A method and system for intelligent identification and statistics of personnel on gangways in a border inspection scene

Abstract

The application relates to the technical field of electric digital data processing, and discloses a gangway personnel intelligent identification and statistics method and system in a border inspection scene, which comprises the following steps: a one-dimensional manifold constraint premise is constructed, target two-dimensional coordinates in gangway video data are transformed into normalized one-dimensional progress coordinates; a bidirectional topological linked list structure is established, so that each virtual slot position can lock a pilot slot position in real time; a nonlinear speed clamping operation is performed, the inertial speed of the virtual slot position in a blind area is nonlinearly clamped based on a logical scalar difference value, and the speed converges to the pilot slot position speed; and a completeness counting signal is generated. Through the establishment of a virtual elastic chain coupling mechanism, the speed of the target in the blind area is calculated and clamped in real time by using the logical spacing between adjacent slot positions, the incompressible characteristics of a physical queue are simulated, and the logical rear-end or position penetration problems caused by simply relying on an inertial prediction model when congestion wave transmission occurs are solved.

Description

Technical Field

[0001] This invention relates to a method and system for intelligent identification and statistics of personnel on gangways in border inspection scenarios, belonging to the field of electronic digital data technology. Background Technology

[0002] In current border inspection port operations, the identification and counting of personnel boarding and alighting via gangways is a core aspect of ensuring accurate duty scheduling and port security. Currently, the most common approach is based on computer vision video analysis technology. This technology relies on the temporal continuity of the spatial appearance features of the target in two-dimensional images. The system uses convolutional neural networks to extract high-dimensional visual features of the target's texture, color, and contour, and calculates the similarity of feature vectors between adjacent video frames to achieve target detection, tracking, and identity association. This approach maintains high statistical accuracy and stability in general monitoring scenarios with uniform lighting, wide field of view, and high target feature discrimination. Border inspection gangways, as special physical spaces, have a high overhead shooting angle and are long and narrow. Due to the closed passageway and strict physical unidirectional characteristics, existing mainstream visual analysis technologies face the risk of fundamental failure under specific scenario constraints. The high downward angle causes the camera to mainly capture information from the top of the head, shoulders, and back of the personnel. In addition, during winter or specific duty periods, personnel generally wear dark coats or uniforms, making the visual features of targets highly homogeneous. Feature extraction algorithms have difficulty distinguishing adjacent targets. The gantry and sunshade facilities on the gangway create fixed visual blind spots, causing physical interruptions in the target's movement trajectory. When personnel reappear after passing through the blind spot, there is a lack of continuous visual feature guidance. Traditional algorithms are prone to identity jumps or trajectory breaks due to feature matching failures, resulting in distorted statistical data.

[0003] To avoid visual processing challenges, existing technologies employ simplified control logic to achieve counting schemes. However, these schemes fall short when dealing with non-steady-state crowd flow. For example, Chinese invention patent CN104700476A discloses a method and device for counting people in barrier-free passages in civil explosives operation areas. This scheme sets up laser beam sensors at both ends of the passage and uses the sequence of sensor blockages, such as A to D or D to A, to determine the direction of people entering and exiting and accumulate the count. This type of scheme belongs to the boundary gating model. Its core flaw is that it treats the inside of the passage as an unperceptible logical black box. Although it can count the throughput at both ends, it lacks the ability to continuously lock onto the microscopic movement state of targets inside the passage. In long passages of border inspection gangways, if people stop midway, congestion causes nonlinear compression of the queue or generates complex accordion-effect travel waves, simply relying on discrete trigger signals at the entrance and exit cannot reproduce the true topological order inside the passage, which can easily lead to logical judgment errors and cannot eliminate false counts caused by non-biological carriers such as luggage carried by people.

[0004] Therefore, how to construct a method that does not rely on continuous visual features, utilizes scene physical topology constraints to maintain the continuity of target identity, and achieves accurate logical tracking and statistical methods under visual blind spots and non-steady-state pedestrian flow conditions has become the technical problem to be solved by this invention. Summary of the Invention

[0005] To address the problems mentioned in the background art, the technical solution of the present invention is as follows: A method for intelligent identification and statistical analysis of personnel on gangways in border inspection scenarios, comprising the following steps:

[0006] The prerequisite steps for constructing one-dimensional manifold constraints are to receive video stream data of the gangway area, transform the two-dimensional pixel coordinates of the detected targets in the video frame into normalized one-dimensional progress coordinates based on the pre-calibrated homography mapping matrix, and construct a scalar data space representing the passage process in memory.

[0007] The step of establishing a bidirectional topological linked list structure is as follows: in response to the newly added virtual slots generated at the beginning of the scalar data space, all active virtual slots are constructed into a logical linked list with a bidirectional index relationship, and each virtual slot locks its immediate preceding slot in real time.

[0008] The nonlinear velocity clamping operation step is executed. When updating the position of the current virtual slot located in the visual blind spot, the logical scalar difference between the current virtual slot and the pilot slot on the one-dimensional progress coordinate is calculated. Based on the logical scalar difference, the inertial prediction velocity of the current virtual slot is nonlinearly clamped. The preset minimum interpersonal distance threshold parameter is called to participate in the logical operation. When the logical scalar difference approaches the minimum interpersonal distance threshold, the propulsion velocity value of the current virtual slot is converged to the instantaneous velocity value of the pilot slot, and the velocity constraint state is synchronously transmitted in the logical linked list.

[0009] The step of generating a completeness counting signal involves monitoring the numerical evolution of the virtual slot in the scalar data space. When the value completes a continuous evolution from the start end to the end end without triggering an abnormal interruption of the topology sequence number, a counting signal is generated.

[0010] Preferably, the steps for constructing one-dimensional manifold constraints specifically include: reading the entrance baseline coordinate parameters and exit baseline coordinate parameters of the gangway area; calling the perspective transformation algorithm to calculate the transformation matrix that maps the entrance baseline coordinate parameters to a progress value of zero and the exit baseline coordinate parameters to a progress value of one; applying the transformation matrix to establish a dimensionality reduction mapping relationship from the image pixel coordinate system to the one-dimensional progress coordinate system, and converting the motion trajectory data of the detected target along the nonlinear physical path into linear scalar change data along a single dimension in the scalar data space.

[0011] Preferably, the steps for establishing a bidirectional topological linked list structure specifically include: instantiating a data structure object containing a progress state variable, a speed state variable, and a topological sequence number variable for each detection target in memory; performing a sorting operation on all data structure objects according to the value of the progress state variable; and chaining the sorted data structure objects together using pointer addresses, so that any data structure object in the middle position contains a physical memory address index pointing to its predecessor object and a physical memory address index pointing to its successor object, forming an associated data structure that transmits logical constraints.

[0012] Preferably, in the nonlinear velocity clamping operation step, the specific logic of the nonlinear clamping is executed through the following mathematical relation: v final =min(v inertia ,v leader +λ max(0,ΔS-S min ), where v final Defined as the final propulsion speed of the current virtual slot after being clamped, v inertia Defined as the inertial predicted velocity of the current virtual slot calculated based on historical states, v leader Defined as the instantaneous velocity of the pilot slot, λ is defined as the preset elastic coefficient constant, and ΔS is defined as the logical scalar difference between the current virtual slot and the pilot slot on the one-dimensional progress coordinate. min Defined as the minimum interpersonal distance threshold.

[0013] Preferably, the nonlinear speed clamping operation step further includes global logic braking processing: when the instantaneous speed value of the pilot slot is zero, the zero speed state is used as a constraint condition to traverse all subsequent virtual slots in reverse through the bidirectional index relationship of the logical linked list; the speed state variables of all subsequent virtual slots in the visual blind zone are overwritten to make them synchronously set to zero, and the topological position of the logical linked list remains unchanged in the scalar data space until the instantaneous speed value of the pilot slot is restored to a positive value.

[0014] Preferably, the method further includes a slot adsorption and state update step: calculating the mapped position value of the newly emerging detection response on the one-dimensional progress coordinate; traversing the currently active virtual slots and calculating the absolute distance between the mapped position value and the current predicted position value of each virtual slot; uniquely binding the detection response to the virtual slot with the smallest absolute distance and whose topological index satisfies the monotonically increasing constraint; and calibrating the progress and speed states of the bound virtual slots using the real-time coordinate data of the detection response.

[0015] Preferably, the method further includes a progress coordinate frequency domain fingerprint filtering step, the specific execution logic of which is as follows: Establishing a micro-motion time-series buffer: A circular data buffer is allocated in memory for each virtual slot, and the instantaneous velocity value of the virtual slot on the one-dimensional progress coordinate is recorded in real time at a preset sampling rate, constructing a micro-velocity time-series signal sequence reflecting the dynamic characteristics of the target's movement process; Performing frequency domain energy feature transformation: A one-dimensional discrete Fourier transform or zero-crossing rate analysis operation is performed on the micro-velocity time-series signal sequence to convert the time-domain velocity fluctuation data into frequency-domain energy distribution data, and the distribution is calculated within a preset biological gait frequency range. The energy amplitude ratio is used to generate a bioactive arbitration tag. The energy amplitude ratio is compared with a preset biorhythm threshold. If the energy amplitude ratio is higher than the biorhythm threshold, an active tag representing a biological target is generated. If the energy amplitude ratio is lower than the biorhythm threshold and the signal exhibits smooth DC characteristics, a non-biological tag representing a mechanically rolling object is generated. Counting logic gating is performed. In the step of generating a completeness counting signal, the bioactive arbitration tag is used as a logic mask. Only virtual slots with active tags are allowed to trigger the counting accumulation operation, while virtual slots with non-biological tags are subjected to statistical bypass processing.

[0016] Preferably, the numerical range of the biological gait frequency interval is set to 1.5 Hz to 3.0 Hz; before performing frequency domain energy characteristic transformation, the micro velocity time sequence signal is first passed through a DC-to-DC filter to eliminate the average moving velocity component of the target, retaining only the fluctuation component characterizing the center of gravity fluctuation.

[0017] Preferably, the step of generating a completeness counting signal further includes abnormal topology circuit breaking: real-time monitoring of the evolution rate of virtual slots on a one-dimensional progress coordinate; if the increment of the progress value of any virtual slot exceeds a preset maximum physical displacement threshold within a single frame, or a non-monotonic abrupt change occurs in the progress value and its topology sequence number, a logic penetration anomaly is determined to have occurred; the abnormal virtual slot is cancelled and the associated logical linked list node link relationship is reset, no counting signal is generated, and the numerical range of the scalar data space is normalized to a closed interval [0,1], where the value 0 corresponds to the entrance baseline of the gangway area and the value 1 corresponds to the exit baseline of the gangway area; the triggering condition for generating the counting signal is specifically limited to: the progress value of the virtual slot monotonically increases from a value less than a preset exit determination threshold to a value greater than or equal to 1.

[0018] A smart identification and statistics system for personnel on gangways in border inspection scenarios, the system includes:

[0019] The one-dimensional manifold mapping module is used to receive video stream data of the gangway area. Based on the pre-calibrated homography mapping matrix, it transforms the two-dimensional pixel coordinates of the detected target in the video frame into normalized one-dimensional progress coordinates and constructs a scalar data space in memory to represent the passage process.

[0020] The doubly linked topology list construction module is used to construct all active virtual slots into a logical linked list with a bidirectional index relationship in response to the newly added virtual slots generated at the beginning of the scalar data space. Each virtual slot locks its immediate preceding slot in real time.

[0021] The nonlinear velocity clamping module is used to calculate the logical scalar difference between the current virtual slot and the pilot slot on the one-dimensional progress coordinate when performing position updates for the current virtual slot located in the visual blind zone; based on the logical scalar difference, nonlinear clamping is performed on the inertial prediction velocity of the current virtual slot, and a preset minimum interpersonal distance threshold parameter is called to participate in the logical operation. When the logical scalar difference approaches the minimum interpersonal distance threshold, the propulsion velocity value of the current virtual slot is converged to the instantaneous velocity value of the pilot slot, and the velocity constraint state is synchronously transmitted in the logical linked list.

[0022] And a completeness count generation module, used to monitor the numerical evolution of virtual slots in scalar data space. When the value completes the continuous evolution from the start end to the end end without triggering an abnormal interruption of the topology sequence number, a count signal is generated.

[0023] Compared with the prior art, the beneficial effects of the present invention are:

[0024] 1. In the intelligent identification of personnel on the gangway, by establishing a homography mapping relationship between image pixel coordinates and one-dimensional progress coordinates, and instantiating virtual slots carrying topological sequence numbers in the progress coordinate system, the problem of image tracking relying on high-dimensional visual features continuity is transformed into a problem of rigid one-dimensional data operation relying on queue logic. Based on the topological order data processing mechanism, when the system processes the confined space scenario of the border inspection gangway, it uses the physical structure constraints of the channel to lock the target identity. Under the conditions of the target facing a complete occlusion blind spot, drastic changes in lighting conditions, or visual feature failure caused by the high homogeneity of personnel clothing, the system maintains the continuity and uniqueness of the target data object, eliminates identity jumps or trajectory breaks caused by confusion of appearance features, and ensures the logical completeness of passage statistics.

[0025] 2. A virtual elastic chain coupling mechanism is adopted to construct a dynamic linked list structure with bidirectional logical constraints. The logical spacing between adjacent virtual slots is used to calculate and constrain the recursive speed of targets in the blind zone in real time. This logic simulates the incompressible characteristics of a physical queue. When a preceding target in the visible area undergoes nonlinear speed change or sudden stop, the state change is transmitted to the subsequent node in the blind zone through the linked list constraint relationship, and synchronous logical braking is executed. The data constraint mechanism solves the problem of logical tailing or position penetration caused by relying solely on the inertial prediction model when congestion waves are transmitted, ensuring the system's positioning accuracy and counting accuracy of targets in the blind zone under non-steady-state pedestrian flow conditions.

[0026] 3. Utilizing the frequency domain fingerprint filtering mechanism of the progress coordinates, frequency domain feature transformation is performed on the instantaneous velocity sequence of the virtual slot in the time dimension. This extracts specific frequency band energy features representing biological gait rhythms from the one-dimensional motion signal. The signal processing method avoids computationally intensive object recognition algorithms, analyzes the fluctuation patterns of microscopic motion data, and distinguishes between periodic gait fluctuations of human targets and mechanical smooth motion features of luggage or trolleys, which are non-biological carriers. By statistically bypassing non-biological target features, the system utilizes existing sensor hardware and, with low edge computing resource consumption, suppresses false counting interference caused by large objects carried by people, thus improving the anti-interference ability of statistical results in complex clearance scenarios. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating the intelligent identification and statistical method for personnel on gangways in a border inspection scenario according to the present invention.

[0028] Figure 2 This is a comparison chart of the position error evolution of different control strategies under unsteady congestion conditions according to the present invention;

[0029] Figure 3 This is a schematic diagram of the overall system architecture of the present invention, which integrates physical acquisition and core data processing modules;

[0030] Figure 4 This is a schematic diagram of the logical architecture and data flow of the intelligent identification and statistics system for personnel on the gangway of the present invention. Detailed Implementation

[0031] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0032] This invention discloses an intelligent identification and counting method and system for personnel on gangways in border inspection scenarios. Executed by an electronic digital data processing terminal, it maps two-dimensional video image data into one-dimensional scalar data and maintains a virtual slot linked list with bidirectional constraints in memory. This enables logical tracking and counting even when visual features fail or targets are occluded. In the preliminary step of constructing one-dimensional manifold constraints, the system receives video stream data from the gangway area. Based on pre-calibrated geometric parameters of the gangway area in memory, including entrance and exit baseline coordinates, the system uses perspective transformation algorithms or homography matrix operations to establish a mapping relationship from the image pixel coordinate system (u,v) to a normalized one-dimensional progress coordinate system s. In this mapping logic, the entrance baseline corresponds to a progress value of 0, the exit baseline corresponds to a progress value of 1, and the coordinates of any detected target within the gangway passage are... It is converted into a scalar value between 0 and 1 to represent the relative position of the target in the passage process, thereby eliminating the interference of lateral displacement noise on the logical judgment. A scalar data space representing the passage process is constructed in memory. When the system is deployed, the homography mapping matrix is ​​constructed and the calibration is performed based on the physical reference. At least four sets of physical position known feature markers such as gangway joints, anti-slip strips or pre-set calibration targets are selected as control points along the geometric center axis of the gangway passage. The physical geodetic distance of each control point relative to the entrance baseline is measured using a laser rangefinder to establish a distance vector table. The processor solves the linear equation system containing the correspondence between the image coordinates (u,v) of the feature point and the physical geodetic distance d to calculate the parameter solution of the projection transformation matrix. The real-time calculated physical distance d is normalized to the interval [0,1] to generate the progress coordinates s. This process establishes a rigid linear relationship between the pixel plane and the normalized scalar space.

[0033] In the step of establishing the bidirectional topology linked list structure, in response to the newly added virtual slot generated at the beginning of the scalar data space, the system instantiates the corresponding virtual slot object in the memory stack. This object contains progress state variables, speed state variables, and topology sequence number variables. The system iterates through all active virtual slots in real time, performs a sorting operation based on the value of the progress state variables, and concatenates the sorted data structure objects using memory address indexes. Any virtual slot in the middle position contains a memory address index pointing to its immediately preceding leader slot and a memory address index pointing to its subsequent object, forming a bidirectional associative data structure that transmits logical constraints. In the step of performing nonlinear speed clamping calculation, when updating the position of the current virtual slot located in the visual blind zone, the processor obtains the real-time state of its leader slot through the linked list index and calculates the logical scalar difference ΔS between the current virtual slot and the leader slot on the one-dimensional progress coordinate. The system performs nonlinear clamping on the inertial prediction speed of the current virtual slot based on this difference, and its logic follows the following mathematical relationship: v final =min(v inertia ,v leader +λ max(0,ΔS-S min ), where v final v represents the final propulsion speed of the current virtual slot after being clamped. inertia v is the inertial predicted velocity calculated based on historical states. leader Let λ be the instantaneous velocity of the pilot slot, λ be the preset elastic coefficient constant, and S be the instantaneous velocity of the pilot slot. min As a preset minimum interpersonal distance threshold, this calculation logic ensures that when the distance between the two approaches the safety threshold, the velocity of the subsequent slot converges to the velocity of the preceding slot. Regarding the value of the elastic coefficient λ in the nonlinear velocity clamping calculation, it is performed based on the controlled damping test site parameters. In the initialization phase, a virtual test slot is simulated and generated at a standard walking speed of 1.2 m / s. A virtual obstacle point with zero instantaneous velocity is set 2.0 m ahead of the logical path. The velocity decay response curve of the test slot as it approaches the obstacle point is monitored. The value of λ is iteratively adjusted using a bisection method until the second derivative of the velocity decay curve in time, that is, the peak deceleration, converges to the preset physical human body emergency stop comfort threshold range, which is usually set to 0.8 m / s. 2 up to 1.5m / s 2 The λ value is locked as the operating parameter for a specific physical scenario, and the logical stiffness and response sensitivity of the virtual queue in response to congestion wave transmission are defined.

[0034] Furthermore, the nonlinear speed clamping operation step also includes global logic braking processing. When the instantaneous speed value of the leading slot at the forefront of the visible area returns to zero, the system uses the bidirectional index relationship of the logical linked list to trigger a reverse traversal operation, passing the zero speed state as a constraint condition to all subsequent virtual slots. The processor overwrites the speed state variables of all subsequent virtual slots in the visual blind zone, making them synchronously set to zero, maintaining the topological spacing of the logical linked list unchanged in the scalar data space until the instantaneous speed value of the leading slot returns to a positive value. The method in this embodiment also includes a progress coordinate frequency domain fingerprint filtering step to distinguish between biological targets and non-biological carriers. The system opens a circular data buffer in memory for each virtual slot, and records the progress of the virtual slot in one dimension in real time at a preset sampling rate. The instantaneous velocity values ​​on the degree coordinate are used to construct a micro-velocity time-series signal sequence. The processor eliminates the average moving velocity component in the signal through a DC removal filter, retaining the fluctuation component representing the center of gravity fluctuation. A one-dimensional discrete Fourier transform or zero-crossing rate analysis is performed on the sequence to calculate the energy amplitude ratio of the signal within a preset biological gait frequency range. In this embodiment, the biological gait frequency range is set to 1.5Hz to 3.0Hz. If this ratio is higher than a preset biological rhythm threshold, the system generates an active tag representing a biological target; if the energy amplitude ratio is lower than the threshold and the signal exhibits smooth DC characteristics, a non-biological tag representing a mechanically rolling object is generated. In the progress coordinate frequency domain fingerprint filtering step, the sampling processing of the micro-velocity time-series signal sequence follows spectral analysis constraints, with a sampling frequency f...s The video stream frame rate is locked at, for example, 25fps, and the sliding time window length N is set to meet the frequency resolution requirements. k The values ​​are typically taken as 64 frames, corresponding to a duration of approximately 2.56 seconds. The raw velocity data within the window is processed by removing DC and Hanning window weighting before being input into the Fast Fourier Transform (FFT) module. The output discrete spectrum modulus is only the energy integral within the corresponding index range of 1.5Hz to 3.0Hz. The ratio of the integral value to the total energy of the entire frequency band serves as the basis for the quantitative determination of the bioactivity arbitration label, without involving nondeterministic machine learning reasoning processes.

[0035] In the step of generating a completeness counting signal, the system performs slot adsorption and state update operations. When a newly appearing detection response is mapped to a one-dimensional progress coordinate, the system calculates its absolute distance to the current predicted position of each active virtual slot and uniquely binds the detection response to the virtual slot with the smallest distance and a topology number that satisfies the monotonically increasing constraint. Real-time observation data is used to calibrate the progress and speed status of the bound slots. Simultaneously, the system continuously monitors the numerical evolution of virtual slots in the scalar data space. When the progress value of a virtual slot completes a continuous evolution from the start to the end, and no abnormal interruption of the topology number or being marked as a non-biological tag is triggered during the entire evolution process, the system generates a personnel access counting signal. For virtual slots with non-biological tags, the system performs statistical bypass processing. Furthermore, the step of generating a completeness counting signal also includes... The system includes abnormal topology circuit breaker handling logic: the system monitors the evolution rate of virtual slots on a one-dimensional progress coordinate in real time through the processor. If the increment of the progress value of any virtual slot exceeds the preset maximum physical displacement threshold within a single frame, or if a non-monotonic abrupt change occurs in the progress value and its topology sequence number, a logic penetration anomaly is determined to have occurred. At this time, the system cancels the abnormal virtual slot and resets the associated logical linked list node link relationship, without generating a counting signal. In this embodiment, the numerical range of the scalar data space is normalized to a closed interval [0,1], where the value 0 corresponds to the entrance baseline of the gangway area and the value 1 corresponds to the exit baseline of the gangway area. The trigger condition for generating a counting signal is specifically limited to: the progress value of the virtual slot is detected to monotonically increase from a value less than the preset exit determination threshold to a value greater than or equal to 1.

[0036] Example 1: In a border inspection port gangway monitoring scenario located in a high-latitude region, the environment exhibits typical characteristics of a high-angle, narrow physical space. During winter, passengers generally wear dark, heavy clothing, leading to highly homogenized visual textures among targets. Furthermore, a gantry structure approximately 2 meters long obstructs the view in the middle of the gangway, creating a fixed blind spot. Under these conditions, a dense flow of people, including passengers carrying large suitcases, continuously enters the gangway. One passenger at the front of the queue, just before exiting the blind spot, experiences a sudden stop due to their documents falling, causing the following passengers to be caught off guard. The square queue forms a sudden nonlinear congestion wave in the visual blind spot; the system maps the bottom center point of all detection boxes in the video stream to a normalized one-dimensional progress coordinate system s based on a preset homography matrix. Although the shoulder and head features of people in the video frame are occluded by perspective distortion, the system still instantiates a virtual slot for each passenger entering the channel in memory based on the detection response at the entrance baseline, and sets the topology number according to the entry time. At this time, the visual features are no longer used as the main index for identity association, and the relative order of each virtual slot in the scalar data space maintains the continuity of the target identity.

[0037] When the aforementioned emergency stop occurs, the speed of the lead slot corresponding to the passenger at the forefront of the visible area drops instantly to zero. Multiple virtual slots immediately following and located in the blind spot calculate the logical scalar difference ΔS between the current virtual slot in the blind spot and the stopped lead slot ahead. As this difference approaches the preset minimum interpersonal distance threshold S... min Formula v final= min(v inertia ,v leader +λ max(0,ΔS-S min The clamping option in )) takes effect, forcing the virtual slot speed v in the blind zone to be activated. final The system converges to the zero velocity of the pilot slot. This zero velocity state is instantaneously reversed to the entire blind zone queue through a bidirectional topological linked list structure, enabling all virtual slots to synchronously perform logical braking in the data space. This replicates the incompressible characteristics of the physical queue, preventing trajectory overlap or identity loss. Simultaneously, for large suitcases mixed in the queue, the system performs a progress coordinate frequency domain fingerprint filtering step in parallel. Although the suitcases are highly similar to the children following behind in size and movement trajectory, by analyzing the microscopic velocity time sequence signal of the corresponding virtual slot within the sampling window, it is detected that the energy amplitude of the signal in the 1.5Hz to 3.0Hz range is lower than the biological rhythm threshold and exhibits smooth DC motion characteristics. Based on this, the slot is determined to be a non-biological carrier, and a non-biological tag is generated. When the queue resumes flow and the target crosses the exit baseline one by one, counting is triggered only for those virtual slots that have completed the continuous evolution from progress value 0 to 1 and do not carry non-biological tags.

[0038] Example 2: This experiment constructs a digital twin simulation platform based on the physical parameters of a real border inspection gangway, integrating a fluid dynamics crowd simulation model and an optical imaging simulation module. The experimental platform is configured to simulate a 15-meter-long, 1.2-meter-wide one-way closed passage. The middle section of the passage, s∈[0.4,0.6], is set as a complete visual blind zone. The crowd model is set as a non-ideal fluid with random walking speed (mean 1.2m / s, standard deviation 0.3m / s) and random stopping probability. Large objects such as suitcases are simulated to be mixed in at a ratio of 5% to simulate signal interference in a real industrial environment. The output of the optical detection module is actively superimposed with position jitter noise (Gaussian distribution, σ=0.05m) and the probability of missed detection (5% outside the blind zone).

[0039] The purpose of this experiment is to verify the effect of nonlinear velocity clamping operation on the suppression of congestion wave propagation, and to determine the key parameter, the minimum interpersonal distance threshold S. min The optimal range of values ​​for S min The setting involves an essential technical trade-off: if S min If S is too small, virtual slots are prone to overlap or even penetration during congestion, leading to confusion in the identity sequence; if S min If the value is too large, it will cause the system to misjudge normal high-density queuing, resulting in the slot lagging behind the actual target. The experiment was designed with three parallel control groups: Control group A (existing technology): only linear Kalman filtering is used for blind zone position prediction, without slot constraints; Control group B (partially missing): linear slot constraints are used, that is, only the following vehicle is restricted from overtaking the preceding vehicle, without an elastic buffer mechanism; The sample group of this invention: adopts a complete nonlinear speed clamping mechanism, and S min Gradient parameter optimization was performed; the experimental process is as follows: at simulation time t=10s, the foremost target at the exit performed an emergency stop operation, and resumed walking after 5 seconds. This operation caused a congestion wave to propagate backward in the queue. The system recorded the real position and algorithm-predicted position of all targets in the blind zone, and calculated the average position error and the number of identity jumps.

[0040] Table 1: Comparison of Congestion Wave Response Performance under Different Control Strategies and Parameters

[0041]

[0042] Referring to Table 1, in control group A, after an emergency stop, the MAE of the target within the blind zone rapidly diverged to 1.8m and experienced 12 identity jumps, indicating that pure inertial prediction cannot cope with nonlinear congestion. Although control group B eliminated identity jumps, the MAE remained at 0.9m, demonstrating the hysteresis of rigid constraints in dealing with elastic compression. In contrast, the sample group of this invention showed that in S... minWhen the distance was set to 0.4m (corresponding to the typical interpersonal spacing in close queuing), the MAE decreased to 0.2m, with no identity jump. Further gradient experiments showed that when S min When deviating from the optimal range, performance exhibits a significant non-linear degradation: when S min When set to 0.2m (below the lower limit), the slot overlap rate surges, and MAE rebounds to 0.6m; when S min When the value is set to 0.8m (above the upper limit), the system triggers braking overly sensitively, causing the predicted position to lag significantly behind the actual position, and the MAE deteriorates to 1.1m. In addition, to verify the effectiveness of the frequency domain fingerprint filtering of the progress coordinate, 100 sets of microscopic velocity time-series signals containing pedestrians and suitcases were collected. The unprocessed raw signals contained a large amount of high-frequency detection noise, which was difficult to distinguish directly. After the DC removal and frequency domain transformation processing of this invention, the average energy ratio of pedestrian targets in the 1.5Hz to 3.0Hz frequency band was 0.65, which was higher than that of suitcase targets (0.12). By setting the biological rhythm threshold to 0.3, the system effectively eliminated non-biological carriers, and the false alarm rate was reduced from the original 15% to below 1%.

[0043] Example 3: This example combines Figures 1 to 4 This document describes a method and system for intelligent identification and statistics of personnel using gangways in a border inspection scenario, such as... Figure 1 As shown, the system first receives video stream data from the gangway area as the processing source. Then, it executes a step to construct a one-dimensional manifold constraint. This step transforms the two-dimensional pixel coordinates of the detected target into normalized one-dimensional progress coordinates based on a homography mapping matrix and constructs a scalar data space in memory. Following this, the system enters a step to establish a bidirectional topological linked list structure. The system generates an active virtual slot logical linked list, ensuring that each virtual slot is locked to its immediately preceding leader slot in real time and establishing a bidirectional index. Next, a nonlinear velocity clamping operation is performed on the virtual slots within the visual blind zone. By calculating the logical scalar difference, when this difference approaches the minimum interpersonal distance threshold, the velocity of the blind zone slot is forced to converge to the velocity of the leader slot, and the velocity constraint state is transmitted. Finally, a completeness counting signal generation step is executed. The system monitors the evolution of the virtual slot values. Only when the monitored values ​​have completed a continuous evolution from the start to the end without triggering an abnormal interruption is the counting condition satisfied, and the counting signal is finally output.

[0044] like Figure 2 As shown in the figure, the graph constructs a coordinate system with time (seconds) on the horizontal axis and the average positional error (MAE) in meters on the vertical axis, intuitively displaying the error evolution curves under three different strategies: Control group A, using Kalman filtering, saw its error curve rise sharply with time, reaching a peak of 1.8 meters near 10 seconds; Control group B, using linear constraints, saw its error curve, although lower than Control group A, still exhibited fluctuations; while the sample group of this invention, with the minimum interpersonal distance threshold S... minWith a setting of 0.4m, the error curve remained at an extremely low and stable level throughout; for example... Figure 3 As shown, the overall architecture of the system consists of a field physical acquisition layer and a core data processing layer. The field physical acquisition layer covers the physical environment including blind spots such as gantry cranes or sunshades and the gangway passageway at entrance and exit baselines. It is equipped with high-view surveillance cameras to output RTSP video streams, which are then transmitted to the core data processing layer deployed on the terminal or server. The core data processing layer runs intelligent recognition and statistics system software, which integrates a one-dimensional manifold mapping module, a bidirectional topology linked list construction module, a nonlinear velocity clamping module, a progress coordinate frequency domain fingerprint filtering module, and a completeness count generation module corresponding to the method flow. It also relies on a memory data buffer to store historical status and micro-time series data. Finally, the count or alarm signal generated by the processing layer is sent to the duty management application layer and visualized on the large screen of the duty command center or on the field handheld verification terminal.

[0045] like Figure 4 As shown, the logical architecture of the intelligent identification and statistics system for gangway personnel in this border inspection scenario begins with the video stream data of the gangway area at the data input end. This data is transmitted to the one-dimensional manifold mapping module, which performs a coordinate transformation operation based on the homography matrix to realize the conversion of two-dimensional pixels into one-dimensional progress coordinates and constructs a scalar data space in memory. Subsequently, the data flows to the bidirectional topology linked list construction module, which is responsible for responding to the addition of virtual slots, constructing active objects into a bidirectional indexed logical linked list, and locking each node to the pilot slot in real time. At the same time, it outputs topology status feedback through the virtual line radial output stage. Next, the nonlinear velocity clamping module performs a scalar difference calculation operation based on the input logical constraints to process the blind zone inertial prediction velocity clamping and forces the target velocity to converge to the pilot slot velocity. The final processing result is fed into the completeness count generation module. This module triggers the count generation signal when the conditions are met by monitoring the evolution of scalar space values ​​and verifying the continuity of topology numbers, thereby outputting a completeness count signal from the signal output end.

[0046] Example 4: This example demonstrates the transparency and engineering calibration of the input variables, calculation path, and boundary condition processing for this computational logic. Specifically, the execution of this operation depends on the inertial prediction velocity v. inertia With nonlinear clamping speed v clamp Real-time arbitration, in which v inertia The calculation is not a simple linear extrapolation, but rather based on the historical displacement data of the current virtual slot within a preset time window, such as the last 30 frames of data in the last 1.5 seconds. The instantaneous velocity estimate at the current moment is obtained by performing least squares linear regression or a first-order Kalman filter algorithm.

[0047] Simultaneously, the nonlinear clamping speed v clampThe calculation introduces a dynamic elasticity coefficient λ. In a specific implementation, λ is defined as a preset constant. This embodiment provides a dynamic adjustment procedure based on local flow density. The system defines a local density index ρ, which is the number of active virtual slots within a preset progress range (e.g., ±0.05 progress units) before and after the current virtual slot. Then, the value of λ is mapped in real time according to the following piecewise linear function: when ρ is less than a preset sparsity threshold (e.g., 3 people), λ is set to a base value (e.g., 2.0) to maintain a weak elastic constraint; when ρ is greater than or equal to the sparsity threshold, λ increases linearly with ρ until it reaches a saturation value (e.g., 5.0). The calculation formula is λ = min(5.0, 2.0 + 0.5). (ρ-3)), this procedure ensures that targets have greater freedom of motion under low density conditions, while imposing stronger velocity synchronization constraints under high density congestion conditions.

[0048] Example 5: This example constructs a standardized parameter calibration and system initialization procedure for field data, and designs it as a mandatory pre-deployment step after system deployment. Through standardized data acquisition and closed-loop feedback logic, the optimal operating point of the aforementioned key parameters is automatically established; regarding the elasticity coefficient... The system was calibrated and executed a congestion response stress test procedure. During the system initialization phase, a controlled flow of people containing at least 5 testers was selected. The system was required to execute a preset emergency stop-and-resume sequence within a visual blind zone. The system recorded the actual positions of the targets within the blind zone (obtained through temporary auxiliary sensors or manual annotation) and the predicted positions of the algorithm under different λ values, constructing a parameter scanning space. The specific decision-making logic was based on a dual-objective optimization function of prediction error and jitter cost. As λ increases, the system's response speed to congestion increases, and the prediction error decreases. However, excessively large λ introduces non-physical high-frequency speed jitter. The system calculated the comprehensive cost function J(λ)=w1 under different λ values. MAE+w2 Jerk, where MAE is the mean absolute position error, Jerk is the root mean square value of the rate of change of velocity, and w1 and w2 are preset weights. The system automatically selects the λ value that minimizes J(λ) as the operating parameter for this specific scenario, thereby achieving the optimal balance between response agility and trajectory smoothness verified by engineering quantification.

[0049] In addition to calibrating the biological rhythm threshold, the system executes an adaptive learning procedure based on the environmental background to eliminate frequency domain feature drift caused by differences in camera installation height, pitch angle, and focal length. The system acquisition time is T. cal For example, using 10 minutes of clean background pedestrian traffic data (containing only normal pedestrians) and typical interference data (such as pushing empty and fully loaded standard suitcases), the system calculates the energy proportion of each collected trajectory in the 1.5Hz to 3.0Hz frequency band, and constructs pedestrian sample distribution P for each.human and the distribution of interfering samples P noise The system calculates the intersection of two distribution curves as the initial biological rhythm threshold based on the Bayesian minimum error rate criterion, and introduces a safety margin coefficient. For example, in 1.2, the final threshold is set as Threshold = Mean(P noise )+α Std(P noise ), where Mean(P noise ) represents the distribution of interfering samples P noise The arithmetic mean of the energy percentage within the biological gait frequency range, Std(P) noise ) represents the distribution of interfering samples P noise The standard deviation of energy proportion within the biological gait frequency range ensures that interfering samples can be eliminated with a confidence level of 99.7%. This procedure transforms fuzzy empirical judgments into deterministic calculations based on on-site statistical data, guaranteeing the generalization ability of the classifier in different deployment environments. In addition, to cope with parameter drift that may occur during long-term operation, the system has a built-in online status monitoring and abnormal circuit breaker mechanism. The system monitors the average survival time and average travel speed of all active virtual slots in real time. If it detects that the average travel speed is lower than 20% of the historical average within N consecutive time windows, such as 5 minutes, or the average survival time exceeds 3 times the normal passage, the system determines that parameter mismatch or abnormal congestion may occur. At this time, the system automatically triggers the degraded operation mode, temporarily relaxes the constraint strength of λ, marks the data of this period as low confidence, and generates an alarm log to prompt the operation and maintenance personnel to review the parameters.

[0050] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0051] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for intelligent identification and statistics of personnel on gangways in a border inspection scene, characterized in that, Includes the following steps: The prerequisite steps for constructing one-dimensional manifold constraints are to receive video stream data of the gangway area, transform the two-dimensional pixel coordinates of the detected target into normalized one-dimensional progress coordinates based on the pre-calibrated homography mapping matrix, and construct a scalar data space. The step of establishing a bidirectional topological linked list structure is as follows: in response to the newly added virtual slots generated at the beginning of the scalar data space, the virtual slots are constructed into a logical linked list with bidirectional indexing relationship, and each virtual slot locks the leader slot in real time. The nonlinear velocity clamping operation step is executed. The position of the current virtual slot in the visual blind zone is updated, and the logical scalar difference between the current virtual slot and the pilot slot in the one-dimensional progress coordinate is calculated. Based on the logical scalar difference, the inertial prediction velocity of the current virtual slot is nonlinearly clamped. The preset minimum interpersonal distance threshold parameter is called to participate in the logical operation. When the logical scalar difference approaches the minimum interpersonal distance threshold, the propulsion velocity value of the current virtual slot is converged to the instantaneous velocity value of the pilot slot, and the velocity constraint state is synchronously transmitted in the logical linked list. The specific logic of the nonlinear clamping is executed through the following mathematical relation: v final =min(v inertia ,v leader +λ max(0,ΔS-S min ), where v final Defined as the final propulsion speed of the current virtual slot after being clamped, v inertia Defined as the inertial predicted velocity of the current virtual slot calculated based on historical states, v leader Defined as the instantaneous velocity of the pilot slot, λ is defined as the preset elastic coefficient constant, and ΔS is defined as the logical scalar difference between the current virtual slot and the pilot slot on the one-dimensional progress coordinate. min Defined as the minimum interpersonal distance threshold; The step of generating a completeness counting signal involves monitoring the numerical evolution of the virtual slots. When the continuous evolution from the start end to the end end is completed without triggering an abnormal interruption of the topology sequence number, a counting signal is generated.

2. The method according to claim 1, wherein, The specific steps for constructing one-dimensional manifold constraints include: reading the entrance baseline coordinate parameters and exit baseline coordinate parameters of the gangway area; calling the perspective transformation algorithm to calculate the transformation matrix that maps the entrance baseline coordinate parameters to a progress value of zero and the exit baseline coordinate parameters to a progress value of one; applying the transformation matrix to establish a dimensionality reduction mapping relationship from the image pixel coordinate system to the one-dimensional progress coordinate system, and converting the motion trajectory data of the detected target along the nonlinear physical path into linear scalar change data along a single dimension in the scalar data space.

3. The intelligent identification and statistical method for personnel on gangways in a border inspection scenario according to claim 1, characterized in that, The steps for establishing a bidirectional topological linked list structure specifically include: instantiating a data structure object containing progress status variables, speed status variables, and topological sequence number variables for each detection target in memory; sorting all data structure objects according to the value of the progress status variables; and chaining the sorted data structure objects together using pointer addresses, so that any data structure object in the middle position contains a physical memory address index pointing to its predecessor object and a physical memory address index pointing to its successor object, forming an associative data structure that transmits logical constraints.

4. The method according to claim 1, wherein, The nonlinear speed clamping operation also includes global logic braking processing: when the instantaneous speed value of the pilot slot is zero, the zero speed state is used as a constraint condition to traverse all subsequent virtual slots in reverse through the bidirectional index relationship of the logical linked list; the speed state variables of all subsequent virtual slots in the visual blind zone are overwritten to make them synchronously set to zero, and the topological position of the logical linked list is kept unchanged in the scalar data space until the instantaneous speed value of the pilot slot is restored to a positive value.

5. The method according to claim 1, wherein, The method also includes slot adsorption and state update steps: calculating the mapped position value of the newly emerging detection response on the one-dimensional progress coordinate; traversing the currently active virtual slots and calculating the absolute distance between the mapped position value and the current predicted position value of each virtual slot; uniquely binding the detection response to the virtual slot with the smallest absolute distance and whose topology number satisfies the monotonically increasing constraint; and calibrating the progress and speed states of the bound virtual slots using the real-time coordinate data of the detection response.

6. The method according to claim 1, wherein, The method also includes a progress coordinate frequency domain fingerprint filtering step, the specific execution logic of which is as follows: Establish a micro-motion timing buffer: A circular data buffer is allocated in memory for each virtual slot, and the instantaneous velocity value of the virtual slot on the one-dimensional progress coordinate is recorded in real time at a preset sampling rate, constructing a micro-velocity timing signal sequence reflecting the dynamic characteristics of the target's movement process; Perform frequency domain energy feature transformation: Perform one-dimensional discrete Fourier transform or zero-crossing rate analysis on the micro-velocity time-series signal sequence to convert the time-domain velocity fluctuation data into frequency-domain energy distribution data, and calculate the proportion of energy amplitude within the preset biological gait frequency range. Generate bioactive arbitration tags: Compare the energy amplitude percentage with a preset biorhythm threshold. If the energy amplitude percentage is higher than the biorhythm threshold, generate an active tag representing a biological target. If the energy amplitude percentage is lower than the biorhythm threshold and the signal exhibits smooth DC characteristics, generate a non-biological tag representing a mechanically rolling object. Execute counting logic gating: In the step of generating a completeness counting signal, the bioactive arbitration tag is used as a logic mask to allow only virtual slots with active tags to trigger the counting accumulation operation, while performing statistical bypass processing on virtual slots with non-biological tags.

7. The method according to claim 6, wherein, The numerical range of the biological gait frequency interval is set to 1.5 Hz to 3.0 Hz; before performing frequency domain energy characteristic transformation, the micro velocity time sequence signal is first passed through a DC-to-DC filter to eliminate the average moving velocity component of the target, retaining only the fluctuation component characterizing the center of gravity fluctuation.

8. The method according to claim 1, wherein, The step of generating a completeness counting signal also includes abnormal topology circuit breaking: real-time monitoring of the evolution rate of virtual slots on a one-dimensional progress coordinate; if the increment of the progress value of any virtual slot exceeds the preset maximum physical displacement threshold within a single frame, or a non-monotonic abrupt change occurs in the progress value and its topology sequence number, a logic penetration anomaly is determined; the abnormal virtual slot is cancelled and the associated logical linked list node link relationship is reset, no counting signal is generated, and the numerical range of the scalar data space is normalized to a closed interval [0,1], where the value 0 corresponds to the entrance baseline of the gangway area and the value 1 corresponds to the exit baseline of the gangway area; the triggering condition for generating the counting signal is specifically limited to: the progress value of the virtual slot monotonically increases from a value less than the preset exit determination threshold to a value greater than or equal to 1.

9. A smart identification and statistics system for personnel on gangways in border inspection scenarios, used to implement the method described in any one of claims 1 to 8, characterized in that, The system includes: The one-dimensional manifold mapping module is used to receive video stream data of the gangway area. Based on the pre-calibrated homography mapping matrix, it transforms the two-dimensional pixel coordinates of the detected target in the video frame into normalized one-dimensional progress coordinates and constructs a scalar data space in memory to represent the passage process. The doubly linked topology list construction module is used to construct all active virtual slots into a logical linked list with a bidirectional index relationship in response to the newly added virtual slots generated at the beginning of the scalar data space. Each virtual slot locks its immediate preceding slot in real time. The nonlinear velocity clamping module is used to calculate the logical scalar difference between the current virtual slot and the pilot slot on the one-dimensional progress coordinate when performing position updates for the current virtual slot located in the visual blind zone; based on the logical scalar difference, nonlinear clamping is performed on the inertial prediction velocity of the current virtual slot, and a preset minimum interpersonal distance threshold parameter is called to participate in the logical operation. When the logical scalar difference approaches the minimum interpersonal distance threshold, the propulsion velocity value of the current virtual slot is converged to the instantaneous velocity value of the pilot slot, and the velocity constraint state is synchronously transmitted in the logical linked list. And a completeness count generation module, used to monitor the numerical evolution of virtual slots in scalar data space. When the value completes the continuous evolution from the start end to the end end without triggering an abnormal interruption of the topology number, a count signal is generated.

Images