An intercity traffic security inspection system based on edge computing and multi-sensor cooperation

Abstract

This invention discloses an intercity transportation security inspection system based on edge computing and multi-sensor collaboration, belonging to the field of security inspection system technology. The system includes: a data receiving and synchronization module for receiving multi-source raw security inspection data and performing time synchronization and standardization processing; a trajectory generation module for detecting and tracking passenger and baggage targets in the body security gate area and baggage inspection entrance area based on the raw security inspection sequence data; a disturbance identification and filtering module for analyzing continuous spatiotemporal trajectory data and identifying multiple queue disturbance behaviors; a consistency fusion and judgment module for performing multi-sensor consistency fusion and one-to-one attribution judgment based on candidate correspondence data and continuous spatiotemporal trajectory data; and a handling instruction push module for aggregating body security inspection risks and baggage inspection risks into risk event handling instruction data based on passenger and baggage correspondence data and pushing it to the on-site terminal.

Description

Technical Field

[0001] This application relates to the field of security inspection system technology, and more specifically, to an intercity transportation security inspection system based on edge computing and multi-sensor collaboration. Background Technology

[0002] With the continuous expansion of the railway passenger network and the increasing passenger flow at stations, railway stations face operational pressures of high throughput, timeliness, and high security, especially during holidays, peak hours, and large-scale events. Railway security checks are typically conducted in conjunction with body scanners and baggage screening equipment (such as X-ray machines and CT scanners). Body scanners are responsible for detecting metal objects, prohibited items, or abnormal body characteristics carried by passengers, while baggage screening is responsible for identifying dangerous goods, prohibited items, and suspicious structures in packages and bags. Together, they constitute a key link in the station's security defense line and are important infrastructure for ensuring railway transportation order, passenger safety, and public safety.

[0003] In traditional security check processes, systems typically assume that passengers passing through the body scanner and their carry-on luggage entering the machine inspection lane are approximately synchronous in time. Therefore, they often employ simple association methods based on temporal proximity or sequential consistency, binding the body inspection result at a given moment to the luggage inspection result at a similar moment. This approach can generally meet the needs in scenarios with low traffic, low disturbance, and relatively stable queue order. However, in actual railway station operations, especially during peak hours, the passage of passengers and luggage is often not strictly synchronized. This can lead to existing systems incorrectly associating risk events with other passengers, resulting in false interceptions and chaotic on-site handling.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This application provides an intercity traffic security inspection system based on edge computing and multi-sensor collaboration to solve the above-mentioned technical problems.

[0006] This application provides an intercity transportation security inspection system based on edge computing and multi-sensor collaboration, comprising: a data receiving and synchronization module for receiving multi-source security inspection raw data, performing time synchronization and standardization processing to obtain security inspection raw sequence data with a unified spatiotemporal reference; a trajectory generation module for detecting and tracking passenger and baggage targets in the body security gate area and baggage inspection entrance area based on the security inspection raw sequence data, generating continuous spatiotemporal trajectory data including passenger movement status, baggage transport status, and cross-device location mapping relationship; a disturbance identification and filtering module for analyzing the continuous spatiotemporal trajectory data, identifying multiple queue disturbance behaviors, and filtering according to potentially related person-baggage combinations to obtain candidate correspondence data; a consistency fusion and judgment module for performing multi-sensor consistency fusion and one-to-one attribution judgment based on the candidate correspondence data and the continuous spatiotemporal trajectory data to obtain passenger-baggage correspondence data; and a handling instruction push module for aggregating body security inspection risks and baggage inspection risks into risk event handling instruction data based on the passenger-baggage correspondence data and pushing it to the on-site terminal.

[0007] Based on the embodiments provided in this application, the data receiving and synchronization module performs time synchronization and standardization processing on the multi-source security inspection raw data. Combined with the trajectory generation module, it detects and tracks passenger and baggage targets in the body security gate area and baggage inspection entrance area, generating continuous spatiotemporal trajectory data containing cross-device location mapping relationships. This allows the system to move beyond simple binding based on temporal proximity or sequential consistency, and instead accurately reconstruct the actual passage process of passengers and their corresponding baggage in the spatiotemporal dimension based on the independent continuous movement trajectories of passengers and baggage, fundamentally avoiding the situation of incorrectly associating risk events with other passengers. The disturbance identification and filtering module analyzes the continuous spatiotemporal trajectory data, identifies multiple queue disturbance behaviors, and filters based on potentially associated person-baggage combinations to obtain candidate correspondence data. When queue disturbances occur, such as passenger queue jumping, baggage delays, or passengers and baggage passing through in sequence, the system can filter out reasonable candidate person-baggage correspondences from the disturbed trajectories, rather than directly causing association failure or errors due to temporal disorder, thus ensuring the reliability of association under high throughput and high time pressure.

[0008] Through a consistency fusion and judgment module, candidate correspondence data and continuous spatiotemporal trajectory data are combined to perform multi-sensor consistency fusion and one-to-one attribution determination. This process utilizes multi-sensor trajectory information from both the passenger security check area and the baggage inspection area for cross-validation, determining a unique baggage attribution for each passenger. This solves the problem that existing simple association methods cannot determine unique correspondences in complex flow scenarios, ensuring that the final passenger baggage correspondence data has high confidence. The handling instruction push module, based on the accurately established passenger baggage correspondence data, aggregates passenger security check risks and baggage inspection risks into risk event handling instruction data for specific passengers and pushes it to the on-site terminal. This allows on-site personnel to clearly know the specific passenger and baggage corresponding to the risk event, avoiding false interceptions and on-site chaos caused by incorrect associations, and improving the accuracy of risk event handling and order maintenance efficiency at the security checkpoint. Attached Figure Description

[0009] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 This is a structural diagram of an optional intercity traffic security inspection system based on edge computing and multi-sensor collaboration according to an embodiment of this application; Figure 2 This is a flowchart illustrating an optional output prediction confidence level according to an embodiment of this application; Figure 3 This is a schematic diagram of the architecture and data flow of an optional intercity traffic security inspection system based on edge computing and multi-sensor collaboration according to an embodiment of this application; Figure 4 This is a schematic diagram of an optional passenger and baggage association decision-making process according to an embodiment of this application.

[0010] 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

[0011] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0012] According to one aspect of the embodiments of this application, such as Figure 1As shown, this application provides an intercity traffic security inspection system based on edge computing and multi-sensor collaboration, including: The data receiving and synchronization module 101 is used to receive multi-source security inspection raw data, perform time synchronization and standardization processing, and obtain security inspection raw sequence data with a unified spatiotemporal reference. Specifically, the multi-source security inspection raw data includes at least the following five categories: First, data on the passage of passengers through the security gate, namely, the passage time signal, alarm level markings, and preliminary human image slices output by the equipment when passengers pass through the metal detector gate or millimeter-wave body scanner; second, data on the transport of baggage inspection equipment, including the baggage entry time triggered by the photoelectric sensor at the entrance of the baggage X-ray machine or CT machine, the real-time speed and position pulses fed back by the conveyor belt encoder, and the baggage departure signal triggered by the photoelectric sensor at the exit; third, video stream data of the security gate area, collected by a global monitoring camera deployed above the security gate, used to capture a continuous sequence of images of passengers entering the security gate, undergoing inspection, and leaving the security gate and moving towards the buffer zone; fourth, video stream data of the baggage inspection entrance area, collected by a close-up camera facing the baggage inspection entrance, used to capture visual information of passengers placing baggage, baggage sliding into the conveyor belt, and subsequent baggage lingering or shifting at the entrance; and fifth, time synchronization status data of edge computing nodes, namely, clock calibration instructions and synchronization feedback records periodically sent by the edge server to each distributed sensor, used to ensure the consistency of the time base for all subsequent processing steps. Upon receiving raw security check data from multiple sources, the challenges include inconsistent time bases, heterogeneous data formats, and inconsistent spatial coordinate systems. Directly performing joint analysis on this data would lead to disordered temporal relationships between passenger and baggage events and failure to match spatial locations. Therefore, the data receiving and synchronization module performs the following three levels of processing.

[0013] At the time synchronization level, because sensors such as security gates, baggage screening equipment, and video cameras operate independently, their internal crystal clocks have inherent drift, and network transmission latency fluctuates randomly. Directly using the original timestamps of these devices for correlation may cause event sequence errors or matching discrepancies. To address this issue, this system configures edge computing nodes as the master clock for time synchronization, using a time protocol or network time protocol to periodically send synchronization commands to all connected sensors. Upon receiving the synchronization command, each sensor calculates the deviation between its own clock and the master clock through protocol interaction and performs local clock compensation. During subsequent data reporting, the sensors simultaneously carry the original device timestamp and a unified timestamp after synchronization calibration. Upon receiving data, the data receiving and synchronization module uniformly uses the calibrated timestamp as the time reference for subsequent processing, ensuring that the times of passenger passage through the security gate, baggage entry into the machine screening entrance, and video frame capture are strictly aligned on the same timeline. During this process, the time synchronization status data of the edge computing nodes is continuously recorded as key metadata for system operation. This includes the synchronization initiation time, the local clock deviation value reported by each device, and the synchronization success flag. This data is not only used to monitor the health status of time synchronization, but also provides a traceable calibration basis for subsequent time series analysis.

[0014] At the level of format standardization, due to significant differences in the data formats output by sensors from different manufacturers and of different types, directly sending them to subsequent modules for processing would greatly increase the complexity of the processing logic. Therefore, the module encapsulates various types of data in a unified format. For data on security gate passage events and baggage inspection equipment transport events, the module directly extracts the event type, equipment identifier, original timestamp, and status value, and encapsulates them into a unified event message body, enabling subsequent modules to access event types and parameters in a consistent manner. For video stream data from the security gate area and the baggage inspection entrance area, the module uses a combination of hardware and software decoding to decompose the original video stream into continuous image frames. Based on the processing requirements of the subsequent trajectory generation module, the module adaptively scales the image resolution and converts the image data into a unified color space and bit depth format.

[0015] At the spatial alignment level, because the installation positions, viewing angles, and imaging parameters of the cameras in the security gate area and the baggage inspection entrance area are different, the target coordinates detected in the two areas are located in their respective pixel coordinate systems, making direct spatial correlation impossible. Therefore, the module utilizes pre-completed on-site calibration results to establish a spatial mapping relationship between the two areas. Specifically, by deploying calibration objects with known spatial coordinates on-site, which are simultaneously captured by cameras in both areas, the homography matrix between the two cameras is calculated. Based on this matrix, the module can transform the coordinates of passenger foot points detected in the security gate area video and the coordinates of the center point of the baggage outline detected in the baggage inspection entrance area video from their respective pixel coordinate systems to a unified world coordinate system. This world coordinate system uses the actual physical space of the security checkpoint as a reference, assigning clear two-dimensional or three-dimensional coordinates to key locations such as the security gate exit, baggage inspection entrance, and buffer zone areas. Through this spatial alignment process, passengers and baggage, originally isolated in the fields of view of two independent cameras, are mapped onto the same global plane map, providing a precise spatial reference for the subsequent trajectory generation module to establish cross-device positional mapping relationships.

[0016] Through the aforementioned processing at three levels—time synchronization, format standardization, and spatial alignment—the original data, originally scattered across different devices, coordinate systems, and time bases, has been integrated into a unified spatiotemporal sequence of security check data. This sequence data not only preserves the high-fidelity details of the original events but also provides a standardized input that can be directly used for subsequent passenger trajectory generation, baggage trajectory generation, and cross-device correlation analysis.

[0017] The trajectory generation module 102 is used to detect and track passenger and baggage targets in the security gate area and baggage inspection entrance area based on the original security inspection sequence data, and generate continuous spatiotemporal trajectory data including passenger movement status, baggage transport status and cross-device position mapping relationship; After obtaining the raw security check sequence data with a unified spatiotemporal reference, one of the core issues that the trajectory generation module needs to address is establishing the spatial correspondence between the body security gate area and the baggage inspection entrance area. Since the events of a passenger passing through the body security gate and baggage entering the X-ray machine are physically separated and captured by different types of sensors, it is impossible to achieve spatiotemporal correlation between people and baggage if they are tracked independently only within their respective sensor fields of view.

[0018] To address this, the trajectory generation module utilizes the spatial alignment results already achieved by the data receiving and synchronization module to correlate the detection results of two isolated areas to the same spatial reference system. Specifically, the module first acquires the pre-calibrated homography matrix between the cameras in the security gate area and the baggage inspection entrance area. This matrix describes the projection transformation relationship between the imaging planes of the two cameras. Based on this matrix, the module projects the coordinates of the passenger's feet detected in the security gate area video onto a unified world coordinate system, and simultaneously projects the coordinates of the center point of the baggage outline detected in the baggage inspection entrance area video onto the same world coordinate system. In this way, even if the passenger and baggage have not yet physically intersected, the system can grasp their relative positions in real time on the same global plane.

[0019] Furthermore, the module models the physical passageway between the security checkpoint exit and the baggage screening entrance on a global plane map. This path is not a simple straight line connection, but rather includes potential corners, buffer zones, queuing areas, and other physical structures based on the actual site layout. On this basis, the module pre-calculates and stores the historical expected travel time for this path, i.e., the average time a passenger takes from exiting the security checkpoint to arriving at the baggage screening entrance. For each passenger passing through the security checkpoint, the module overlays the historical expected travel time from the security checkpoint exit to the baggage screening entrance with their passage time, generating a deterministic time window with upper and lower boundaries. This time window not only predicts when a passenger will arrive at the baggage screening entrance area, but more importantly, it serves as the core basis for subsequent screening of passenger-baggage candidate relationships: only baggage entering the baggage screening entrance within this deterministic time window will be included in the passenger's candidate association set.

[0020] Through the aforementioned cross-device location mapping and time window constraints, the trajectory generation module effectively narrows the matching search space for people and luggage. The complex problem that originally required combining and matching all passengers and all luggage globally has been transformed into a problem of making association judgments on a limited set of candidates within a local spatiotemporal neighborhood, thus achieving efficient initial screening of correspondences in high-passenger-flow scenarios at intercity transportation hubs.

[0021] The disturbance identification and filtering module 103 is used to analyze continuous spatiotemporal trajectory data, identify multiple queue disturbance behaviors, and filter them according to possible related person-bag combinations to obtain candidate correspondence data. Multiple queue disturbance behaviors include queue jumping, taking on behalf of others, re-checking and re-jumping, brief pauses, and turning back; The consistency fusion and determination module 104 is used to perform multi-sensor consistency fusion and one-to-one attribution determination based on candidate correspondence data and continuous spatiotemporal trajectory data to obtain passenger baggage correspondence data. The handling instruction push module 105 is used to aggregate personal security inspection risks and baggage inspection risks into risk event handling instruction data based on passenger baggage correspondence data and push it to the on-site terminal.

[0022] Furthermore, the trajectory generation module is one of the core components of this system for realizing the spatiotemporal correlation between passengers and baggage. Based on the unified spatiotemporal reference of the security check raw sequence data output by the data receiving and synchronization module, this module continuously detects and tracks passenger and baggage targets in the body security gate area and the baggage inspection entrance area, generating continuous spatiotemporal trajectory data including passenger movement status, baggage transport status, and cross-equipment position mapping relationships. To achieve the above functions, this module contains four collaborative units: a rigid topology modeling unit, a conveyor belt kinematics embedding unit, a passenger trajectory estimation unit, and a cross-equipment mapping unit. The trajectory generation module includes: The rigid topology modeling unit is used to model the passenger passage path from the exit of the security gate area to the baggage inspection entrance area, as well as the baggage transport path inside the baggage inspection entrance area, into a directed acyclic graph. The nodes of the directed acyclic graph include the security gate exit, buffer area, baggage inspection entrance, baggage inspection internal passage, and baggage inspection exit. The edges represent the forced passage path, and the historical expected passage time of each edge is stored. The rigid topology modeling unit is responsible for abstracting the physical space of the security checkpoint area into a structured model that can be used for trajectory generation and correlation analysis. In the intercity transportation security check scenario, the movement paths of passengers from exiting the body security gate to arriving at the baggage screening entrance, and the movement of baggage from entering the screening entrance to completing scanning, are constrained by physical facilities and have a deterministic passage order. Therefore, this unit models the passenger passage path from the body security gate area exit to the baggage screening entrance area, as well as the baggage transport path within the baggage screening entrance area, as a directed acyclic graph.

[0023] The nodes of this directed acyclic graph are set based on the physical layout of the site and the mandatory requirements of the security check process. Specifically, they include: a human body security gate exit node, representing the physical location where passengers leave the security gate after completing the body check; a buffer area node, representing the passageway passengers pass through from the security gate exit to the baggage screening entrance, which typically includes queuing space, guide rails, and other physical structures; a baggage screening entrance node, representing the starting position where passengers place their baggage and the baggage slides onto the conveyor belt; a baggage screening internal passage node, representing the enclosed passageway through which baggage moves along the conveyor belt inside the X-ray or CT machine; and a baggage screening exit node, representing the physical location where baggage leaves the screening equipment after scanning. The edges between these nodes represent mandatory passage paths, i.e., the passage order that passengers or baggage must follow in the normal security check process. For example, from the human body security gate exit, one must pass through the buffer area to reach the baggage screening entrance, and after entering the screening entrance, baggage must pass through the internal passageway in sequence to reach the exit.

[0024] To support subsequent time window predictions, this unit stores historical expected travel times on each edge of the directed acyclic graph (DAG). Specifically, for edges on passenger travel paths (such as from the security checkpoint exit to the baggage checkpoint entrance), the system collects the actual travel time of each passenger completing that segment in real time and dynamically updates the historical expected travel time using an exponentially weighted moving average. This ensures that the statistical value reflects both long-term stable travel patterns and responds appropriately to recent changes in passenger flow. For edges on baggage transport paths (such as the time baggage takes to travel in internal passageways), since the conveyor belt speed is relatively constant, the historical expected travel time is mainly calculated based on the conveyor belt length and nominal speed, and stored as Gaussian distribution parameters (mean and standard deviation) for subsequent confidence assessment. This DAG structure provides clear path topological constraints for passenger trajectory estimation and cross-device mapping, enabling trajectory generation and correlation analysis to be performed within well-defined physical boundaries.

[0025] The conveyor belt kinematic embedding unit is used to acquire the conveyor belt speed parameters of the baggage inspection equipment and establish a deterministic position function of the baggage in the baggage inspection channel. The deterministic position function takes the moment when the baggage enters the baggage inspection entrance as the starting point and the conveyor belt speed as a constant rate to predict the longitudinal position of the baggage at any time. The conveyor belt kinematic embedding unit is responsible for quantifying the movement of baggage within the inspection equipment into a predictable deterministic function. Since the movement of baggage after entering the X-ray or CT scanner entrance is entirely driven by the conveyor belt, and the conveyor belt speed is kept relatively constant by the equipment control system, the longitudinal position of the baggage within the aisle can be represented as a deterministic function of time.

[0026] Specifically, this unit first obtains the conveyor belt speed parameters from the baggage inspection equipment. These parameters are typically calculated from the encoder pulse frequency fed back in real-time by the equipment control system, and are expressed in meters per second. For any piece of baggage, when it triggers the photoelectric sensor at the baggage inspection entrance, the system records its entry time. Subsequently, the longitudinal position of the baggage within the inspection tunnel is determined by the following deterministic position function: , in, Indicates that the luggage is The vertical position at any given time, in meters; This indicates the current operating speed of the conveyor belt, expressed in meters per second. This indicates the moment the baggage enters the machine inspection entrance, in seconds. Longitudinal position is defined as the distance from the front end or center of gravity of the baggage along the conveyor belt direction to the plane of the machine inspection channel entrance. hour, This indicates that the luggage has just entered the entrance; as time goes on, The baggage continues to grow until it reaches the machine inspection exit.

[0027] This deterministic position function provides a physical driving model for baggage trajectory prediction. In actual operation, this unit also receives hard anchor point signals triggered by sensors at the machine inspection exit: when baggage triggers the exit photoelectric sensor, the system forcibly resets the baggage's longitudinal position to the full length of the machine inspection channel at that moment and outputs a trajectory completion marker to calibrate any accumulated errors in the prediction model. Furthermore, this unit calculates the prediction confidence level of the baggage trajectory based on the degree of deviation between the deterministic position function and the video observation position (provided by visual tracking results in the video stream of the baggage inspection entrance area). This confidence level reflects the reliability of the current baggage trajectory; it is high when visual observation and physical model prediction consistently match, and decreases when anomalies occur (such as baggage seemingly stuck in the channel, conveyor belt slippage, etc.). This confidence level is then output to the consistency fusion and judgment module as an important basis for association decisions.

[0028] The passenger trajectory estimation unit is used to divide the buffer area into discrete location grids along the channel direction and establish a hidden Markov model of the passenger on the discrete location grids. The state transition probability of the hidden Markov model is determined by the average walking speed and the variance of the crowd density in the current time period. The passenger trajectory estimation unit is responsible for continuously tracking and predicting the location of passengers in spaces without mandatory physical constraints, such as buffer zones. Unlike the deterministic movement of luggage on the conveyor belt, the walking speed of passengers in the buffer zone exhibits individual differences and is easily affected by factors such as queuing congestion and passengers stopping ahead, displaying strong randomness. Therefore, this unit adopts an estimation method based on discretized spatial grids and implicit Markov models.

[0029] This unit first divides the buffer area along the aisle direction into discrete position grids. The granularity of the grid is determined based on the physical dimensions of average human shoulder width and stride, typically set to 0.4 to 0.6 meters, so that each grid cell roughly corresponds to the space occupied by an adult passenger in the aisle. This granularity ensures the positioning accuracy of passengers while avoiding the state space explosion problem caused by overly fine division. In the direction perpendicular to the aisle, since physical facilities such as railings restrict the lateral movement of passengers, only a single column of grid is usually divided. If the site has a multi-aisle parallel layout, the number of columns is determined according to the actual number of aisles.

[0030] On the aforementioned discrete location grid, this unit establishes a hidden Markov model of passenger location. The state variable of the model is the passenger's current grid position, and the observation variable is the passenger detection result extracted from the video stream of the security gate area. The state transition probability is jointly determined by the average walking speed and the variance of crowd density in the current time period: when the crowd density is low, passengers tend to move at a speed close to the average, and the probability of transitioning to the adjacent grid ahead is higher; when the crowd density is high or congestion occurs, the probability of passengers staying in place or moving slowly increases significantly. The observation probability is defined based on the confidence score of the target detector: when the video detector detects a passenger target near the current grid position and the detection confidence is high, that grid has a higher probability of becoming the current observation as a hidden state; if no target is detected or the detection confidence is low, the observation probability decreases accordingly. Through the forward-backward algorithm, this unit can estimate the most likely location of the passenger in the buffer area and its uncertainty distribution in real time.

[0031] For example, the forward-backward algorithm includes: setting the initial state probability distribution for each grid position based on the initial time and initial position of the passenger entering the buffer area, typically setting the grid corresponding to the exit of the security gate as the initial high-probability position. At each observation time, the algorithm performs forward recursion, predicting the prior probability distribution of the passenger in each grid at the current time based on the state probability and state transition probability of each grid at the previous time; then, combined with the observation probability at the current time (i.e., the detection confidence output by the video detector at each grid position), it calculates the posterior probability distribution of each grid at the current time, i.e., the forward probability. Simultaneously, the forward-backward algorithm performs backward recursion from future time to the current time, calculating the backward probability, which represents the impact of the future state distribution on the current state under the current observation sequence conditions. Finally, the forward and backward probabilities are combined to obtain the smooth probability distribution of the passenger in each grid position at the current time, where the grid with the highest probability is the most likely position of the passenger at that time, and the dispersion of the probability distribution reflects the uncertainty of the position estimation. Through the recursive calculation at each time, this unit realizes real-time estimation and tracking of the continuous positions of passengers within the buffer area.

[0032] In addition, this unit also receives conveyor belt status markers from the consistency fusion and decision module. When the conveyor belt status markers indicate an anomaly (such as conveyor belt speed fluctuations, baggage backlog, etc.), this unit expands the state transition variance parameter of the hidden Markov model and switches to a high-uncertainty operating mode, allowing for a larger range of jumps in passenger location estimation to cope with changes in passenger behavior patterns caused by baggage system anomalies. When the conveyor belt status markers return to normal, this unit restores the standard state transition variance parameter and returns to the normal operating mode. This adaptive adjustment mechanism enables passenger trajectory estimation to remain synchronized with the operating status of the baggage system.

[0033] The cross-device mapping unit is used to generate a deterministic time window for each passenger passing through the security gate based on the historical expected passage time. The boundary of the deterministic time window is calculated by superimposing the historical expected passage time with the allowable deviation. Baggage entering the baggage inspection entrance within the deterministic time window is included in the passenger's candidate association set.

[0034] The cross-device mapping unit is responsible for generating a set of candidate baggage associations for each passenger passing through the security gate, thus achieving a spatiotemporal bridge between the security check area and the baggage inspection area. The core mechanism of this unit is to generate a deterministic time window based on historical expected passage time, thereby reducing the search scope of passenger and baggage associations from global to local spatiotemporal neighborhood.

[0035] For each passenger passing through the security gate, this unit first obtains the precise time of their passage through the security gate, denoted as . Subsequently, from the directed acyclic graph stored in the rigid topology modeling unit, the historical expected passage time of the edge corresponding to the exit node of the security gate and the entrance node of the baggage inspection machine is read, denoted as . This expected travel time reflects the average time it takes for passengers to complete this route under normal passenger flow conditions. However, due to factors such as individual walking speed differences and variations in congestion levels in buffer zones, the actual travel time will inevitably fluctuate around the expected value. Therefore, this unit introduces an allowable deviation parameter, denoted as . This is used to quantify the reasonable range of such fluctuations.

[0036] The upper and lower boundaries of the deterministic time window are determined by the following formula: , , in, This indicates the lower boundary of the deterministic time window, which is the earliest time a passenger might arrive at the baggage check entrance area; This represents the upper boundary of the deterministic time window, which is the latest time a passenger might arrive at the baggage checkpoint area. This is the average of historical expected travel times, with the same unit as time, to ensure that the dimensions on both sides of the equation are consistent. It represents the standard deviation of historical travel time, and also has a time dimension, reflecting the degree of fluctuation in travel time; This is the confidence coefficient, typically set to 2 or 3, indicating that the time window is extended to cover approximately 95% or 99% of normal traffic conditions. Tolerance deviation. The introduction of this feature allows the width of the time window to be adaptively adjusted according to the dispersion of historical data: when the travel time fluctuates greatly, the time window automatically widens to accommodate more possibilities; when the travel time is highly stable, the time window narrows to achieve more precise correlation constraints.

[0037] The physical meaning of the aforementioned deterministic time window is clear: only baggage entering the baggage screening entrance within this time window has the potential to have a spatiotemporal relationship with the corresponding passenger. This unit compares the time window boundary with the moment baggage enters the screening entrance, and identifies all baggage that meets the criteria... The baggage is included in the passenger's candidate association set, among which This is the moment when the baggage triggers the entrance photoelectric sensor. This candidate association set is then output to the consistency fusion and determination module as input data for subsequent one-to-one attribution determination.

[0038] Furthermore, such as Figure 2 As shown, the conveyor belt kinematic embedding unit is also used for: S201 receives the hard anchor point signal triggered by the machine inspection exit sensor. When the hard anchor point signal is detected, the longitudinal position of the luggage is reset to the full length of the machine inspection channel, and the trajectory completion mark is output. At the exit of the baggage inspection equipment, this system deploys through-beam photoelectric sensors as hard anchor signal sources. The sensor consists of a transmitter and a receiver, installed on opposite sides of the conveyor belt exit channel. When baggage moves to the exit position with the conveyor belt, the baggage itself blocks the light path, causing a signal level change at the receiver and triggering an exit event signal. This signal is acquired via the digital input interface of the equipment controller and reported in real-time to the edge computing node via industrial Ethernet. The data receiving and synchronization module encapsulates this signal with a unified timestamp and transmits it to the conveyor belt kinematics embedding unit. Upon detecting the hard anchor signal, this unit immediately performs a position reset operation: forcibly setting the current longitudinal position of the baggage to the full length of the inspection channel, which is the physical length of the conveyor belt between the inlet and outlet photoelectric sensors, in meters. Simultaneously, this unit outputs a trajectory completion marker, indicating that the baggage has completed its entire journey within the inspection channel. Subsequent modules can use this marker to release tracking resources associated with the baggage or perform final confirmation.

[0039] During the movement of luggage along the conveyor belt, this unit maintains two position estimation paths simultaneously: one based on the physical model's predicted position output by a deterministic position function, and the other based on the video observation position output by the visual tracking algorithm in the baggage inspection entrance area video stream. The estimated values ​​of both paths are recorded at every moment, with the video observation position transformed to the same longitudinal coordinate system as the physical model using camera calibration parameters. This unit calculates the absolute value of the deviation between the two, i.e., the longitudinal distance difference between the physical model's predicted position and the video observation position, in meters. This deviation value reflects the consistency between the physical model and the visual observation: a small deviation indicates smooth conveyor belt operation, no abnormal luggage slippage, and high prediction reliability; a large deviation may indicate conveyor belt speed fluctuations, luggage deviation or jamming, or visual tracking interference. Based on this deviation value, this unit calculates the prediction confidence level, ranging from 0 to 1. When the deviation approaches zero, the confidence level approaches 1; when the deviation exceeds a preset deviation tolerance threshold, the confidence level drops to zero. The prediction confidence level is then output to the consistency fusion and judgment module as an important part of the calculation of the association evidence score, enabling subsequent modules to dynamically adjust the weight of the association decision based on the credibility of the trajectory.

[0040] S202, based on the degree of deviation between the deterministic position function and the video observation position, calculates the prediction confidence of the luggage trajectory and outputs the prediction confidence to the consistency fusion and judgment module.

[0041] When this unit outputs a trajectory completion marker, this marker, along with the final trajectory data of the baggage, is transmitted to the consistency fusion and judgment module. Upon receiving the trajectory completion marker, this module performs two operations: First, for baggage that has completed the machine inspection process, if it has been associated with a passenger, the association status is upgraded from "temporary association" to "confirmed association," and the passenger's status is transferred to the next stage. Second, if the baggage has not been associated with any passenger during the machine inspection process, it is marked as "isolated baggage," triggering an anomaly handling process, and the handling instruction push module sends a manual review prompt to the on-site terminal. Furthermore, the trajectory completion marker can also be used to notify the trajectory generation module to release tracking context resources related to the baggage, including video tracker instances and physical model state caches, to reduce the memory footprint of edge computing nodes.

[0042] Furthermore, the passenger trajectory estimation unit is also used for: The system receives the conveyor belt status flags from the consistency fusion and judgment module. When the conveyor belt status flags indicate an anomaly, the system increases the state transition variance parameter of the implicit Markov model and switches to a high uncertainty operation mode. The conveyor belt status flag is generated by the global status maintenance unit in the consistency fusion and judgment module and fed back to the passenger trajectory estimation unit. This flag indicates the current operational health status of the baggage conveyor system. Specifically, the global status maintenance unit continuously monitors the conveyor belt encoder data reported by the baggage inspection equipment, as well as the actual time taken for multiple pieces of baggage to pass through the inspection channel. When the conveyor belt speed fluctuation exceeds a preset fluctuation threshold, the conveyor belt starts and stops multiple times within a unit of time, or the actual transmission time of multiple pieces of baggage deviates significantly from the historical average within the same time period and is accompanied by a general decrease in prediction confidence, the global status maintenance unit sets the conveyor belt status flag to abnormal. This flag is fed back to the passenger trajectory estimation unit via an internal data channel as a trigger signal for adjusting the trajectory estimation parameters.

[0043] In the implicit Markov model maintained by the passenger trajectory estimation unit, the state transition variance parameter controls the possible range of changes in passenger position between adjacent time steps. The physical meaning of this parameter lies in quantifying the system's confidence in the prior assumption that passengers move at a normal, uniform speed. Specifically, when the state transition variance is small, the model assumes that passengers can only move to a few adjacent grids within a unit of time, limiting the range of speed changes. This corresponds to the scenario of passengers walking smoothly under normal passenger flow conditions. When the state transition variance increases, the model allows passengers to traverse more grids within a unit of time, accepting a wider range of speed changes, thus adapting to non-uniform movement patterns caused by abnormal situations such as congestion ahead, sudden stops, rapid queue jumping, or turning back.

[0044] The passenger trajectory estimation unit switches between two operating modes based on the received conveyor belt status markers. When the conveyor belt status marker is abnormal, the unit expands the state transition variance parameter of the implicit Markov model to several times the standard value, switching to a high-uncertainty operating mode. In this mode, passenger position estimation no longer strictly relies on the assumption of uniform movement, but instead assigns higher weight to the observed data, allowing the estimation results to respond more quickly to changes in actual passenger behavior. However, to avoid frequent mode switching due to brief conveyor belt signal jitter, the unit implements a switching hysteresis mechanism: switching from the standard mode to the high-uncertainty mode is only performed when the conveyor belt status marker remains abnormal for more than a preset abnormal duration threshold (e.g., 3 seconds); similarly, the standard state transition variance parameter is restored and the unit switches back to the standard operating mode only when the conveyor belt status marker returns to "normal" and remains so for more than a preset stable duration threshold (e.g., 5 seconds). This hysteresis mechanism effectively suppresses oscillations between modes, ensuring the stability of passenger trajectory estimation.

[0045] When the conveyor belt status indicator shows normal, restore the standard state transition variance parameters of the implicit Markov model.

[0046] Furthermore, the consistency fusion and determination module includes: The global status maintenance unit is used to maintain the passenger status table and baggage occupancy table. The passenger status table records the current status of each passenger and the associated spatial interest area parameters. The baggage occupancy table records the associated baggage identifiers in a hard-marking manner. The global state maintenance unit is responsible for maintaining two core data structures: the passenger status table and the baggage occupancy table. These two tables run as in-memory databases on edge computing nodes, supporting high-frequency read and write operations, and are used to record the real-time status of all passengers and baggage in the security check process.

[0047] The passenger status table uses the passenger identifier as the primary key. This identifier is assigned by the trajectory generation module when the passenger first passes through the security gate and is typically in a globally unique identifier format. Each record in the table contains the following fields: passenger identifier, current status, spatial focus area parameters, deterministic time window boundaries, associated baggage identifier, and status update timestamp. The current status field uses enumerated values ​​to represent the stage the passenger is in during the security check process, including initial status, waiting for baggage, in progress during machine inspection, in progress during re-inspection, and re-inspection completed. The system's focus and behavioral expectations for the passenger differ in different states.

[0048] The baggage occupancy table uses the baggage identifier as the primary key, which is assigned by the trajectory generation module when the baggage first triggers the baggage inspection sensor at the machine entrance. Each record in the table contains the following fields: baggage identifier, occupancy flag, occupant passenger identifier, and occupancy timestamp. The hard-marking method is implemented as follows: when a piece of baggage is associated with a passenger, the system sets the occupancy flag of that baggage to true and records the passenger identifier that occupies the baggage. Once baggage is occupied, subsequent association decisions by any other passenger cannot include that baggage in the candidate set, thus ensuring one-to-one exclusivity.

[0049] The association evidence calculation unit is used to calculate the association evidence score of each piece of luggage in the candidate association set for passengers who are waiting for luggage. The association evidence score is obtained by multiplying the trajectory confidence score output by the trajectory generation module with the time proximity weight. The correlation evidence calculation unit is responsible for calculating the correlation evidence score for each piece of luggage in the candidate correlation set for each passenger who is waiting for luggage. This score comprehensively reflects the credibility of the luggage trajectory and the temporal proximity between the luggage and the passenger.

[0050] The formula for calculating the relevance score is as follows: , in, denoted as the evidence score for the association between passenger i and baggage j, dimensionless, with a range of [0,+∞). The confidence score of the trajectory of luggage j is represented by the output of the conveyor belt kinematic embedding unit. It is dimensionless and has a value range of [0,1], reflecting the reliability of the luggage trajectory prediction. This indicates the moment when passenger i passed through the security gate, provided by the security gate passage event data; The time when baggage j enters the baggage inspection entrance is indicated by a trigger signal from the photoelectric sensor at the baggage inspection entrance.

[0051] This represents the time proximity weighting function, which is dimensionless, and its input parameters are... This is the time difference between the moment baggage enters the machine inspection entrance and the moment the passenger passes through the security gate, expressed in seconds.

[0052] Time proximity weighting function The specific form is designed based on the statistical characteristics of deterministic time windows, using a Gaussian kernel function, and its expression is: , in, This represents the average historical expected travel time, in seconds. The standard deviation of historical transit times is given in seconds. Both parameters are derived from statistical information stored on the edges of the directed acyclic graph by the rigid topology modeling units. The physical meaning of this Gaussian kernel function is clear: the difference between the time baggage arrives and the time passenger passes through. Exactly equal to the historical expected travel time At that time, the weight is set to the maximum value of 1, indicating that the baggage perfectly matches the passenger's time; as... Deviation The weights decay smoothly in the form of a Gaussian curve, and the decay rate is determined by the standard deviation. Control, when historical travel time fluctuations are small, The weight is relatively small, decays rapidly, and strict screening is applied to luggage that deviates from the designated time window; when historical travel times fluctuate significantly... The weights are relatively large, and the weight decays gradually, maintaining a certain tolerance for luggage that deviates from the time window. This design ensures that the weight function is consistent with the statistical basis of the time window, while avoiding the discontinuity problem of piecewise functions at the boundaries, thus ensuring the smoothness and stability of the associated decisions.

[0053] The hard boundary check unit is used to query the baggage occupancy table before calculating the associated evidence score. If the candidate baggage has been marked as occupied, the associated evidence score of the candidate baggage is forced to be set to negative infinity. The hard boundary check unit intervenes after the associated evidence calculation unit completes the score calculation and before the greedy selection unit executes the selection. This unit receives the candidate baggage score list output by the associated evidence calculation unit and queries the baggage occupancy table to perform a hard boundary check before sorting the candidate sets for each passenger.

[0054] The specific timing of the check is as follows: When traversing the candidate baggage set for passenger i, for each baggage j in the set, the hard boundary check unit first queries the corresponding occupancy flag field in the baggage occupancy table based on the baggage identifier. If the occupancy flag is true, it means that the baggage has been occupied by another passenger. At this time, the hard boundary check unit forcibly sets the association evidence score of the baggage to negative infinity. Negative infinity is usually represented by a minimum value, and its purpose is to ensure that the baggage will never be selected in the subsequent greedy selection process. If the occupancy flag is false, the original score remains unchanged. Through this hard boundary check mechanism, the system strictly guarantees that a baggage can only be associated with at most one passenger, thus preventing conflicting associations from occurring at the data level.

[0055] The greedy selection unit is used to select the maximum value among the associated evidence scores. If the maximum value exceeds the confidence threshold and the corresponding luggage is not occupied, a state transition is performed and the luggage is marked as occupied. If multiple associated evidence scores are close and the corresponding luggage is not occupied, a delayed decision instruction is output to the trajectory generation module to trigger the observation accumulation mode. If there is still no clear maximum value after the waiting time limit is exceeded, a state reset is performed and the occupation mark is released.

[0056] The greedy selection unit is responsible for performing the final one-to-one allocation determination for each passenger's candidate baggage set based on hard boundary checks, and outputting the decision result.

[0057] The confidence threshold is set using a preset fixed value, determined based on actual operational experience and configurable during system initialization. In typical intercity transportation security check scenarios, the confidence threshold is usually set to 0.7, meaning that the system considers the association sufficiently credible and can immediately confirm it only when the score of the associated evidence reaches 0.7 or higher. For scenarios with high security requirements or dense passenger flow, the threshold can be appropriately increased to 0.85 to reduce the risk of false associations; for periods with sparse passenger flow and low waiting tolerance, the threshold can be appropriately decreased to 0.6 to speed up association confirmation and reduce passenger waiting time. This threshold can be dynamically adjusted through the management interface during system operation.

[0058] For each passenger in the waiting baggage state, the greedy selection unit first obtains the score list of their candidate baggage after hard boundary checks and finds the maximum value. If the maximum value exceeds the confidence threshold and the corresponding baggage is not occupied (the hard boundary check guarantees this condition), a state transition is performed: the passenger's state is changed from waiting baggage to baggage under inspection, the associated baggage identifier is recorded in the passenger state table, and the baggage's occupancy flag is set to true in the baggage occupancy table, while the occupancy passenger identifier is recorded. This process achieves a one-to-one confirmation association between the passenger and the baggage.

[0059] When multiple pieces of evidence with similar scores exist in the candidate baggage set, the greedy selection unit does not make a decision immediately but instead outputs a delayed decision instruction to the trajectory generation module. The specific triggering condition is that the difference between the maximum and second-largest scores is less than a preset discrimination margin threshold, and the maximum score itself is not significantly higher than a confidence threshold. The delayed decision instruction carries the passenger identifier and a list of identifiers for the current candidate baggage set, instructing the trajectory generation module to accumulate more observation data for that passenger before recalculating the evidence score. Upon receiving the delayed decision instruction, the trajectory generation module enters an observation accumulation mode: on the one hand, it waits for the next frame of video data to obtain more precise visual observation positions of the baggage to update the trajectory confidence score; on the other hand, it waits for the baggage to continue moving within the inspection channel. When the baggage triggers the next sensor (such as the photoelectric sensor in the middle of the inspection channel), the system can obtain richer spatiotemporal observation data, thereby updating the baggage trajectory confidence score. After a preset observation accumulation period, the trajectory generation module recalculates the evidence score and pushes it to the consistency fusion and judgment module, triggering a new round of greedy selection judgment.

[0060] To prevent passengers from waiting indefinitely in the buffer area due to delayed baggage confirmation, this unit sets a waiting time limit. This waiting time limit is the maximum allowed waiting time for a passenger to make a baggage confirmation decision from the moment they enter the baggage waiting state. This time limit is determined based on human factors engineering and on-site operational experience, comprehensively considering passengers' psychological tolerance for waiting, security checkpoint throughput requirements, and baggage inspection equipment processing cycles. In intercity transportation security scenarios, this time limit is typically set to 5 to 8 seconds. If a passenger fails to complete the confirmation process within the waiting time limit while in the baggage waiting state, the system performs a status reset: marking the passenger's status as abnormally terminated, releasing any temporary resources occupied by the passenger, and sending a manual intervention request to the on-site terminal via the disposal instruction push module, prompting security personnel to manually confirm the passenger's baggage ownership. Simultaneously, if there are unoccupied bags in the passenger's candidate baggage set, the system will release the candidate association markers for these bags, allowing them to be matched with other subsequent passengers, avoiding single-point delays that could affect the overall security check process.

[0061] In one specific implementation, such as Figure 3 As shown, the system in this application deploys multiple types of sensors at the physical device layer, including a security gate (outputting passage event data, human body scan data, and body temperature data), a baggage X-ray machine (outputting image sequences, photoelectric trigger signals, and conveyor belt encoding data), a video surveillance system (including security gate area cameras, machine inspection entrance cameras, and buffer channel cameras), and auxiliary sensors (such as facial recognition cameras at the re-inspection channel entrance). The multi-source data collected by these devices is uniformly accessed by the edge computing node.

[0062] The edge computing node contains a data receiving and synchronization module. This module performs time synchronization, format standardization, and spatial coordinate alignment on multi-source data, providing raw sequence data with a unified spatiotemporal reference for subsequent processing. The data then flows into the trajectory generation module, which consists of four collaborative units: a rigid topology modeling unit abstracts the physical path of the security check area into a directed acyclic graph (DAG), with nodes including the exit of the security gate, buffer area, baggage inspection entrance, baggage inspection internal passage, and baggage inspection exit, used to describe the mandatory passage path; a conveyor belt kinematics embedding unit establishes a deterministic position function based on the conveyor belt speed and baggage entry time, calibrates the position using hard anchor point signals, and outputs baggage trajectory confidence; a passenger trajectory estimation unit divides the buffer area into a discrete grid, constructs an implicit Markov model, and executes a forward-backward algorithm to achieve continuous estimation of passenger positions; and a cross-device mapping unit generates a deterministic time window for each passenger based on the statistical characteristics of historical passage times, including baggage entering within the time window into a candidate association set. The continuous spatiotemporal trajectory data output by the trajectory generation module is simultaneously fed into a disturbance identification and filtering module and a consistency fusion and judgment module.

[0063] The consistency fusion and judgment module includes a global state maintenance unit (maintaining the passenger status table and baggage occupancy table), a correlation evidence calculation unit (calculating correlation evidence scores based on baggage trajectory confidence and time proximity weights), a hard boundary check unit (querying the baggage occupancy table and invalidating the scores of occupied baggage), and a greedy selection unit (performing one-to-one correlation confirmation based on confidence thresholds and outputting passenger baggage correspondence). The disturbance identification and filtering module analyzes trajectory data for temporal violation, queue order, accompaniment, state conflict, and timeout delay to identify abnormal behavior. The handling instruction push module integrates risk events and their corresponding relationships into handling instructions and pushes them to on-site terminals (such as handheld terminals for security personnel and channel displays), completing the closed loop from perception to handling.

[0064] Furthermore, the global state maintenance unit is also used for: Based on the current status of the passenger, a spatial interest region clipping instruction is dynamically generated and output to the trajectory generation module. The spatial interest region clipping instruction is used to instruct the implicit Markov model to perform the forward-backward algorithm only on the clipped sub-grid. Receive the trajectory confidence score output by the trajectory generation module and update the transfer ready flag in the passenger status table.

[0065] The global state maintenance unit dynamically generates spatial interest region clipping instructions based on the passenger's current state and outputs them to the trajectory generation module. Specifically, when a passenger is waiting for baggage, their interest region is the buffer channel between the security gate exit and the baggage screening entrance, which is a long and narrow strip. When the passenger enters the baggage screening process, the interest region switches to the internal baggage screening channel. When the passenger enters the re-screening process, the interest region switches to the re-screening channel. The clipping instructions carry the grid index range of the target region. After receiving the instructions, the trajectory generation module clips the state space of its implicit Markov model from the complete grid set covering the entire buffer region to the specified sub-grid set. The forward-backward algorithm only performs inference on this sub-grid, thus significantly reducing the computational load.

[0066] The transition ready flag is updated by the trajectory generation module based on the trajectory confidence score. When a passenger's trajectory confidence score remains stable above a preset threshold for multiple consecutive time steps, the trajectory generation module sets the passenger's transition ready flag to valid and pushes it to the global state maintenance unit. This flag is used to trigger the state transition decision of the greedy selection unit, indicating that the passenger has met the trajectory quality conditions for entering the next state.

[0067] Furthermore, the global state maintenance unit is also used for: Aggregate the local conveyor belt speed observations output by the trajectory generation module corresponding to multiple passengers, and calculate the group speed deviation; If the group speed deviation exceeds the abnormal deviation threshold and continues to exceed the preset baggage cycle number, the conveyor belt status mark will be set to abnormal and output to the trajectory generation module and the field terminal. If the group speed deviation returns to normal and continues to exceed the preset stable cycle number, the conveyor belt status flag is set to normal, and the global conveyor belt speed belief value is updated.

[0068] The global state maintenance unit aggregates local conveyor belt speed observations from multiple bags in the trajectory generation module to form a group speed perception. Specifically, for all bags currently in the inspection channel, the module extracts the instantaneous conveyor belt speed observations output by the conveyor belt kinematic embedding unit and calculates the median of these observations as a representative value of the group speed. The group speed deviation is defined as the absolute difference between this representative value and the rated speed of the conveyor belt, in meters per second.

[0069] The abnormal deviation threshold is set to ±15% of the rated speed. For example, if the rated speed is 0.3 m / s, an anomaly is triggered when the group speed is below 0.255 m / s or above 0.345 m / s. Similarly, when the rated speed is 0.4 m / s, the anomaly threshold is below 0.34 m / s or above 0.46 m / s. The preset baggage cycle count is set to 3 cycles, where each baggage cycle is defined as the average time interval from when a piece of baggage enters the inspection entrance to when it triggers the exit sensor. An abnormal state lasting more than 3 baggage cycles will trigger the conveyor belt status flag to be set to abnormal, to avoid false triggers caused by jamming of a single piece of baggage or momentary sensor vibration. The stabilization cycle count is set to 5 cycles, meaning the status flag will only be restored to normal after the group speed deviation has returned to normal for 5 baggage cycles.

[0070] The global conveyor belt speed belief value is a fusion estimate of the current actual conveyor belt speed, obtained by the global state maintenance unit through exponential weighted smoothing based on the group speed median and historical belief values. This belief value is fed back to the conveyor belt kinematic embedding unit to correct the conveyor belt speed parameters in the deterministic position function, enabling the physical model to adaptively track the actual operating state of the conveyor belt.

[0071] Furthermore, the passenger status table includes: The initial state indicates that the passenger has not passed through the security gate; The "Waiting for Baggage" status indicates that the passenger has passed through the security gate and the definitive time window has opened. The corresponding spatial focus area is from the security gate exit to the baggage inspection entrance. The "Under Inspection Status" indicates that the baggage has entered the baggage inspection machine and the associated information has been locked. The corresponding area of ​​interest is the baggage inspection machine's internal passageway. The status of re-inspection indicates that the baggage inspection alarm has been triggered and the passenger has entered the re-inspection channel. The corresponding spatial focus area is from the re-inspection channel to the body security gate. The "Re-inspection Completed" status indicates that the re-inspection process is complete and the passenger's baggage has been confirmed.

[0072] The passenger status table defines five states, each corresponding to a specific physical stage in the security check process. The initial state indicates the passenger has not yet passed through the body scanner; the system only performs preliminary detection and identification. When the body scanner passage event is triggered, the passenger enters the baggage waiting state. At this point, the passenger has passed through the scanner and is waiting for their baggage in the buffer area. The system opens a deterministic time window and establishes a candidate association set. Once the greedy selection unit completes the association locking, the passenger enters the machine inspection state. At this point, the baggage is already in the machine inspection channel, and the system continuously tracks the baggage's location and awaits the inspection results. If the baggage inspection equipment outputs an alarm signal, the passenger enters the re-inspection state. The system guides the passenger to the re-inspection channel and switches the focus area to the area between the re-inspection channel and the body scanner. After the re-inspection is completed, the passenger enters the re-inspection completed state, and the system confirms the passenger and baggage correspondence and releases relevant resources.

[0073] The design of the spatial focus area and state binding relationship is based on physical constraints: when waiting for luggage, passengers must be in the buffer channel; when undergoing machine inspection, passengers have left the buffer area and are waiting to retrieve their bags; and when undergoing re-inspection, passengers are in the re-inspection channel. This binding allows the trajectory generation module to focus its computing resources on the actual spatial range where passengers may appear, avoiding indiscriminate detection and tracking throughout the entire area.

[0074] Furthermore, the global state maintenance unit is also used for: When a passenger is in the re-inspection process, the system receives a reverse detection signal triggered by the sensor at the entrance of the re-inspection channel, extracts the currently detected passenger identity features, and calculates the similarity between the passenger's original inspection record in the historical passenger feature database. If the similarity calculation result exceeds the identity confirmation threshold, the passenger is confirmed to be the same physical person as the original inspection person. The passenger status in the passenger status table is transferred to the re-inspection completed status, and the passenger baggage correspondence established during the original inspection is retained as a valid association. If the similarity calculation result does not exceed the identity verification threshold, it is determined to be a misidentification. A new initial status record is created for the passenger in the passenger status table, and the original status record in the re-inspection is marked as abnormally terminated and the baggage association mark it occupies is released.

[0075] A facial recognition camera is deployed at the entrance of the re-inspection channel as a sensor. When a passenger enters the re-inspection channel, the camera captures the passenger's facial image and triggers a reverse detection signal, indicating that the passenger is returning from the re-inspection channel to the security check area for a second inspection. The system extracts facial feature vectors from the captured facial images, typically using a deep neural network to generate 128-dimensional or 512-dimensional feature representations.

[0076] The historical passenger feature database stores the facial feature vectors collected when each passenger passes through the security gate, along with their associated original inspection records. When a re-inspection process is triggered, the system quickly retrieves the passenger's original inspection feature vector using the passenger identifier as an index, and performs a cosine similarity calculation with the currently collected feature vector. The similarity value ranges from -1 to 1, with a higher value indicating a higher degree of identity matching.

[0077] The identity verification threshold is set to 0.75, meaning that when the cosine similarity reaches 0.75 or higher, the passenger is confirmed as the same person. For high-security scenarios, the threshold can be increased to 0.85 to reduce the false recognition rate; for scenarios with high passenger flow and requiring rapid passage, it can be reduced to 0.70 to improve throughput. If the similarity calculation result exceeds the threshold, the passenger's identity is confirmed, and the status transitions to "re-inspection completed," retaining the passenger and baggage association established during the original inspection. If the threshold is not exceeded, it is determined to be a false identity recognition. The system creates a new initial status record for the passenger in the passenger status table, marks the original re-inspection status record as abnormally terminated, and releases the baggage association marker occupied by that record, allowing the baggage to be associated with the correct new passenger.

[0078] Furthermore, the disturbance identification and filtering module includes: The timing violation calculation unit is used to calculate the degree of deviation between the time when a passenger passes through the security gate and the time when their baggage enters the machine inspection entrance, relative to the historical expected passage time, and to identify cases of early arrival or late arrival based on the direction of deviation. The queue order verification unit is used to maintain the queue of passengers who have passed through the security gate and are waiting for their luggage. The queue is sorted by the time of passage. When the luggage enters the machine inspection entrance, the associated evidence scores are verified in the order of the queue. If the first passenger to be successfully matched is not at the head of the queue, the queue-jumping behavior is marked. The accompanying quantification unit is used to calculate the spatial accompanying duration and distance variance of luggage with each neighboring passenger in the buffer area, and to identify the proxy carrying behavior based on the concentration of the accompanying duration distribution. The state conflict detection unit is used to detect when the same passenger identifier triggers the security gate passage event twice within a time shorter than the normal process time threshold, or to detect when the passenger trajectory reverses direction, and to mark the re-intrusion behavior or the return behavior respectively. The overstay detection unit is used to detect when the duration of a passenger waiting for luggage exceeds the upper limit of a deterministic time window, and to mark the short stay behavior.

[0079] The timing violation calculation unit uses historical expected transit time as a benchmark to calculate the degree of deviation between the time a passenger passes through the security gate and the time their baggage enters the machine inspection entrance. When the interval is less than the historical expected transit time minus the allowable deviation, it is judged as early arrival; when the interval is greater than the historical expected transit time plus the allowable deviation, it is judged as late arrival. The degree of deviation is quantified as the ratio of the deviation value to the standard deviation.

[0080] The queue order verification unit maintains a first-in-first-out queue, recording the identifiers of passengers who have passed through the security gate and are waiting for their luggage, sorted by their passage time. When luggage enters the machine inspection entrance, the system traverses the queue and verifies the associated evidence scores in turn. If the first successfully matched passenger is not at the head of the queue, it is marked as queue-jumping behavior.

[0081] The accompanying quantification unit calculates the spatial companionship duration and distance variance of luggage within the buffer zone and its proximity to other passengers. Companionship duration is the continuous time during which luggage and passengers are within a spatial distance of less than a threshold (e.g., 1.5 meters); distance variance indicates the degree of fluctuation in the distance between luggage and passengers within this duration. The concentration of the companionship duration distribution is measured using standard deviation; when the standard deviation is less than the threshold, it is considered a proxy carrying behavior.

[0082] The state conflict detection unit sets the normal process time threshold to 10 to 15 seconds. For example, if the time interval between a passenger entering the security gate and triggering the security gate again is less than 10 seconds, it is marked as a re-intrusion behavior; if it is less than 12 seconds, the detection is triggered. The detection of direction reversal points is based on the dot product of the trajectory direction vectors. When the dot product of the direction vectors of two adjacent time steps is negative, it indicates that the passenger's direction of travel has reversed, and it is marked as a backtracking behavior.

[0083] The overstay detection unit uses the upper boundary of the deterministic time window in right 2 as the judgment criterion. When the duration of a passenger waiting for luggage exceeds this upper boundary, it is marked as a short-term stay. After the short-term stay is marked, the system sends a prompt message to the on-site terminal through the handling instruction push module, and dispatches security personnel to pay attention or intervene manually.

[0084] In one specific implementation, such as Figure 4 The diagram illustrates the entire process from event triggering to final association confirmation or anomaly handling. This process begins with two parallel events: when a passenger passes through the security gate, the passage time is recorded, a passenger identifier is assigned, and the status is set to "waiting for baggage"; when baggage enters the machine inspection entrance, the entry time is recorded, a baggage identifier is assigned, and baggage location prediction is initiated. The cross-device mapping unit generates a deterministic time window for the passenger based on the statistical characteristics of historical passage times, including baggage entering within that time window in the passenger's candidate association set. Subsequently, the association evidence calculation unit calculates the association evidence score by combining baggage trajectory confidence and time proximity weights. The hard boundary check unit queries the baggage occupancy table; if a candidate baggage is already occupied, its score is invalidated, ensuring the exclusivity of one-to-one attribution.

[0085] The greedy selection unit makes a decision based on the highest score in the candidate set: if the highest score exceeds the preset confidence threshold, the association is confirmed, the passenger status is changed to "under machine inspection", the corresponding baggage is marked as occupied, and the passenger baggage correspondence is output; if the highest score is close to the second highest score, a delayed decision instruction is output, the observation accumulation mode is entered, and the calculation is recalculated after a new video frame or sensor trigger is waited for; if the confirmation is not made after the waiting time limit is exceeded, the status is reset to abnormal termination, temporary resources are released and a manual intervention instruction is sent.

[0086] When the baggage inspection alarm sounds, the passenger status changes to "in re-inspection," and the focus area switches to the re-inspection lane. Upon triggering the facial recognition camera at the re-inspection lane entrance, the system extracts the passenger's identity features and compares them with the original inspection record in the historical feature database. If the similarity exceeds the identity verification threshold, the passenger is confirmed as the same person, the status changes to "re-inspection complete," and the original association is retained. If the similarity does not exceed the threshold, it is determined to be a misidentification, a new passenger record is created, the original "in re-inspection" status is marked as abnormally terminated, and the baggage tag it occupied is released.

[0087] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. An intercity traffic security inspection system based on edge computing and multi-sensor collaboration, characterized in that, include: The data receiving and synchronization module is used to receive multi-source security inspection raw data, perform time synchronization and standardization processing, and obtain security inspection raw sequence data with a unified spatiotemporal reference. The trajectory generation module is used to detect and track passenger and baggage targets in the security gate area and baggage inspection entrance area based on the original security sequence data, and generate continuous spatiotemporal trajectory data including passenger movement status, baggage transport status and cross-device position mapping relationship; The disturbance identification and filtering module is used to analyze the continuous spatiotemporal trajectory data, identify multiple queue disturbance behaviors, and filter them according to possible related person-bag combinations to obtain candidate correspondence data. The consistency fusion and determination module is used to perform multi-sensor consistency fusion and one-to-one attribution determination based on the candidate correspondence data and the continuous spatiotemporal trajectory data to obtain passenger baggage correspondence data. The handling instruction push module is used to aggregate personal security inspection risks and baggage inspection risks into risk event handling instruction data based on the passenger baggage correspondence data and push it to the on-site terminal; The trajectory generation module includes: A rigid topology modeling unit is used to model the passenger passage path from the exit of the security gate area to the baggage inspection entrance area, as well as the baggage transport path inside the baggage inspection entrance area, into a directed acyclic graph. The nodes of the directed acyclic graph include the security gate exit, buffer area, baggage inspection entrance, baggage inspection internal passage, and baggage inspection exit. The edges represent the forced passage path, and the historical expected passage time of each edge is stored. The conveyor belt kinematic embedding unit is used to acquire the conveyor belt speed parameters of the baggage inspection equipment and establish a deterministic position function of the baggage in the baggage inspection channel. The deterministic position function takes the moment when the baggage enters the baggage inspection entrance as the starting point and the conveyor belt speed as a constant rate to predict the longitudinal position of the baggage at any time. The passenger trajectory estimation unit is used to divide the buffer area into discrete location grids along the channel direction and establish a hidden Markov model of the passenger on the discrete location grids. The state transition probability of the hidden Markov model is determined by the average walking speed and the variance of the crowd density in the current time period. The cross-device mapping unit is used to generate a deterministic time window for each passenger passing through the security gate based on the historical expected passage time. The boundary of the deterministic time window is calculated by superimposing the historical expected passage time with the allowable deviation. The baggage entering the baggage inspection entrance within the deterministic time window is included in the passenger's candidate association set.

2. The intercity traffic security inspection system based on edge computing and multi-sensor collaboration according to claim 1, characterized in that, The conveyor belt kinematic embedding unit is also used for: Receive the hard anchor point signal triggered by the sensor at the machine inspection exit. When the hard anchor point signal is detected, reset the longitudinal position of the luggage to the full length of the machine inspection channel and output the trajectory completion mark. Based on the degree of deviation between the deterministic position function and the video observation position, the prediction confidence of the luggage trajectory is calculated, and the prediction confidence is output to the consistency fusion and judgment module.

3. The intercity traffic security inspection system based on edge computing and multi-sensor collaboration according to claim 1, characterized in that, The passenger trajectory estimation unit is also used for: The system receives the conveyor belt status flag fed back by the consistency fusion and judgment module. When the conveyor belt status flag indicates an anomaly, the system expands the state transition variance parameter of the implicit Markov model. When the conveyor belt status indicator indicates normal, the standard state transition variance parameters of the implicit Markov model are restored.

4. The intercity traffic security inspection system based on edge computing and multi-sensor collaboration according to claim 1, characterized in that, The consistency fusion and determination module includes: The global status maintenance unit is used to maintain the passenger status table and the baggage occupancy table. The passenger status table records the current status of each passenger and the associated spatial interest area parameters, and the baggage occupancy table records the baggage identifiers that have been associated and occupied. The association evidence calculation unit is used to calculate the association evidence score of each piece of luggage in the candidate association set for passengers who are waiting for luggage. The association evidence score is obtained by multiplying the trajectory confidence score output by the trajectory generation module by the time proximity weight. The hard boundary check unit is used to query the baggage occupancy table before calculating the associated evidence score. If the candidate baggage has been marked as occupied, the associated evidence score of the candidate baggage is set to negative infinity.

5. The intercity traffic security inspection system based on edge computing and multi-sensor collaboration according to claim 4, characterized in that, The global state maintenance unit is also used for: Based on the current status of the passenger, a spatial attention region clipping instruction is dynamically generated and output to the trajectory generation module. The spatial attention region clipping instruction is used to instruct the implicit Markov model to perform a forward-backward algorithm on the clipped sub-grid. Receive the trajectory confidence score output by the trajectory generation module and update the transfer ready flag in the passenger status table.

6. The intercity traffic security inspection system based on edge computing and multi-sensor collaboration according to claim 4, characterized in that, The passenger status table includes: The initial state indicates that the passenger has not passed through the security gate; The "Waiting for Baggage" status indicates that the passenger has passed through the security gate and the definitive time window has opened. The corresponding spatial focus area is from the security gate exit to the baggage inspection entrance. The "Under Inspection Status" indicates that the baggage has entered the baggage inspection machine and the associated information has been locked. The corresponding area of ​​interest is the baggage inspection machine's internal passageway. The status of re-inspection indicates that the baggage inspection alarm has been triggered and the passenger has entered the re-inspection channel. The corresponding spatial focus area is from the re-inspection channel to the body security gate. The "Re-inspection Completed" status indicates that the re-inspection process is complete and the passenger's baggage has been confirmed.

7. The intercity traffic security inspection system based on edge computing and multi-sensor collaboration according to claim 6, characterized in that, The global state maintenance unit is also used for: When a passenger is in the re-inspection state, the system receives a reverse detection signal triggered by the re-inspection channel entrance sensor, extracts the currently detected passenger identity features, and calculates the similarity between the passenger's original inspection record in the historical passenger feature database. If the similarity calculation result exceeds the identity confirmation threshold, the passenger is confirmed to be the same physical person as the original inspection person. The passenger status in the passenger status table is transferred to the re-inspection completed status, and the passenger baggage correspondence established during the original inspection is retained as a valid association. If the similarity calculation result does not exceed the identity verification threshold, it is determined to be a misidentification. A new initial status record is created for the passenger in the passenger status table, and the original status record in the re-inspection is marked as abnormally terminated and the baggage association mark it occupies is released.

8. The intercity traffic security inspection system based on edge computing and multi-sensor collaboration according to claim 1, characterized in that, The disturbance identification and filtering module includes: The timing violation calculation unit is used to calculate the degree of deviation between the time when a passenger passes through the security gate and the time when their baggage enters the machine inspection entrance, relative to the historical expected passage time, and to identify cases of early arrival or late arrival based on the direction of deviation. The queue order verification unit is used to maintain the queue of passengers who have passed through the security gate and are waiting for their luggage. The queue is sorted by the time of passage. When the luggage enters the machine inspection entrance, the associated evidence scores are verified in the order of the queue. If the first passenger to be successfully matched is not at the head of the queue, the queue-jumping behavior is marked. The accompanying quantification unit is used to calculate the spatial accompanying duration and distance variance of luggage with each neighboring passenger in the buffer area, and to identify the proxy carrying behavior based on the concentration of the accompanying duration distribution. The state conflict detection unit is used to detect when the same passenger identifier triggers the security gate passage event twice within a time shorter than the normal process time threshold, or to detect when the passenger trajectory reverses direction, and to mark the re-intrusion behavior or the return behavior respectively. The overstay detection unit is used to detect when the duration of a passenger waiting for luggage exceeds the upper limit of a deterministic time window, and to mark the short stay behavior.

9. The intercity traffic security inspection system based on edge computing and multi-sensor collaboration according to claim 1, characterized in that, The multi-source security inspection raw data includes personal security gate passage event data, baggage inspection equipment conveying event data, personal security gate area video stream data, baggage inspection entrance area video stream data, and edge computing node time synchronization status data.

Images