Passenger flow detection method and system

Abstract

The present invention relates to the technical field of passenger flow detection and discloses a passenger flow detection method and system, comprising collecting movement direction data of human bodies in a monitoring area through a sensor array, generating a time series data stream after signal conditioning; initializing a passenger flow density network parameter population of a passenger flow density network using a chaotic map, optimizing the passenger flow density network parameter population through a nonlinear decreasing pitch, and injecting random disturbances into the passenger flow density network; processing the time series data stream through the optimized passenger flow density network, outputting a real-time passenger flow density value and a future passenger flow change trend; generating a graded warning instruction through a dynamic combination relationship between the passenger flow density value and the future passenger flow change trend, wherein the grade of the graded warning instruction increases nonlinearly with the density increase rate; executing the graded warning instruction and collecting actual passenger flow change data, and periodically re-optimizing the passenger flow density network parameters based on the feedback data, thereby realizing multi-scenario intelligent passenger flow detection.

Description

Technical Field

[0001] The invention relates to the technical field of passenger flow detection and discloses a passenger flow detection method and system. Background Art

[0002] With the advancement of smart city construction, passenger flow monitoring has become a key requirement for public place management. Traditional passenger flow monitoring technologies such as infrared sensing and video analysis have problems such as low accuracy, poor real-time performance, and weak adaptability. In complex passenger flow scenarios, such as holiday peaks and sudden events, traditional solutions are difficult to accurately predict the changing trends of passenger flow density and cannot adjust monitoring strategies in a timely manner. At the same time, existing passenger flow monitoring systems usually use fixed parameter models and lack adaptive optimization capabilities in the face of dynamic changes in passenger flow, resulting in delayed warnings and high false alarm rates, making it difficult to meet the actual needs of safety management and resource allocation. Summary of the Invention

[0003] The purpose of this section is to summarize some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the abstract and title of this application to avoid obscuring the purpose of this section, the abstract and the title of the invention, and such simplifications or omissions should not be used to limit the scope of the present invention.

[0004] To solve the above technical problems, the main purpose of the present invention is to provide a passenger flow detection method, wherein a passenger flow detection method includes:

[0005] The sensor array collects the movement direction data of the human body in the monitoring area and generates a time series data stream after signal conditioning;

[0006] The passenger flow density network parameter population of the passenger flow density network is initialized by chaotic mapping, the passenger flow density network parameter population is optimized by nonlinear decreasing pitch, and random disturbance is injected into the passenger flow density network.

[0007] The optimized passenger flow density network processes the time series data stream and outputs the real-time passenger flow density value and the future trend of passenger flow;

[0008] The hierarchical warning instructions are generated through the dynamic combination relationship between the passenger flow density value and the future change trend of the passenger flow. The level of the hierarchical warning instructions increases nonlinearly with the rate of increase of density.

[0009] Execute graded warning instructions and collect actual passenger flow change data, and periodically re-optimize passenger flow density network parameters based on feedback data.

[0010] As a preferred solution of a passenger flow detection method of the present invention, wherein:

[0011] The method for generating a time series data stream includes:

[0012] The human body position coordinates are calculated by weighting the human body's movement direction data and correlating the motion trajectory in continuous time frames;

[0013] Dynamically determine the movement direction category based on the motion trajectory displacement vector and generate a semantic direction label;

[0014] Encapsulate timestamps, device identifiers, direction tags, and location coordinates into standardized data units;

[0015] Pushing data units at fixed time intervals forms a time-series data stream.

[0016] As a preferred solution of a passenger flow detection method of the present invention, wherein:

[0017] The method for initializing the passenger flow density network parameter population using chaotic mapping includes:

[0018] By deterministically traversing passenger flow characteristics through chaotic mapping, a chaotic sequence is generated;

[0019] Map the chaotic sequence to the adjustable parameter definition domain of the passenger flow density network, so that the initial passenger flow population is evenly distributed in the solution space;

[0020] The parameter boundary constraints of the passenger flow population are preset to generate a set of passenger flow density network diversity parameter combinations.

[0021] As a preferred solution of a passenger flow detection method of the present invention, wherein:

[0022] The method for optimizing the passenger flow density network parameter population by nonlinear decreasing pitch includes:

[0023] An initial pitch factor is set in the passenger flow density network to enable the passenger flow population to perform a wide-area search in the solution space;

[0024] A pitch factor is attenuated according to the progress of the wide-area search of the passenger flow population in the solution space, wherein the rate of attenuation decreases as the number of iterations increases;

[0025] After optimizing the passenger flow density network, the passenger flow population is determined by the minimum pitch factor to perform development on the local optimal area.

[0026] As a preferred solution of a passenger flow detection method of the present invention, wherein:

[0027] The injecting random disturbance into the passenger flow density network comprises:

[0028] When the population fitness variance is lower than the convergence threshold, a heavy-tailed random perturbation is injected into the current optimal passenger flow parameters;

[0029] The amplitude of the random perturbation of the distribution is adjusted along with the optimization process, breaking through the potential local optimum through large perturbations, and then ensuring the global optimum of the passenger flow parameters through micro-perturbations;

[0030] After the disturbance injection is completed, the passenger flow population fitness is re-evaluated.

[0031] As a preferred solution of a passenger flow detection method of the present invention, wherein:

[0032] The processing of the time series data stream by the optimized passenger flow density network includes:

[0033] The entry and exit balance coefficient is dynamically calculated based on the movement direction label, where entry events are positively accumulated and exit events are negatively offset, generating a continuous time series variable that represents the changes in regional passenger flow.

[0034] By utilizing the spatiotemporal separation characteristics of the depthwise separable convolutional structure, the coupled characteristics of short-term sudden fluctuations and long-term gradual trends are captured while reducing computational complexity.

[0035] Dynamically evaluate feature importance weights through gated activation units to enhance the feature responses of key passenger flow aggregation / dissipation patterns and suppress noise signals caused by environmental disturbances.

[0036] Generate high-dimensional feature vectors as passenger flow status representation;

[0037] The convolution kernel size, dilation factor and gating weight parameters of the passenger flow density network are configured in real time by an adaptive optimization process.

[0038] As a preferred solution of a passenger flow detection method of the present invention, wherein:

[0039] The output of real-time passenger flow density value and future passenger flow change trend includes:

[0040] The filtered feature vector is input into the fully connected regression layer of the passenger flow density network, and a standardized density scalar value is generated through nonlinear mapping. The physical meaning of this value is the passenger flow load rate per unit area.

[0041] The first-order derivative of the density scalar is calculated within the sliding time window of the passenger flow density network. Combining the dual criteria of the derivative sign and amplitude, the decreasing trend, stable trend and increasing trend are discretized and output.

[0042] The downward trend is that the derivative is continuously negative and the amplitude is greater than the minimum threshold of passenger flow dissipation;

[0043] The stable trend is when the derivative is zero and the amplitude is between the minimum threshold of passenger flow dissipation and the maximum threshold of passenger flow gathering;

[0044] The upward trend is that the derivative is continuously positive and the amplitude is greater than the maximum threshold of passenger flow aggregation;

[0045] A hierarchical early warning decision-making mechanism is constructed by combining real-time passenger flow density values ​​and discrete trend states.

[0046] As a preferred solution of a passenger flow detection method of the present invention, wherein:

[0047] Generating a graded warning instruction through a dynamic combination relationship includes:

[0048] A two-dimensional decision space is established with real-time passenger flow density as the horizontal axis and change trend as the vertical axis. The density value is divided into three intervals: loose, normal, and crowded. The trend status is divided into three categories: declining, stable, and rising.

[0049] When the density value is in the loose range, the trend change is ignored and no warning is triggered;

[0050] When the density value enters the normal range and the trend is rising, the warning level increases in a square relationship with the density increase rate;

[0051] When the density value exceeds the congestion threshold, a basic warning is immediately triggered, and an acceleration factor for the passenger flow surge trend is added;

[0052] Generate progressive commands from observation level 1 to emergency level 3, where Level 3 commands force the activation of emergency equipment.

[0053] As a preferred solution of a passenger flow detection method of the present invention, wherein:

[0054] The periodic re-optimization based on feedback data includes:

[0055] Compare the actual density change rate before and after the early warning instruction is executed to calculate the strategy effectiveness coefficient , is the actual passenger flow change, △ represents the change mark, To predict the change in passenger flow density, D represents the passenger flow density index.

[0056] The re-optimization of passenger flow density network parameters includes timing triggering and event triggering. The timing triggering is to perform passenger flow density network parameter optimization every fixed period. The event triggering is to immediately start passenger flow network parameter optimization if the effectiveness coefficient of the strategy is not greater than the minimum parameter optimization threshold.

[0057] As a preferred solution of a passenger flow detection system of the present invention, wherein:

[0058] Acquisition module, calculation module, prediction module, early warning module and feedback module;

[0059] The acquisition module is used to collect human body movement direction data and generate a structured time series data stream containing movement characteristics through signal processing;

[0060] The calculation module generates an initial parameter group through parameter initialization, optimizes the parameters of the passenger flow density network by adopting a strategy that can dynamically adjust the optimization step size according to data changes, and performs a perturbation test on the passenger flow density network by introducing a perturbation mechanism that can change the parameter search direction;

[0061] The prediction module uses the collected time series data stream as the data input of the optimized passenger flow density network, outputs the real-time passenger flow density value and change trend, and predicts the future change trend of passenger flow, assigning different weights to the future change trend of passenger flow according to the freshness of the data;

[0062] The warning module dynamically adjusts the warning threshold value and generates a graded warning instruction through the dynamic combination relationship between the real-time passenger flow density value and the change trend and the historical warning data.

[0063] The feedback module is used to execute the graded warning instructions and collect actual passenger flow change data, analyze the execution effect through the evaluation mechanism, and drive the calculation module to re-optimize the network parameters in a cycle that gradually shortens as the system operation time.

[0064] Beneficial effects of the present invention:

[0065] The passenger flow detection method of the present invention collects data through a sensor array and generates a time-series data stream. It combines chaotic map initialization and nonlinear decreasing pitch optimization of the passenger flow density network, effectively avoiding parameters from falling into local optimality and improving the generalization ability of the model.

[0066] When processing time series data streams, this application uses depthwise separable convolution and gated activation units to capture the spatiotemporal characteristics of passenger flow, enabling accurate prediction of real-time passenger flow density and future change trends.

[0067] This application uses hierarchical warning instructions generated through dynamic combination to quickly respond to changes in passenger flow and trigger emergency measures in a timely manner. Combined with the periodic re-optimization mechanism of feedback data, the passenger flow density network has real-time optimization capabilities and can adapt to passenger flow changes in different scenarios. It significantly improves the accuracy of passenger flow prediction and warning efficiency, and reduces the false alarm rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort. Among them:

[0069] Figure 1 This is a flow chart of a passenger flow detection method of the present invention;

[0070] Figure 2 This is a composition diagram of a passenger flow detection system of the present invention;

[0071] Figure 3 A passenger flow density detection diagram from Monday to Saturday of a passenger flow detection method of the present invention;

[0072] Figure 4 This is a flow chart of a passenger flow detection method of the present invention that outputs real-time passenger flow density values ​​and future passenger flow change trends. DETAILED DESCRIPTION

[0073] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0074] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0075] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0076] Example 1

[0077] like Figure 1 As shown, a passenger flow detection method includes:

[0078] The sensor array collects the movement direction data of the human body in the monitoring area and generates a time series data stream after signal conditioning;

[0079] The method for generating a time series data stream includes:

[0080] The human body position coordinates are calculated based on the weighted spatial distribution of heat sources and associated with the motion trajectory in continuous time frames;

[0081] Specifically, in the present application, a preferred sensor array can sense the heat source signal emitted by the human body in the monitoring area. For example, the infrared sensor is sensitive to the thermal radiation of the human body. The heat source signal intensity received by sensors at different positions varies. The closer the sensor is to the human body, the stronger the signal received. By performing weighted calculation on the heat source signal intensity collected by multiple sensors, sensors with high signal intensity are given higher weights, so that the specific position coordinates of the human body in the monitoring area can be determined.

[0082] Dynamically determine the movement direction category based on the trajectory displacement vector and generate a semantic direction label;

[0083] Specifically, the sensor array continuously collects data at a preset frequency to form continuous time frames. The human body position coordinates are obtained through weighted calculation in each frame of data. The position coordinates in these different time frames are connected in sequence to determine the movement trajectory of the human body within the time series.

[0084] Furthermore, by connecting the position coordinates in these different time frames in sequence, we can depict the movement trajectory of the human body over a period of time.

[0085] Encapsulate timestamps, device identifiers, direction tags, and location coordinates into standardized data units;

[0086] The motion trajectory is composed of position coordinates, and displacement vectors are formed between adjacent position coordinates. By analyzing the direction of the displacement vector, the direction of movement of the human body can be determined. For example, if the displacement vector points to the upper left, it can be determined that the human body is moving in the upper left direction.

[0087] The timestamp records the specific moment of data collection, and the device identifier clearly identifies the source of the data;

[0088] The orientation label and position coordinates are encapsulated into a standardized data unit.

[0089] Pushing data units at fixed time intervals forms a time-series data stream.

[0090] At pre-set fixed time intervals, the encapsulated data units are pushed out one by one. As time goes by, one data unit after another is continuously output, forming a data stream with a time sequence.

[0091] The passenger flow density network parameter population of the passenger flow density network is initialized by chaotic mapping, the passenger flow density network parameter population is optimized by nonlinear decreasing pitch, and random disturbance is injected into the passenger flow density network.

[0092] The method of initializing the passenger flow density network parameter population by using chaotic mapping includes:

[0093] By deterministically traversing passenger flow characteristics through chaotic mapping, a chaotic sequence is generated;

[0094] Specifically, in this application, the principle of initializing the passenger flow density network parameter population using chaotic mapping includes:

[0095] Chaotic mapping is a mathematical system that is deterministic but exhibits random characteristics. In passenger flow monitoring, the specific implementation methods of chaotic mapping to traverse passenger flow characteristics and generate chaotic sequences include:

[0096] S11 has the traversal capability of traversing the passenger flow state space through chaotic mapping, covering each area of ​​the passenger flow state space without repetition, and generating a chaotic sequence with traversal through chaotic mapping.

[0097] S12 takes passenger flow characteristics including density change frequency, directional distribution, peak and valley periodicity as input conditions or constraint parameters of chaotic mapping, so that the generated chaotic sequence can reflect the characteristics of passenger flow density data stream.

[0098] During the chaotic traversal process, S13 dynamically adjusts the parameters of the chaotic mapping according to the variance and skewness of the real-time passenger flow density data to enhance the adaptability of the sequence to passenger flow changes.

[0099] Map the chaotic sequence to the adjustable parameter definition domain of the passenger flow density network, so that the initial passenger flow population is evenly distributed in the solution space;

[0100] Specifically, the chaotic sequence is mapped to the adjustable parameter definition domain and evenly distributed.

[0101] The solution space refers to a multidimensional space composed of all possible parameter combinations in the passenger flow density network. All possible values ​​of each parameter within its value range constitute the coordinate axes of this space, and the passenger flow density parameter combinations correspond to points in the multidimensional space.

[0102] Furthermore, the domain of adjustable parameters for mapping chaotic sequences to passenger flow density networks includes:

[0103] First, the value range of the chaotic sequence is mapped to the actual definition domain of each parameter through linear or nonlinear transformation.

[0104] Through the uniform distribution characteristics of chaotic sequences, the mapped parameter points are ensured to be evenly distributed in the solution space, avoiding aggregation in certain local areas.

[0105] For each adjustable parameter in the passenger flow density network, including weight, bias, convolution kernel size, etc., independent mapping operations are performed to ensure that the distribution of each parameter within its definition domain is not affected by other parameters.

[0106] The diversity of parameter combinations is enhanced by introducing random perturbations or averaging multiple mappings.

[0107] The parameter boundary constraints of the passenger flow population are preset to generate a set of passenger flow density network diversity parameter combinations.

[0108] Specifically, presetting parameter boundary constraints and generating a diverse parameter combination set includes presetting parameter boundary constraints and generating diverse parameter combinations;

[0109] Furthermore, the specific implementation method of parameter boundary constraint preset includes:

[0110] Based on the expertise and experience in the field of passenger flow monitoring, the value range of each parameter in the passenger flow density network is pre-set. For example, the convolution kernel size is an odd number and the learning rate does not exceed the convergence threshold of the network.

[0111] Furthermore, the upper and lower limits of the parameters of the passenger flow density network are set. For example, too many network layers will lead to excessive computational complexity.

[0112] Furthermore, the parameter boundaries are dynamically adjusted based on the statistical characteristics of historical passenger flow data and the model training results. For example, in scenarios with large passenger flow fluctuations, the parameter value range is relaxed.

[0113] Specifically, the specific implementation method for generating a combination of diverse parameters includes:

[0114] Multiple sample points are extracted from the mapped chaotic sequence, and each point corresponds to a complete set of network parameters.

[0115] Orthogonal calculation is used to calculate the sample points of the chaotic sequence and select parameter combinations to ensure that the combination of each parameter level is representative.

[0116] The generated parameter combinations are analyzed and combinations with too high similarity are eliminated.

[0117] Injecting random perturbations into the selected parameter combinations further increases the diversity of the population and improves the probability of the algorithm searching for the global optimal solution.

[0118] Through the ergodic nature of chaotic sequences, the algorithm can explore the solution space more comprehensively and reduce the risk of falling into local optimality. The chaotic mapping parameters can be dynamically adjusted according to passenger flow characteristics, making the initialization process more suitable for actual application scenarios. The uniformly distributed initial parameter population can accelerate the network training process and reduce the number of iterations required for convergence. The diverse parameter combinations make the model more adaptable and robust to different passenger flow scenarios.

[0119] The method for optimizing the passenger flow density network parameter population by nonlinear decreasing pitch includes:

[0120] An initial pitch factor is set in the passenger flow density network, and the passenger flow population is searched to perform a wide-area search in the solution space;

[0121] Specifically, the pitch factor is used to search for the key parameter of the step size. In the passenger density network optimization, the pitch factor is used to determine the amplitude of each parameter update in the passenger density network:

[0122] Furthermore, a larger pitch factor is set initially so that the passenger density network can perform a large-scale jumping search in the solution space, covering multiple areas and avoiding falling into the local optimum.

[0123] As the optimization of the passenger flow density network progresses, the pitch factor gradually decreases, and the search step size decreases as the pitch factor gradually decreases, so that the parameters of the passenger flow density network are searched in a local area.

[0124] In this application, the pitch factor is dynamically associated with the spatiotemporal characteristics of passenger flow data. For example, in areas where passenger flow fluctuates violently, the pitch factor adjustment is more sensitive to adapt to the rapidly changing passenger flow pattern.

[0125] A pitch factor is attenuated according to the progress of the wide-area search of the passenger flow population in the solution space, wherein the rate of attenuation decreases as the number of iterations increases;

[0126] Specifically, the distribution of passenger flow populations in the solution space and the changing trend of the objective function value are monitored in real time. When it is found that the improvement of the objective function value gradually decreases, the search has entered a better area and the pitch factor begins to be attenuated.

[0127] The decay rate decreases with the increase of the number of iterations, forming a decay characteristic that is fast in the early stage and slow in the later stage, so that the passenger flow density network can quickly narrow the search range in the early stage and maintain the search time when approaching the optimal solution.

[0128] The attenuation process is also dynamically adjusted based on the real-time characteristics of passenger flow data. For example, during peak passenger flow periods, in order to quickly respond to passenger flow changes, the pitch factor attenuation rate is appropriately reduced to maintain a certain degree of search flexibility.

[0129] After optimizing the passenger flow density network, the passenger flow population is determined by the minimum pitch factor to perform development on the local optimal area.

[0130] In this application, a specific implementation method of a pitch factor optimized passenger flow density network includes:

[0131] A larger initial pitch factor is set, and the algorithm performs a random walk search in the solution space, trying different parameter combinations.

[0132] Combined with the random sequence generated by the chaotic map, the parameters in the current passenger flow density network are perturbed to further expand the search range. Furthermore, the perturbation amplitude is proportional to the pitch factor, which is used to ensure that the solution space is fully explored in the wide-area search stage.

[0133] When the pitch factor decays to the preset minimum value, the passenger flow density network optimization enters the local optimization stage, and the parameter optimal price search step is extremely small.

[0134] When the passenger flow density network optimization enters the local optimization stage, the passenger flow density network fine-tunes parameters along the gradient descent direction according to the gradient information of the pitch factor attenuation, and gradually approaches the global optimal solution.

[0135] Through multiple rounds of iterations, the minimum pitch factor is used to repeatedly verify and adjust the parameters to ensure that the final solution of the passenger flow density network parameters is optimal in the local area.

[0136] When the pitch factor is small, the amplitude of the random disturbance is also reduced accordingly, avoiding excessive disturbances that destroy the parameter combination that is close to the optimal one.

[0137] If the number of iterations reaches the maximum number of iterations, the optimization of the passenger flow density network parameters is stopped.

[0138] The injecting random disturbance into the passenger flow density network comprises:

[0139] When the population fitness variance is lower than the convergence threshold, a heavy-tailed random perturbation is injected into the current optimal passenger flow parameters;

[0140] The amplitude of the random perturbation of the distribution is adjusted along with the optimization process, reducing the potential local optimum through large perturbations, and then ensuring the global optimum of the passenger flow parameters through small perturbations;

[0141] After the perturbation injection is completed, the passenger flow population fitness is re-evaluated and the perturbation results that are effective in the optimization direction are retained;

[0142] The perturbation mechanism and the pitch factor work together to ensure the global convergence of the parameter optimization process.

[0143] Specifically, the population fitness variance is an indicator to measure the diversity of the passenger flow parameter optimization group, reflecting the degree of dispersion of the performance of all current parameter groups in the task of predicting passenger flow density. Among them, the fitness is a comprehensive score of performance indicators such as prediction accuracy and trend judgment reliability of each parameter group in the passenger flow density network. The variance is the degree of deviation of the fitness values ​​of all parameter groups from the average value. The smaller the variance, the smaller the difference between the parameter groups, and the group tends to be homogeneous.

[0144] Furthermore, the heavy-tailed distribution is a probability distribution characterized by a thicker tail probability than the normal distribution, that is, the probability of extreme values ​​occurring is higher.

[0145] like Figure 3 As shown, this is the passenger flow density of the station from Monday to Saturday. It can be seen that the passenger flow from Monday to Thursday is lower than that from Friday to Saturday.

[0146] In this application, the extreme value characteristics of the heavy-tailed distribution are used to make random disturbances change significantly, helping the parameters in the passenger flow density network to jump out of the local optimal area.

[0147] Furthermore, the injection method includes generating random offsets according to the heavy-tailed distribution for each dimension of the current optimal passenger flow density network parameter group, such as network weight, bias, etc., and superimposing these offsets on the original parameters to form a new candidate parameter group.

[0148] In the early stages of passenger flow density network optimization, the parameters of the passenger flow density network are far from the global optimal solution. At this time, a large offset is generated through the heavy-tailed distribution, allowing the parameters to jump over a large range in the solution space. As the passenger flow density network is optimized, the parameters approach the global optimal solution. At this time, the generated offset gradually decreases, and fine-tuning is performed to avoid parameters jumping out of the optimal area due to excessive disturbance, thereby ensuring convergence accuracy.

[0149] The disturbance amplitude decreases with the increase of the number of iterations, and the decreasing rate is fast in the early stage and slow in the later stage. The nonlinear attenuation characteristics enable the passenger flow density network optimization to achieve a smooth transition from coarse search to fine positioning in different optimization stages.

[0150] In this application, the specific implementation method of the setting principle of large disturbance and small disturbance includes:

[0151] The principle of large perturbation setting is that when the population fitness variance is lower than the threshold, or the fitness improvement stagnates after multiple consecutive iterations, a heavy-tailed distribution is used to generate extreme perturbation values ​​with a higher probability. For example, a large offset is imposed on the convolution kernel parameters of the passenger flow density network, allowing the parameters to perform a leapfrog search in the solution space and explore new areas far away from the current location.

[0152] Furthermore, the principle of small perturbation setting is that when the parameters have entered the area near the global optimal solution, for example, the fitness variance and gradient changes can be used to determine whether they have entered the area near the optimal solution.

[0153] Generate extremely small perturbation values ​​to fine-tune the network bias parameters. Small perturbations are used to eliminate the influence of local small fluctuations, so that the parameters can stably converge to the precise position of the global optimal solution.

[0154] The switching between large disturbance and small disturbance is controlled by the pitch factor. When the pitch factor is large, large disturbance is given priority; when the pitch factor is small, it automatically switches to small disturbance.

[0155] By using random perturbations in heavy-tailed distributions, the problem of traditional optimization algorithms easily falling into local optimality is effectively solved. The perturbation amplitude is dynamically adjusted with the optimization process, and it has both global exploration and local development capabilities. In complex passenger flow scenarios, it can quickly adjust parameters and maintain prediction accuracy.

[0156] The optimized passenger flow density network processes the time series data stream and outputs the real-time passenger flow density value and the future trend of passenger flow;

[0157] The processing of the time series data stream by the optimized passenger flow density network includes:

[0158] The entry and exit balance coefficient is dynamically calculated based on the movement direction label, where entry events are positively accumulated and exit events are negatively offset, generating a continuous time series variable that represents the changes in regional passenger flow.

[0159] Direction tags are output by the infrared sensor array in real time as directional semantic identifiers. Each identifier represents a category of human movement direction: an entry event is when the target moves from outside the monitoring area to inside, a departure event is when the target moves from within the monitoring area to outside, and an in-area movement is when the target does not enter or exit the area. Tags are output once at fixed intervals to form a continuously updated data stream that reflects the status of personnel flow in real time.

[0160] The principle of the entry and exit balance coefficient is that if the number of people in the area increases, the entry event is given a positive increment, and if the number of people in the area decreases, it is a leaving event, which is given a negative offset. If the area moves, the value remains unchanged.

[0161] Furthermore, all event quantization values ​​are accumulated in chronological order to generate a continuously changing equilibrium value sequence.

[0162] Continuous time series variables are generated. The specific implementation method can be to fill the gaps between events by performing time interpolation on discrete equilibrium values, thus forming a dense sampling sequence with multiple data points per second.

[0163] Time window smoothing technology is used to eliminate jitter caused by single event mutations and retain the macro trend of passenger flow changes.

[0164] By utilizing the spatiotemporal separation characteristics of the depthwise separable convolutional structure, the coupled characteristics of short-term sudden fluctuations and long-term gradual trends are captured while reducing computational complexity.

[0165] Dynamically evaluate feature importance weights through gated activation units to enhance the feature responses of key passenger flow aggregation / dissipation patterns and suppress noise signals caused by environmental disturbances.

[0166] Generate high-dimensional feature vectors as passenger flow status representation;

[0167] In this application, a door control activation unit includes dynamic weight evaluation and passenger density event reinforcement;

[0168] Specifically, dynamic weight evaluation includes passenger flow pattern recognition and environmental noise suppression;

[0169] Furthermore, passenger flow pattern recognition is used to enhance features related to passenger flow aggregation and dissipation, such as the time interval between consecutive entry events and the burst intensity of departure events.

[0170] Furthermore, environmental noise suppression is used to suppress interference signals that are unrelated to actual passenger flow changes, such as sensor false triggering and small animal movement. By learning the noise characteristic pattern, the interference weight is automatically reduced.

[0171] Specific passenger flow density event enhancements include increasing the weight of features when the frequency of entry events increases and the balance coefficient continues to rise; and increasing the sensitivity of corresponding features when high-intensity departure events occur continuously, providing early warning of area clearance risks.

[0172] Specifically, in this application, a preferred method for generating and characterizing a high-dimensional feature vector by fusion of multi-dimensional features includes:

[0173] Timing characteristics are used to balance the rate of change of coefficients, acceleration, periodic patterns, etc.

[0174] Spatial features are used to analyze the differences in passenger flow density distribution in different monitoring areas and migration trajectories in hot spots.

[0175] Statistical features are used for frequency distribution, time interval distribution, and burst intensity statistics of entry and exit events.

[0176] The physical meaning of eigenvectors.

[0177] Each dimension corresponds to a specific attribute of the passenger flow state. For example, one dimension represents the gathering rate of the entrance area during the morning rush hour; another dimension represents the evacuation uniformity of each exit when the event ends.

[0178] like Figure 4 As shown, the output of real-time passenger flow density value and future passenger flow change trend includes:

[0179] The filtered feature vector is input into the fully connected regression layer of the passenger flow density network, and a standardized density scalar value is generated through nonlinear mapping. The physical meaning of this value is the passenger flow load rate per unit area.

[0180] The first-order derivative of the density scalar is calculated within the sliding time window of the passenger flow density network. Combining the dual criteria of the derivative sign and amplitude, the decreasing trend, stable trend and increasing trend are discretized and output.

[0181] The downward trend is when the derivative is continuously negative and the amplitude is greater than the minimum threshold of passenger flow dissipation, which is marked as a passenger flow dissipation state;

[0182] The stable trend is a state of passenger flow dynamic equilibrium when the derivative is zero and the amplitude is between the minimum threshold of passenger flow dissipation and the maximum threshold of passenger flow gathering;

[0183] The upward trend is when the derivative is continuously positive and the amplitude is greater than the maximum threshold of passenger flow aggregation, which is marked as an accelerated passenger flow aggregation state;

[0184] A hierarchical early warning decision-making mechanism is constructed by combining real-time passenger flow density values ​​and discrete trend states.

[0185] In this application, a specific output of real-time passenger flow density value and future passenger flow change trend includes:

[0186] Before the feature vector is input into the fully connected regression layer, the high-dimensional feature vector will be screened through the gated activation unit, and weights will be assigned to each feature based on its contribution to the passenger flow density prediction. Redundant features will be removed, and key features that are sensitive to changes in passenger flow density, such as changes in the number of people entering and leaving within a specific time period and the length of time spent in the area, will be retained.

[0187] The fully connected regression layer receives the filtered feature vectors, and its internal neurons are interconnected through a weight matrix. The input features are weighted summed and nonlinearly transformed. The weight parameters obtained through pre-training optimization can map the feature vectors to a numerical range related to passenger flow density. Finally, the standardized density scalar value is output, which intuitively reflects the passenger flow load rate per unit area and facilitates subsequent analysis and comparison.

[0188] Furthermore, the analysis of future passenger flow trends includes: defining a sliding time window in the passenger flow density network. The window size is flexibly set based on actual needs and the frequency of passenger flow changes. The window will slide forward at fixed time intervals. Each slide will incorporate a new density scalar value while excluding the earliest data to ensure that the window always contains the latest and most timely passenger flow data for trend analysis.

[0189] Furthermore, the first-order derivative of the density scalar value sequence within the sliding time window is calculated. The first-order derivative reflects the rate of change of passenger flow density over time. By calculating the ratio of the difference between the density scalar values ​​of adjacent time points and the time interval, the derivative of each time point is obtained, thereby quantifying the speed of change of passenger flow density.

[0190] Furthermore, the passenger flow trend is discretized based on the dual criteria of the derivative sign and amplitude, and the minimum threshold for passenger flow dissipation and the maximum threshold for passenger flow aggregation are set as the decision boundaries:

[0191] When the derivative is continuously negative and the amplitude is greater than the minimum threshold for passenger flow dissipation, it is considered a downward trend, indicating that the passenger flow is dissipating rapidly, such as the stage when the crowd leaves after the event.

[0192] When the derivative is zero and the amplitude is between the minimum threshold for passenger flow dissipation and the maximum threshold for passenger flow aggregation, it is determined to be a stable trend, which means that the passenger flow is dynamically balanced and the number of people entering and leaving is basically the same;

[0193] When the derivative is continuously positive and its amplitude is greater than the maximum threshold of customer flow aggregation, it is determined to be an upward trend, indicating that the customer flow is in an accelerated aggregation state, such as when a large number of customers flock to the mall at the beginning of a promotion.

[0194] With the real-time passenger flow density value as the horizontal axis and the future passenger flow change trend as the vertical axis, a two-dimensional decision space is constructed. The density value is divided into three intervals: loose, normal, and crowded. The trend status is divided into three categories: declining, stable, and rising, forming different combination scenarios.

[0195] Different warning rules are set for different scenarios. For example, when the density value is in the relaxed range, passenger flow pressure is considered low and no warning is triggered. When the density value enters the normal range and the trend is rising, the warning level is increased in a square relationship based on the density increase rate, emphasizing the amplifying effect of the upward trend on risk. When the density value exceeds the congestion threshold, a basic warning is immediately triggered, and an acceleration factor is superimposed based on the passenger flow surge trend to increase the urgency of the warning, ultimately generating a progressively graded warning instruction from observation level to emergency level.

[0196] The hierarchical warning instructions are generated through the dynamic combination relationship between the passenger flow density value and the future change trend of the passenger flow. The level of the hierarchical warning instructions increases nonlinearly with the rate of increase of density.

[0197] Generating a graded warning instruction through a dynamic combination relationship includes:

[0198] A two-dimensional decision space is established with real-time passenger flow density as the horizontal axis and change trend as the vertical axis. The density value is divided into three intervals: loose, normal, and crowded. The trend status is divided into three categories: declining, stable, and rising.

[0199] When the density value is in the loose range, the trend change is ignored and no warning is triggered;

[0200] When the density value enters the normal range and the trend is rising, the warning level increases in a square relationship with the density increase rate;

[0201] When the density value exceeds the congestion threshold, a basic warning is immediately triggered, and an acceleration factor for the passenger flow surge trend is added;

[0202] Generate progressive commands from observation level 1 to emergency level 3, where Level 3 commands force the activation of emergency equipment.

[0203] In this application, a preferred two-dimensional decision space construction implementation method includes:

[0204] The density value is divided into three intervals: loose, normal, and crowded. The threshold of each interval is determined based on the statistical analysis of historical passenger flow data and the carrying capacity of the venue.

[0205] Furthermore, the two-dimensional decision space presents the passenger flow status in a visual and structured manner, which facilitates the system to quickly locate the current passenger flow status and provides a framework for the generation of early warning instructions.

[0206] Furthermore, when the passenger flow density value is in a loose range, it indicates that there are fewer people in the current area. Even if there is a future trend of passenger flow changes, the overall risk is still at a low level. Therefore, the trend changes are ignored and no warnings are triggered, thereby reducing unnecessary prompts and reducing operational interference.

[0207] If the density value enters the normal range and the trend is upward, as the passenger flow density increases at an accelerated rate, the potential risk increases nonlinearly. The faster the passenger flow density increases, the higher the possibility of reaching a crowded state in a short period of time, and the risk level increases faster.

[0208] If the passenger flow density value exceeds the congestion threshold, the passenger flow in the area has reached a high risk level, and the basic warning is immediately triggered. If the passenger flow is in a rapid gathering state, the acceleration factor increases and the warning level is raised.

[0209] Specifically, in the warning level, progressive instructions are generated from observation level 1 to emergency level 3, and different levels correspond to different response strategies;

[0210] Specifically, Level 1 is the observation level, which only prompts people to pay attention to changes in passenger flow; Level 2 is the warning level, which recommends taking measures such as adding guide personnel; Level 3 is the emergency level, which forces emergency equipment to be linked, such as the automatic closing of flow-limiting gates and emergency notifications on the broadcasting system.

[0211] Continuously monitor passenger flow density values ​​and trend changes, and update the position in the two-dimensional decision space in real time. Once the passenger flow status changes, immediately re-evaluate and adjust the warning level to ensure that the warning instructions always fit the actual risk situation.

[0212] Execute graded warning instructions and collect actual passenger flow change data, and periodically re-optimize passenger flow density network parameters based on feedback data.

[0213] The periodic re-optimization based on feedback data includes:

[0214] Compare the actual density change rate before and after the early warning instruction is executed to calculate the strategy effectiveness coefficient;

[0215] The re-optimization of passenger flow density network parameters includes timed triggering and event triggering. The timed triggering is to perform passenger flow density network parameter optimization every fixed period. The event triggering is to immediately start passenger flow network parameter optimization when the effectiveness coefficient of the strategy is not greater than the minimum parameter optimization threshold.

[0216] The execution of the hierarchical warning instructions and periodic re-optimization based on feedback data includes:

[0217] Trigger differentiated response operations based on warning levels, including information broadcasting, channel control, and personnel guidance;

[0218] Synchronously collect the actual passenger flow density change rate within the set time window after the instruction is executed;

[0219] Compare the density drop rate expected by the warning with the actual drop rate and calculate the effectiveness coefficient η;

[0220] When η is not greater than the minimum parameter optimization threshold, it is marked as a low-efficiency response event;

[0221] The network parameter update is started after every N warning executions, and is started immediately when K low-efficiency response events have accumulated.

[0222] Example 2

[0223] like Figure 2 As shown, a passenger flow detection system, the specific implementation method includes: an acquisition module, a calculation module, a prediction module, an early warning module and a feedback module;

[0224] Specifically, the acquisition module includes a sensor array and a signal conditioning unit. The sensor array collects human body movement direction data, and the signal conditioning unit processes the signal to generate a structured time series data stream containing motion characteristics.

[0225] The calculation module includes an optimization unit and a lightweight network container. The optimization unit generates an initial parameter set through parameter initialization, dynamically adjusts the optimization step size strategy based on changes in passenger flow data, optimizes the passenger flow density network parameters, and introduces a perturbation mechanism that can change the parameter search direction to prevent the passenger flow density network from falling into a local optimal solution.

[0226] The prediction module includes an inflow and outflow balance analysis unit and a density mapping unit. The balance analysis unit uses the collected time series data stream as the data input of the optimized passenger flow density network, converting it into real-time passenger flow density values ​​and changing trends. During the trend prediction process, the density mapping unit assigns different weights based on the data's freshness, highlighting the impact of recent data on the prediction results.

[0227] The early warning module includes a dynamic threshold calculation unit and a hierarchical logic unit. Based on the dynamic combination of real-time passenger flow density values ​​and change trends, combined with historical early warning data and actual conditions, it dynamically adjusts the early warning threshold and generates hierarchical early warning instructions.

[0228] The feedback module includes a device control interface and an effect collection unit. The device control interface is used to execute graded warning instructions and collect actual passenger flow change data. The feedback data is analyzed through a specific evaluation mechanism to analyze the execution effect, driving the edge computing module to re-optimize network parameters according to a cycle that gradually shortens as the system runs, forming a closed-loop self-evolution system.

[0229] It is important to note that the configuration and arrangement of the present application, as illustrated in various exemplary embodiments, are illustrative only. Although only two embodiments are described in detail in this disclosure, those reading this disclosure will readily appreciate that numerous modifications are possible without materially departing from the novel teachings and advantages of the subject matter described herein. For example, variations in the size, dimensions, structure, shape, and proportions of various components, as well as parameter values ​​(e.g., temperature, pressure, etc.), mounting arrangements, use of materials, color, orientation, and the like, are possible. For example, components shown as integrally formed may be comprised of multiple parts or components, the positions of components may be inverted or otherwise altered, and the nature, number, or position of discrete components may be modified or changed. Therefore, all such modifications are intended to be encompassed within the scope of this invention. The order or sequence of any process or method steps may be altered or reordered according to alternative embodiments. Any "means-plus-function" clause is intended to cover structures that perform the functions described herein, and not only structural equivalence but also structural equivalents. Other substitutions, modifications, changes, and omissions may be made in the design, operating conditions, and arrangement of the exemplary embodiments without departing from the scope of this invention. Therefore, the invention is not limited to the specific embodiments, but extends to various modifications that still fall within the scope of the appended claims.

[0230] Additionally, in order to provide a concise description of exemplary embodiments, all features of an actual embodiment may not be described (ie, those features that are not relevant to the best mode presently contemplated for carrying out the invention or those that are not relevant to implementing the invention).

[0231] It should be understood that in the development of any actual embodiment, as in any engineering or design project, numerous implementation-specific decisions may be made. Such a development effort may be complex and time-consuming, but for those of ordinary skill having the benefit of this disclosure, the development effort will be a routine task of design, fabrication, and production without undue experimentation.

[0232] 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 the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A passenger flow detection method, characterized in that: include: The sensor array collects the movement direction data of the human body in the monitoring area and generates a time series data stream after signal conditioning; The passenger flow density network parameter population of the passenger flow density network is initialized by chaotic mapping, the passenger flow density network parameter population is optimized by nonlinear decreasing pitch, and random disturbance is injected into the passenger flow density network. The method for initializing the passenger flow density network parameter population using chaotic mapping includes: By deterministically traversing passenger flow characteristics through chaotic mapping, a chaotic sequence is generated; Map the chaotic sequence to the adjustable parameter definition domain of the passenger flow density network, so that the initial passenger flow population is evenly distributed in the solution space; Preset the parameter boundary constraints of passenger flow populations and generate a set of passenger flow density network diversity parameter combinations; The method for optimizing the passenger flow density network parameter population by nonlinear decreasing pitch includes: An initial pitch factor is set in the passenger flow density network to enable the passenger flow population to perform a wide-area search in the solution space; A pitch factor is attenuated according to the progress of the wide-area search of the passenger flow population in the solution space, wherein the rate of attenuation decreases as the number of iterations increases; After optimizing the passenger flow density network, the passenger flow population is determined by the minimum pitch factor to perform development on the local optimal area; The injecting random disturbance into the passenger flow density network comprises: When the population fitness variance is lower than the convergence threshold, a heavy-tailed random perturbation is injected into the current optimal passenger flow parameters; The amplitude of the random perturbation of the distribution is adjusted along with the optimization process, breaking through the potential local optimum through large perturbations, and then ensuring the global optimum of the passenger flow parameters through perturbations; After the perturbation is injected, the passenger flow population fitness is re-evaluated; The optimized passenger flow density network processes the time series data stream and outputs the real-time passenger flow density value and the future trend of passenger flow; The hierarchical warning instructions are generated through the dynamic combination relationship between the passenger flow density value and the future change trend of the passenger flow. The level of the hierarchical warning instructions increases nonlinearly with the rate of increase of density. Execute graded warning instructions and collect actual passenger flow change data, and periodically re-optimize passenger flow density network parameters based on feedback data.

2. A passenger flow detection method according to claim 1, characterized in that: The method for generating a time series data stream includes: The human body position coordinates are calculated by weighting the human body's movement direction data and correlating the motion trajectory in continuous time frames; Dynamically determine the movement direction category based on the motion trajectory displacement vector and generate a semantic direction label; Encapsulate timestamps, device identifiers, direction tags, and location coordinates into standardized data units; Pushing data units at fixed time intervals forms a time-series data stream.

3. A passenger flow detection method according to claim 1, characterized in that: The processing of the time series data stream by the optimized passenger flow density network includes: The entry and exit balance coefficient is dynamically calculated based on the movement direction label, where entry events are positively accumulated and exit events are negatively offset, generating a continuous time series variable that represents the changes in regional passenger flow. By utilizing the spatiotemporal separation characteristics of the depthwise separable convolutional structure, the coupled characteristics of short-term sudden fluctuations and long-term gradual trends are captured while reducing computational complexity. Dynamically evaluate feature importance weights through gated activation units to enhance the feature responses of key passenger flow aggregation / dissipation patterns and suppress noise signals caused by environmental disturbances. Generate high-dimensional feature vectors as passenger flow status representation.

4. A passenger flow detection method according to claim 3, characterized in that: The output of real-time passenger flow density value and future passenger flow change trend includes: The filtered feature vector is input into the fully connected regression layer of the passenger flow density network, and a standardized density scalar value is generated through nonlinear mapping. The physical meaning of this value is the passenger flow load rate per unit area. The first-order derivative of the density scalar is calculated within the sliding time window of the passenger flow density network. Combining the dual criteria of the derivative sign and amplitude, the decreasing trend, stable trend and increasing trend are discretized and output. The downward trend is that the derivative is continuously negative and the amplitude is greater than the minimum threshold of passenger flow dissipation; The stable trend is when the derivative is zero and the amplitude is between the minimum threshold of passenger flow dissipation and the maximum threshold of passenger flow gathering; The upward trend is that the derivative is continuously positive and the amplitude is greater than the maximum threshold of passenger flow aggregation; A hierarchical early warning decision-making mechanism is constructed by combining real-time passenger flow density values ​​and discrete trend states.

5. The passenger flow detection method according to claim 1, wherein: Generating a graded warning instruction through a dynamic combination relationship includes: A two-dimensional decision space is established with real-time passenger flow density as the horizontal axis and change trend as the vertical axis. The density value is divided into three intervals: loose, normal, and crowded. The trend status is divided into three categories: declining, stable, and rising. When the density value is in the loose range, the trend change is ignored and no warning is triggered; When the density value enters the normal range and the trend is rising, the warning level increases in a square relationship with the density increase rate; When the density value exceeds the congestion threshold, a basic warning is immediately triggered, and an acceleration factor for the passenger flow surge trend is added; Generate progressive commands from observation level 1 to emergency level 3, where Level 3 commands force the activation of emergency equipment.

6. A passenger flow detection method according to claim 1, characterized in that: The periodic re-optimization based on feedback data includes: Compare the actual density change rate before and after the early warning instruction is executed to calculate the strategy effectiveness coefficient , is the actual passenger flow change, △ represents the change mark, To predict the change in passenger flow density, D represents the passenger flow density index; The re-optimization of passenger flow density network parameters includes timing triggering and event triggering. The timing triggering is to perform passenger flow density network parameter optimization every fixed period. The event triggering is to immediately start passenger flow network parameter optimization if the effectiveness coefficient of the strategy is not greater than the minimum parameter optimization threshold.

7. A passenger flow detection system, based on a passenger flow detection method according to any one of claims 1 to 6, characterized in that: include: Acquisition module, calculation module, prediction module, early warning module and feedback module; The acquisition module is used to collect human body movement direction data and generate a structured time series data stream containing movement characteristics through signal processing; The calculation module generates an initial parameter group through parameter initialization, optimizes the parameters of the passenger flow density network by adopting a strategy that can dynamically adjust the optimization step size according to data changes, and performs a perturbation test on the passenger flow density network by introducing a perturbation mechanism that can change the parameter search direction; The prediction module uses the collected time series data stream as the data input of the optimized passenger flow density network, outputs the real-time passenger flow density value and change trend, and predicts the future change trend of passenger flow, assigning different weights to the future change trend of passenger flow according to the freshness of the data; The warning module dynamically adjusts the warning threshold value and generates graded warning instructions through the dynamic combination relationship between the real-time passenger flow density value and the change trend and the historical warning data; The feedback module is used to execute the graded warning instructions and collect actual passenger flow change data, analyze the execution effect through the evaluation mechanism, and drive the calculation module to re-optimize the network parameters in a cycle that gradually shortens as the system operation time.

Images