A data processing method and system for aerial passenger controllers

Abstract

The present application relates to the technical field of data processing, and particularly relates to a data processing method and system for an overhead passenger controller. The method comprises the following steps: acquiring gondola position information data and controller acquisition data of an overhead passenger controller where the gondola is located; confirming a gondola running track based on the gondola position information data, and generating gondola three-dimensional modeling data by performing three-dimensional modeling on the controller acquisition data according to the gondola running track; respectively analyzing inertia disturbance characteristics, inertia resistance delay characteristics and intermittent load mutation characteristics of the gondola three-dimensional modeling data, and performing data feature distribution fusion to generate gondola controller behavior feature data set. The present application realizes intelligent identification and early warning control of abnormal behaviors of the overhead passenger controller by fusing three-dimensional modeling, dynamic feature analysis and image recognition, and solves problems such as single perception and response lag of the traditional system.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a data processing method and system for an aerial passenger controller. Background Technology

[0002] Traditional controllers generally employ PLC programmable logic control, which, while capable of basic start / stop control, operation monitoring, and alarm functions, suffers from response lag and insufficient accuracy in handling complex operating states and identifying abnormal behaviors. In recent years, advancements in sensor technology, embedded systems, and data communication technologies have enabled controllers to integrate inertial measurement units (IMUs), multi-source signal acquisition modules, and edge computing capabilities, giving them real-time data processing and behavior recognition capabilities. Simultaneously, the introduction of big data analytics and artificial intelligence algorithms has propelled the evolution of controllers from rule-driven to data-driven, enabling precise modeling and intelligent identification of features such as inertial disturbances, load anomalies, and signal mutations.

[0003] However, traditional monitoring methods have delays or even misses in identifying dynamic anomalies such as inertial disturbances and sudden load changes, which cannot meet the requirements for immediate response in high-risk operating environments. At the same time, they cannot integrate controller operation behavior and occupant action behavior data, leading to misjudgments and omissions, especially in non-standard operation or sudden behavior scenarios where timely response is not possible. Summary of the Invention

[0004] Therefore, it is necessary to provide a data processing method and system for an aerial passenger controller to solve at least one of the above-mentioned technical problems.

[0005] To achieve the above objectives, a data processing method for an aerial passenger controller is provided, the method comprising the following steps:

[0006] Step S1: Obtain the location information data of the monkey bus and the controller data collected by the overhead passenger controller where the monkey bus is located; confirm the running trajectory of the monkey bus based on the location information data, and perform three-dimensional modeling on the controller data based on the running trajectory of the monkey bus to generate three-dimensional modeling data of the monkey bus;

[0007] Step S2: Analyze the inertial disturbance characteristics, inertial stagnation delay characteristics, and discontinuous load mutation characteristics of the 3D modeling data of the monkey car, and perform data feature distribution fusion to generate a monkey car controller behavior feature dataset.

[0008] Step S3: Acquire images of the occupants of the monkey bus; identify the behavioral characteristics of the occupants in the images, and perform behavioral pairing and model building between the behavioral characteristics of the monkey bus controller and the behavioral characteristics of the occupants to generate a controller response judgment model;

[0009] Step S4: Use the controller response judgment model to identify abnormal behavior of the controller. When abnormal behavior is detected, issue speed limit, stop or alarm commands to the overhead passenger controller where the monkey bus is located to execute the corresponding early warning control operation.

[0010] This invention establishes a multi-dimensional model incorporating motion and human interaction features by integrating the location information of the passenger transport vehicle, controller data, and occupant image behavior information, making abnormal behavior identification more comprehensive and accurate. Targeted analysis and modeling of common phenomena during passenger transport vehicle operation, such as inertial disturbances, inertial hysteresis, and intermittent load mutations, effectively captures abnormal operating signs and improves the system's adaptability to nonlinear and complex working conditions. By extracting behavioral features from occupant images and further matching them with controller feature data, the invention compensates for the shortcomings of traditional controller monitoring data in interpreting abnormal behavior, enhancing the intuitiveness and logical rationality of judgments. The controller response model built based on multimodal features can intelligently identify and rapidly respond to abnormal behavior, improving the system's proactive prevention and control capabilities and reducing the accident rate. Once abnormal behavior is detected, the system can automatically issue speed limit, shutdown, or alarm commands, achieving an integrated closed-loop response of "identification-decision-control," significantly improving the real-time early warning capability and overall safety of the passenger transport vehicle system. This method is particularly suitable for mining environments with dense personnel, complex terrain, and frequent operations, improving the operational stability of overhead passenger transport systems and enhancing the life safety of workers. Therefore, by integrating 3D modeling, dynamic feature analysis and image recognition, this invention achieves intelligent identification and early warning control of abnormal behavior of the aerial passenger controller, solving the problems of single perception and delayed response in traditional systems.

[0011] Preferably, step S1 includes the following steps:

[0012] Step S11: Obtain the location information data of the monkey bus and the controller data collected by the overhead passenger controller where the monkey bus is located;

[0013] Step S12: Reconstruct the coordinate trajectory of the monkey vehicle's location information data to generate the monkey vehicle's three-dimensional running trajectory data;

[0014] Step S13: Extract the controller timestamp from the controller's collected data, and combine it with the monkey car's 3D running trajectory data to perform trajectory point mapping and fusion, generating a trajectory-bound controller dataset;

[0015] Step S14: Perform spatial data interpolation and parameter normalization based on the trajectory binding controller dataset to generate trajectory control parameter distribution data; perform spatial mesh modeling based on the trajectory control parameter distribution data to generate 3D modeling data for the monkey car.

[0016] This invention processes the monkey vehicle's location information and controller-collected data step-by-step, especially in S12 where coordinate trajectory reconstruction is achieved. This allows for accurate reconstruction of the monkey vehicle's 3D operating trajectory, ensuring that subsequent analysis is based on real and high-precision trajectory data. Mapping and fusing timestamp-synchronized controller data with 3D trajectory points breaks the limitations of a single data source, forming a trajectory-bound controller dataset, providing a solid data foundation for subsequent comprehensive modeling. Spatial data interpolation and parameter normalization effectively fill in data gaps and discrepancies, eliminate the influence of noise and outliers, improve data continuity and consistency, and enhance model stability. Spatial grid modeling based on trajectory control parameter distribution generates 3D modeling data of the monkey vehicle with spatial structure information, providing three-dimensional, multi-dimensional data support for monkey vehicle behavior analysis and anomaly detection. Refined processing steps improve data quality and expressive power, promoting the accuracy and real-time performance of subsequent steps such as inertial disturbance feature extraction and behavioral feature fusion, supporting intelligent early warning decision-making. This method fully considers data characteristics in both spatial and temporal dimensions, providing a solid technical guarantee for monitoring monkey vehicle operation in complex terrain and variable working conditions in mining areas, improving system robustness and safety.

[0017] Preferably, step S14, which involves spatial grid modeling based on trajectory control parameter distribution data, includes:

[0018] Based on the trajectory control parameter distribution data, spatial discretization is performed to divide it into several three-dimensional grid units to generate initial spatial grid data.

[0019] Boundary conditions are identified in the initial spatial grid data, trajectory-related grid cells are labeled, and trajectory-constrained grid data is generated.

[0020] Based on the trajectory-constrained grid data, the spatial attribute parameters of each grid cell are calculated, including position coordinates and motion state indicators, to generate grid attribute parameter data;

[0021] By using grid attribute parameter data to establish topological connections between grid cells, a complete spatial grid model is constructed, generating 3D modeling data for the monkey car.

[0022] This invention achieves fine-grained modeling of the manned vehicle's operating space by dividing the trajectory control parameter distribution data into multiple three-dimensional grid cells, improving the accuracy and resolution of spatial representation and facilitating the capture of subtle changes in complex trajectories. Boundary condition identification and annotation of trajectory-related grid cells ensure that the modeling process closely follows the manned vehicle's trajectory, enhancing the spatial model's ability to express trajectory constraints and improving the targeting of abnormal behavior detection. By calculating the spatial position and motion state indices of each grid cell, multi-dimensional attributes are assigned to the spatial grid, enabling a deep characterization of the manned vehicle's dynamic motion features and providing rich data support for subsequent behavior analysis. The use of grid attribute parameters for topological connections between cells forms a continuous and logically rigorous spatial grid model, improving the model's structural integrity and coherence, which is beneficial for dynamic analysis and simulation. Detailed grid modeling and attribute assignment give the manned vehicle's three-dimensional data model excellent visualization effects and data analysis performance, facilitating intuitive monitoring of the manned vehicle's motion status and abnormal point distribution. This spatial grid modeling method has good flexibility, allowing adjustment of grid division and attribute calculation according to different trajectories and control parameters, adapting to complex mining environments and diverse operating conditions.

[0023] Preferably, the inertial disturbance characteristics of the monkey car 3D modeling data analyzed in step S2 include:

[0024] Acceleration and angular velocity signals are extracted from the 3D modeling data of the monkey vehicle to generate inertial motion signal data;

[0025] The inertial motion signal data is subjected to time-domain filtering to remove high-frequency noise and generate filtered inertial signal data.

[0026] The rate of change of acceleration and the rate of change of angular velocity are calculated using the filtered inertial signal data to obtain the amplitude data of inertial disturbance.

[0027] Statistical analysis was performed on the amplitude data of inertial disturbances to extract characteristic inertial disturbance indicators and obtain the characteristics of inertial disturbances.

[0028] This invention extracts acceleration and angular velocity signals from 3D modeling data to capture the real-time motion changes of the scooter in space, ensuring the authenticity and timeliness of the inertial motion signals. Time-domain filtering removes high-frequency noise, reducing environmental interference and measurement errors, ensuring that subsequent calculations and analyses are based on high-quality inertial signal data and improving the accuracy of feature extraction. The rate of change of acceleration and angular velocity reflects subtle changes in the motion state, sensitively capturing fluctuations in the amplitude of inertial disturbances, which helps to identify abnormal motion behavior in a timely manner. Statistical analysis of the inertial disturbance amplitude yields representative characteristic indicators, facilitating a quantitative description of the scooter's inertial disturbance characteristics and providing a valid basis for subsequent anomaly detection and behavior discrimination. Based on detailed inertial disturbance characteristic data, the system can identify inertial anomalies in scooter operation earlier, improving the sensitivity and accuracy of early warnings and ensuring the safe operation of the scooter. Inertial disturbance characteristics, as important physical quantities, are combined with other behavioral characteristic data to construct a comprehensive judgment model, enhancing the system's ability to understand and predict complex operating states.

[0029] Preferably, the inertial lag delay characteristics of the monkey vehicle 3D modeling data analyzed in step S2 include:

[0030] Acceleration and angular velocity signals are extracted from the 3D modeling data of the monkey vehicle to generate inertial motion signal data;

[0031] Based on inertial motion signal data, the time delay of the inertial lag response is calculated to obtain inertial lag delay data;

[0032] Time-domain and frequency-domain analyses were performed on the inertial hysteresis delay data to extract delay characteristic parameters and generate inertial hysteresis characteristic parameter data.

[0033] By combining the inertial stagnation characteristic parameter data, the stagnation effect on the motion state of the monkey car is evaluated, and the inertial stagnation delay characteristics are obtained.

[0034] This invention extracts acceleration and angular velocity signals to calculate the time delay of inertial lag response, accurately reflecting the dynamic lag characteristics of the scooter caused by lag effects during motion, providing key time-scale information for motion state analysis. Combining time-domain and frequency-domain analysis methods, it delves into the temporal and frequency characteristics of inertial lag delay data, enhancing the ability to identify lag phenomena and improving the descriptive accuracy and richness of feature data. Using inertial lag characteristic parameter data, it comprehensively evaluates the constraints and influences of lag effects on the scooter's motion state, revealing the performance of inertial lag in actual operation and assisting in the accurate judgment of motion anomalies. Inertial lag delay characteristics, as an important indicator of abnormal signals, help to detect lag anomalies and operational delays in scooter motion at an early stage, enhancing the early warning system's response capability to abnormal behavior under complex operating conditions. This analysis process, based on the time delay and frequency characteristics of physical signals, endows the model with stronger physical interpretability, improving the credibility and stability of the behavior discrimination model. Combining inertial delay features with other behavioral features such as inertial perturbation features enriches the descriptive dimensions of the monkey car's motion state, which helps to build a more comprehensive dataset of controller behavioral features and improves model robustness.

[0035] Preferably, the discontinuous load mutation characteristics of the monkey cart 3D modeling data analyzed in step S2 include:

[0036] Load change time series is extracted from the 3D modeling data of the monkey cart to generate load time series data;

[0037] Perform segmented statistical analysis on load time-series data to identify discontinuous nodes of load changes and generate discontinuous load node data.

[0038] Calculate the load fluctuation amplitude at the discontinuous load node and generate load fluctuation amplitude data;

[0039] Abnormal fluctuations are detected based on load mutation amplitude data, and significant discontinuous load mutation characteristics are screened out to obtain discontinuous load mutation characteristics.

[0040] This invention extracts load change time series based on 3D modeling data to construct continuous load time series data, which can comprehensively and dynamically reflect the load fluctuation of the manned vehicle during operation, facilitating subsequent behavior analysis. Using segmented statistical analysis technology, discontinuous nodes in the load time series are identified, helping to quickly locate the location of abrupt changes and improving the efficiency and accuracy of feature extraction. By calculating the magnitude of the change at each discontinuous node, the intensity of the change can be effectively measured, providing a quantitative indicator for determining whether it belongs to an abnormal state, enhancing the objectivity and data support capability of anomaly detection. Combined with anomaly fluctuation detection methods, significant features are screened from the change magnitude, effectively avoiding false positives and false negatives, and improving the sensitivity of the early warning system to sudden load anomalies. Discontinuous load abrupt changes are often highly correlated with risk events such as hook slippage, personnel getting on and off the vehicle, and obstacle collisions in actual operation. By extracting such features, the ability to perceive potential abnormal behaviors can be significantly improved. Combining discontinuous load abrupt change features with other behavioral features such as inertial disturbances and inertial stagnation enhances the feature dimension and discrimination accuracy of the controller response model, improving the model's adaptability to complex scenarios. This feature analysis method is applicable to sudden load fluctuations in mining areas caused by terrain changes, non-standard personnel operations, or equipment failures. It helps to identify and respond to sudden risks in a timely manner, thereby improving system security and robustness.

[0041] Preferably, step S3, which involves identifying the behavioral characteristic data of the passengers in the images captured by the monkey bus, and performing behavioral pairing and model building between the monkey bus controller behavioral characteristic data and the personnel behavioral characteristic data, includes:

[0042] Image preprocessing is performed on images of passengers in the monkey bus to generate images with improved clarity. Image preprocessing includes image brightness enhancement, image geometric transformation, and image resolution enhancement.

[0043] Based on the image data with improved clarity, key points of human posture are extracted to generate human posture feature data.

[0044] Perform behavioral classification on personnel posture feature data to generate personnel behavioral feature data;

[0045] The human behavior characteristic data is synchronized with the monkey bus controller behavior characteristic data in time and the behavior is matched to generate behavior matching data.

[0046] A controller response judgment model is constructed based on behavior pairing data, and the controller response judgment model is generated.

[0047] This invention significantly improves image clarity and stability by performing image preprocessing operations such as brightness enhancement, geometric transformation, and resolution enhancement on captured images, providing high-quality input images for subsequent posture recognition and behavior extraction. Based on the optimized image data, key points are extracted to generate human posture features, which comprehensively reflect the dynamic changes in personnel's posture during the operation of the manned vehicle, providing intuitive and structured data support for behavior judgment. By classifying the posture feature data into behaviors such as standing, sitting, getting on and off the vehicle, and abnormal shaking, high-level semantic labels for personnel behavior are formed, enhancing behavior understanding capabilities. Through time synchronization and behavior pairing of personnel behavior features and controller behavior features, the relationship between personnel operation and equipment response can be accurately mapped, identifying potential abnormal human-machine interaction behaviors. A controller response judgment model is trained based on behavior pairing data, enabling the model to jointly infer the equipment response state from image behavior and controller features, improving the intelligence level of anomaly detection. This method can effectively identify abnormal control behaviors caused by improper personnel operation, realizing the linkage analysis of human factors and equipment behavior, and adapting to the complex, changeable, and high-risk operating environment of mining areas. Through a closed-loop process of image preprocessing, feature extraction, behavior matching, and model building, the entire chain from raw visual data to intelligent discrimination model is automated, improving the flexibility of system deployment and its promotional value.

[0048] Preferably, the time synchronization and behavior pairing of personnel behavior feature data and controller behavior feature data includes:

[0049] The behavioral characteristic data of personnel and the behavioral characteristic data of the monkey bus controller are divided into time windows to generate multi-level time slice data.

[0050] Time alignment is performed on multi-level time slice data to handle the asynchrony of the two sets of data on the time scale and generate a preliminary time alignment index.

[0051] Complex action sequences from personnel behavior feature data and monkey bus controller behavior feature data are decomposed based on a predefined micro-event dictionary to extract extremely fine-grained micro-event sequence data. Each micro-event contains action type, duration, and intensity features, generating a behavioral micro-event time series.

[0052] By combining the action type, intensity, and duration characteristics of micro-events, a three-dimensional similarity metric is designed, and a custom multi-dimensional behavior similarity matrix is ​​constructed.

[0053] Iterative weighted matching is performed based on the initial time alignment index and the multidimensional behavior similarity matrix to identify the best micro-event corresponding pairing path and generate behavior pairing data.

[0054] This invention effectively addresses the time scale differences between human and controller behaviors by dividing them into time windows and constructing multi-level time slices, combined with a preliminary time alignment index. This achieves high-precision time synchronization. A predefined micro-event dictionary decomposes complex action sequences into micro-event sequences containing elements such as action type, duration, and intensity, significantly improving the granularity and structure of behavioral expression, facilitating the accurate capture of subtle abnormal actions. A three-dimensional behavioral similarity index is designed based on action type, duration, and intensity, generating a multi-dimensional behavioral similarity matrix. This ensures that behavioral matching relies not only on time features but also on semantics, amplitude, and duration, improving the reliability and accuracy of behavioral pairing. Iterative weighted matching of the preliminary time index and the behavioral similarity matrix effectively avoids the influence of short-term errors and transient disturbances, extracting stable and logically consistent behavioral pairing paths, enhancing the model's ability to uncover real-world behavioral linkages. This method not only identifies the surface-level relationship between human actions and controller responses but also deeply mines potential micro-behavioral linkage patterns, providing a logically rigorous pairing basis for behavior-driven controller response models. By conducting in-depth matching at the micro-event level, subtle correspondences between "normal operation - abnormal response" or "abnormal action - abnormal response" can be identified, greatly enhancing the system's ability to perceive and recognize complex abnormal linkage events. The multi-dimensional similarity matching framework and micro-event structure constructed by this method can be flexibly extended and are applicable to different types of behavioral data matching scenarios, possessing good algorithmic versatility and cross-application transferability.

[0055] Preferably, step S4 includes the following steps:

[0056] Step S41: Based on the controller response judgment model, perform abnormal feature matching on the real-time behavior data of the controller to generate abnormal behavior discrimination result data;

[0057] Step S42: Assess the risk level of the abnormal behavior identification results data and generate abnormal behavior risk level data;

[0058] Step S43: Based on the abnormal behavior risk level data, determine the corresponding early warning control instruction type and generate early warning control instruction data, wherein the early warning control instruction type includes speed limit, shutdown or alarm;

[0059] Step S44: Send a warning control command to the overhead passenger controller where the monkey bus is located via the warning control command data to trigger the corresponding warning control operation.

[0060] This invention utilizes a controller response judgment model to match abnormal features in real-time collected controller behavior data, enabling rapid and automatic identification of potential abnormal control behaviors. This avoids delayed manual judgment in the event of emergencies and improves system response efficiency. A risk level assessment mechanism is introduced, quantifying risk levels based on the severity, scope, and duration of abnormal behavior. This transitions from "whether it is abnormal" to "the degree of abnormality," providing a basis for tiered response decisions. Different types of control commands (speed limit, shutdown, alarm) are generated based on different risk levels, enabling differentiated handling strategies. This ensures that high-risk anomalies are dealt with decisively, while low-risk anomalies are handled with flexible control or alerts, maintaining a balance between operational efficiency and safety. By issuing and executing control commands in real time, such as speed limit, shutdown, and alarms, a closed-loop mechanism of "identification-assessment-response" is effectively established. This ensures that the system can immediately trigger a response once abnormal behavior is identified, significantly improving the timeliness and automation level of overall safety control. Automatic decision-making on control strategies based on model judgment and risk assessment reduces reliance on human experience and effectively avoids safety accidents such as equipment collisions, personnel injuries, or system downtime caused by misoperation or delayed judgment. The early warning control commands can be flexibly configured according to actual needs (such as adding types like "deceleration" and "early warning broadcast"), possessing good scalability and adaptability, suitable for complex operating scenarios of different mining areas and different equipment models. The entire process data of each anomaly identification, risk level assessment, and control command execution is structured and saved, forming a traceable data chain, which facilitates subsequent safety analysis, model optimization, and operational strategy adjustment.

[0061] This specification provides a data processing system for an aerial passenger controller, used to execute the aforementioned data processing method for an aerial passenger controller. The data processing system for the aerial passenger controller includes:

[0062] The 3D modeling module is used to acquire the location information data of the monkey bus and the controller data collected by the overhead passenger controller where the monkey bus is located; based on the location information data of the monkey bus, the running trajectory of the monkey bus is confirmed, and 3D modeling is performed on the controller data based on the running trajectory of the monkey bus to generate 3D modeling data of the monkey bus;

[0063] The control behavior feature module is used to analyze the inertial disturbance features, inertial stagnation delay features, and discontinuous load mutation features of the 3D modeling data of the monkey car, and to perform data feature distribution fusion to generate a monkey car controller behavior feature dataset.

[0064] The user behavior feature module is used to acquire images of the occupants of the monkey bus; identify the human behavior feature data in the images of the occupants of the monkey bus; and perform behavior matching and model building between the monkey bus controller behavior feature data and the human behavior feature data to generate a controller response judgment model.

[0065] The early warning control module is used to identify abnormal behavior of the controller by using the controller response judgment model. When abnormal behavior is detected, it sends speed limit, shutdown or alarm commands to the overhead passenger controller where the monkey bus is located to execute the corresponding early warning control operation.

[0066] The beneficial effects of this invention lie in the fact that, through a 3D modeling module, the trajectory of the maneuver vehicle is fused with data collected by the controller to establish a high-precision, dynamic, and real-time 3D operating model of the maneuver vehicle, providing a realistic and structured spatial semantic foundation for subsequent behavior analysis. The control behavior feature module performs in-depth analysis of dynamic features such as inertial disturbances, inertial delay, and intermittent load mutations, and integrates their data distribution to effectively characterize the controller's operating state under complex conditions, significantly improving the identifiability and distinguishability of abnormal control behaviors. The user behavior feature module performs feature pairing and model construction between human image behavior and controller behavior, innovatively introducing "modeling of the impact of human behavior on controller response," opening up a human-machine collaborative anomaly analysis path, and significantly enhancing the system's intelligent identification capability for operational risks. The early warning control module uses the controller response judgment model to discriminate behavior and automatically executes control commands such as speed limits, shutdowns, or alarms based on the judgment results, forming a real-time, automatic, and precisely responsive closed-loop early warning system, improving the overall safety level. The system makes decisions and responds based on data, reducing reliance on human experience and judgment, and lowering the risk of control anomalies and accidents caused by human factors such as operational errors and reaction delays. The modules are clearly decoupled, supporting flexible expansion and replacement. For example, it can adapt to different types of image recognition algorithms, anomaly detection models, and control strategies to meet the safety control needs of various complex scenarios such as mines and industrial transportation. Therefore, this invention, by integrating 3D modeling, dynamic feature analysis, and image recognition, achieves intelligent identification and early warning control of abnormal behavior in aerial passenger transport controllers, solving the problems of single-sensorship and delayed response in traditional systems. Attached Figure Description

[0067] Figure 1 This is a flowchart illustrating the steps of a data processing method for an aerial passenger controller.

[0068] Figure 2 for Figure 1 A detailed flowchart illustrating the implementation steps of step S1.

[0069] Figure 3 for Figure 1 A detailed flowchart illustrating the implementation steps of step S4.

[0070] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0071] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0072] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0073] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0074] To achieve the above objectives, please refer to Figures 1 to 3 A data processing method for an aerial passenger controller, the method comprising the following steps:

[0075] Step S1: Obtain the location information data of the monkey bus and the controller data collected by the overhead passenger controller where the monkey bus is located; confirm the running trajectory of the monkey bus based on the location information data, and perform three-dimensional modeling on the controller data based on the running trajectory of the monkey bus to generate three-dimensional modeling data of the monkey bus;

[0076] Step S2: Analyze the inertial disturbance characteristics, inertial stagnation delay characteristics, and discontinuous load mutation characteristics of the 3D modeling data of the monkey car, and perform data feature distribution fusion to generate a monkey car controller behavior feature dataset.

[0077] Step S3: Acquire images of the occupants of the monkey bus; identify the behavioral characteristics of the occupants in the images, and perform behavioral pairing and model building between the behavioral characteristics of the monkey bus controller and the behavioral characteristics of the occupants to generate a controller response judgment model;

[0078] Step S4: Use the controller response judgment model to identify abnormal behavior of the controller. When abnormal behavior is detected, issue speed limit, stop or alarm commands to the overhead passenger controller where the monkey bus is located to execute the corresponding early warning control operation.

[0079] In this embodiment of the invention, a positioning device (such as UWB, inertial navigation, or laser ranging equipment) installed in the mine roadway acquires real-time position information data of the manned vehicle, including dynamic coordinate data in the horizontal, vertical, and longitudinal directions. The sampling frequency can be set to more than 20 times per second to ensure trajectory continuity and accuracy. Simultaneously, controller data collected by the overhead passenger controller where the manned vehicle is located is acquired. This controller data includes traction current, voltage changes, braking command response time, equipment operating status codes, etc. Based on the manned vehicle's position information data, its running trajectory model in a three-dimensional roadway environment is constructed. The controller data is aligned with the timestamp and position information, combined with roadway cross-section and slope information, to perform three-dimensional modeling calculations. Finally, three-dimensional modeling data of the manned vehicle is generated for subsequent behavioral feature extraction. The three-dimensional modeling data constructed in step S1 is analyzed to extract the following three types of controller behavioral features: identifying speed fluctuations of the manned vehicle under non-braking conditions, such as sudden acceleration or deceleration, and calculating parameters such as peak acceleration, rate of change, and duration. Analyzing the delay in the actual actions of the manned vehicle after the controller issues a run or stop command. Extract indicators such as command response time difference and inertial gliding distance. Determine if the monkey car experiences sudden loading or unloading by monitoring fluctuations in current, voltage, and traction power. If the power change exceeds a preset value per unit time, it is considered a load mutation. These three types of features are distributed according to spatial location and time sequence, and then fused to form a structured controller behavior feature dataset containing feature point coordinates, feature type, feature value, and timestamp. Acquire image data of the monkey car occupants using image acquisition devices installed inside or around the monkey car. The image sampling frequency is recommended to be no less than 10 frames per second to ensure continuous motion capture. Utilize video behavior recognition algorithms to process the image data and identify occupant behavior characteristics, including pulling, jumping, kicking, and abnormal postures (such as standing or suddenly getting up). Extract corresponding time, spatial location, and action category information for each type of behavior to generate personnel behavior feature data. Pair the controller behavior feature data obtained in step S2 with the personnel behavior feature data according to the time axis and spatial location to construct a relationship model between personnel behavior and controller response. A controller response judgment model is trained using machine learning models (such as random forests or long short-term memory neural networks) to identify which human behaviors are likely to cause abnormal equipment responses or pose potential risks. The system runs the controller response judgment model in real time, jointly analyzing the currently collected controller behavior characteristics and human behavior characteristics. If the model's judgment result shows that the current behavior combination is highly similar to historical abnormal patterns, it is judged as abnormal behavior.Upon detecting abnormal behavior, the system automatically selects the corresponding control response strategy based on the anomaly level: for minor anomalies, such as short-term coasting delays, the system issues a speed limit command to reduce the speed of the passenger car; for moderate anomalies, such as sudden load changes accompanied by abnormal passenger actions, the system issues a shutdown command and cuts off the main power supply; for severe anomalies, such as inertial disturbances occurring simultaneously with violent passenger actions, the system immediately triggers an audible and visual alarm and uploads the data to the mine dispatch system or the central control room. All control responses are transmitted in real time via the communication bus of the passenger car's overhead passenger controller and recorded in the log for easy post-event traceability and model optimization.

[0080] As an example of the present invention, reference is made to Figure 2 As shown, in this example, step S1 includes:

[0081] Step S11: Obtain the location information data of the monkey bus and the controller data collected by the overhead passenger controller where the monkey bus is located;

[0082] Step S12: Reconstruct the coordinate trajectory of the monkey vehicle's location information data to generate the monkey vehicle's three-dimensional running trajectory data;

[0083] Step S13: Extract the controller timestamp from the controller's collected data, and combine it with the monkey car's 3D running trajectory data to perform trajectory point mapping and fusion, generating a trajectory-bound controller dataset;

[0084] Step S14: Perform spatial data interpolation and parameter normalization based on the trajectory binding controller dataset to generate trajectory control parameter distribution data; perform spatial mesh modeling based on the trajectory control parameter distribution data to generate 3D modeling data for the monkey car.

[0085] In this embodiment of the invention, a high-precision positioning system (such as a UWB positioning module, inertial measurement unit (IMU), or lidar) is deployed in the monkey bus operating environment to collect real-time position information data of the monkey bus within the tunnel. This data includes coordinate values ​​in three dimensions: lateral, longitudinal, and height, as well as timestamp information, for subsequent trajectory reconstruction. Simultaneously, controller-collected data is extracted from the overhead passenger controller where the monkey bus is located. Controller-collected data includes: controller operating status codes (such as start, brake, stop); electrical parameters such as current, voltage, and traction power; control command execution timestamps; and sensor input data (such as tilt angle, speed, load changes, etc.). Through a unified time synchronization mechanism, the controller data and positioning data are time-aligned to ensure data consistency and fusion. Using the monkey bus position information data obtained in step S11, the coordinate points are reconstructed into a trajectory. An interpolation smoothing algorithm (such as B-spline interpolation) is used to eliminate jitter noise and restore the continuous operating path of the monkey bus in the tunnel. The trajectory data is organized in chronological order, and each trajectory point includes position coordinates (lateral, longitudinal, and height), speed change trend, and acquisition time. The final output is the 3D trajectory data of the maneuvering vehicle, used for subsequent spatial fusion modeling. The timestamp of each control operation is extracted from the controller's data and matched with the time axis of the 3D trajectory data to bind the controller's action data to the corresponding trajectory point. During the binding process, a time window matching method is used to ensure that each trajectory point can be associated with the controller's state in the most recent or recent time period. After fusion, each trajectory point not only contains spatial location information but also embeds the controller's operating parameters such as control current, load state, and speed commands at that time. The final result is a trajectory-bound controller dataset with the following data structure: trajectory point location information (3D coordinates), control parameters (current, voltage, load, speed, etc.), and corresponding timestamps. Based on the trajectory-bound controller dataset, the following two processing steps are performed: interpolation is performed to complete the control parameters between trajectory points to compensate for the sparsity caused by the data sampling interval. A normalization method is used to unify the scale of multi-dimensional control parameters such as current, voltage, and speed to facilitate spatial distribution modeling. The entire trajectory data is projected onto a spatial grid, for example, based on grid cells with a 1-meter interval, and the average control parameters within each grid are statistically analyzed and aggregated. A spatial grid model containing spatial coordinates and control parameter feature values ​​is constructed. The output 3D modeling data of the monkey car is a structured mesh dataset, in which each cell contains: a 3D spatial location index, parameters such as the average speed, current, and load corresponding to that location, implicit feature indicators such as control response delay and inertial trend. This modeling result can provide spatial basic data support for subsequent anomaly identification, path optimization, and equipment status monitoring.

[0086] Preferably, step S14, which involves spatial grid modeling based on trajectory control parameter distribution data, includes:

[0087] Based on the trajectory control parameter distribution data, spatial discretization is performed to divide it into several three-dimensional grid units to generate initial spatial grid data.

[0088] Boundary conditions are identified in the initial spatial grid data, trajectory-related grid cells are labeled, and trajectory-constrained grid data is generated.

[0089] Based on the trajectory-constrained grid data, the spatial attribute parameters of each grid cell are calculated, including position coordinates and motion state indicators, to generate grid attribute parameter data;

[0090] By using grid attribute parameter data to establish topological connections between grid cells, a complete spatial grid model is constructed, generating 3D modeling data for the monkey car.

[0091] In this embodiment of the invention, the entire spatial region is discretized based on the geometric dimensions and trajectory density characteristics of the monkey car's operating area. A reasonable grid size (e.g., 1 meter × 1 meter × 1 meter) is set to divide the three-dimensional space into multiple cubic grid units. During spatial discretization, the trajectory control parameter distribution data is mapped to the corresponding grid units according to their corresponding three-dimensional coordinate positions, forming initial spatial grid data. Each grid unit records the normalized control parameter values ​​such as speed, current, and load carried by the trajectory points within that region. Trajectory boundary condition identification is performed on the initial spatial grid data, including: detecting all grid units traversed by the trajectory path; marking grids containing monkey car trajectory points as "trajectory-related grid units"; and buffering and expanding the surrounding grids (e.g., expanding by one grid unit) to support subsequent topology modeling and anomaly analysis. This process generates trajectory constraint grid data, which clarifies which grids belong to the valid trajectory region, helping to limit the analysis scope and improve modeling efficiency. Based on trajectory-constrained mesh data, key spatial attributes of each mesh cell are statistically analyzed and calculated, including: center position coordinates (e.g., X, Y, Z values ​​of the mesh's geometric center); mean and distribution characteristics of control parameters (e.g., average velocity, current, control delay, etc. of all trajectory points within the mesh); motion state indicators, such as the velocity change trend, acceleration range, and inertial disturbance amplitude of trajectory points within the region; and trajectory density or frequency indicators, used to assess the importance or activity of the mesh. Finally, the mesh attribute parameter data for each cell is output, providing a basis for subsequent topology modeling and anomaly identification. When constructing a complete 3D model, based on the mesh attribute parameter data, the topological structure between spatial meshes is established in the following manner: an adjacency matrix is ​​constructed according to mesh adjacency relationships (e.g., six-neighborhood, twenty-six-neighborhood); the connection status of each mesh cell with its neighboring meshes is recorded; if the trajectory attributes of adjacent mesh cells are continuous (e.g., consistent velocity direction, consistent control response), a connectivity relationship is established; a spatial graph structure is constructed, forming a topological mesh graph with path continuity and coherent control behavior. The final generated 3D modeling data of the monkey vehicle is a structured mesh graph model, which includes both trajectory path and control parameters, and also possesses spatial continuity and topological logical relationships.

[0092] Preferably, the inertial disturbance characteristics of the monkey car 3D modeling data analyzed in step S2 include:

[0093] Acceleration and angular velocity signals are extracted from the 3D modeling data of the monkey vehicle to generate inertial motion signal data;

[0094] The inertial motion signal data is subjected to time-domain filtering to remove high-frequency noise and generate filtered inertial signal data.

[0095] The rate of change of acceleration and the rate of change of angular velocity are calculated using the filtered inertial signal data to obtain the amplitude data of inertial disturbance.

[0096] Statistical analysis was performed on the amplitude data of inertial disturbances to extract characteristic inertial disturbance indicators and obtain the characteristics of inertial disturbances.

[0097] In this embodiment of the invention, velocity, acceleration, and attitude change information containing time series is extracted from the constructed 3D modeling data of the monkey vehicle. This information is then combined with sensor data recorded in the controller's data collection, such as the three-axis acceleration values ​​output by the accelerometer and the angular velocity signals (pitch, roll, and yaw rates) provided by the gyroscope. This information is integrated to generate inertial motion signal data containing timestamps, 3D coordinates, linear acceleration, and angular velocity, which serves as the basic input for subsequent analysis. Due to the presence of noise and mechanical jitter in the collected data, time-domain filtering is required to improve the accuracy of inertial feature analysis. This involves using moving average, low-pass filtering, or Kalman filtering to suppress high-frequency jitter and abrupt errors; setting a reasonable sliding window width (e.g., 0.2 to 0.5 seconds) to smooth the inertial change curve; and outputting the processed stable acceleration and angular velocity curves to generate filtered inertial signal data. Based on the filtered signal data, the dynamic rate of change of inertial disturbance is calculated to quantify the impact characteristics and abrupt changes in the movement of the scooter. This mainly includes: the difference in acceleration over a continuous time period to identify instantaneous acceleration or sudden stop behavior; and the identification of rotational impacts, attitude changes, etc., especially applicable to uphill, downhill, or curved sections. The above rate of change is normalized to unify the unit standard, and statistical values ​​such as maximum, average, and variance are calculated to output the inertial disturbance amplitude data. The disturbance amplitude data is further analyzed to extract key indicators reflecting the smoothness and safety of the scooter's operation, including but not limited to: identifying whether inertial impacts exceeding a set threshold occur; determining whether the inertial disturbance is instantaneous or continuous; the number of times the disturbance exceeds the set threshold per unit time; and statistically analyzing the direction in which the inertial disturbance mainly occurs (up / down / left / right / forward / backward). The fluctuation energy characteristics reflect the energy density of the overall inertial disturbance and are used to assess ride comfort or equipment wear risk. The final output is the inertial disturbance characteristics.

[0098] Preferably, the inertial lag delay characteristics of the monkey vehicle 3D modeling data analyzed in step S2 include:

[0099] Acceleration and angular velocity signals are extracted from the 3D modeling data of the monkey vehicle to generate inertial motion signal data;

[0100] Based on inertial motion signal data, the time delay of the inertial lag response is calculated to obtain inertial lag delay data;

[0101] Time-domain and frequency-domain analyses were performed on the inertial hysteresis delay data to extract delay characteristic parameters and generate inertial hysteresis characteristic parameter data.

[0102] By combining the inertial stagnation characteristic parameter data, the stagnation effect on the motion state of the monkey car is evaluated, and the inertial stagnation delay characteristics are obtained.

[0103] In this embodiment of the invention, the 3D modeling data of the monkey bus is processed using an inertial signal processing tool. The 3D modeling data originates from the sequence of coordinate points on the overhead running path of the monkey bus and its corresponding controller sampling data. Each trajectory point contains spatial position (3D coordinates x, y, z) and timestamp information. By calculating the rate of change of position between adjacent time periods for continuous trajectory points, the velocity vector of each trajectory point is first calculated using the formula v(t) = [x(t+1) -x(t)] / Δt, where Δt is the sampling time interval, uniformly set to 0.1 seconds. Subsequently, the change of velocity vector over time is differentially calculated using a(t) = [v(t+1) -v(t)] / Δt to obtain the three-axis acceleration values ​​Ax, Ay, and Az. The angular velocity is obtained based on the change of spatial direction between trajectory points. By calculating the rate of change of the angle between the vectors formed by three adjacent trajectory points, converting it to Euler angles using the Rodrigues rotation formula, and then differentiating, the angular velocities Gx, Gy, and Gz around the three coordinate axes are obtained. Finally, the aforementioned acceleration and angular velocity time series data were recorded as inertial motion signal data in a unified structured format. Each time point contained six inertial dimensions: Ax, Ay, Az, Gx, Gy, and Gz, with a uniform frequency of 10Hz, covering the entire trajectory. To determine the inertial response delay of the maneuver vehicle to control commands, the command sequence in the controller's collected data first needed to be classified and identified. Controller commands included three categories: "acceleration command," "deceleration command," and "stop command," all accompanied by a timestamp accurate to milliseconds. For example, the "acceleration command" was recorded at time T0. The controller command time was aligned with the inertial motion signal time series, and the response start point of each type of inertial index was detected one by one after the command was issued. Taking acceleration Ax as an example, a window detection was performed on the trend of Ax after T0. When, in three consecutive time points (e.g., T1, T2, T3), the increase in Ax was not less than 0.2 m / s² each time and the single difference was greater than 0.1 m / s², T1 was determined to be the response start point. The delay ΔT = T1 - T0. This method is applied separately to each type of command, extracting multiple ΔT values ​​in each response cycle to form a complete inertial hysteresis delay data set. The above processing flow uses fixed response threshold values: acceleration threshold is 0.2 m / s², angular velocity threshold is 10 deg / s, minimum effective response window length is 3 time points (i.e., 0.3 seconds), and maximum observation duration is 3 seconds. Finally, the delay result between each control command and its response time is recorded as an inertial hysteresis delay data. Time-domain analysis of the delay data is achieved by calculating statistical characteristic indicators such as average response time, maximum delay time, and standard deviation of response time.The delay data sets corresponding to each type of control command are statistically analyzed. For example, for all delay results of acceleration commands, the average (e.g., 1.45 seconds), maximum (e.g., 2.75 seconds), minimum (e.g., 0.65 seconds), and standard deviation (e.g., 0.58 seconds) are calculated. Simultaneously, the abnormal delay rate is calculated, which is the ratio of the number of samples with a delay exceeding 2.5 seconds to the total number of samples, with a threshold of 10%. Frequency domain analysis uses the Fast Fourier Transform (FFT) algorithm to perform spectral analysis on the acceleration signal sequence, with the frequency analysis range set from 0.1Hz to 5Hz, corresponding to a wavelength range of 2 seconds to 10 seconds. The dominant frequency component and its amplitude in each inertial signal channel (Ax, Ay, Az) are taken as frequency domain features; for example, the dominant frequency is 0.45Hz, and the amplitude is -35dB. All extracted feature parameters, including but not limited to average delay, maximum delay, dominant frequency, peak amplitude, and abnormal delay ratio, are uniformly encoded and named as feature vectors to constitute complete inertial hysteresis feature parameter data. Based on the inertial lag characteristic parameter data obtained above, the lag response status of the manned train at different operating stages is further analyzed and evaluated. By mapping the characteristic parameters back to the original trajectory node positions, the average delay changes corresponding to different positions or sections (e.g., tunnel entrances, turning points, ramp start sections) are compared to mark high lag response areas. The evaluation indicators include the following: First, the average change rate of trajectory segment lag, calculated as ΔT2-ΔT1, where ΔT2 and ΔT1 are the average delay values ​​of two adjacent trajectory segments; Second, the maximum number of lag responses and the proportion of abnormal responses within each trajectory segment are recorded. If the abnormal response rate within a certain trajectory segment exceeds 20%, that segment is marked as a high lag area; Third, the correlation between structural lag and control lag is analyzed by comparing the lag characteristics of the same control command under different control cycles. Finally, the results of the above evaluation process are uniformly encoded and output to generate inertial lag delay features, which are used for subsequent matching with occupant behavior data or training of controller response models.

[0104] Preferably, the discontinuous load mutation characteristics of the monkey cart 3D modeling data analyzed in step S2 include:

[0105] Load change time series is extracted from the 3D modeling data of the monkey cart to generate load time series data;

[0106] Perform segmented statistical analysis on load time-series data to identify discontinuous nodes of load changes and generate discontinuous load node data.

[0107] Calculate the load fluctuation amplitude at the discontinuous load node and generate load fluctuation amplitude data;

[0108] Abnormal fluctuations are detected based on load mutation amplitude data, and significant discontinuous load mutation characteristics are screened out to obtain discontinuous load mutation characteristics.

[0109] In this embodiment of the invention, the real-time load records of the scooter during its entire operation are extracted by calling the data channel bound to the load monitoring module in the controller acquisition system. The load data unit is kilograms (kg), and the sampling frequency is set to 1Hz, meaning one data point is recorded per second. Each data point includes a collection timestamp and the corresponding load value. Each trajectory node in the 3D modeling data contains a controller number and a time identifier. The load data is synchronized and matched with the trajectory time axis to construct a load change time series that is completely aligned with the scooter trajectory. This time series is uniformly named "Load Time Series Data" and has the following structure: {t1,L1},{t2,L2},...,{t n ,L n} where t is a time point and L is the actual load value at that time. This data sequence ensures it covers the entire operating section of the bus from departure to arrival, with at least 95% of the data being valid across consecutive time periods. After preprocessing the load time series data, a fixed sliding window differencing algorithm is used to identify areas of sudden load changes. The sliding window width is set to 5 seconds (i.e., 5 sampling points). The load change amplitude within each window is statistically analyzed, and the change amplitude ΔL is defined as the difference between the load value at the end of the window and the load value at the beginning. If the absolute value of ΔL is greater than a set threshold of 10 kg, the window is considered to contain discontinuous load change nodes. To pinpoint the exact location of the discontinuous nodes, point-to-point differencing is further performed within the window. The load difference between two consecutive time points is calculated. If the load change between any two points is greater than 15 kg, and no similar fluctuations repeat within adjacent 10 seconds, then that time point is marked as a "discontinuous load node". All marked time points are aggregated to form a discontinuous load node dataset. Each record point in this dataset includes a timestamp, the corresponding load value, and the direction of change (increase or decrease). Based on intermittent load node data, load values ​​are extracted from several time points (within ±3-second time windows) before and after each node. The stable load value L0 before the mutation and the new value L1 after the mutation are recorded. The load mutation amplitude F = |L1 - L0| is calculated, and the direction of change is recorded. The mutation amplitude data structure is set as a triple: {t, F, dir}, where t is the time of the mutation node, F is the mutation amplitude (in kg), and dir is the direction of change indicator ("increase" or "decrease"). The validity of the mutation amplitude must meet the following requirements: after the mutation, it must remain within the new load range for at least 5 seconds (fluctuation not exceeding ±5 kg); otherwise, it is judged as short-term noise and not accepted. For cases where the interval between consecutive mutation nodes is less than 10 seconds, only the change with the larger amplitude is retained. All valid mutation amplitude samples are uniformly numbered and classified for storage, forming a complete load mutation amplitude dataset. Statistical classification and anomaly detection processing are performed on the mutation amplitude dataset. First, the mean μ and standard deviation σ of the overall mutation amplitude are calculated, and the anomaly judgment threshold is set to μ + 1.5σ. All data samples with mutation amplitudes greater than this threshold are marked as "significant mutations". Simultaneously, combining information from the monkey train's operational phases, such as acceleration, deceleration, and turning, the distribution frequency of mutation nodes in each phase is statistically analyzed. If the mutation density within a certain period exceeds three times the normal level (based on a normal level of no more than one mutation per minute), that period is marked as a "high-frequency mutation segment." Finally, samples meeting one of the following two conditions are selected: the mutation amplitude is greater than the anomaly detection threshold; or the mutation occurs within a high-frequency mutation segment, thus defining them as "intermittent load mutation features." These features are uniformly encoded, including the mutation time, mutation amplitude, direction of change, and the corresponding trajectory segment, constituting the final intermittent load mutation feature data used as input to the behavior judgment model.

[0110] Preferably, step S3, which involves identifying the behavioral characteristic data of the passengers in the images captured by the monkey bus, and performing behavioral pairing and model building between the monkey bus controller behavioral characteristic data and the personnel behavioral characteristic data, includes:

[0111] Image preprocessing is performed on images of passengers in the monkey bus to generate images with improved clarity. Image preprocessing includes image brightness enhancement, image geometric transformation, and image resolution enhancement.

[0112] Based on the image data with improved clarity, key points of human posture are extracted to generate human posture feature data.

[0113] Perform behavioral classification on personnel posture feature data to generate personnel behavioral feature data;

[0114] The human behavior characteristic data is synchronized with the monkey bus controller behavior characteristic data in time and the behavior is matched to generate behavior matching data.

[0115] A controller response judgment model is constructed based on behavior pairing data, and the controller response judgment model is generated.

[0116] In this embodiment of the invention, continuous images of the passengers in the monkey transport vehicle are acquired, with an image sampling frequency set to 25 frames per second and a minimum image resolution of 1920×1080 pixels. During the image preprocessing stage, a high-performance image processing module is used to perform the following operations on the original image: histogram equalization is used to improve the overall brightness and contrast of the image, ensuring clear details in dark areas, and adjusting parameters including controlling the brightness increase within the range of 20% to 40% to avoid overexposure. Affine transformation is used to correct tilt and rotation in the image, with the correction angle not exceeding ±5 degrees to maintain image stability and realism. Bilinear interpolation is used to perform super-resolution resampling of the image, increasing the resolution to 1.5 times the original, ensuring clear identification of key features. An open-source multi-human pose recognition algorithm module (such as OpenPose or similar open-source libraries) is used to detect pose key points in the preprocessed image. This algorithm extracts the joint positions of the human body through a convolutional neural network, with key points including the top of the head, shoulders, elbows, wrists, hips, knees, and ankles, totaling 18 standard joint points. The algorithm takes a single-frame RGB image as input and outputs the two-dimensional coordinates and confidence scores of 18 keypoints. Coordinate data is represented using pixel positions, and the confidence threshold is set to 0.6; keypoints below this threshold are automatically removed or interpolated. The personnel posture feature data structure consists of a set of 18 joint coordinates per frame, stored in conjunction with confidence information, to describe the spatial distribution of the human body's current posture. Behavior classification employs a time-series-based machine learning method, performing feature engineering on the time series of posture keypoints over several consecutive frames (30 frames recommended, approximately 1.2 seconds), including joint angle change rates, relative position changes between joints, movement speed, and acceleration, forming feature vectors. The classification algorithm is based on a trained Support Vector Machine (SVM) or Random Forest model, outputting specific behavior category labels such as "standing normally," "walking," "climbing," and "falling." The classification decision threshold is set to a confidence probability of 0.8 or higher; low-confidence results are further confirmed using data from neighboring time frames. The final generated personnel behavior feature data includes the behavior category, start and end timestamps, duration of the behavior, and corresponding posture feature descriptions. Using a unified time base, the timestamps of personnel behavior characteristic data are aligned with the timestamps of the monkey bus controller behavior characteristic data. Time synchronization errors are controlled within ±10 milliseconds to ensure high-precision pairing. A time window matching method is used to filter corresponding controller behavior characteristic events within the time period of personnel behavior, forming behavior event pairs. The behavior pairing process employs a time-series-based cross-matching algorithm, mapping personnel behavior categories to various indicators of controller behavior characteristic data (such as inertial disturbance characteristics, load mutation characteristics, etc.) according to time periods. The behavior pairing data structure includes: personnel behavior category, corresponding time period, controller behavior characteristic indicator value and its time range, achieving multi-dimensional joint description.Using a paired behavior dataset, a multivariate discriminant model is constructed, employing a decision tree algorithm or a rule-based expert system to determine the controller response. Model input includes personnel behavior characteristic categories and their time periods, and a set of controller behavior characteristic parameters (inertial disturbance, load change, stall delay, etc.). Output is the controller response status, such as "normal response," "abnormal response," or "warning response." Model parameters are determined through training with historical data. Thresholds, such as the abnormal response trigger threshold, are set when the controller's inertial disturbance index exceeds 0.75 standard deviations, the load change exceeds 20 kg, and the abnormal behavior category is considered as triggering the response. The controller response judgment results output by the model will be used for subsequent control strategy adjustments and abnormal alarm generation.

[0117] Preferably, the time synchronization and behavior pairing of personnel behavior feature data and controller behavior feature data includes:

[0118] The behavioral characteristic data of personnel and the behavioral characteristic data of the monkey bus controller are divided into time windows to generate multi-level time slice data.

[0119] Time alignment is performed on multi-level time slice data to handle the asynchrony of the two sets of data on the time scale and generate a preliminary time alignment index.

[0120] Complex action sequences from personnel behavior feature data and monkey bus controller behavior feature data are decomposed based on a predefined micro-event dictionary to extract extremely fine-grained micro-event sequence data. Each micro-event contains action type, duration, and intensity features, generating a behavioral micro-event time series.

[0121] By combining the action type, intensity, and duration characteristics of micro-events, a three-dimensional similarity metric is designed, and a custom multi-dimensional behavior similarity matrix is ​​constructed.

[0122] Iterative weighted matching is performed based on the initial time alignment index and the multidimensional behavior similarity matrix to identify the best micro-event corresponding pairing path and generate behavior pairing data.

[0123] In this embodiment of the invention, timestamp sequences of human behavior feature data and monkey bus controller behavior feature data are acquired separately. The two sets of data typically have different temporal resolutions: human behavior features are sampled at the second level (approximately 25 frames per second), while controller data is sampled at the millisecond level (sampling period of 10 milliseconds). Multi-level time window division is performed on both sets of data, with time window lengths categorized by level: coarse-grained (e.g., 1 second), medium-grained (200 milliseconds), and fine-grained (50 milliseconds). Within each time window, statistical values ​​(such as mean, peak, and frequency) or event markers of the corresponding behavioral features are extracted to generate a behavioral summary corresponding to the time window, forming a multi-level time slice data set. The time window boundaries employ an overlapping sliding window design, with an overlap ratio set to 50% to ensure the continuity and integrity of events at the time boundaries. Since the sampling clocks of the two sets of data differ, and there are network delays and sensor sampling deviations, time synchronization correction is first performed. Cross-correlation analysis is conducted on the synchronization event points in the two time series (such as the start and end of marked actions or changes in control commands) to calculate the time offset and eliminate system clock differences. For the corrected data, the Dynamic Time Warping (DTW) algorithm is used to time-align the multi-level time-slice data sequences. DTW allows for non-linear time scaling, accommodating the time differences between gait and control response. An initial time alignment index is output, indicating the correspondence between the human behavior time window and the controller behavior time window, providing a temporal basis for subsequent micro-event matching. A micro-event dictionary is established, containing common action types (such as "raising an arm," "bending over," "turning around," etc.) and corresponding controller response events (such as "acceleration peak," "motor load mutation"), with each micro-event defined by an action type identifier, a duration range (e.g., 50 ms to 500 ms), and force characteristics (e.g., acceleration amplitude or motor torque magnitude). For human behavior sequences, attitude key point time series analysis is used to segment them into micro-event units according to action type and duration thresholds. For controller behavior data, based on features such as inertial disturbance peaks and load mutation points, corresponding micro-event responses are marked, and control micro-event sequences within the same time period are extracted. A micro-event time series is generated, including action type ID, start and end timestamps, and intensity parameters, providing a structured representation of fine-grained actions and control behaviors. The three-dimensional similarity metrics are defined as follows: Action type similarity: Boolean matching based on whether the action type IDs are the same; a match scores 1 point, and a different match scores 0 points. Intensity similarity: Calculated using Euclidean distance of intensity features; the smaller the distance, the higher the similarity. The normalization range is 0 to 1, with a distance threshold of 0.2. Duration similarity: The absolute difference in duration between two micro-events is calculated, normalized, and the similarity is 1 when the time difference is less than 50 milliseconds, linearly decaying to 0. The three-dimensional similarity is calculated through a weighted summation, with weights set to 0.5, 0.3, and 0.2, respectively.Using the sequences of micro-events related to human behavior and the sequences of micro-events related to controller behavior as rows and columns, a similarity matrix is ​​calculated, with each element representing the comprehensive similarity score between the two micro-events. An improved dynamic programming algorithm is employed to search for paths based on the multi-dimensional similarity matrix within the time frame constrained by the initial time alignment index. The objective is to maximize the cumulative similarity along the path, with path nodes representing the pairing relationship between micro-event pairs. During iterative weighted matching, the time window weights are adjusted based on the previous matching results, dynamically narrowing the search space for poorly matched areas and improving matching accuracy. The final output is the optimal micro-event pairing path sequence, forming structured behavior pairing data. This data includes: micro-event sequence number, start and end times of the matched micro-events, action type matching identifier, intensity matching score, and duration difference. This behavior pairing data is used for subsequent controller response model training and abnormal behavior detection.

[0124] As an example of the present invention, reference is made to Figure 3 As shown, step S4 in this example includes:

[0125] Step S41: Based on the controller response judgment model, perform abnormal feature matching on the real-time behavior data of the controller to generate abnormal behavior discrimination result data;

[0126] Step S42: Assess the risk level of the abnormal behavior identification results data and generate abnormal behavior risk level data;

[0127] Step S43: Based on the abnormal behavior risk level data, determine the corresponding early warning control instruction type and generate early warning control instruction data, wherein the early warning control instruction type includes speed limit, shutdown or alarm;

[0128] Step S44: Send a warning control command to the overhead passenger controller where the monkey bus is located via the warning control command data to trigger the corresponding warning control operation.

[0129] In this embodiment of the invention, multi-dimensional behavioral data of the aerial passenger controller of the monkey bus is collected in real time, including but not limited to motor torque, joint angles, load changes, and inertial sensor data. The collected real-time data is input into a pre-trained controller response judgment model at a preset sampling frequency (e.g., every 10 milliseconds). This model is based on machine learning classification algorithms (e.g., support vector machines, random forests, or lightweight neural networks) and has pre-learned the feature distributions of normal and abnormal control behaviors. For the input real-time data, the model performs multi-feature fusion and matching operations, and determines whether the current behavior deviates from the normal range through an anomaly score function in the feature space. The anomaly threshold is set to an anomaly score exceeding 0.7 (out of 1), which is considered an abnormal behavior. The abnormal behavior discrimination result data is output, including an anomaly timestamp, anomaly feature type (e.g., overload, stall, oscillation, etc.), anomaly score, and associated parameter values, forming a structured anomaly discrimination log. Based on the abnormal behavior discrimination result, a hierarchical risk assessment algorithm is used to quantify the severity of the abnormal behavior. The risk level is divided into three levels: Level 1 (High Risk): Anomaly score ≥ 0.9, and the abnormal characteristics involve safety-related indicators, such as motor overload exceeding limits, current abnormal fluctuations exceeding 20%, etc. Level 2 (Medium Risk): Anomaly score between 0.7 and 0.9, and the abnormal characteristics are functional fluctuations, such as short-term stall or slight vibration. Level 3 (Low Risk): Anomaly score below 0.7, but there are deviations in non-critical indicators. The duration of the anomaly is considered when assessing the risk level; a continuous anomaly exceeding 500 milliseconds will increase the level. Combining the anomaly type and duration, abnormal behavior risk level data is generated, including the risk level number, assessment basis, and recommended handling level. A predefined early warning control instruction strategy library is matched based on the risk level data. There are three types of early warning control instructions: Speed ​​Limit Instruction: Applicable to Level 2 risk situations, controlling the speed of the train to reduce by 30% to 50%, with the specific speed limit ratio determined according to the actual risk level classification. Stop Instruction: Applicable to Level 1 risk, immediately stopping the train operation to prevent accidents. Alarm Commands: Applicable to Level 3 risks, triggering remote or on-site alarm notifications without affecting current operation. The command generation module combines the current controller status and safety specifications to generate structured early warning control command data, including command type, effective time, target device ID, and relevant execution parameters. Early warning control commands are sent to the overhead passenger controller where the passenger transport vehicle is located via a real-time communication link (such as industrial Ethernet or a dedicated wireless channel). The communication protocol uses a secure and certified transport layer protocol to ensure data integrity and timeliness. After receiving the command, the controller parses the command type and parameters and immediately executes the corresponding early warning control operation, including issuing speed limit control signals, triggering mechanical emergency stops, or activating alarm audible and visual devices. Feedback signals during execution are monitored in real time to confirm the success of command execution; abnormal feedback information is recorded for subsequent fault diagnosis. After the early warning control operation is completed, the system automatically enters a monitoring state, awaiting subsequent commands to ensure the safe operation of the passenger transport vehicle.

[0130] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0131] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A data processing method for an aerial passenger controller, characterized in that, Includes the following steps: Step S1: Obtain the location information data of the monkey bus and the controller data collected by the overhead passenger controller where the monkey bus is located; confirm the running trajectory of the monkey bus based on the location information data, and perform three-dimensional modeling on the controller data based on the running trajectory of the monkey bus to generate three-dimensional modeling data of the monkey bus; Step S2: Analyze the inertial disturbance characteristics, inertial stagnation delay characteristics, and discontinuous load mutation characteristics of the 3D modeling data of the monkey car, and perform data feature distribution fusion to generate a monkey car controller behavior feature dataset. Step S3: Acquire images of the occupants of the monkey bus; identify the behavioral characteristics of the occupants in the images, and perform behavioral pairing and model building between the behavioral characteristics of the monkey bus controller and the behavioral characteristics of the occupants to generate a controller response judgment model; Step S4: Use the controller response judgment model to identify abnormal behavior of the controller. When abnormal behavior is detected, issue speed limit, stop or alarm commands to the overhead passenger controller where the monkey bus is located to execute the corresponding early warning control operation.

2. The data processing method for an aerial passenger controller according to claim 1, characterized in that, Step S1 includes the following steps: Step S11: Obtain the location information data of the monkey bus and the controller data collected by the overhead passenger controller where the monkey bus is located; Step S12: Reconstruct the coordinate trajectory of the monkey vehicle's location information data to generate the monkey vehicle's three-dimensional running trajectory data; Step S13: Extract the controller timestamp from the controller's collected data, and combine it with the monkey car's 3D running trajectory data to perform trajectory point mapping and fusion, generating a trajectory-bound controller dataset; Step S14: Perform spatial data interpolation and parameter normalization based on the trajectory binding controller dataset to generate trajectory control parameter distribution data; perform spatial mesh modeling based on the trajectory control parameter distribution data to generate 3D modeling data for the monkey car.

3. The data processing method for an aerial passenger controller according to claim 2, characterized in that, Step S14, which involves spatial grid modeling based on trajectory control parameter distribution data, includes: Based on the trajectory control parameter distribution data, spatial discretization is performed to divide it into several three-dimensional grid units to generate initial spatial grid data. Boundary conditions are identified in the initial spatial grid data, trajectory-related grid cells are labeled, and trajectory-constrained grid data is generated. Based on the trajectory-constrained grid data, the spatial attribute parameters of each grid cell are calculated, including position coordinates and motion state indicators, to generate grid attribute parameter data; By using grid attribute parameter data to establish topological connections between grid cells, a complete spatial grid model is constructed, generating 3D modeling data for the monkey car.

4. The data processing method for an aerial passenger controller according to claim 1, characterized in that, Step S2 analyzes the inertial perturbation characteristics of the monkey car's 3D modeling data, including: Acceleration and angular velocity signals are extracted from the 3D modeling data of the monkey vehicle to generate inertial motion signal data; The inertial motion signal data is subjected to time-domain filtering to remove high-frequency noise and generate filtered inertial signal data. The rate of change of acceleration and the rate of change of angular velocity are calculated using the filtered inertial signal data to obtain the amplitude data of inertial disturbance. Statistical analysis was performed on the amplitude data of inertial disturbances to extract characteristic inertial disturbance indicators and obtain the characteristics of inertial disturbances.

5. The data processing method for an aerial passenger controller according to claim 1, characterized in that, Step S2 analyzes the inertial lag delay characteristics of the 3D modeling data of the monkey train, including: Acceleration and angular velocity signals are extracted from the 3D modeling data of the monkey vehicle to generate inertial motion signal data; Based on inertial motion signal data, the time delay of the inertial lag response is calculated to obtain inertial lag delay data; Time-domain and frequency-domain analyses were performed on the inertial hysteresis delay data to extract delay characteristic parameters and generate inertial hysteresis characteristic parameter data. By combining the inertial stagnation characteristic parameter data, the stagnation effect on the motion state of the monkey car is evaluated, and the inertial stagnation delay characteristics are obtained.

6. The data processing method for an aerial passenger controller according to claim 1, characterized in that, Step S2 analyzes the discontinuous load mutation characteristics of the monkey cart 3D modeling data, including: Load change time series is extracted from the 3D modeling data of the monkey cart to generate load time series data; Perform segmented statistical analysis on load time-series data to identify discontinuous nodes of load changes and generate discontinuous load node data. Calculate the load fluctuation amplitude at the discontinuous load node and generate load fluctuation amplitude data; Abnormal fluctuations are detected based on load mutation amplitude data, and significant discontinuous load mutation characteristics are screened out to obtain discontinuous load mutation characteristics.

7. The data processing method for an aerial passenger controller according to claim 1, characterized in that, Step S3 involves identifying the behavioral characteristics of the passengers in the images captured by the monkey bus, and then matching the behavioral characteristics of the monkey bus controller with the behavioral characteristics of the passengers to build a model. Image preprocessing is performed on images of passengers in the monkey bus to generate images with improved clarity. Image preprocessing includes image brightness enhancement, image geometric transformation, and image resolution enhancement. Based on the image data with improved clarity, key points of human posture are extracted to generate human posture feature data. Perform behavioral classification on personnel posture feature data to generate personnel behavioral feature data; The human behavior characteristic data is synchronized with the monkey bus controller behavior characteristic data in time and the behavior is matched to generate behavior matching data. A controller response judgment model is constructed based on behavior pairing data, and the controller response judgment model is generated.

8. The data processing method for an aerial passenger controller according to claim 7, characterized in that, Synchronizing and matching personnel behavior data with controller behavior data in time includes: The behavioral characteristic data of personnel and the behavioral characteristic data of the monkey bus controller are divided into time windows to generate multi-level time slice data. Time alignment is performed on multi-level time slice data to handle the asynchrony of the two sets of data on the time scale and generate a preliminary time alignment index. Complex action sequences from personnel behavior feature data and monkey bus controller behavior feature data are decomposed based on a predefined micro-event dictionary to extract extremely fine-grained micro-event sequence data. Each micro-event contains action type, duration, and intensity features, generating a behavioral micro-event time series. By combining the action type, intensity, and duration characteristics of micro-events, a three-dimensional similarity metric is designed, and a custom multi-dimensional behavior similarity matrix is ​​constructed. Iterative weighted matching is performed based on the initial time alignment index and the multidimensional behavior similarity matrix to identify the best micro-event corresponding pairing path and generate behavior pairing data.

9. The data processing method for an aerial passenger controller according to claim 1, characterized in that, Step S4 includes the following steps: Step S41: Based on the controller response judgment model, perform abnormal feature matching on the real-time behavior data of the controller to generate abnormal behavior discrimination result data; Step S42: Assess the risk level of the abnormal behavior identification results data and generate abnormal behavior risk level data; Step S43: Based on the abnormal behavior risk level data, determine the corresponding early warning control instruction type and generate early warning control instruction data, wherein the early warning control instruction type includes speed limit, shutdown or alarm; Step S44: Send a warning control command to the overhead passenger controller where the monkey bus is located via the warning control command data to trigger the corresponding warning control operation.

10. A data processing system for an aerial passenger controller, characterized in that, For performing the data processing method for an aerial passenger controller as described in claim 1, the data processing system for the aerial passenger controller includes: The 3D modeling module is used to acquire the location information data of the monkey bus and the controller data collected by the overhead passenger controller where the monkey bus is located; based on the location information data of the monkey bus, the running trajectory of the monkey bus is confirmed, and 3D modeling is performed on the controller data based on the running trajectory of the monkey bus to generate 3D modeling data of the monkey bus; The control behavior feature module is used to analyze the inertial disturbance features, inertial stagnation delay features, and discontinuous load mutation features of the 3D modeling data of the monkey car, and to perform data feature distribution fusion to generate a monkey car controller behavior feature dataset. The user behavior feature module is used to acquire images of the occupants of the monkey bus; identify the human behavior feature data in the images of the occupants of the monkey bus; and perform behavior matching and model building between the monkey bus controller behavior feature data and the human behavior feature data to generate a controller response judgment model. The early warning control module is used to identify abnormal behavior of the controller by using the controller response judgment model. When abnormal behavior is detected, it sends speed limit, shutdown or alarm commands to the overhead passenger controller where the monkey bus is located to execute the corresponding early warning control operation.

Images