Method for automatically identifying, tracking and positioning vehicles in railway station

Abstract

The invention discloses a method for automatically identifying, tracking and positioning a vehicle in a railway station, relates to the technical field of railway transportation management, and solves the problems that a three-level judgment mechanism is difficult to construct to identify the vehicle in combination with multi-sensor data, and automatic conversion from coordinates to tags is difficult to realize through a digital twin model. The technical problem of lack of prediction variance-based weight adjustment and joint loss function for balancing the multi-task training effect is solved; comprising the steps that multi-source data are collected by constructing a comprehensive monitoring system, and automatic vehicle recognition is achieved in combination with a three-level judgment mechanism; calculating vehicle position information by using a Kalman filtering algorithm and a railway yard special coordinate system, and mapping coordinates into semantic position tags through a digital twin model; a multi-task neural network framework is constructed based on historical data and semantic tags, the weight of a loss function is dynamically adjusted to optimize a joint loss function, and efficient semantic prediction and classification of vehicle positions are achieved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of railway transportation management, and particularly relates to a method for automatic identification and tracking positioning of vehicles in a railway station yard. BACKGROUND

[0002] With the rapid development of the railway transportation industry, the vehicle flow in the railway station yard is increasing, and the traditional manual identification and positioning method has been difficult to meet the efficient and accurate operation requirements of modern railway transportation. In the complex station yard environment, accurately and quickly identifying vehicle information and real-time mastering the vehicle position and driving trajectory are crucial for optimizing station yard scheduling, improving operation efficiency and ensuring transportation safety.

[0003] However, although some existing vehicle identification and positioning technologies achieve vehicle identification and positioning to some extent, they have the following problems: relying on a single sensor, resulting in a decrease in identification accuracy; lacking a special coordinate system and a multi-sensor fusion algorithm, resulting in a large position calculation error, and a semantic location label requiring manual intervention, which is inefficient; and not considering the correlation between tasks, resulting in model parameter redundancy, resource waste, and difficulty in dynamically adapting to data distribution changes.

[0004] Therefore, there is an urgent need for a vehicle identification and tracking positioning method that fuses multi-source data, has semantic mapping capability and supports dynamic optimization. SUMMARY

[0005] The present application aims to at least solve one of the technical problems in the prior art; for this purpose, the present application proposes a method for automatic identification and tracking positioning of vehicles in a railway station yard, which is used to solve the following technical problems:

[0006] Relying on a single sensor, resulting in a decrease in identification accuracy; lacking a special coordinate system and a multi-sensor fusion algorithm, resulting in a large position calculation error, and a semantic location label requiring manual intervention, which is inefficient; and not considering the correlation between tasks, resulting in model parameter redundancy, resource waste, and difficulty in dynamically adapting to data distribution changes.

[0007] To solve the above problems, the present application provides a method for automatic identification and tracking positioning of vehicles in a railway station yard, comprising the following steps:

[0008] S1: constructing an internal comprehensive monitoring system of the railway station yard for collecting various types of data and performing synchronous fusion processing thereon;

[0009] S2: setting a three-level judgment mechanism to calculate first, second and third matching degree scores, performing weighted addition to obtain a comprehensive matching degree score, and performing vehicle automatic identification on the comparison result of the comprehensive matching degree score and a comprehensive matching degree threshold value;

[0010] S3: After the vehicle automatically identifies successfully, a tracking database is constructed, a railway track coordinate system is established, and Kalman filtering algorithm is used to integrate multi-source data to obtain two-dimensional plane coordinates and height information of the vehicle in the railway station coordinate system;

[0011] S4: The vehicle coordinates are dynamically obtained by using the RFID reader and high-definition industrial camera distributed in the railway station, and are synchronized to the tracking database. The vehicle coordinates are mapped to the function area and associated track number marked by artificial labeling through the digital twin model of the railway station, and the complete semantic information is generated by combining the state label;

[0012] S5: A multi-task neural network framework is constructed by historical data and semantic location labels, and a joint optimization loss function is defined by dynamic weight adjustment and joint loss optimization.

[0013] Preferably, the internal comprehensive monitoring system of the railway station comprises a high-definition industrial camera, a laser sensor, a radar, an RFID tag and a reader / writer, and a Beidou positioning terminal.

[0014] Preferably, the three-level judgment mechanism in step S2 comprises the following steps:

[0015] The data collected by the internal comprehensive monitoring system of the railway station are preprocessed respectively;

[0016] A three-level judgment mechanism is set, specifically:

[0017] The vehicle visual features including the color of the vehicle body, the identification pattern, the number of windows, and the shape of the vehicle head are extracted, matched with the pre-established vehicle feature database, and the first visual judgment is made according to the matching result;

[0018] The vehicle contour, size and distance from the track measured by the laser sensor are matched with the standard vehicle parameter library, and the second physical judgment is made according to the matching result;

[0019] The completeness and shielding degree of the vehicle label information read by the RFID reader are used for the third information judgment;

[0020] The visual judgment, physical judgment and RFID information are weighted and fused to obtain a comprehensive matching degree score.

[0021] Preferably, the weighted fusion of the visual judgment, physical judgment and RFID information to obtain a comprehensive matching degree score comprises the following steps:

[0022] The extracted vehicle visual features are matched with the pre-set vehicle feature database to calculate the first matching degree score; the physical parameters measured by the laser sensor are matched with the standard vehicle parameter library to calculate the second matching degree score; and the label information read by the RFID is used as a reference to calculate the third matching degree score according to the label completeness and shielding degree.

[0023] According to the historical data, weights are assigned to each data source, and the comprehensive matching degree score is obtained by weighted addition, specifically:

[0024]

[0025] Wherein, G is the comprehensive matching degree score, G1, G2 and G3 are the first matching degree score, the second matching degree score and the third matching degree score respectively, and are the corresponding weights respectively;

[0026] Based on the pre-set confidence, the confidence interval of the comprehensive matching degree score is determined through historical training data, and the comprehensive matching degree threshold is defined, specifically:

[0027] G T =μ-k c ·σ

[0028] Wherein G T is the comprehensive matching degree threshold, μ is the mean of the historical comprehensive matching degree score, σ is the standard deviation, and k c is the confidence coefficient;

[0029] If G>G T , the vehicle identification result is output; otherwise, it is determined that there is a matching conflict, and a matching inconsistency processing procedure is triggered.

[0030] Preferably, the third matching degree score is calculated according to the label completeness and the occlusion degree, comprising the following steps:

[0031] The vehicle label information including the vehicle number and the vehicle type key field is read by the RFID reader, and at the same time, the completeness of the label and the occlusion degree are obtained;

[0032] The completeness is determined by judging whether the label is complete and intact; the occlusion degree is determined by calculating the proportion of the label being occluded by an object;

[0033] According to the label completeness and the occlusion degree, the label state is divided into the following levels, and different weights are assigned according to the label quality level, specifically:

[0034] First-class quality: completeness ≥ 95%, occlusion degree ≤ 5%, corresponding weight: 1 is set;

[0035] Second-class quality: completeness ≥ 70%, occlusion degree ≤ 30%, corresponding weight: 0.7 is set;

[0036] Third-class quality: completeness < 70%, occlusion degree > 30%, corresponding weight: 0.4 is set;

[0037] According to the label quality level judgment corresponding weight, multiplied by 100 to get the third matching degree score.

[0038] Preferably, the trigger matching inconsistency processing flow comprises the following steps:

[0039] S201: Data review, including reviewing the data collected by the comprehensive monitoring system inside the railway station, checking whether there is data collection error or omission, at the same time, starting the standby monitoring equipment to obtain vehicle information;

[0040] S202: Cross verification, including matching process of three-level judgment mechanism for the reviewed data and newly collected data again;

[0041] S203: Artificial auxiliary judgment, including introducing artificial auxiliary judgment link for final vehicle identification if there is still matching inconsistency after data review and cross verification.

[0042] Preferably, the step S3 comprises the following steps:

[0043] After the vehicle is successfully identified, a tracking database is constructed for it;

[0044] The tracking database records the identification information and semantic location label of the vehicle;

[0045] A railway station coordinate system is established with the track starting point as the origin and the track direction as the reference axis;

[0046] The two-dimensional plane coordinates and height information of the vehicle in the railway station coordinate system are calculated by fusing multi-sensor data through Kalman filtering algorithm.

[0047] Preferably, the step S4 comprises the following steps:

[0048] The vehicle position coordinates are updated in real time by the RFID reader and high-definition industrial camera distributed in the railway station;

[0049] The position coordinates are mapped into functional area label, state label and associated label in real time through data twin model, and are synchronized to the tracking database;

[0050] The tracking positioning system updates the position coordinates of the vehicle in the tracking file according to the received position, and after updating the vehicle coordinates, the coordinates are mapped into functional area label and associated label by real-time querying the digital twin model of the railway station, and are stored in the tracking database synchronously with the state label;

[0051] The railway station digital twin model is obtained by converting a railway station CAD drawing into a three-dimensional grid model, manually labeling the function areas of each grid, and matching a preset label according to a grid index where a vehicle coordinate is located.

[0052] Preferably, the step S5 of constructing the multi-task neural network framework through the historical data and the semantic location label includes the following steps:

[0053] The vehicle data with the location coordinates is collected from the historical operation data of the railway station and the manually labeled semantic location label, and the semantic location label includes a function area label, a state label and an association label.

[0054] The neural network is used to extract the location coordinates, the time stamp, the sensor original data and the external data features as input features, and the neural network automatically learns the inter-task association through shared bottom layer parameters.

[0055] The input features are processed through a shared preprocessing layer, including normalizing the coordinate values, extracting the time features and encoding the classification features.

[0056] A multi-task branch structure is used, including a function area label branch, a state label branch and an association label branch.

[0057] The function area label branch: inputting the coordinates, the time and the vehicle type features, and outputting the grid function classification.

[0058] The state label branch: inputting the speed, the historical trajectory and the location features, and outputting the state label.

[0059] The association label branch: inputting the coordinates and the track state, and outputting the current track number.

[0060] Preferably, the step S5 of defining the joint optimization loss function includes the following steps:

[0061] The loss function weight is dynamically adjusted according to the prediction variance of each task, and specifically,

[0062]

[0063] where ω f is the weight coefficient corresponding to the function area label, σ f is the variance of the prediction result of the function area label, σ s is the variance of the prediction result of the state area label, and σ a is the variance of the prediction result of the association label.

[0064] The same method is used to calculate the weight coefficients corresponding to the state area label and the association label, respectively.

[0065] The label corresponding weight coefficient is added to the label corresponding loss function to construct a joint optimization loss function;

[0066] The joint optimization loss function is specifically:

[0067] L=ω f ·L f +ω s ·L s +ω a ·L a

[0068] Wherein, L is a joint optimization loss function, ω f , ω s and ω a are label corresponding weight coefficients, L f , L s and L a are label corresponding loss functions, wherein L f is a focal loss function, L s and L a are cross-entropy loss functions.

[0069] The beneficial effects of the present application are:

[0070] The present application fuses multi-source data through a three-level judgment mechanism (vision, physics, RFID), improves vehicle recognition accuracy, and realizes accurate calculation of vehicle two-dimensional plane coordinates and height information based on Kalman filtering algorithm and special coordinate system;

[0071] The present application maps vehicle coordinates into functional areas, states and associated track labels by using a digital twin model, reduces manual annotation cost, improves label generation efficiency, automatically learns task association by combining multi-task neural networks, and realizes end-to-end semantic prediction of "position-function-state";

[0072] The present application updates vehicle coordinates in real time by RFID and high-definition cameras, synchronously stores semantic labels in a tracking database, supports fast query and track backtracking, is suitable for complex railway station environments, improves the intelligent management level of railway stations, and dynamically adjusts loss function weights according to prediction variance, thereby enhancing the robustness of the model to difficult classification samples. BRIEF DESCRIPTION OF DRAWINGS

[0073] Figure 1 The present application is a method flowchart;

[0074] Figure 2 The present application is a railway station internal comprehensive monitoring system module flowchart;

[0075] Figure 3 The present application is a three-level judgment mechanism method flowchart. DETAILED DESCRIPTION

[0076] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work are within the protection scope of the present application.

[0077] Please refer to Figure 1 The present application is a method for automatic identification and tracking positioning of vehicles in a railway yard, comprising the following steps:

[0078] S1: Construct an internal comprehensive monitoring system of the railway yard for collecting various types of data and performing synchronous fusion processing thereon;

[0079] S2: Set up a three-level judgment mechanism to calculate the first, second and third matching degree scores, add them after weighting to obtain a comprehensive matching degree score, and perform vehicle automatic identification on the comparison result of the comprehensive matching degree score and a comprehensive matching degree threshold value;

[0080] S3: After successful vehicle automatic identification, construct a tracking database and establish a railway track coordinate system, and use Kalman filtering algorithm to integrate multi-source data to obtain two-dimensional plane coordinates and height information of the vehicle in the railway yard coordinate system;

[0081] S4: Use the RFID reader and high-definition industrial camera distributed in the railway yard to dynamically obtain vehicle coordinates and synchronize them to the tracking database, map the vehicle coordinates to the function areas and associated track numbers manually labeled by the railway yard digital twin model, and generate complete semantic information in combination with the state label;

[0082] S5: Construct a multi-task neural network framework through historical data and semantic location labels, and define a joint optimization loss function through dynamic weight adjustment and joint loss optimization.

[0083] In one embodiment of the present application, the internal comprehensive monitoring system of the railway yard comprises a high-definition industrial camera, a laser sensor, a radar, an RFID tag and reader, and a Beidou positioning terminal.

[0084] Specifically, high-definition industrial cameras and light supplementing devices are installed at a high position on both sides of the tracks in the railway station (from a bird's eye view), covering multiple tracks and avoiding lens obstruction, to collect images of the vehicle head and carriages entering and leaving the station as well as marker information; laser sensors are installed on both sides or the top of the tracks to synchronously trigger the high-definition industrial cameras, for detecting vehicle signals, speed and distance information; radars are installed at the ends of the tracks or at bends for long-distance vehicle detection; RFID tags are pasted on the bottom of the vehicles to store unique identifiers such as vehicle number and vehicle type; corresponding RFID readers are arranged at key nodes of the tracks (such as the entrance to the station and the shunting area); and Beidou positioning terminals are arranged on some vehicles or fixed base stations; NTP or PTP is used to align the clocks of all devices to ensure that the error of the data timestamp is less than or equal to 10 ms.

[0085] In one embodiment of the present application, the three-level judgment mechanism in step S2 includes the following steps:

[0086] The data collected by the comprehensive monitoring system inside the railway station is preprocessed respectively;

[0087] A three-level judgment mechanism is set, specifically as follows:

[0088] The visual features of the vehicle, including the color of the vehicle body, the identification pattern, the number of windows and the shape of the vehicle head, are extracted and matched with the pre-established vehicle feature database, and the first visual judgment is made according to the matching result;

[0089] The vehicle contour, size and distance from the track measured by the laser sensor are matched with the standard vehicle parameter library, and the second physical judgment is made according to the matching result;

[0090] The completeness and obstruction degree of the vehicle tag information read by the RFID reader are used for the third information judgment;

[0091] The visual judgment, physical judgment and RFID information are weighted and fused to obtain a comprehensive matching degree score.

[0092] Specifically, the visual data collected by the high-definition industrial camera is subjected to image denoising, grayscale, edge detection, extraction of vehicle body color (HSV color space analysis), identification pattern (template matching or feature point recognition), number of vehicle windows (morphological analysis), and vehicle head shape (contour comparison); the laser sensor data is subjected to point cloud filtering and extraction of vehicle contour, calculation of vehicle external dimensions (length, width, and height), centroid position, and distance from the track; decoding of vehicle label information, including vehicle number, vehicle type, and other key fields, evaluation of label integrity (whether missing or damaged) and occlusion degree (area ratio of occlusion); coordinate conversion and timestamp alignment of Beidou positioning data; a three-level judgment mechanism is set, wherein the first level is visual feature matching, the pre-processed visual features (color, identification, vehicle window, and vehicle head shape) are input, the extracted visual features are compared with the pre-set vehicle feature database, and the first matching degree score is calculated; the second level is physical parameter matching, the vehicle contour, size, and distance from the track measured by the laser sensor are input, the vehicle external shape parameters are matched with the standard vehicle parameter library, whether the distance between the vehicle and the track conforms to the standard safety range is calculated, and the second matching degree score is generated; the third level is RFID information judgment, the label information (vehicle number, vehicle type) read by the RFID is input, the integrity and occlusion degree are divided into quality grades, and the third matching degree score is calculated; wherein the pre-set vehicle feature database collects vehicle type samples, covers all common vehicle types in the station (such as freight locomotives, passenger cars, and switching locomotives), including images under different angles, lighting, and weather conditions, the high-definition industrial camera is used to shoot the key parts of the vehicle, such as the front, side, and head, to ensure the completeness of the features, the vehicle body color, identification pattern, number of vehicle windows, and head shape are labeled, the image processing software is used to extract the features and store them as structured data, and a standardized process is performed, including conversion of color to HSV color space, statistics of the proportion of dominant color, identification of specific identification through template matching or SIFT feature points, statistics of the number of vehicle window regions based on morphological analysis (contour detection), and extraction of the head contour through edge detection, comparison with the pre-defined shape library (trapezoidal, streamlined, etc.), periodic supplement of newly collected image features, and elimination of old vehicle model data; the standard vehicle parameter library obtains the vehicle design manual from the vehicle manufacturer, extracts the standard dimensions, centroid position, wheelbase, and other parameters, uses the laser sensor or three-dimensional measuring instrument to scan the actual vehicle, obtains the external contour point cloud data, calculates the actual dimensions through point cloud segmentation and fitting algorithm, and compares and verifies with the design value, the standardized process includes taking the maximum vehicle external contour as the standard, eliminating the telescopic components, calculating the vehicle and track safety distance range according to the design document or through laser point cloud centroid, and periodically re-measuring the dimensions of the in-service vehicles and updating the parameters of the aging or modified vehicle models in the database.

[0093] In one embodiment of the present application, the comprehensive matching degree score obtained by weighting and fusing the visual judgment, physical judgment and RFID information comprises the following steps:

[0094] The extracted vehicle visual features are matched with a preset vehicle feature database to calculate a first matching degree score; the physical parameters measured by the laser sensor are matched with a standard vehicle parameter library to calculate a second matching degree score; and the label information read by the RFID is taken as a reference to calculate a third matching degree score according to the label completeness and the shielding degree;

[0095] According to historical data, weights are assigned to each data source, and the comprehensive matching degree score is obtained by weighted addition, specifically as follows:

[0096]

[0097] wherein G is the comprehensive matching degree score, G1, G2 and G3 are the first matching degree score, the second matching degree score and the third matching degree score respectively, and are the corresponding weights respectively;

[0098] Based on the preset confidence, the confidence interval of the comprehensive matching degree score is determined through historical training data, and the comprehensive matching degree threshold is defined, specifically as follows:

[0099] G T =μ-k c ·σ

[0100] wherein G T is the comprehensive matching degree threshold, μ is the mean of the historical comprehensive matching degree score, σ is the standard deviation, and k c is the confidence coefficient;

[0101] If G > G T , the vehicle recognition result is output; otherwise, it is determined that there is a matching conflict, and a matching inconsistency processing procedure is triggered.

[0102] Specifically, the first matching degree score is calculated by weighting the matching degree of the visual features, including the vehicle body color, the identification pattern, the number of windows and the shape of the vehicle head, with the vehicle feature database, specifically as follows:

[0103]

[0104] wherein G1 is the first matching degree score, α i is the weight of the i-th feature, and sim i is the similarity of the i-th feature, wherein the values of α1, α2, α3 and α4 are 0.2, 0.3, 0.2 and 0.3 respectively;

[0105] The second matching degree score is obtained by matching the vehicle shape parameters (length, width, height, and centroid position) measured by the laser sensor with the standard vehicle parameter library, and the matching degree score is weighted and calculated, and the same method as the first matching degree is used to calculate the matching degree score, and the weights of the length, width, height, and centroid position are 0.4, 0.3, 0.2, and 0.1, respectively.

[0106] and The weights of the first, second, and third matching degree scores are 0.5, 0.3, and 0.2, respectively.

[0107] In one embodiment of the present application, the third matching degree score is calculated according to the label integrity and the blocking degree, comprising the following steps:

[0108] The vehicle label information is read by the RFID reader, including the vehicle number and the vehicle type key field, and the integrity and the blocking degree of the label are obtained at the same time;

[0109] The integrity is determined by judging whether the label is complete and intact, and the blocking degree is calculated by calculating the proportion of the label blocked by the object.

[0110] According to the label integrity and the blocking degree, the label state is divided into the following levels, and different weights are given according to the label quality level, specifically:

[0111] First-class quality: integrity ≥ 95%, blocking degree ≤ 5%, corresponding weight is set to 1;

[0112] Second-class quality: integrity ≥ 70%, blocking degree ≤ 30%, corresponding weight is set to 0.7;

[0113] Third-class quality: integrity < 70%, blocking degree > 30%, corresponding weight is set to 0.4;

[0114] According to the label quality level, the corresponding weight is multiplied by 100 to obtain the third matching degree score.

[0115] In one embodiment of the present application, the matching inconsistency processing flow is triggered, comprising the following steps:

[0116] S201: data review, including reviewing the data collected by the railway station internal comprehensive monitoring system, checking whether there is data collection error or omission, and starting the standby monitoring equipment to obtain vehicle information;

[0117] S202: cross verification, including the matching process of the three-level judgment mechanism for the reviewed data and the newly collected data again;

[0118] S203: Artificially assisted determination, including, if there are still matching inconsistencies after data review and cross-validation, introducing an artificially assisted determination link to conduct the final vehicle identification.

[0119] In one embodiment of the present application, the step S3 comprises the following steps:

[0120] After successful identification of the vehicle, a tracking database is constructed for it;

[0121] The tracking database records the identification information and semantic location label of the vehicle;

[0122] A railway station coordinate system is established using a railway station special coordinate system, with the track starting point as the origin and the track direction as the reference axis;

[0123] The two-dimensional plane coordinates and height information of the vehicle in the railway station coordinate system are calculated by fusing multi-sensor data through Kalman filtering algorithm.

[0124] Specifically, the tracking database includes vehicle ID, identification time, semantic location label, two-dimensional plane coordinates, height, speed, direction angle and confidence, and a new record is inserted each time the vehicle is successfully identified and the coordinates are updated, and if the old record timestamp of the same vehicle ID exceeds the threshold (such as 10 minutes), it is updated, and the "vehicle ID" and "identification time" are indexed to speed up the historical trajectory query; wherein the track starting point is the intersection of the station signal building center line and the track, and the local coordinate system of the laser radar, camera and other sensors is converted into the railway station coordinate system, specifically: P1=T*P2+R, wherein P1 is the railway station coordinate system, P2 is the local coordinate system of the sensor, T is the rotation matrix calculated according to the installation angle, and R is the translation vector; multi-sensor data including laser radar data, high-definition industrial camera data and track encoder data are fused through Kalman filtering algorithm.

[0125] In one embodiment of the present application, the step S4 comprises the following steps:

[0126] The vehicle position coordinates are updated in real time by using the RFID reader and high-definition industrial camera distributed in the railway station;

[0127] The position coordinates are mapped into functional area label, state label and association label in real time through the data twin model, and are synchronized to the tracking database;

[0128] The tracking positioning system updates the position coordinates of the vehicle in the tracking file according to the received position, and after updating the vehicle coordinates, the coordinates are mapped into functional area label and association label in real time by querying the railway station digital twin model, and are stored synchronously to the tracking database with the state label;

[0129] The railway station digital twin model is obtained by converting a railway station CAD drawing into a three-dimensional grid model, manually labeling the function areas of each grid, and matching a preset label according to a grid index in which a vehicle coordinate is located.

[0130] Specifically, the RFID reader and the industrial camera respectively send vehicle position data to the edge computing node, the edge node fuses multi-source data, fuses RFID, camera and track encoder data through Kalman filtering algorithm, outputs vehicle position coordinates, and the data twin model mapping includes locating the grid index in the three-dimensional grid model according to the vehicle position coordinates (x, y), querying the rule base to return the function label, the state label generation includes marking “overstay” if the speed v≤0.5m / s and lasts >60 seconds, marking “path deviation” if the trajectory deviates from the planned path by >10 meters, the associated coordinate generation includes querying the train dispatching system, matching the task corresponding to the current vehicle ID, and associating the track occupancy state; the station CAD drawing is imported into the BIM software to generate a three-dimensional grid model, the grid accuracy is divided according to the complexity of the function area, for example, the grid of the turnout area ≤1m×1m, and the grid of the track ≤5m×5m, each grid is given a function label and an associated attribute, the grid in which the vehicle position coordinates (x, y) are located is queried to return the preset label.

[0131] In one embodiment of the present application, the step S5 of constructing a multi-task neural network framework through historical data and semantic location labels includes the following steps:

[0132] Collect vehicle data with position coordinates from historical operation data of the railway station and manually labeled semantic location labels, the semantic location labels including function area labels, state labels and associated labels;

[0133] Use a neural network to extract position coordinates, timestamps, sensor raw data and external data features as input features, the neural network automatically learns the inter-task association through shared bottom layer parameters;

[0134] The input features are processed through a shared preprocessing layer, including standardizing coordinate values, extracting time features and encoding classification features;

[0135] Use a multi-task branch structure, including a function area label branch, a state label branch and an associated label branch;

[0136] The function area label branch: input coordinates, time and vehicle type features, and output grid function classification;

[0137] The state label branch: input speed, historical trajectory and position features, and output state labels;

[0138] The association label branch: input coordinates and track state, output current track number.

[0139] Specifically, the functional area label, the state label and the association label are collected from the tracking database, the sensor raw data includes RFID trigger record, high-definition industrial camera track coordinates, speed sensor data and track encoder state, the external data includes train scheduling plan and signal state; the data is cleaned, including removing duplicate records and filling missing values, the position coordinates, the timestamp, the sensor raw data and the external data features including position features, time features, sensor features, vehicle type features, historical track features and external features are extracted, the semantic position label includes the functional area label, the state label and the association label, wherein the functional area label directly uses the grid functional label generated by the digital twin model, the state label is defined according to the business rules, for example, static is 0, moving is 1 and overspeed is 2; the association label is the track number; a neural network model is constructed, the shared preprocessing layer design includes that the input layer receives all input features, the input dimension is 11 dimensions, [coordinate x, coordinate y, hour, minute, second, speed, acceleration, direction angle, vehicle type code, track entropy, track state code], the shared fully connected layer outputs 128-dimensional hidden layer representation, which is used to extract general features across tasks; the multi-task branch structure includes the functional area label branch, the state label branch and the association label branch; the loss function of the functional area label branch and the association label branch is the cross-entropy loss function, the loss function of the state label branch is the focal loss function, the optimizer is Adam, and lr is 0.001; the trained model is quantized and deployed to the station edge device, the input includes the coordinates, the speed, the time and other features of the vehicle tracking database received in real time, and the output includes the prediction results and the confidence of the three labels; the prediction label and the confidence are written into the tracking database, and the vehicle profile is updated.

[0140] In one embodiment of the present application, the step S5 of defining the joint optimization loss function comprises the following steps:

[0141] The loss function weight is dynamically adjusted according to the prediction variance of each task, specifically:

[0142]

[0143] Wherein, ω f is the weight coefficient corresponding to the functional area label, σ f is the variance of the prediction result of the functional area label, σ s is the variance of the prediction result of the state area label, σ a is the variance of the prediction result of the association label.

[0144] The weight coefficients corresponding to the state area label and the association label are calculated respectively using the same method.

[0145] The label corresponding weight coefficient is added to the label corresponding loss function to construct a joint optimization loss function;

[0146] The joint optimization loss function is specifically:

[0147] L = ω f ·L f + ω s ·L s + ω a ·L a

[0148] Wherein, L is a joint optimization loss function, ω f , ω s and ω a are label corresponding weight coefficients, L f , L s and L a are label corresponding loss functions, wherein, L f is a focal loss function, and L s and L a are cross-entropy loss functions.

[0149] The above examples are only used to illustrate the technical method of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical method of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present application.

Claims

1. A method for automatic vehicle identification, tracking, and positioning within a railway station, characterized in that, Includes the following steps: S1: Construct a comprehensive monitoring system for railway station areas to collect various types of data and perform synchronous fusion processing; S2: Set up a three-level judgment mechanism to calculate the first, second and third matching degree scores respectively, add them with weights to obtain the comprehensive matching degree score, and automatically identify the vehicle based on the comparison result of the comprehensive matching degree score and the comprehensive matching degree threshold. S3: After successful automatic vehicle identification, a tracking database is constructed and a railway track coordinate system is established. The Kalman filter algorithm is used to integrate multi-source data to obtain the two-dimensional plane coordinates and height information of the vehicle in the railway station coordinate system. S4: Utilize RFID readers and high-definition industrial cameras distributed within railway stations to dynamically acquire vehicle coordinates and synchronize them to the tracking database. Through the digital twin model of the railway station, the vehicle coordinates are mapped to manually labeled functional areas and associated track numbers. At the same time, complete semantic information is generated by combining status tags. S5: Construct a multi-task neural network framework using historical data and semantic location labels, and define a joint optimization loss function through dynamic weight adjustment and joint loss optimization.

2. The method for automatic vehicle identification and tracking positioning within a railway station according to claim 1, characterized in that, The proposed integrated monitoring system for railway station interiors includes high-definition industrial cameras, laser sensors, radar, RFID tags and readers, and BeiDou positioning terminals.

3. The method for automatic vehicle identification and tracking positioning within a railway station according to claim 1, characterized in that, The three-level judgment mechanism in step S2 includes the following steps: The data collected by the integrated monitoring system inside the railway station are preprocessed respectively; A three-level judgment mechanism is set up, specifically as follows: Extract vehicle visual features, including vehicle body color, logo patterns, number of windows and shape of the front of the vehicle, match them with a pre-established vehicle feature database, and make a first visual judgment based on the matching results; The vehicle's outline, dimensions, and distance from the track measured by the laser sensor are matched with a standard vehicle parameter library, and a second physical judgment is made based on the matching results. The integrity and occlusion of vehicle tag information read by RFID readers are used to make third-party information judgments; The overall matching score is obtained by weighted fusion of visual judgment, physical judgment and RFID information.

4. The method for automatic vehicle identification and tracking positioning within a railway station according to claim 3, characterized in that, The process of weightedly fusing visual judgment, physical judgment, and RFID information to obtain a comprehensive matching score includes the following steps: The extracted vehicle visual features are matched with a preset vehicle feature database to calculate the first matching score; the physical parameters measured by the laser sensor are matched with a standard vehicle parameter database to calculate the second matching score; and the tag information read by RFID is used as the basis to calculate the third matching score based on the tag integrity and occlusion. Weights are assigned to each data source based on historical data, and the weighted sums are then used to obtain a comprehensive matching score, specifically: Where G represents the overall matching score, and G1, G2, and G3 represent the first, second, and third matching scores, respectively. and These are the corresponding weights; Based on a pre-set reliability, the confidence interval for the overall matching score is determined using historical training data, and the overall matching score threshold is defined as follows: G T =μ-k c ·s Among them G T Here, μ is the overall matching score threshold, μ is the mean of historical overall matching scores, σ is the standard deviation, and k is the mean of the overall matching scores. c Confidence coefficient; If G > G T If the match is successful, the vehicle recognition result will be output; otherwise, it will be determined as a matching conflict and the matching inconsistency handling process will be triggered.

5. The method for automatic vehicle identification and tracking positioning within a railway station according to claim 4, characterized in that, The calculation of the third matching score based on tag integrity and occlusion includes the following steps: The system reads vehicle tag information using an RFID reader, including key fields such as vehicle number and vehicle model, and also obtains the completeness and occlusion of the tag. The integrity level is determined by whether the label is intact and undamaged; the occlusion level is determined by calculating the proportion of the label that is obscured by an object. Based on label integrity and occlusion, label status is categorized into the following levels, with different weights assigned according to each quality level: Level 1 Quality: Integrity ≥ 95%, Occlusion ≤ 5%, corresponding weight: 1; Level 2 Quality: Integrity ≥ 70%, Occlusion ≤ 30%, corresponding weight: 0.7; Level 3 Quality: Integrity < 70%, Occlusion > 30%, corresponding weight set to 0.4; The corresponding weight is determined based on the tag quality level, and multiplied by 100 to obtain the third matching score.

6. The method for automatic vehicle identification and tracking positioning within a railway station according to claim 4, characterized in that, The process for triggering inconsistency matching includes the following steps: S201: Data review, including reviewing the data collected by the integrated monitoring system inside the railway station to check for any data collection errors or omissions, and at the same time, activating backup monitoring equipment to obtain vehicle information; S202: Cross-validation, which includes the process of matching the reviewed data and the newly collected data again using a three-level judgment mechanism; S203: Manual assistance in judgment, including, if there are still inconsistencies in matching after data review and cross-validation, a manual assistance in judgment is introduced to perform the final vehicle identification.

7. The method for automatic vehicle identification and tracking positioning within a railway station according to claim 1, characterized in that, Step S3 includes the following steps: After a vehicle is successfully identified, a tracking database is built for it; The tracking database records vehicle identification information and semantic location tags; A railway station-specific coordinate system is adopted, with the track start point as the origin and the track direction as the reference axis. By fusing data from multiple sensors using the Kalman filter algorithm, the two-dimensional plane coordinates and height information of the vehicle in the railway station coordinate system are calculated.

8. The method for automatic vehicle identification and tracking positioning within a railway station according to claim 1, characterized in that, Step S4 includes the following steps: The vehicle location coordinates are updated in real time using RFID readers and high-definition industrial cameras distributed within the railway station. The location coordinates are mapped in real time to functional area labels, status labels, and associated labels using a data twin model, and then synchronized to the tracking database. The tracking and positioning system updates the vehicle's position coordinates in the tracking file based on the received location. At the same time, after updating the vehicle coordinates, it queries the railway station digital twin model in real time, maps the coordinates to functional area labels and associated labels, and stores them synchronously with the status labels in the tracking database. The railway station digital twin model is created by converting railway station CAD drawings into a three-dimensional mesh model, manually annotating the functional areas of each mesh, and matching preset labels according to the mesh index where the vehicle coordinates are located.

9. A method for automatic vehicle identification and tracking positioning within a railway station according to claim 1, characterized in that, Step S5 involves constructing a multi-task neural network framework using historical data and semantic location labels, including the following steps: Vehicle data with location coordinates is collected by combining historical operational data of railway stations with manually labeled semantic location tags, which include functional area tags, status tags, and association tags. The neural network extracts location coordinates, timestamps, raw sensor data, and external data features as input features. The neural network automatically learns the correlation between tasks by sharing underlying parameters. The input features are processed through a shared preprocessing layer, which includes standardizing coordinate values, extracting temporal features, and encoding classification features. Use a multi-task branching structure, including: functional area label branches, status label branches, and related label branches; The functional area label branch: input coordinates, time and vehicle type characteristics, output grid function classification; The status label branch takes speed, historical trajectory, and location features as inputs and outputs status labels. The associated label branch: input coordinates and track status, output current track number.

10. A method for automatic vehicle identification and tracking positioning within a railway station according to claim 1, characterized in that, Step S5 defines the joint optimization loss function, including the following steps: The weights of the loss function are dynamically adjusted based on the prediction variance of each task, specifically as follows: Where, ω f σ represents the weight coefficient corresponding to the functional area label. f σ represents the variance of the functional area label prediction results. s σ represents the variance of the state region label prediction results. a The variance of the associated label prediction results; Calculate the weight coefficients corresponding to the state region labels and associated labels using the same method; The weight coefficients corresponding to the labels are weighted and added to the loss functions corresponding to the labels to construct a joint optimization loss function; The joint optimization loss function is specifically as follows: L=ω f ·L f +oh s ·L s +oh a ·L a Where L is the joint optimization loss function, ω f ω s and ω a These are the weight coefficients corresponding to the labels, L f L s and L a These are the loss functions corresponding to the labels, where L f L is the focus loss function. s and L a Both are cross-entropy loss functions.

Images