same lane, compares the character sequences and time intervals of the two records, and if a short-term abnormal jump occurs and the current device reports a jitter flag, the credibility level of this record is downgraded but not directly rejected. The downgrade information is written to the annotation field of the verification record to provide context for subsequent anomaly analysis.

[0109] Subsequently, anomaly analysis is performed on the verification records to construct a preliminary draft of the evidence chain entries. Anomaly analysis is triggered by mismatch markers, low confidence levels, and window offsets, and cross-validates execution snapshots and device logs to generate an anomaly summary, a list of involved devices, and a minimum necessary set of image pointers. This draft is not stored in plaintext; only read-only pointers and hash digests are saved, awaiting further classification in the alarm process.

[0110] Based on the anomaly summary, preset alarm rules are applied to generate an alarm trigger table. The trigger table distinguishes between structural mismatch and time window inconsistency according to severity level, and provides suggested handling actions and callback entry points. For each anomaly in the trigger table, combined with lane policy domain and management preferences, hierarchical processing is performed to generate an alarm information package containing session number, lane number, cause code, and minimum necessary evidence pointer, while recording the push channel and receipt placeholder.

[0111] The verification records and alarm information packets enter a distributed encrypted storage process to generate an evidence chain database. During storage, the identity field, character sequence hash, execution snapshot pointer, and alarm information are encrypted separately, and a key validity period and revocation pointer are bound to them. A timestamp and integrity verification value are written to each field to form a verifiable chain index. After successful writing, a read-only index key is returned, which is then written back to the trajectory node and the review entry of the management platform.

[0112] After the evidence chain database is generated, the core fields of the verification record are synchronized to the session loop table, the loop status is updated to "compared", and the alarm push result and receipt time are recorded. The loop table uses the session number as the primary key to link the vehicle data packets, comparison results and alarm information, ensuring that subsequent queries can replay the decision trajectory of the same session with one click, and read the necessary metadata without additional decryption.

[0113] Finally, the verification records and evidence chain index output in this step are used as the reading interface for downstream statistics and review. The statistics module calculates the lane-level mismatch rate and equipment jitter distribution based on this; when the review module receives an alarm information packet, it retrieves the minimum necessary evidence through the read-only index for manual review, and updates the handling status of the trigger table through the callback entry after the review is completed, forming a complete closed loop from comparison and judgment to the landing of evidence information.

[0114] To effectively address the shortcomings of traditional technologies in data processing, multimodal verification, and passage control, and to provide technical support for level crossing verification, this application provides an embodiment of a level crossing intelligent verification and risk blocking device based on multimodal feature association and dynamic threshold determination, used to implement all or part of the aforementioned level crossing intelligent verification and risk blocking method based on multimodal feature association and dynamic threshold determination. See [link to embodiment]. Figure 2 The intelligent checkpoint verification and risk prevention device based on multimodal feature association and dynamic threshold determination specifically includes the following components: The data acquisition module 10 is used to collect multidimensional data based on the intelligent level crossing device to generate a raw dataset. The multidimensional data includes license plate images, RFID card numbers, vehicle roof images, vehicle under images, face images, and device identifiers. The module performs format and integrity checks on the raw dataset to generate a standardized data packet. The module performs hierarchical encryption processing on the standardized data packet to generate a safety dataset. Based on the safety dataset, a feature index table is established according to vehicle identifiers. The feature index table is used to construct a state mapping model according to lane numbers. The feature recognition module 20 is used to perform multimodal verification of passing vehicles based on the state mapping model to generate a verification dataset, perform dual recognition on the license plate images in the verification dataset to generate license plate feature vectors, associate and match the license plate feature vectors with RFID card numbers to generate a vehicle-license mapping table, perform contraband analysis on the roof and undercarriage images in the verification dataset based on a preset model library to generate a risk assessment table, construct a verification state matrix based on the risk assessment table, and generate a verification result set by judging the verification state matrix according to a dynamic threshold. The intelligent verification and risk blocking module 30 for road crossings based on multimodal feature association and dynamic threshold determination is used to control the road crossing equipment to generate a passage instruction according to the verification result set. When the passage instruction is to allow passage, the gate equipment is opened, and the rear of the passing vehicle is captured to generate a vehicle data packet. The vehicle data packet is correlated and compared with the verification status matrix to generate a verification record. When the verification record shows a mismatch, an alarm message is generated and pushed to the management platform. The verification record and the alarm message are encrypted and stored to generate an evidence chain database.

[0115] As described above, the intelligent checkpoint verification and risk blocking device based on multimodal feature association and dynamic threshold determination provided in this application can effectively protect information through format verification and security encryption. A verification mechanism is constructed, combining multimodal analysis and risk assessment to establish a reliable determination strategy. Access control is introduced, ensuring verification accuracy through record comparison and alarm push notifications. This method effectively solves the shortcomings of traditional technologies in data processing, multimodal verification, and access control, providing technical support for intelligent checkpoint verification and risk blocking based on multimodal feature association and dynamic threshold determination.

[0116] From a hardware perspective, in order to effectively address the shortcomings of traditional technologies in data processing, multimodal verification, and passage control, and to provide technical support for level crossing verification, this application provides an embodiment of an electronic device for implementing all or part of the aforementioned intelligent level crossing verification and risk blocking method based on multimodal feature association and dynamic threshold determination. 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 realize information transmission between the intelligent checkpoint verification and risk blocking device based on multimodal feature association and dynamic threshold determination and core business systems, user terminals, and related databases and other related equipment; 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 intelligent checkpoint verification and risk blocking method based on multimodal feature association and dynamic threshold determination, and the embodiments of the intelligent checkpoint verification and risk blocking device based on multimodal feature association and dynamic threshold determination, the contents of which are incorporated herein by reference, and repeated details will not be described again.

[0117] 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.

[0118] In practical applications, the intelligent verification and risk blocking method for road crossings based on multimodal feature association and dynamic threshold determination can be partially executed on the electronic device side as described above, or all operations can be completed in the client device. The specific choice depends 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.

[0119] 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.

[0120] 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 3This is an example; other types of structures can also be used to supplement or replace this structure to achieve telecommunications functions or other functions.

[0121] In one embodiment, the intelligent checkpoint verification and risk blocking method based on multimodal feature association and dynamic threshold determination can be integrated into the central processing unit 9100. The central processing unit 9100 can be configured to perform the following controls: Step S101: Generate a raw dataset by collecting multidimensional data from the intelligent toll gate device. The multidimensional data includes license plate images, RFID card numbers, vehicle roof images, vehicle under-vehicle images, face images, and device identifiers. Perform format and integrity checks on the raw dataset to generate a standardized data packet. Perform hierarchical encryption processing on the standardized data packet to generate a safety dataset. Based on the safety dataset, establish a feature index table according to vehicle identifiers. Construct a state mapping model based on the feature index table according to lane numbers. Step S102: Based on the state mapping model, perform multimodal verification on passing vehicles to generate a verification dataset. Perform dual recognition on the license plate images in the verification dataset to generate license plate feature vectors. Associate and match the license plate feature vectors with RFID card numbers to generate a vehicle-license mapping table. Based on a preset model library, perform contraband analysis on the roof and undercarriage images in the verification dataset to generate a risk assessment table. Construct a verification state matrix based on the risk assessment table. Generate a verification result set by judging the verification state matrix according to dynamic thresholds. Step S103: Based on the verification result set, control the level crossing device to generate a passage instruction. When the passage instruction is to allow passage, open the gate device, capture the rear of the passing vehicle to generate a vehicle data packet, associate and compare the vehicle data packet with the verification status matrix to generate a verification record, generate an alarm message when the verification record shows a mismatch and push it to the management platform, and encrypt and store the verification record and the alarm message to generate an evidence chain database.

[0122] As described above, the electronic device provided in this application embodiment effectively protects information through format verification and security encryption. A verification mechanism is constructed, combining multimodal analysis and risk assessment to establish a reliable judgment strategy. Access control is introduced, ensuring verification accuracy through record comparison and alarm push notifications. This method effectively solves the shortcomings of traditional technologies in data processing, multimodal verification, and access control, providing technical support for intelligent level crossing verification and risk blocking based on multimodal feature association and dynamic threshold determination.

[0123] In another embodiment, the intelligent checkpoint verification and risk blocking device based on multimodal feature association and dynamic threshold determination can be configured separately from the central processing unit 9100. For example, the intelligent checkpoint verification and risk blocking device based on multimodal feature association and dynamic threshold determination can be configured as a chip connected to the central processing unit 9100, and the function of the intelligent checkpoint verification and risk blocking method based on multimodal feature association and dynamic threshold determination can be realized through the control of the central processing unit.

[0124] 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.

[0125] 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.

[0126] 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.

[0127] 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.

[0128] 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.

[0129] 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.).

[0130] 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.

[0131] 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.

[0132] Embodiments of this application also provide a computer-readable storage medium capable of implementing all steps of the intelligent checkpoint verification and risk blocking method based on multimodal feature association and dynamic threshold determination, where the execution subject is a server or client, as described in the above embodiments. The computer-readable storage medium stores a computer program that, when executed by a processor, implements all steps of the intelligent checkpoint verification and risk blocking method based on multimodal feature association and dynamic threshold determination, where the execution subject is a server or client, as described in the above embodiments. For example, when the processor executes the computer program, it implements the following steps: Step S101: Generate a raw dataset by collecting multidimensional data from the intelligent toll gate device. The multidimensional data includes license plate images, RFID card numbers, vehicle roof images, vehicle under-vehicle images, face images, and device identifiers. Perform format and integrity checks on the raw dataset to generate a standardized data packet. Perform hierarchical encryption processing on the standardized data packet to generate a safety dataset. Based on the safety dataset, establish a feature index table according to vehicle identifiers. Construct a state mapping model based on the feature index table according to lane numbers. Step S102: Based on the state mapping model, perform multimodal verification on passing vehicles to generate a verification dataset. Perform dual recognition on the license plate images in the verification dataset to generate license plate feature vectors. Associate and match the license plate feature vectors with RFID card numbers to generate a vehicle-license mapping table. Based on a preset model library, perform contraband analysis on the roof and undercarriage images in the verification dataset to generate a risk assessment table. Construct a verification state matrix based on the risk assessment table. Generate a verification result set by judging the verification state matrix according to dynamic thresholds. Step S103: Based on the verification result set, control the level crossing device to generate a passage instruction. When the passage instruction is to allow passage, open the gate device, capture the rear of the passing vehicle to generate a vehicle data packet, associate and compare the vehicle data packet with the verification status matrix to generate a verification record, generate an alarm message when the verification record shows a mismatch and push it to the management platform, and encrypt and store the verification record and the alarm message to generate an evidence chain database.

[0133] As described above, the computer-readable storage medium provided in this application embodiment effectively protects information through format verification and secure encryption. A verification mechanism is constructed, combining multimodal analysis and risk assessment to establish a reliable judgment strategy. Access control is introduced, ensuring verification accuracy through record comparison and alarm push notifications. This method effectively addresses the shortcomings of traditional technologies in data processing, multimodal verification, and access control, providing technical support for intelligent level crossing verification and risk prevention based on multimodal feature association and dynamic threshold determination.

[0134] Embodiments of this application also provide a computer program product capable of implementing all steps in the intelligent checkpoint verification and risk blocking method based on multimodal feature association and dynamic threshold determination, where the execution subject is a server or client, as described in the above embodiments. When executed by a processor, this computer program / instruction implements the steps of the intelligent checkpoint verification and risk blocking method based on multimodal feature association and dynamic threshold determination. For example, the computer program / instruction implements the following steps: Step S101: Generate a raw dataset by collecting multidimensional data from the intelligent toll gate device. The multidimensional data includes license plate images, RFID card numbers, vehicle roof images, vehicle under-vehicle images, face images, and device identifiers. Perform format and integrity checks on the raw dataset to generate a standardized data packet. Perform hierarchical encryption processing on the standardized data packet to generate a safety dataset. Based on the safety dataset, establish a feature index table according to vehicle identifiers. Construct a state mapping model based on the feature index table according to lane numbers. Step S102: Based on the state mapping model, perform multimodal verification on passing vehicles to generate a verification dataset. Perform dual recognition on the license plate images in the verification dataset to generate license plate feature vectors. Associate and match the license plate feature vectors with RFID card numbers to generate a vehicle-license mapping table. Based on a preset model library, perform contraband analysis on the roof and undercarriage images in the verification dataset to generate a risk assessment table. Construct a verification state matrix based on the risk assessment table. Generate a verification result set by judging the verification state matrix according to dynamic thresholds. Step S103: Based on the verification result set, control the level crossing device to generate a passage instruction. When the passage instruction is to allow passage, open the gate device, capture the rear of the passing vehicle to generate a vehicle data packet, associate and compare the vehicle data packet with the verification status matrix to generate a verification record, generate an alarm message when the verification record shows a mismatch and push it to the management platform, and encrypt and store the verification record and the alarm message to generate an evidence chain database.

[0135] As described above, the computer program product provided in this application embodiment effectively protects information through format verification and security encryption. A verification mechanism is constructed, combining multimodal analysis and risk assessment to establish a reliable judgment strategy. Access control is introduced, ensuring verification accuracy through record comparison and alarm push notifications. This method effectively solves the shortcomings of traditional technologies in data processing, multimodal verification, and access control, providing technical support for intelligent verification and risk blocking at level crossings based on multimodal feature association and dynamic threshold determination.

[0136] 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.

[0137] 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.

[0138] 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.

[0139] 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.

[0140] 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 intelligent verification and risk prevention at road crossings based on multimodal feature association and dynamic threshold determination, characterized in that, The method includes: A raw dataset is generated by collecting multidimensional data from intelligent level crossing equipment. The multidimensional data includes license plate images, RFID card numbers, vehicle roof images, vehicle under-vehicle images, face images, and equipment identifiers. The raw dataset is then subjected to format and integrity checks to generate standardized data packets. The standardized data packets are then subjected to hierarchical encryption to generate a safety dataset. Based on the safety dataset, a feature index table is established according to vehicle identifiers. The feature index table is then used to construct a state mapping model according to lane numbers. Based on the state mapping model, multimodal verification of passing vehicles is performed to generate a verification dataset. The license plate images in the verification dataset are subjected to dual recognition to generate license plate feature vectors. The license plate feature vectors are associated and matched with RFID card numbers to generate a vehicle-license mapping table. Based on a preset model library, the roof and undercarriage images in the verification dataset are analyzed for prohibited items to generate a risk assessment table. A verification state matrix is ​​constructed based on the risk assessment table. The verification state matrix is ​​then judged according to dynamic thresholds to generate a verification result set. Based on the verification result set, the control gate equipment generates a passage instruction. When the passage instruction is to allow passage, the gate equipment is opened, and the rear of the passing vehicle is captured to generate a vehicle data packet. The vehicle data packet is correlated and compared with the verification status matrix to generate a verification record. When the verification record shows a mismatch, an alarm message is generated and pushed to the management platform. The verification record and the alarm message are encrypted and stored to generate an evidence chain database.

2. The intelligent checkpoint verification and risk blocking method based on multimodal feature association and dynamic threshold determination according to claim 1, characterized in that, The process involves collecting multidimensional data from intelligent level crossing devices to generate a raw dataset. This multidimensional data includes license plate images, RFID card numbers, vehicle roof images, vehicle under-car images, facial images, and device identifiers. The raw dataset undergoes format and integrity checks to generate a standardized data package, including: Based on a distributed acquisition mechanism, multiple data streams are acquired from intelligent checkpoint equipment to generate a data collection set. The multiple data streams are then segmented according to a preset acquisition cycle to generate segmented data groups. Fields are extracted and encoded from the segmented data groups according to a preset data template to generate a device data package. The license plate image, RFID card number, roof image, undercarriage image, face image, and device identifier in the device data package are organized to generate an acquisition record table. The data collection record table is validated to generate an original dataset. The data items in the data collection record table are validated according to the field integrity rules to generate a validation result set. Based on the validation result set, the data completion interface is called to obtain missing field information and generate a completion data package. The validated data is reorganized according to the preset field specifications to generate an original dataset containing license plate images, RFID card numbers, vehicle roof images, vehicle under images, face images and device identifiers.

3. The intelligent checkpoint verification and risk blocking method based on multimodal feature association and dynamic threshold determination according to claim 1, characterized in that, The step of generating a security dataset by performing hierarchical encryption on the standardized data packets, establishing a feature index table based on the security dataset according to vehicle identifiers, and constructing a state mapping model based on the feature index table according to lane numbers includes: A hierarchical encryption module is constructed based on the data security level to generate an encryption policy table. Standardized data packets are divided into data security levels according to preset classification rules to generate hierarchical data packets. Keys are allocated to sensitive fields in the hierarchical data packets to generate encryption parameter sets. Based on the encryption parameter sets, the hierarchical data packets are encrypted to generate a secure dataset. The secure dataset is used to establish a data mapping table based on vehicle identifiers and timestamps. Based on the data mapping table, a feature index system is constructed to generate a state mapping model. The vehicle identification information in the data mapping table is serialized to generate a vehicle feature sequence. The vehicle feature sequence is grouped and mapped according to lane number to generate an index relationship table. Based on the index relationship table, a lane-level feature mapping is established to generate a state mapping model. The state mapping model is used for subsequent vehicle verification analysis.

4. The intelligent checkpoint verification and risk blocking method based on multimodal feature association and dynamic threshold determination according to claim 1, characterized in that, The process of performing multimodal verification of passing vehicles based on the state mapping model to generate a verification dataset, performing dual recognition on license plate images in the verification dataset to generate license plate feature vectors, and associating and matching the license plate feature vectors with RFID card numbers to generate a vehicle-license mapping table includes: A multimodal verifier is constructed based on a state mapping model to generate a verification process table. Passing vehicle data is collected from multiple sources according to preset collection rules to generate a real-time data stream. The real-time data stream is time-aligned to generate a synchronous data packet. The synchronous data packet is then feature-extracted according to the verification process table to generate a verification dataset. The verification dataset is used to generate a vehicle-license mapping table by performing identity feature recognition. The license plate images in the verification dataset are recognized by high-position and low-position cameras respectively to generate dual license plate features. The dual license plate features are then used to perform consistency verification to generate a license plate feature vector. Based on the license plate feature vector and the RFID card number, a spatiotemporal correlation calculation is performed to generate the vehicle-license mapping table.

5. The intelligent verification and risk blocking method for level crossings based on multimodal feature association and dynamic threshold determination according to claim 1, characterized in that, The step involves analyzing the vehicle roof and undercarriage images in the verification dataset based on a preset model library to generate a risk assessment table. A verification status matrix is ​​then constructed based on the risk assessment table. Finally, the verification status matrix is ​​used to generate a verification result set according to dynamic thresholds, including: Based on a pre-defined model library, a prohibited item identifier is constructed to generate an analysis rule table. The images of the vehicle roof and the vehicle bottom in the verification dataset are pre-processed according to the pre-defined resolution rules to generate a standard image set. The standard image set is then subjected to target detection based on a deep learning prohibited item identification model to generate a detection feature set. Based on the detection feature set, a risk probability score is calculated to generate a risk assessment table. Based on the risk assessment table, a verification result set is generated by determining the verification status. The risk scores in the risk assessment table are quantified according to preset weight rules to generate a verification status matrix. The verification status matrix is ​​then dynamically analyzed based on an adaptive threshold model to generate a threshold determination result. Based on the threshold determination result, the vehicle traffic status is classified and labeled to generate a verification result set.

6. The intelligent verification and risk blocking method for level crossings based on multimodal feature association and dynamic threshold determination according to claim 1, characterized in that, The step of controlling the level crossing equipment to generate a passage command based on the verification result set, and opening the gate equipment when the passage command is to allow passage, and capturing the rear of the passing vehicle to generate a vehicle data packet, includes: Based on the verification result set, a control controller is constructed to generate a control instruction table. The verification result set is mapped to a state according to a preset access rule to generate an access state package. The access state package is verified for permissions to generate an authorization result set. Based on the authorization result set, an access instruction is generated according to a preset control strategy. Based on the access instruction, the barrier gate device is driven to perform opening and closing actions. Based on the vehicle-triggered lane-breaking mechanism to generate vehicle data packets, the status information of the barrier gate equipment and vehicle passage information are collected in real time to generate a status data stream. The rear images of the vehicles in the status data stream are captured and processed to generate a rear feature set. The license plate information and timestamp are extracted from the rear feature set to generate a vehicle data packet.

7. The intelligent verification and risk blocking method for level crossings based on multimodal feature association and dynamic threshold determination according to claim 1, characterized in that, The process of associating and comparing the vehicle data packet with the verification status matrix to generate a verification record, generating an alarm message and pushing it to the management platform when the verification record shows a mismatch, and encrypting and storing the verification record and the alarm message to generate an evidence chain database includes: A comparison and verification device is established based on the vehicle data packet to generate a matching rule table. The license plate information and time information in the vehicle data packet are extracted according to the preset mapping rules to generate a vehicle feature set. The vehicle feature set and the verification state matrix are spatiotemporally correlated to generate a comparison result set. The verification record is generated according to the comparison result set and the preset threshold conditions. Anomaly analysis is performed on the verification records to generate an evidence chain database. Based on preset alarm rules, mismatch analysis is performed on the verification records to generate an alarm trigger table. The abnormal information in the alarm trigger table is processed in a hierarchical manner to generate an alarm information package. The verification records and the alarm information package are distributed and encrypted for storage to generate an evidence chain database.

8. A smart checkpoint verification and risk prevention device based on multimodal feature association and dynamic threshold determination, characterized in that, The device includes: The data acquisition module is used to collect multidimensional data based on the intelligent level crossing equipment to generate a raw dataset. The multidimensional data includes license plate images, RFID card numbers, vehicle roof images, vehicle under images, face images, and equipment identifiers. The module performs format and integrity checks on the raw dataset to generate a standardized data packet. The standardized data packet is then subjected to hierarchical encryption to generate a safety dataset. Based on the safety dataset, a feature index table is established according to vehicle identifiers. The feature index table is then used to construct a state mapping model according to lane numbers. The feature recognition module is used to perform multimodal verification of passing vehicles based on the state mapping model to generate a verification dataset, perform dual recognition on the license plate images in the verification dataset to generate license plate feature vectors, associate and match the license plate feature vectors with RFID card numbers to generate a vehicle-license mapping table, perform contraband analysis on the vehicle roof and vehicle bottom images in the verification dataset based on a preset model library to generate a risk assessment table, construct a verification state matrix based on the risk assessment table, and generate a verification result set by judging the verification state matrix according to a dynamic threshold. The intelligent verification and risk blocking module for road crossings based on multimodal feature association and dynamic threshold determination is used to control the road crossing equipment to generate passage instructions according to the verification result set. When the passage instruction is to allow passage, the gate equipment is opened, and the rear of the passing vehicle is captured to generate a vehicle data packet. The vehicle data packet is correlated and compared with the verification status matrix to generate a verification record. When the verification record shows a mismatch, an alarm message is generated and pushed to the management platform. The verification record and the alarm message are encrypted and stored to generate an evidence chain database.

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 intelligent checkpoint verification and risk blocking method based on multimodal feature association and dynamic threshold determination as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the intelligent checkpoint verification and risk blocking method based on multimodal feature association and dynamic threshold determination as described in any one of claims 1 to 7.

Images