Method and system for acquiring category information of entrance and exit of target place

Abstract

The invention discloses a method and a system for acquiring category information of an entrance and an exit of a target place, particularly relates to the technical field of traffic monitoring and vehicle information fusion, and is used for solving the problem of inaccurate vehicle identification caused by mismatching of multi-source time sequence data. On the basis of the track, different types of monitoring information accurately correspond to the same vehicle, so that the problem of data dislocation caused by conditions such as acceleration, deceleration or lane change of the vehicle is effectively avoided; besides, by performing unified coordination and comprehensive analysis on various monitoring data and intelligently judging the reliability of the data, the situation that wrong matching possibly exists can be actively screened out and re-checked in time, the situation of vehicle identification errors and misjudgment is remarkably reduced, it is ensured that information is accurate and reliable when the vehicle enters and exits a place, and the safety of the vehicle is improved. And the vehicle passing efficiency and the accuracy and effectiveness of on-site supervision are improved.

Description

Technical Field

[0001] This invention relates to the field of traffic monitoring and vehicle information fusion technology, and more specifically, to a method and system for obtaining category information of entrances and exits of target locations. Background Technology

[0002] The entrance to the large logistics park is equipped with license plate recognition cameras, remote sensing emission monitoring beams, and embedded dynamic weighing plates, aiming to allow transport vehicles to simultaneously complete emission standard determination and axle load verification while passing at a constant speed. The license plate frame streams, emission spectra, and axle load features generated at each detection stage are transmitted to the fusion platform in real time through edge nodes. The platform uses a sliding time window algorithm to integrate multiple data streams, generate vehicle category labels, and drive the barrier gate and monitoring records.

[0003] In dense traffic environments, acceleration, deceleration, and lane-changing behaviors constantly alter vehicle speed and trajectory, causing timestamps at various detection stages to fluctuate. Cameras experience exposure delays due to lighting fluctuations; emission beams require spectral analysis before results are transmitted; and weighing plates output weight instantly. The timing of these outputs is difficult to synchronize, leading to repeated misalignments of records for the same vehicle. When the platform writes data to the table in arrival order, emission records have not yet been returned, while subsequent vehicle weight information has already arrived. Aggregation logic in congested traffic easily mixes data from different vehicles, ultimately resulting in the swapping of labels for over-emission light vehicles and compliant heavy vehicles, accidental triggering of gates, and distortion of the enforcement chain. Traditional association methods rely solely on license plate character consistency, ignoring spatiotemporal drift in high-density traffic and lacking real-time trajectory prediction compensation. This vulnerability is a core obstacle to the implementation of multi-source detection fusion solutions.

[0004] To address the aforementioned problems, a technical solution is provided. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method and system for acquiring category information of entrances and exits of target locations. By first clearly tracking the vehicle's trajectory when it enters the location, and then accurately mapping different types of monitoring information to the same vehicle based on this trajectory, the method effectively avoids data misalignment caused by vehicle acceleration, deceleration, or lane changes. Furthermore, by uniformly coordinating and comprehensively analyzing various monitoring data, the system intelligently judges data reliability, proactively filters out potential mismatches, and promptly verifies them. This significantly reduces vehicle identification errors and misjudgments, ensuring accurate and reliable information when vehicles enter and exit the location, improving vehicle traffic efficiency and the accuracy and effectiveness of on-site supervision, thereby solving the problem of inaccurate vehicle identification caused by mismatches in multi-source time-series data.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for obtaining category information of entrances and exits of a target location, comprising the following steps: S1: The entrance laser micro-measurement array continuously extracts the vehicle motion vector and writes the motion vector primary key into the temporary trajectory index table; S2: After capturing the character frame, the license plate recognition module calls the motion vector primary key, maps the primary key to the license plate record, and synchronously inserts the sequence timestamp; S3: When the remote sensing emission resolver completes the resolution, it retrieves the primary key of the motion vector, predicts the arrival delay based on the motion vector, rewrites the emission record, and pushes it to the fusion buffer. S4: After the dynamic weighing plate outputs the axle load value in real time, it references the same motion vector primary key and performs time difference correction and order rearrangement with the emission records in the fusion buffer to generate a complete multi-mode entry. S5: The fusion engine obtains the intersection density factor and the scattering symmetry factor from the multi-modal entries, and derives the fusion confidence factor through machine learning. If the fusion confidence factor is higher than the preset threshold, the vehicle category label is updated and sent to the gate control terminal; otherwise, the entry is transferred to the review buffer and archived to the regulatory database.

[0007] In a preferred embodiment, step S1 includes the following: Multiple high-precision laser sensors scan vehicles in real time to capture their position and speed information and generate motion vectors. A unique motion vector key is generated for each vehicle, based on the initial time and position of the vehicle entering the detection area, and its uniqueness is ensured by a hash function. The motion vector key and the corresponding trajectory data are written into a temporary trajectory index table, which contains the motion vector key, timestamp, position coordinates, and motion vector fields, and is dynamically updated as the vehicle moves.

[0008] In a preferred embodiment, step S2 includes the following: After capturing character frames and recognizing license plate numbers, the vehicle trajectory entry closest to the capture time is retrieved from the temporary trajectory index table to obtain the corresponding motion vector primary key. Then, the motion vector primary key is paired with the license plate number to generate an associated record. At the same time, a sequence timestamp is generated to record the time when recognition is completed. Finally, the motion vector primary key, license plate number, and sequence timestamp are integrated into a license plate record and stored in the license plate database.

[0009] In a preferred embodiment, step S3 includes the following: After completing the analysis, the remote sensing emission analyzer records the analysis completion time and retrieves the vehicle trajectory entry closest to the analysis completion time from the temporary trajectory index table, extracting the motion vector primary key; it predicts the vehicle's position at the analysis completion time based on the motion vector sequence and calculates the arrival delay from the emission collection point to the predicted position.

[0010] In a preferred embodiment, step S3 further includes the following: Subtracting the arrival delay from the parsing completion time yields the corrected emission collection timestamp. The motion vector primary key, emission data, and emission collection timestamp are then integrated into an emission record and pushed to the fusion buffer.

[0011] In a preferred embodiment, step S4 includes the following: A dynamic weighing plate measures the axle load of a vehicle and records the measurement time. It retrieves the motion trajectory record closest to the measurement time from a temporary trajectory index table and extracts the motion vector primary key. Based on the motion vector primary key, it retrieves emission records from the fusion buffer, calculates the time difference between the axle load measurement time and the emission collection timestamp, uses the motion vector sequence to predict the vehicle's position at the measurement time, and adjusts the emission record timestamp to achieve time alignment. The motion vector primary key, license plate number, corrected emission data, axle load sequence, and measurement time are integrated into a multi-modal entry and arranged in ascending order of measurement time to generate an ordered multi-modal entry list.

[0012] In a preferred embodiment, step S5 includes the following: The intersection density factor and the scatter symmetry factor are extracted from the multi-modal entries. The intersection density factor and the scatter symmetry factor are input into the pre-trained gradient boosting ensemble classifier to output the fusion confidence factor. The fusion engine decides to update the vehicle category label or transfer the multi-modal entries to the verification buffer based on the comparison between the fusion confidence factor and the preset threshold.

[0013] In a preferred embodiment, step S5 further includes the following: The intersection density factor is obtained by dividing the total length of the intersection segment between the predicted trajectory and the adjacent trajectory by the length of the target trajectory, while the scattering symmetry factor is obtained by calculating the dispersion of the multi-mode timestamps and the ratio of the maximum to the minimum deviation.

[0014] In a preferred embodiment, step S5 further includes the following: When the fusion confidence factor is higher than the preset threshold, the fusion engine determines the vehicle category based on the license plate record, emission record and axle load measurement data in the multi-mode entry and issues it to the gate control terminal. When the fusion confidence factor is not higher than the preset threshold, the fusion engine transfers the multi-mode entry to the review buffer and archives it to the regulatory database.

[0015] A system for obtaining category information of entrances and exits of a target location, comprising: The entrance laser macro measurement array continuously extracts the vehicle motion vector and writes the motion vector primary key into a temporary trajectory index table; After capturing a character frame, the license plate recognition module calls the motion vector primary key, maps the primary key to the license plate record, and synchronously inserts a sequence timestamp. When the remote sensing emission parser completes the parsing, it retrieves the primary key of the motion vector, predicts the arrival delay based on the motion vector, rewrites the emission record, and pushes it to the fusion buffer. After the dynamic weighing plate outputs the axle load value in real time, it references the same motion vector primary key and performs time difference correction and order rearrangement with the emission records in the fusion buffer to generate a complete multi-mode entry. The fusion engine obtains the intersection density factor and the scattering symmetry factor from the multi-modal entries, and derives the fusion confidence factor through machine learning. If the fusion confidence factor is higher than the preset threshold, the vehicle category label is updated and sent to the gate control terminal; otherwise, the entry is transferred to the review buffer and archived to the regulatory database.

[0016] The technical effects and advantages of the method and system for obtaining category information of entrances and exits of target locations according to the present invention are as follows: This invention first tracks the vehicle's trajectory when it enters a location, and then accurately associates different types of monitoring information with the same vehicle based on that trajectory. This effectively avoids data misalignment caused by vehicle acceleration, deceleration, or lane changes. Furthermore, by coordinating and comprehensively analyzing various monitoring data, the invention intelligently assesses data reliability, proactively identifies potential mismatches, and promptly verifies them. This significantly reduces vehicle identification errors and misjudgments, ensuring accurate and reliable information when vehicles enter and exit locations, and improving vehicle traffic efficiency and the accuracy and effectiveness of on-site supervision. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a method for obtaining category information of entrances and exits of a target location according to the present invention.

[0018] Figure 2 This is a schematic diagram of the structure of a target location entrance / exit category information acquisition system according to the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1: Figure 1 The present invention provides a method for obtaining category information of entrances and exits of a target location, comprising: S1: The entrance laser micro-measurement array continuously extracts the vehicle motion vector and writes the motion vector primary key into the temporary trajectory index table; S2: After capturing the character frame, the license plate recognition module calls the motion vector primary key, maps the primary key to the license plate record, and synchronously inserts the sequence timestamp; S3: When the remote sensing emission resolver completes the resolution, it retrieves the primary key of the motion vector, predicts the arrival delay based on the motion vector, rewrites the emission record, and pushes it to the fusion buffer. S4: After the dynamic weighing plate outputs the axle load value in real time, it references the same motion vector primary key and performs time difference correction and order rearrangement with the emission records in the fusion buffer to generate a complete multi-mode entry. S5: The fusion engine obtains the intersection density factor and the scattering symmetry factor from the multi-modal entries, and derives the fusion confidence factor through machine learning. If the fusion confidence factor is higher than the preset threshold, the vehicle category label is updated and sent to the gate control terminal; otherwise, the entry is transferred to the review buffer and archived to the regulatory database.

[0021] In modern traffic management scenarios, such as entrances and exits of toll stations and parking lots, accurate vehicle identification and classification are core requirements for ensuring traffic efficiency and regulatory reliability. Traditional methods typically rely on a single data source (such as license plate recognition or weighing systems). However, under complex traffic conditions, such as multiple vehicles driving side-by-side, vehicles accelerating, or changing lanes, data misalignment or misjudgment can easily occur, leading to inaccurate vehicle category information. This invention proposes a method for acquiring category information at entrances and exits of target locations. It integrates multiple monitoring methods through multi-modal fusion technology, including laser macro measurement, license plate recognition, remote sensing emission analysis, and dynamic weighing, using motion vectors as the primary key to associate data from different sources. This method can effectively track vehicle trajectories and ensure data accuracy, thereby improving identification precision and reliability.

[0022] Step S1 serves as the starting point of the entire process, responsible for extracting vehicle motion vectors and establishing a temporary trajectory index table, providing a crucial data foundation for subsequent steps.

[0023] S1.1, Extraction of motion vectors: At the entrance and exit areas of the target site, the entrance laser macro measurement array consists of multiple high-precision laser sensors. These sensors are strategically positioned along the entrance and exit paths to cover all possible passing vehicles. Each laser sensor can scan and detect passing vehicles in real time, capturing their spatial position information. To ensure the accuracy of position recording, a timestamp is appended to each detected position data point, with a timestamp recording accuracy down to the millisecond level. This high-precision time recording ensures the continuity and accuracy of vehicle position data, facilitating subsequent analysis of vehicle motion states.

[0024] To calculate the vehicle's velocity, position data from two consecutive time points are processed. Assume that at a certain time point, the vehicle's position coordinates are recorded as the first x-coordinate and the first y-coordinate; at the next immediately following time point, the vehicle's position coordinates are recorded as the second x-coordinate and the second y-coordinate. The calculation process first determines the time difference between the two time points, which is obtained by subtracting the timestamp of the first time point from the timestamp of the second time point. Then, the change in the vehicle's position in the horizontal direction is calculated, i.e., the horizontal displacement is obtained by subtracting the first x-coordinate from the second x-coordinate; similarly, the change in the vertical direction is calculated, i.e., the vertical displacement is obtained by subtracting the first y-coordinate from the second y-coordinate. Next, the horizontal displacement is removed by the time interval to obtain the vehicle's velocity component in the horizontal direction; the vertical displacement is removed by the time interval to obtain the vehicle's velocity component in the vertical direction. These two velocity components together describe the vehicle's direction and magnitude of motion within that time period, forming a motion vector.

[0025] A laser sensor continuously scans passing vehicles at a fixed high frequency, such as 100 times per second, generating a series of discrete position and corresponding speed data. These data are arranged chronologically to form a sequence of vehicle motion vectors. In this way, the vehicle's trajectory can be completely recorded, providing precise spatial and temporal information for subsequent vehicle identification and data matching.

[0026] S1.2, Generation of motion vector primary keys: To accurately distinguish different vehicles in subsequent processing, a unique identifier, called the motion vector key, is generated for each vehicle entering the detection area. The process of generating the motion vector key is based on the time and location information of the vehicle's first detection. The specific generation method is as follows: When a vehicle enters the detection range of the entrance laser macro measurement array, the time of its first detection is recorded, called the initial time, along with the corresponding position coordinates, called the initial x-coordinate and initial y-coordinate. The initial time, initial x-coordinate, and initial y-coordinate are combined into a data sequence in a pre-defined order. Then, a specific hash calculation method is applied to this data sequence; for example, each value in the data sequence is converted into a fixed-length number string, and these number strings are concatenated sequentially and then rounded to the nearest integer, ultimately generating a unique result as the vehicle's motion vector key.

[0027] By combining initial time and initial location information and performing hash calculations, the motion vector key of each vehicle can be ensured to be unique within the system. This uniqueness avoids data confusion between different vehicles, enabling subsequent steps to accurately link relevant information of each vehicle, such as trajectory data or other monitoring data, through the motion vector key.

[0028] S1.3, Creation and writing of the temporary trajectory index table: To dynamically record and manage vehicle motion information, a temporary trajectory index table was designed to store the motion vector primary key and its associated trajectory data for each vehicle. The temporary trajectory index table contains multiple fields, such as the motion vector primary key, timestamp, position coordinates, and motion vector, to comprehensively describe the vehicle's motion state.

[0029] When a vehicle enters the detection area, a motion vector primary key for the vehicle is generated according to the aforementioned method, and a new record entry is created for the vehicle in the temporary trajectory index table. The initial record entry contains the motion vector primary key, the timestamp of the vehicle's first detection, and the corresponding initial position coordinates and motion vector. As the vehicle moves within the detection area, the entrance laser macro measurement array continuously captures new position data and motion vectors, appending this new data to the corresponding record entry for the vehicle in chronological order, maintaining continuous updates to the trajectory data. The temporary trajectory index table is stored in an in-memory database. This storage method supports fast data writing and efficient data reading, ensuring the system can process a large amount of vehicle motion information in real time.

[0030] Through the dynamic updating and efficient storage of the temporary trajectory index table, the movement process of each vehicle can be fully tracked, and accurate data association can be provided for subsequent processing steps such as license plate recognition, emissions analysis, or weighing.

[0031] Step S1 continuously extracts vehicle motion vectors using an entrance laser macro measurement array and generates a unique motion vector primary key for each vehicle. This information is then written into a temporary trajectory index table. This process enables precise capture and dynamic management of vehicle motion states at the entrance and exit of the target location, ensuring the uniqueness and continuity of vehicle data and providing reliable technical support for subsequent comprehensive analysis by the system.

[0032] In a multi-modal fusion system, license plate recognition, as a crucial link connecting vehicle identity and movement trajectory, must be precise and efficient in its processing logic to provide a reliable data foundation for subsequent steps. Step S2 focuses on the specific processing procedures of the license plate recognition module, aiming to achieve accurate association between license plate information and vehicle trajectory by capturing character frames, calling the motion vector primary key, and inserting sequence timestamps.

[0033] Step S2: After capturing the character frame, the license plate recognition module calls the motion vector primary key, maps the primary key to the license plate record, and synchronously inserts the sequence timestamp.

[0034] S2.1, Character Frame Capture and Recognition: When a vehicle passes through the entrance / exit of a target location, the camera needs to quickly capture a clear image of the license plate to ensure the reliability of subsequent processing. The license plate recognition module uses cameras installed at the entrance / exit to continuously monitor vehicle traffic. When a vehicle is detected entering the field of view, the camera captures images at a frequency of multiple frames per second, selecting image frames containing the license plate area, called character frames. The captured character frames then enter the image processing and recognition process to extract the license plate number.

[0035] First, the character frame undergoes image preprocessing to optimize its quality. The color image is converted to grayscale by extracting the brightness value of each pixel, ignoring color information, thus generating a grayscale image that only reflects brightness variations. Next, a noise reduction algorithm is applied to smooth the image. Specifically, the median brightness value of each pixel's neighborhood is taken and replaced with the original value to reduce the impact of background noise. Then, edge detection enhances the clarity of the license plate region's outline. This involves comparing the brightness differences between adjacent pixels; when the difference exceeds a preset threshold, it is marked as an edge point, generating an image that highlights the license plate boundary.

[0036] After preprocessing, the license plate region is located. Based on the edge detection results, morphological operations are applied to optimize the integrity of the license plate region. First, a dilation operation is performed, which expands the area around the edge points to fill in the broken parts of the license plate region. Then, an erosion operation is performed, which removes small isolated points outside the edges to generate a coherent license plate region outline. After localization, the system crops the license plate region and retains this part of the image for subsequent analysis.

[0037] Next, the cropped license plate area is divided into individual character sub-regions. Specifically, the image is scanned horizontally, changes in brightness are detected, and the gaps between characters are identified. When a gap width exceeds a preset value, it is determined as a character demarcation point, thus segmenting each character sub-region. Each character sub-region is input into a pre-trained optical character recognition model. This model compares the features of the character image with those of training samples and outputs the corresponding text character. After all characters are recognized, the system sequentially combines these characters to generate the complete license plate number.

[0038] This process, through multi-layered image processing and recognition, ensures accurate extraction of license plate numbers even under changing lighting conditions or vehicle movement, providing a reliable basis for vehicle identification.

[0039] S2.2, call the motion vector primary key: To achieve accurate matching between license plate information and vehicle trajectory, the identified license plate number needs to be associated with the vehicle's motion data. After recognizing the license plate number, the license plate recognition module retrieves the primary key of the motion vector corresponding to the current vehicle from the temporary trajectory index table generated in the previous steps, establishing a mapping relationship between the two.

[0040] First, the precise moment the character frame was captured is recorded, called the capture time. Specifically, this is the timestamp of the time the camera captured the frame, accurate to the millisecond level. Then, a temporary trajectory index table is used to find the vehicle trajectory entry closest to the capture time. This temporary trajectory index table contains multiple vehicle trajectory records, each including the vehicle's motion vector key and the corresponding trajectory timestamp. The difference between the capture time and the trajectory timestamp in each trajectory record is calculated, and the trajectory record with the smallest difference is selected as the matching object.

[0041] The motion vector primary key is extracted from the matched trajectory records. This primary key is a unique identifier used to represent the motion trajectory of a specific vehicle. The extracted motion vector primary key is paired with the identified license plate number to generate an associated record, ensuring a one-to-one correspondence between the license plate number and the vehicle trajectory at the data level. During the pairing process, if multiple trajectory records with similar time differences exist, vehicle location information can be further combined to prioritize the trajectory record whose location is closest to the entrance / exit.

[0042] This association mechanism solves the data confusion problem that may occur when multiple vehicles pass through entrances and exits at the same time by matching time and space, and ensures the accurate correspondence between license plate information and movement trajectory.

[0043] S2.3, Insert sequence timestamp: To ensure the consistency and traceability of license plate records, a precise timestamp needs to be added to each record. After mapping the license plate number to the motion vector primary key, the license plate recognition module generates and inserts a sequence timestamp to record the completion time of the recognition process.

[0044] The sequence timestamp is generated by a high-precision clock. Specifically, it reads the current system time value, accurate to the millisecond, ensuring each timestamp is unique throughout the system. The generated timestamp, along with the motion vector primary key and license plate number, is integrated into a complete license plate record. The license plate record format includes three parts: a motion vector primary key, used to associate the vehicle trajectory; the license plate number, serving as the vehicle's identifier; and a sequence timestamp, marking the exact moment the record was generated.

[0045] After integration, license plate records are stored in a license plate database. The storage process writes records to the database sequentially in chronological order, ensuring data is arranged in the order of generation for easy subsequent querying and analysis. If the database needs to update existing records, the corresponding entry is located based on the motion vector primary key, and relevant fields are replaced or appended.

[0046] The license plate recognition module captures character frames and identifies license plate numbers, retrieves the corresponding motion vector primary key from a temporary trajectory index table, maps it to the license plate record, and inserts a sequence timestamp, thus achieving a precise association between vehicle identity and motion trajectory. From image processing to data matching and time stamping, a complete processing flow is formed. In complex traffic scenarios at target location entrances and exits, it effectively addresses the challenges posed by dynamic vehicle changes, ensuring the temporal and spatial correspondence of data and providing support for subsequent vehicle monitoring and data analysis.

[0047] Step S1 extracts the vehicle motion vector using the inlet laser macro measurement array and records it to a temporary trajectory index table. Step S2 uses the license plate recognition module to map the motion vector primary key to the license plate record and synchronize the timestamp, laying the foundation for subsequent data fusion. However, due to the time delay in the collection and parsing of emission data during dynamic driving (such as acceleration, deceleration, or lane change), there may be a spatiotemporal misalignment between the emission record and the actual vehicle position.

[0048] Step S3 focuses on the processing of the remote sensing emission resolver, aiming to ensure the accurate correspondence between emission data and vehicle trajectory by retrieving motion vector primary keys, predicting arrival delays, and supplementing emission records, thus providing a reliable multi-mode data foundation for subsequent steps.

[0049] S3.1, Retrieve the motion vector primary key: In vehicle management scenarios at the entrances and exits of target locations, remote sensing emission analyzers need to accurately correlate the analyzed emission data with the vehicle's movement trajectory to ensure that the data reflects the actual emission situation of a specific vehicle. Therefore, after completing emission data analysis, the precise moment of completion is first recorded, known as the analysis completion time. This time is recorded with millisecond-level precision to ensure the accuracy of the time stamp and facilitate subsequent time matching with the vehicle's trajectory.

[0050] The system then retrieves the vehicle trajectory entry closest to the parsing completion time from the temporary trajectory index table. This table stores multiple vehicle trajectory records, each containing a unique motion vector key and a corresponding trajectory timestamp. The trajectory record with the smallest time difference between the parsing completion time and the timestamp of each record is determined and used as the matching target. The time difference is calculated by subtracting the trajectory timestamp from the parsing completion time, taking the absolute value, and comparing it with all records to find the record with the smallest difference.

[0051] The motion vector primary key is extracted from the matched trajectory records. This primary key is a unique identifier used to identify the motion trajectory of a specific vehicle. Through this process, the remote sensing emissions analyzer can associate emissions data with the correct vehicle trajectory, avoiding data errors caused by time or vehicle confusion, and providing accurate identification basis for subsequent processing.

[0052] S3.2, Predicted arrival delay: In real-world scenarios, vehicles are in motion, and the collection and analysis of emissions data takes time, meaning the vehicle may have already left the emissions collection point by the time analysis is complete. To ensure that the timestamps in the emissions records reflect the actual time the vehicle passed the emissions collection point, it is necessary to predict the vehicle's position at the moment analysis is complete based on its trajectory, and calculate the time required to travel from the emissions collection point to the predicted position, i.e., the arrival delay.

[0053] Retrieve the motion vector sequence corresponding to the primary key of the motion vector from the temporary trajectory index table. This sequence contains the vehicle's velocity and direction components at different points in time. Select the trajectory point that is closest in time to the parsing completion time, called the nearest trajectory point. This point records the vehicle's position coordinates and velocity components at that moment. The position coordinates represent the vehicle's spatial position, and the velocity components represent the vehicle's speed and direction.

[0054] The vehicle's position at the moment of resolution completion is predicted using information from the nearest trajectory point. The specific process is as follows: First, calculate the time interval from the nearest trajectory point's time to the resolution completion time, which is calculated by subtracting the nearest trajectory point's time from the resolution completion time. Then, multiply this time interval by the velocity component of the nearest trajectory point to obtain the position increment, which represents the vehicle's displacement within that time interval. Finally, add the position increment to the coordinates of the nearest trajectory point to obtain the predicted position.

[0055] Next, the arrival delay of the vehicle from the emission collection point to the predicted location is calculated. The emission collection point is a fixed geographical location of the remote sensing equipment. The straight-line distance between the predicted location and the emission collection point is calculated by subtracting the coordinates of the emission collection point from the coordinates of the predicted location, squared the differences in the x-coordinate and y-coordinate respectively, summing the results, and then taking the square root of the result to obtain the straight-line distance. To calculate the arrival delay, the average speed of the vehicle needs to be determined. The average speed is obtained by averaging the values ​​of all velocity components in the motion vector sequence, that is, summing the values ​​of all velocity components in the sequence and dividing by the total number of trajectory points. Dividing the straight-line distance by the average speed yields the arrival delay, which represents the time required for the vehicle to move from the emission collection point to the predicted location.

[0056] This prediction process can accurately correct the timestamps of emission records to match the actual time when the vehicle passes through the emission collection point, thus improving the accuracy of the data in the time dimension.

[0057] S3.3, Rewrite the emission records and push them to the fusion buffer: After the emissions data is parsed and corrected, it needs to be organized into a complete record and transmitted to the next processing stage to achieve integration with other vehicle data. Following the predicted arrival delay, the remote sensing emissions parser adjusts the timestamps of the emissions records and pushes them to the fusion buffer.

[0058] The emissions record includes a motion vector primary key, parsed emissions data, and corrected emissions collection timestamps. The emissions collection timestamp is calculated by subtracting the arrival delay from the parsing completion time to obtain the actual time the vehicle passed the emissions collection point. The motion vector primary key identifies the corresponding vehicle trajectory, and the emissions data are the specific emissions values ​​obtained from the remote sensing equipment.

[0059] The motion vector primary key, emission data, and corrected emission acquisition timestamps are integrated into a single complete emission record, which is then pushed to the fusion buffer. The fusion buffer serves as a temporary storage area to cache emission data, awaiting matching and correction with other data such as dynamic weighing plate data.

[0060] This step ensures that emissions data are aligned with vehicle trajectories in terms of time and labeling, guaranteeing the accuracy of data integration.

[0061] After completing the analysis, the remote sensing emissions analyzer retrieves the motion vector primary key, predicts arrival delay, and adds timestamps to the emissions records, achieving a precise correlation between emissions data and vehicle trajectories. From time matching to location prediction and then to time correction, a complete processing flow is formed. This effectively solves the problem of emissions data latency, ensures the temporal and spatial correspondence of the data, and provides precise technical support for vehicle monitoring and data analysis.

[0062] Step S3 uses a remote sensing emission resolver to generate emission records and pushes them to the fusion buffer. These data are all associated using motion vector primary keys. However, due to complex situations such as speed changes during vehicle dynamics or multiple vehicles driving in parallel, the data acquisition times of different devices may differ, leading to temporal inconsistencies in the multi-mode data.

[0063] Step S4 needs to address the time difference between the axle load value output by the dynamic weighing plate and the emission records in the fusion buffer, and generate complete multi-mode entries by sequential rearrangement to provide an accurate data foundation for subsequent vehicle classification.

[0064] S4.1, The dynamic weighing plate outputs the shaft load value and references the motion vector primary key: To ensure that the axle load values ​​measured when a vehicle passes through the entrance / exit of a target location are accurately correlated with the specific vehicle's trajectory, a dynamic weighing plate needs to match the axle load values ​​with the vehicle's trajectory data after capturing them in real time. The dynamic weighing plate is installed on the ground and contains high-precision sensors. As a vehicle passes, the sensors sequentially detect the weight of each axle, generating an axle load value sequence. This sequence consists of multiple weight values, each corresponding to one axle of the vehicle, measured in tons, and the sequence length is the same as the number of axles on the vehicle. Simultaneously, the dynamic weighing plate records the precise moment each axle load value sequence measurement is completed, known as the axle load measurement time, with millisecond-level accuracy to support subsequent time alignment with emissions data.

[0065] After generating the axle load value sequence, the vehicle trajectory record closest to the axle load value measurement time is retrieved from the temporary trajectory index table. The temporary trajectory index table stores multiple vehicle trajectory records, each containing a unique motion vector primary key and a corresponding trajectory timestamp. The matching process first calculates the time difference between the axle load value measurement time and the trajectory timestamp of each trajectory record. Specifically, the time difference is calculated by subtracting the trajectory timestamp of each trajectory record from the axle load value measurement time and taking the absolute value. Then, all time differences are compared, and the trajectory record corresponding to the smallest time difference is identified as the matching object. The motion vector primary key is extracted from the matched trajectory record; this primary key is a unique identifier used to identify the trajectory of a specific vehicle.

[0066] Subsequently, the extracted motion vector primary key is bound to the axle load value sequence and axle load value measurement time to generate a data pair containing the motion vector primary key, axle load value sequence, and axle load value measurement time. This process, through time matching and primary key referencing, ensures that the axle load value is associated with the correct vehicle's motion trajectory, avoiding data confusion when multiple vehicles pass simultaneously, thus providing accurate identification criteria for multi-mode data integration.

[0067] S4.2, Perform time zone correction: Because the dynamic weighing plate and the remote sensing emission analyzer may have time differences during data acquisition due to equipment response speed or vehicle motion status, the timestamps of emission records and axle load values ​​in the fusion buffer need to be corrected to ensure the temporal consistency of multi-mode data. The fusion buffer stores emission records associated with the motion vector primary key, and each emission record contains emission data and an emission acquisition timestamp.

[0068] First, the emission records with the same motion vector primary key as those extracted in the previous steps are retrieved from the fusion buffer, and the emission acquisition timestamp is extracted from them. Next, the time difference between the axle load measurement time and the emission acquisition timestamp is calculated. Specifically, the emission acquisition timestamp is subtracted from the axle load measurement time to obtain the time difference. This time difference can be positive or negative; a positive value indicates that the axle load measurement time is later than the emission acquisition timestamp, and a negative value indicates that it is earlier than the emission acquisition timestamp.

[0069] To achieve time alignment, a motion vector sequence is used to predict the vehicle's position at the axle load measurement time. The motion vector sequence records the vehicle's speed and direction at different time points. The trajectory point closest in time to the axle load measurement time is selected, and its velocity component and position coordinates are obtained. Then, the time interval from the trajectory point's time to the axle load measurement time is calculated. Specifically, the trajectory point's time is subtracted from the axle load measurement time to obtain the time interval. Next, this time interval is multiplied by the velocity component to obtain the position increment. This position increment is then added to the trajectory point's position coordinates to calculate the predicted position of the vehicle at the axle load measurement time.

[0070] Based on the predicted location, the timestamps of the emission records are adjusted. Specifically, the calculated time difference is added to the emission acquisition timestamp to generate a corrected timestamp, aligning it with the axle load measurement time. The adjusted emission record includes the motion vector primary key, emission data, and the corrected emission acquisition timestamp. This correction process, through time difference calculation and location prediction, eliminates the impact of time misalignment between equipment, ensuring the consistency of emission data and axle load values ​​in time, and providing accurate timing data for multi-mode data integration.

[0071] S4.3, Rearrange the order to generate complete multimodal entries: After time correction, to fully describe the state of vehicles passing through the entrance / exit of the target location, the license plate records associated with the motion vector primary key, the corrected emission records, and the axle load value sequence are integrated into a multi-modal entry and arranged in chronological order to reflect the actual order in which vehicles pass. The multi-modal entry contains the following fields: motion vector primary key, used to identify the vehicle; license plate number, provided by the license plate recognition module; emission data, provided by the remote sensing emission resolver; corrected emission acquisition timestamp; axle load value sequence, provided by the dynamic weighing plate; and axle load measurement time. These fields collectively describe the vehicle's identity, emission status, and weight information.

[0072] To ensure that the multi-modal entries reflect the true order in which vehicles pass through the entrances and exits, all multi-modal entries are sorted in ascending order based on the axle load measurement time. Specifically, multiple multi-modal entries are sorted in ascending order of their respective axle load measurement times, generating an ordered list of multi-modal entries. After sorting, the multi-modal entries are arranged according to the order in which vehicles pass through the entrances and exits, facilitating subsequent data processing and analysis.

[0073] By integrating and rearranging, complete and orderly multimodal entries are generated, which can accurately reflect the multimodal information of vehicles passing through the entrances and exits of target locations, providing structured and consistent data support for vehicle classification, supervision and data analysis.

[0074] After the dynamic weighing plate outputs axle load values ​​in real time, it associates these values ​​with the vehicle's trajectory by referencing the motion vector primary key. Time difference correction eliminates time discrepancies between devices, and then the multi-modal data is integrated through sequential rearrangement to ultimately generate a complete multi-modal entry. From axle load measurement to time alignment, and then to data integration and sorting, each sub-step is tightly linked, forming a complete processing flow. In dynamic traffic scenarios at the entrances and exits of target locations, this technical logic effectively solves the problem of inconsistent data collection times from multiple devices, ensuring the spatiotemporal consistency of multi-modal data and providing accurate and reliable technical support for vehicle monitoring and data analysis.

[0075] Steps S1 to S4, described above, utilize various monitoring methods such as entrance laser macro measurement, license plate recognition, remote sensing emission analysis, and dynamic weighing to generate and integrate multimodal vehicle data, forming complete and ordered multimodal entries. These multimodal entries include the vehicle's motion vector primary key, license plate records, emission records, axle load values, and corresponding timestamps, laying the data foundation for subsequent intelligent classification and reliability assessment. However, in complex traffic scenarios, vehicle dynamic behaviors (such as acceleration, deceleration, or lane changes) and temporal discrepancies and spatial interference during data acquisition may lead to the risk of mixed or incorrect multimodal entries. Step S5, as the core processing step of the fusion engine, aims to extract intersection density factors and scatter symmetry factors, and derive fusion confidence factors using an ensemble classifier to achieve accurate vehicle category updates and effective screening of abnormal entries, ensuring the reliability of classification decisions and traffic efficiency.

[0076] S5.1, the calculation process for obtaining the intersection density factor: To assess the spatial overlap between the target vehicle's trajectory and those of neighboring vehicles, the fusion engine first needs to generate a predicted trajectory for the target vehicle. This predicted trajectory is generated based on the target vehicle's current motion vector sequence. By analyzing the vehicle's position, velocity, and direction changes over a past period and combining this with Kalman filter technology, the engine calculates the set of possible locations the vehicle might traverse within the next few seconds. Specifically, the engine takes the vehicle's current horizontal and vertical coordinates, velocity, and acceleration as input, applies a Kalman filter state prediction step, and calculates the horizontal and vertical coordinates of future locations second by second, forming a continuous predicted path.

[0077] Next, the trajectories of neighboring vehicles that passed through the entrance / exit during the same time period as the target vehicle are extracted from the temporary trajectory index table. The temporary trajectory index table stores the position records of all vehicles within a specific time window. After extraction, the predicted trajectory of the target vehicle is compared with each neighboring vehicle trajectory to determine the spatial overlap, i.e., the intersection segment. The calculation process for the intersection segment is as follows: First, the intersection points of the two trajectory curves are identified. The intersection points are determined by comparing the position coordinates of the corresponding time points on the two trajectories to determine whether there is coordinate overlap or intersection of trajectory segments. Then, for each pair of intersection points, the length of the trajectory segment between them is calculated. Specifically, multiple position points between the intersection points are taken, and the sum of the squares of the differences between the horizontal and vertical coordinates of adjacent position points is calculated for each pair. The square root is taken to obtain the straight-line distance between the two points. Then, the distances between all adjacent points are summed to obtain the length of the intersection segment. Finally, the lengths of all intersection segments are summed to obtain the total length of the intersection segment.

[0078] Subsequently, the total length of the predicted trajectory of the target vehicle is calculated. The calculation method is as follows: take all position points on the predicted trajectory, process adjacent position points one by one, calculate the sum of the squares of the differences between the horizontal and vertical coordinates of each pair of adjacent position points, take the square root to obtain the straight-line distance between the two points, and then add up the distances between all adjacent points in turn to obtain the total length of the predicted trajectory.

[0079] Finally, the intersection density factor is defined as the ratio of the total length of the intersection segment to the total length of the predicted trajectory of the target vehicle. This ratio reflects the density of intersections between the target vehicle's trajectory and the trajectories of neighboring vehicles. The larger the value, the higher the degree of intersection, which facilitates subsequent assessment of traffic congestion and the possibility of data confusion.

[0080] S5.2, the calculation process for obtaining the scatter symmetry factor: To measure the dispersion of timestamps in multi-modal entries, the fusion engine extracts three types of timestamps from the multi-modal entries: license plate record timestamp, emissions record timestamp, and axle load measurement timestamp, collectively referred to as the three-modal timestamps. These timestamps are represented in numerical form, typically in seconds or milliseconds since a certain reference time.

[0081] First, calculate the average of the three timestamps. Specifically, add the timestamps of the license plate record, emissions record, and axle load measurement, then divide the sum by three to obtain an average timestamp representing the center position of the three.

[0082] Next, the dispersion of the three timestamps is calculated. The dispersion is calculated by taking the difference between the license plate record timestamp, emission record timestamp, and axle load measurement timestamp and the average timestamp, squared each difference, summing them, and then dividing the sum by three to obtain a value reflecting the dispersion of the timestamps. This value quantifies the degree to which the timestamps deviate from the average.

[0083] Next, determine the dispersion symmetry factor. The specific steps are as follows: First, find the maximum and minimum values ​​among the three timestamp models; calculate the difference between the maximum value and the average timestamp, and square this difference; then calculate the difference between the minimum value and the average timestamp, and square this difference; finally, divide the square of the difference between the maximum and average timestamps by the square of the difference between the minimum and average timestamps to obtain the dispersion symmetry factor. If the minimum value is exactly equal to the average timestamp, to avoid division by zero, the dispersion symmetry factor is set to a predefined maximum value, such as 1000, indicating a highly asymmetrical timestamp distribution. This factor reflects the uniformity of the timestamp distribution; the larger the value, the higher the dispersion, thus facilitating the assessment of data synchronization.

[0084] S5.3, the process of deriving the fusion confidence factor from the input ensemble classifier: The fusion engine combines the previously calculated intersection density factor and scatter symmetry factor into a feature vector, which contains two values ​​corresponding to the intersection density factor and the scatter symmetry factor, respectively. This feature vector is then fed into a pre-trained gradient boosting ensemble classifier to derive the fusion confidence factor.

[0085] The construction process of the gradient boosting ensemble classifier is as follows. Initially, a decision tree is selected as the base learner. Each decision tree partitions the data and generates prediction results based on the input features, while log loss is used as a metric to measure the deviation between the prediction results and the true values. The training process is iterative. In each iteration, the difference between the predicted values ​​and the true values ​​on the training data of the current model is first calculated, i.e., the residual. Then, a new decision tree is trained to learn and fit these residuals. Next, the prediction results of the new decision tree are added to the current model according to certain weights, which are controlled by the learning rate, usually a positive number less than 1, to gradually optimize the model. Parameters such as the number of iterations, the maximum depth of the decision tree, and the learning rate are determined by adjustment to ensure that the model's performance on the training data and its generalization ability on new data are balanced. During training, a K-fold cross-validation method is employed. The training data is divided into K subsets, and the model is trained sequentially using K-1 subsets, with the remaining subset used for validation. This process is repeated K times, and the average value is used to evaluate performance. Simultaneously, an early stopping mechanism monitors the loss on the validation set. If the loss no longer decreases within several iterations, training is stopped to prevent the model from overfitting the training data. Furthermore, the contribution of each feature to all decision trees is calculated. For example, the relative importance of the intersection density factor and the scatter symmetry factor is evaluated by statistically analyzing the frequency of feature usage when splitting nodes, thus optimizing feature selection. Finally, prediction results are generated on an independent test set, and the number of correct and incorrect classifications is counted. Curves are plotted to evaluate the model's ability to distinguish between positive and negative samples, confirming the model's reliability.

[0086] After training, the fusion engine inputs the feature vectors containing the intersection density factor and the scatter symmetry factor into the gradient boosting ensemble classifier. The classifier outputs a value between 0 and 1, called the fusion confidence factor. This value represents the probability of the reliability of the multi-mode entries, providing a quantitative basis for subsequent decisions.

[0087] S5.4, Decision process for updating vehicle category labels or transferring to review: The fusion engine compares the fusion confidence factor output by the gradient boosting ensemble classifier with a preset threshold, for example, a threshold set to 0.8, to distinguish the reliability of multi-mode entries.

[0088] If the fusion confidence factor value is greater than a preset threshold, the data in the multi-modal entry is deemed to have high reliability. The fusion engine processes the license plate records, emission records, and axle load measurement data contained in the multi-modal entry according to predefined classification rules. For example, by querying the axle load value range to determine the vehicle weight class, and combining the emission records to determine the emission standard, the vehicle category is finally determined, such as a sedan or a truck. The determined vehicle category information is then sent to the gate control terminal for the release operation.

[0089] If the fusion confidence factor is less than or equal to a preset threshold, the data in the multi-modal entry is determined to be abnormal or uncertain. This multi-modal entry is then transferred to the review buffer and simultaneously recorded and archived in the regulatory database, awaiting further manual verification and processing.

[0090] Through the above decision-making process, the fusion engine can flexibly distinguish between reliable and abnormal data based on the value of the fusion confidence factor, ensuring the accuracy of vehicle classification, while providing a verification mechanism for abnormal situations, thereby improving the overall robustness of processing and traffic efficiency.

[0091] The above steps, through the calculation of intersection density factor and scatter symmetry factor, combined with gradient enhancement to improve the predictive ability of the ensemble classifier, ultimately achieve automatic updating of vehicle categories or screening of abnormal data. The entire process, from feature extraction to classification decision and then to anomaly handling, is interconnected, effectively addressing the complexities of dynamic traffic scenarios and ensuring the reliability of gate control.

[0092] Example 2: Figure 2 The present invention provides a system for obtaining category information of entrances and exits of target locations, comprising: The entrance laser macro measurement array continuously extracts the vehicle motion vector and writes the motion vector primary key into a temporary trajectory index table; After capturing a character frame, the license plate recognition module calls the motion vector primary key, maps the primary key to the license plate record, and synchronously inserts a sequence timestamp. When the remote sensing emission parser completes the parsing, it retrieves the primary key of the motion vector, predicts the arrival delay based on the motion vector, rewrites the emission record, and pushes it to the fusion buffer. After the dynamic weighing plate outputs the axle load value in real time, it references the same motion vector primary key and performs time difference correction and order rearrangement with the emission records in the fusion buffer to generate a complete multi-mode entry. The fusion engine obtains the intersection density factor and the scattering symmetry factor from the multi-modal entries, and derives the fusion confidence factor through machine learning. If the fusion confidence factor is higher than the preset threshold, the vehicle category label is updated and sent to the gate control terminal; otherwise, the entry is transferred to the review buffer and archived to the regulatory database.

[0093] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0094] It should be noted that the system of the present invention can be deployed on the device itself to realize embedded applications, or it can run on a PC or other terminal with a user interface, thereby meeting a variety of hardware environments and usage requirements.

[0095] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0096] It should be noted that, in this document, the use of relational terms such as "first" and "second" is merely to distinguish one entity or operation from another, and does not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0097] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for obtaining category information of entrances and exits of a target location, characterized in that, Including the following steps: S1: The entrance laser micro-measurement array continuously extracts the vehicle motion vector and writes the motion vector primary key into the temporary trajectory index table; S2: After capturing the character frame, the license plate recognition module calls the motion vector primary key, maps the primary key to the license plate record, and synchronously inserts the sequence timestamp; S3: When the remote sensing emission resolver completes the resolution, it retrieves the primary key of the motion vector, predicts the arrival delay based on the motion vector, rewrites the emission record, and pushes it to the fusion buffer. S4: After the dynamic weighing plate outputs the axle load value in real time, it references the same motion vector primary key and performs time difference correction and order rearrangement with the emission records in the fusion buffer to generate a complete multi-mode entry. S5: The fusion engine obtains the intersection density factor and the scattering symmetry factor from the multi-modal entries, and derives the fusion confidence factor through machine learning. If the fusion confidence factor is higher than the preset threshold, the vehicle category label is updated and sent to the gate control terminal; otherwise, the entry is transferred to the review buffer and archived to the regulatory database.

2. The method for obtaining category information of entrances and exits of a target location according to claim 1, characterized in that, Step S1 includes the following: Multiple high-precision laser sensors scan vehicles in real time to capture their position and speed information and generate motion vectors. A unique motion vector key is generated for each vehicle, based on the initial time and position of the vehicle entering the detection area, and its uniqueness is ensured by a hash function. The motion vector primary key and the corresponding trajectory data are written into a temporary trajectory index table. The temporary trajectory index table contains the motion vector primary key, timestamp, location coordinates and motion vector fields, and is dynamically updated as the vehicle moves.

3. The method for obtaining category information of entrances and exits of a target location according to claim 2, characterized in that, Step S2 includes the following: After capturing character frames and recognizing license plate numbers, the vehicle trajectory entry closest to the capture time is retrieved from the temporary trajectory index table to obtain the corresponding motion vector primary key. Then, the motion vector primary key is paired with the license plate number to generate an associated record. At the same time, a sequence timestamp is generated to record the time when recognition is completed. Finally, the motion vector primary key, license plate number, and sequence timestamp are integrated into a license plate record and stored in the license plate database.

4. The method for obtaining category information of entrances and exits of a target location according to claim 3, characterized in that, Step S3 includes the following: After completing the analysis, the remote sensing emission analyzer records the analysis completion time and retrieves the vehicle trajectory entry closest to the analysis completion time from the temporary trajectory index table, extracting the motion vector primary key; it predicts the vehicle's position at the analysis completion time based on the motion vector sequence and calculates the arrival delay from the emission collection point to the predicted position.

5. The method for obtaining category information of entrances and exits of a target location according to claim 4, characterized in that, Step S3 also Includes the following: Subtracting the arrival delay from the parsing completion time yields the corrected emission collection timestamp. The motion vector primary key, emission data, and emission collection timestamp are then integrated into an emission record and pushed to the fusion buffer.

6. The method for obtaining category information of entrances and exits of a target location according to claim 5, characterized in that, Step S4 includes the following: A dynamic weighing plate measures the axle load of a vehicle and records the measurement time. It retrieves the motion trajectory record closest to the measurement time from a temporary trajectory index table and extracts the motion vector primary key. Based on the motion vector primary key, it retrieves emission records from the fusion buffer, calculates the time difference between the axle load measurement time and the emission collection timestamp, uses the motion vector sequence to predict the vehicle's position at the measurement time, and adjusts the emission record timestamp to achieve time alignment. The motion vector primary key, license plate number, corrected emission data, axle load sequence, and measurement time are integrated into a multi-modal entry and arranged in ascending order of measurement time to generate an ordered multi-modal entry list.

7. The method for obtaining category information of entrances and exits of a target location according to claim 6, characterized in that, Step S5 includes the following: The intersection density factor and the scatter symmetry factor are extracted from the multi-modal entries. The intersection density factor and the scatter symmetry factor are input into the pre-trained gradient boosting ensemble classifier to output the fusion confidence factor. The fusion engine decides to update the vehicle category label or transfer the multi-modal entries to the verification buffer based on the comparison between the fusion confidence factor and the preset threshold.

8. The method for obtaining category information of entrances and exits of a target location according to claim 7, characterized in that, Step S5 also includes the following: The intersection density factor is obtained by dividing the total length of the intersection segment between the predicted trajectory and the adjacent trajectory by the length of the target trajectory, while the scattering symmetry factor is obtained by calculating the dispersion of the multi-mode timestamps and the ratio of the maximum to the minimum deviation.

9. The method for obtaining category information of entrances and exits of a target location according to claim 8, characterized in that, Step S5 also includes the following: When the fusion confidence factor is higher than the preset threshold, the fusion engine determines the vehicle category based on the license plate record, emission record and axle load measurement data in the multi-mode entry and issues it to the gate control terminal. When the fusion confidence factor is not higher than the preset threshold, the fusion engine transfers the multi-mode entry to the review buffer and archives it to the regulatory database.

10. A system for obtaining category information of entrances and exits of a target location, used to implement the method for obtaining category information of entrances and exits of a target location as described in any one of claims 1-9, characterized in that, include: The entrance laser macro measurement array continuously extracts the vehicle motion vector and writes the motion vector primary key into a temporary trajectory index table; After capturing a character frame, the license plate recognition module calls the motion vector primary key, maps the primary key to the license plate record, and synchronously inserts a sequence timestamp. When the remote sensing emission parser completes the parsing, it retrieves the primary key of the motion vector, predicts the arrival delay based on the motion vector, rewrites the emission record, and pushes it to the fusion buffer. After the dynamic weighing plate outputs the axle load value in real time, it references the same motion vector primary key and performs time difference correction and order rearrangement with the emission records in the fusion buffer to generate a complete multi-mode entry. The fusion engine obtains the intersection density factor and the scattering symmetry factor from the multi-modal entries, and derives the fusion confidence factor through machine learning. If the fusion confidence factor is higher than the preset threshold, the vehicle category label is updated and sent to the gate control terminal; otherwise, the entry is transferred to the review buffer and archived to the regulatory database.

Images