Smart park personnel trajectory tracking method and system based on data analysis

Abstract

The invention relates to the technical field of data processing, and discloses a smart park personnel trajectory tracking method and system based on data analysis. The method comprises the following steps: acquiring position information of personnel in a park through a multi-source sensor and carrying out standardization processing to obtain preprocessed track data; mapping the trajectory data to a three-dimensional tensor space and extracting feature vectors; identifying and coding a personnel behavior state based on the track features; and inputting the trajectory features and the behavior codes into a multi-layer variable neural network model for training to realize real-time trajectory prediction and abnormal behavior alarm. The technical problems of difficulty in multi-source data fusion, low trajectory prediction precision, inaccurate behavior recognition and poor anomaly detection real-time performance in an existing smart park personnel trajectory tracking method are solved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method and system for tracking personnel trajectories in a smart park based on data analysis. Background Art

[0002] Existing smart campus personnel tracking technology primarily relies on single-sensor data collection and simple path analysis methods. This involves deploying devices such as cameras, access control systems, or wireless probes to acquire personnel location information. Trajectory prediction is then performed using Kalman filtering or particle filtering algorithms. Euclidean distance calculations are used to measure trajectory similarity, and linear interpolation algorithms are used to connect discrete trajectory points to form movement paths. Traditional solutions typically represent trajectory data as a two-dimensional vector sequence and use single-layer neural networks based on vector space or simple machine learning algorithms for trajectory analysis and behavior recognition. These approaches primarily rely on statistical analysis to identify basic movement patterns.

[0003] However, existing technologies have many technical defects: first, the data processing capability is insufficient. Traditional vector space computing models cannot effectively process the massive and complex multi-source trajectory data in the park, and cannot fully extract the deep features in the trajectory data; second, the trajectory prediction accuracy is limited. The existing simple neural network structure cannot learn the multi-level and multi-scale features of personnel movement data, and it is difficult to deal with the contingency and randomness of trajectory data; third, there is a lack of multimodal data fusion capability. Traditional methods cannot effectively integrate heterogeneous data from different sensors, affecting the comprehensiveness and accuracy of trajectory tracking; finally, the real-time and adaptability are poor. The existing algorithms respond slowly when faced with large-scale real-time data streams and are difficult to adapt to the complex and changing park environment.

[0004] Traditional single-sensor data acquisition and two-dimensional vector representation methods have problems with information loss and incomplete feature extraction when processing multi-source heterogeneous data. There is an urgent need to solve the problems of effective fusion of multi-source sensor data and complete expression of three-dimensional spatiotemporal features. Based on the acquisition of trajectory features, existing technologies lack a deep understanding of the semantics of human behavior and are unable to correlate and analyze the physical motion characteristics of the trajectory with the behavioral state characteristics. Therefore, it is necessary to solve the technical difficulties of trajectory-behavior dual feature fusion and accurate identification of behavioral state. Furthermore, traditional neural networks show insufficient learning ability and poor convergence when processing trajectory data with spatiotemporal correlation. It is necessary to solve the problems of neural network architecture design and parameter optimization algorithm improvement that adapt to the characteristics of trajectory data. Finally, how to achieve real-time detection and graded warning of abnormal behavior based on accurate trajectory prediction results and establish a complete intelligent monitoring system has become a key technical bottleneck for intelligent park security management. Summary of the Invention

[0005] The present application provides a method and system for tracking personnel trajectories in a smart park based on data analysis, which is used to solve the technical problems of difficulty in multi-source data fusion, low trajectory prediction accuracy, inaccurate behavior recognition, and poor real-time anomaly detection in existing methods for tracking personnel trajectories in smart parks.

[0006] In the first aspect, the present application provides a method for tracking personnel trajectories in a smart park based on data analysis, and the method for tracking personnel trajectories in a smart park based on data analysis includes: collecting and processing the location information of personnel in the park through multi-source sensors to obtain an original trajectory data set, standardizing the original trajectory data set to obtain preprocessed trajectory data; mapping the preprocessed trajectory data to a three-dimensional tensor space for construction processing to obtain a trajectory spatiotemporal tensor, performing feature extraction processing on the trajectory spatiotemporal tensor to obtain a trajectory feature vector; identifying the behavior status of personnel according to the trajectory feature vector to obtain a behavior status classification result, encoding the behavior status classification result to obtain a behavior coding vector; inputting the trajectory feature vector and the behavior coding vector into a multi-layer variable neural network model for training processing to obtain a trained multi-layer variable neural network model, performing real-time prediction processing according to the multi-layer variable neural network model to obtain a personnel location coordinate sequence and abnormal behavior alarm information.

[0007] In a second aspect, the present application provides a smart park personnel trajectory tracking system based on data analysis, the smart park personnel trajectory tracking system based on data analysis includes:

[0008] An acquisition module is used to collect and process the location information of park personnel through multi-source sensors to obtain an original trajectory data set, and to perform standardization processing on the original trajectory data set to obtain pre-processed trajectory data;

[0009] A construction module, configured to map the pre-processed trajectory data into a three-dimensional tensor space for construction processing to obtain a trajectory spatiotemporal tensor, and perform feature extraction processing on the trajectory spatiotemporal tensor to obtain a trajectory feature vector;

[0010] an identification module, configured to identify the behavior state of a person according to the trajectory feature vector to obtain a behavior state classification result, and encode the behavior state classification result to obtain a behavior encoding vector;

[0011] The prediction module is used to input the trajectory feature vector and the behavior coding vector into a multi-layer variable neural network model for training processing to obtain a trained multi-layer variable neural network model, and perform real-time prediction processing based on the multi-layer variable neural network model to obtain a personnel position coordinate sequence and abnormal behavior alarm information.

[0012] On the third aspect, a smart park personnel trajectory tracking device based on data analysis is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the smart park personnel trajectory tracking device based on data analysis executes the above-mentioned smart park personnel trajectory tracking method based on data analysis.

[0013] In a fourth aspect, a computer-readable storage medium is provided, in which instructions are stored. When the computer-readable storage medium is run on a computer, the computer executes the above-mentioned smart park personnel trajectory tracking method based on data analysis.

[0014] In the technical solution provided by this application, the technical feature of collecting, processing and standardizing the location information of park personnel through multi-source sensors solves the problems of incomplete data from traditional single sensors and time asynchrony of multi-source data, realizes the effective fusion of heterogeneous data such as cameras, access control systems and wireless probes, and significantly improves the integrity and reliability of trajectory data. The technical feature of mapping the pre-processed trajectory data to a three-dimensional tensor space for construction and processing breaks through the limitations of traditional two-dimensional vector representation. The construction of the trajectory spatiotemporal tensor can simultaneously retain the spatial position, time series and correlation relationship of multi-source features of the trajectory, and can extract richer trajectory feature information than traditional methods. The technical feature of identifying, processing and encoding the behavior status of personnel based on trajectory feature vectors realizes the effective combination of trajectory physical features and behavioral semantic features. The behavior recognition algorithm through multi-dimensional feature fusion can accurately distinguish between five states such as stillness, slow walking, fast walking, gathering and abnormal behavior. Compared with the traditional simple judgment method based on speed threshold, it has higher recognition accuracy and stronger adaptability.

[0015] The multi-layer variable neural network model and the trajectory tensor high-order backpropagation algorithm play a key role in the application of personnel trajectory tracking in smart parks. The variable weight mechanism can dynamically adjust the weight parameters according to the importance of different trajectory features, allowing the network to better adapt to the complexity and diversity of personnel movement in the park environment, and significantly improves the trajectory prediction accuracy compared to traditional fixed-weight neural networks. The trajectory tensor high-order backpropagation algorithm is specifically optimized for the spatiotemporal correlation characteristics of trajectory data. It considers both temporal continuity and spatial neighborhood constraints during gradient calculation, effectively solving the convergence difficulties of traditional backpropagation algorithms when processing spatiotemporal series data, making network training more stable and efficient. The technical features of real-time prediction processing and abnormal behavior alerts realize the integration of trajectory prediction and security monitoring. Through a multi-level anomaly determination mechanism, including trajectory deviation detection and out-of-bounds behavior detection, it can promptly detect and warn of various abnormal situations, and has stronger proactive prevention capabilities than traditional passive monitoring methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0017] Figure 1 This is a schematic diagram of an embodiment of a method for tracking personnel trajectories in a smart park based on data analysis in an embodiment of the present application;

[0018] Figure 2 This is a schematic diagram of an embodiment of a smart park personnel trajectory tracking system based on data analysis in an embodiment of the present application;

[0019] Figure 3 It is a schematic block diagram of the structure of a smart park personnel trajectory tracking device based on data analysis in an embodiment of the present invention. DETAILED DESCRIPTION

[0020] The embodiments of the present application provide a method and system for tracking personnel trajectories in a smart park based on data analysis. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way are interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or that are inherent to these processes, methods, products or devices.

[0021] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiments of the present application, an embodiment of a method for tracking personnel trajectories in a smart park based on data analysis includes:

[0022] Step S101: collect and process the location information of park personnel through multi-source sensors to obtain an original trajectory data set, and standardize the original trajectory data set to obtain pre-processed trajectory data;

[0023] Step S102: Map the pre-processed trajectory data to a three-dimensional tensor space for construction processing to obtain a trajectory spatiotemporal tensor, perform feature extraction processing on the trajectory spatiotemporal tensor to obtain a trajectory feature vector;

[0024] Step S103: Identify the behavior state of the person according to the trajectory feature vector to obtain a behavior state classification result, and encode the behavior state classification result to obtain a behavior encoding vector;

[0025] Step S104: Input the trajectory feature vector and the behavior coding vector into the multi-layer variable neural network model for training processing to obtain a trained multi-layer variable neural network model. Perform real-time prediction processing based on the multi-layer variable neural network model to obtain a personnel position coordinate sequence and abnormal behavior alarm information.

[0026] It is understandable that the execution subject of this application can be a smart park personnel trajectory tracking system based on data analysis, or a terminal or server, which is not limited here. The embodiment of this application is described by taking the server as the execution subject as an example.

[0027] Specifically, multi-source sensor acquisition processing first locates and scans people within the campus using cameras, access control systems, and wireless probes. Cameras extract people's location coordinates and identity features using target detection algorithms. The access control system records people's entry and exit times and locations, and wireless probes capture mobile device MAC addresses and signal strength values. Time synchronization and correction processing aligns data collected by different sensors according to a unified time base, eliminating time deviations between sensors. Noise identification and rejection processing filters out poor-quality data points by setting signal strength thresholds and distance ranges. Coordinate transformation processing converts the local coordinate system of each sensor into the campus' global coordinate system. Personnel identity grouping processing categorizes trajectory data by individual based on person features identified by the camera and access card information. Sliding window algorithm smoothing processing uses a fixed window size to perform a weighted average of consecutive trajectory points, eliminating sudden changes and jitter in the trajectory.

[0028] The three-dimensional tensor space construction process arranges the pre-processed trajectory data according to the three dimensions of spatial coordinates x, y and time t to form a three-dimensional data structure. The tensor construction algorithm converts discrete trajectory point data into a continuous tensor form through dimensional reorganization. The direction angle calculation process obtains the movement direction vector by calculating the coordinate difference between adjacent time points, and the speed calculation process calculates the speed change rate parameter based on the position change and time interval. The stay frequency statistical processing analyzes the distribution of personnel's stay time in various areas, and the continuity analysis processing calculates the curvature change of the trajectory path. The vector combination processing linearly combines the movement direction vector, speed change rate parameter, stay pattern parameter and path curvature parameter according to preset weights to form a comprehensive trajectory feature vector.

[0029] The behavior state recognition processing performs threshold judgment based on the speed change rate parameter in the trajectory feature vector, and divides the movement speed of the person into different speed state categories. The stop behavior duration statistical processing calculates the time a person stays in a specific area based on the stop pattern parameters to form a stop state classification. The combination matching processing logically combines the speed state and the stop state to identify five basic behavior states: stationary, slow walking, fast walking, gathering, and abnormal behavior. The probability calculation processing calculates the probability of occurrence for each behavior state to form a behavior state probability distribution. The maximum probability judgment processing selects the behavior state with the highest probability as the current behavior classification result. The state verification processing verifies the rationality of the behavior state according to the preset behavior transfer rules. The one-hot encoding algorithm converts the confirmed behavior state into a numerical vector form, and each behavior state corresponds to a unique binary code.

[0030] The feature fusion process combines the trajectory feature vector and the behavior encoding vector in a preset ratio to form a comprehensive feature input vector. The multi-layer variable neural network model consists of an input layer, multiple hidden layers, and an output layer, with each layer connected by a weight matrix. The forward propagation process passes the input data through each network layer, applying an activation function to each layer for nonlinear transformation. The trajectory tensor high-order backpropagation algorithm calculates the gradient of the loss function with respect to the network parameters, while taking into account the spatiotemporal correlation of the trajectory data. The weight parameter update optimization process adjusts the network weights based on the calculated gradient values, and the convergence verification process checks the convergence of the loss function during training. The real-time prediction process inputs new trajectory data into the trained network model and obtains the predicted position coordinates through forward propagation. The deviation comparison process calculates the Euclidean distance between the predicted coordinates and the actual coordinates. When the deviation exceeds a preset threshold, an abnormal behavior alert is triggered.

[0031] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0032] Based on cameras, access control systems, and wireless probes, personnel in the park are located and scanned to obtain multi-source location data. The multi-source location data is then time-synchronized and corrected to obtain a time-aligned location dataset.

[0033] Perform noise identification and elimination on the time-aligned position data set to obtain valid position data, and perform coordinate conversion on the valid position data according to the park coordinate system to obtain standard coordinate trajectory data;

[0034] The standard coordinate trajectory data are grouped according to the personnel identity to obtain individual trajectory data groups, and the individual trajectory data groups are sorted in time series to obtain a continuous time trajectory sequence;

[0035] The continuous time trajectory sequence is smoothed based on the sliding window algorithm to obtain smoothed trajectory data, which is then resampled to obtain preprocessed trajectory data.

[0036] Specifically, the camera uses an image recognition algorithm to detect moving targets within the park, extracting the pixel coordinates and timestamp information of each person, and simultaneously identifying the person's identity code. The access control system records the exact timestamp and physical coordinates of the access point when a person swipes their card to pass through, generating data that associates the identity with the location. The wireless probe receives WiFi or Bluetooth signals sent by the mobile device, calculates the distance between the device and the probe based on the signal strength and propagation time, and infers the approximate location of the mobile device based on the known coordinates of the probe. Time synchronization and correction processing aligns the data collected by the three sensors according to a unified time base, eliminating clock deviations between the sensors and forming a multi-source location data set at the same moment.

[0037] Noise identification identifies and marks anomalous data points by setting a lower signal strength threshold and a reasonable range for location coordinates. Elimination removes data points from the dataset if their signal strength is too low, their coordinates fall outside the campus boundary, or their timestamps are abnormal. Coordinate conversion converts the camera's pixel coordinates into actual physical coordinates using a perspective transformation matrix. The local coordinates of the access control system and wireless probes are uniformly converted to the campus' global coordinate system, creating a unified coordinate representation format. Standard coordinate trajectory data includes location information, corresponding timestamps, and individual identification in a unified coordinate system.

[0038] The identity grouping process categorizes the trajectory data by individual based on the person's signature code, access card number, or mobile device MAC address identified by the camera. Each individual trajectory data group contains all the location records of the person within a specific time period. The time series sorting process arranges the trajectory data of the same individual in ascending order according to the order of the timestamps to form a continuous time trajectory sequence, eliminating the timing confusion caused by transmission delays during the data acquisition process. The sliding window algorithm sets a fixed-size time window, slides on the continuous time trajectory sequence, and calculates the weighted average of the trajectory points within the window. The smoothing process reduces the trajectory jitter caused by sensor accuracy limitations or environmental interference by calculating the weighted average of the coordinates of each trajectory point in the window. The resampling process uniformly samples the smoothed trajectory data at preset time intervals to ensure the temporal density consistency of the trajectory data.

[0039] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0040] The pre-processed trajectory data is arranged in three-dimensional space according to the park grid coordinates and time series to obtain space-time structured data. The space-time structured data is re-dimensionalized based on the tensor construction algorithm to obtain the trajectory space-time tensor.

[0041] The direction angle calculation is performed on the trajectory coordinates of adjacent time points in the trajectory space-time tensor to obtain the moving direction vector. The velocity calculation is performed on the trajectory space-time tensor according to the position change of consecutive time points to obtain the velocity change rate parameter.

[0042] Based on the dwell time distribution in the trajectory space-time tensor, the dwell frequency statistics of each area of ​​the park are processed to obtain the dwell pattern parameters, and the movement direction vector is subjected to continuity analysis to obtain the path curvature parameters;

[0043] The moving direction vector, velocity change rate parameter, dwell mode parameter and path curvature parameter are processed by vector combination to obtain the trajectory feature vector.

[0044] Specifically, the three-dimensional spatial arrangement processing first divides the park into regular grid units, each grid unit represents a spatial position node, and each position coordinate in the preprocessed trajectory data is mapped to the corresponding grid node. The spatial-temporal structured data organizes the trajectory data according to three dimensions: x-coordinate, y-coordinate, and time t to form a three-dimensional matrix structure. The tensor construction algorithm converts the original two-dimensional trajectory point sequence into a three-dimensional tensor structure through dimensional reorganization processing. Each element of the tensor contains the trajectory information of a specific space-time position. The trajectory space-time tensor is a three-dimensional data structure, in which the first dimension represents the spatial position in the x-direction, the second dimension represents the spatial position in the y-direction, and the third dimension represents the time series. Each numerical value in the tensor indicates whether there is a trajectory point at that space-time position and its related attribute information.

[0045] The direction angle calculation process calculates the direction angle between two consecutive position points by extracting the coordinate differences between adjacent time points in the trajectory spatiotemporal tensor. The movement direction vector forms a two-dimensional vector describing the movement direction by calculating the coordinate changes between adjacent trajectory points. The direction of the vector represents the movement direction, and the modulus of the vector represents the distance traveled. The speed calculation process calculates the instantaneous speed value by dividing the position change between consecutive time points by the time interval. The speed change rate parameter reflects the acceleration or deceleration trend of the person's movement speed by calculating the change in speed values ​​within a continuous time period. This parameter is calculated by subtracting the previous speed from the current time period and dividing it by the time interval. The dwell frequency statistics process analyzes the length of time a person spends in each grid area in the trajectory spatiotemporal tensor, counting the number of visits and total dwell time in each area. The dwell pattern parameter consists of three sub-parameters: the spatial distribution of dwell areas, the dwell duration distribution, and the dwell frequency distribution. Statistical analysis of dwell data reveals the spatial movement patterns of the person. The continuity analysis process calculates the directional changes of the movement direction vector within a continuous time period and assesses the curvature of the trajectory path by calculating the change in the angle between adjacent direction vectors. The path curvature parameter reflects the curvature characteristics of the trajectory path by calculating the angle change formed by three consecutive trajectory points. The larger the curvature value, the more curved the path, and the smaller the curvature value, the straighter the path.

[0046] Vector combination processing linearly combines the four characteristic parameters according to preset weights to form a comprehensive trajectory feature vector. The movement direction vector serves as the directional feature component, the velocity change rate parameter serves as the dynamic feature component, the dwell pattern parameter serves as the spatial behavior feature component, and the path curvature parameter serves as the trajectory shape feature component. The trajectory feature vector combines these four feature components in series to form a multi-dimensional feature description vector that comprehensively describes the spatial, temporal, and behavioral characteristics of a person's trajectory.

[0047] For example, during the trajectory processing of Li, an employee of a certain campus, moving from the office building to the cafeteria, the preprocessed trajectory data was mapped to the campus's grid coordinate system, and a tensor construction algorithm converted the trajectory data into a three-dimensional tensor structure. Direction angle calculations revealed that Li's direction vector pointed southeast when he started from the office building, then changed direction midway to due south, and finally pointed southwest when he arrived at the cafeteria. Speed ​​calculations showed that Li moved slowly near the office building, quickly on the campus's main roads, and then slowed down again as he approached the cafeteria. The speed change rate parameter reflects this acceleration and deceleration process. Stop frequency statistics revealed that Li made significant stops at the office building entrance and the cafeteria entrance. The stop pattern parameter recorded these stops and their duration. Continuity analysis revealed that Li's movement path was relatively straight along the main road, with significant directional changes at bends, with the path curvature parameter peaking at the bends. Vector combination processing combines direction features, speed features, stay features and curvature features into a comprehensive feature vector, which fully describes the trajectory characteristics of Li from the office building to the cafeteria, solving the technical problems of incomplete trajectory feature extraction and difficult multi-scale feature fusion in traditional methods.

[0048] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0049] Based on the speed change rate parameter in the trajectory feature vector, the threshold judgment processing of the personnel movement speed is performed to obtain the speed state classification. According to the stay mode parameter in the trajectory feature vector, the duration of the personnel's stay behavior is statistically processed to obtain the stay state classification;

[0050] The speed state classification and the stop state classification are combined and matched to obtain five basic behavior states: stationary, slow walking, fast walking, gathering and abnormal behavior. The probability of the five basic behavior states is calculated to obtain the behavior state probability distribution.

[0051] Based on the probability distribution of behavior status, the current behavior of the park personnel is judged with the maximum probability to obtain the behavior status classification result, and the behavior status classification result is verified according to the preset behavior transfer rules to obtain the confirmed behavior status classification result;

[0052] The confirmed behavior state classification results are input into the one-hot encoding algorithm for numerical conversion processing to obtain the behavior encoding vector.

[0053] Specifically, the threshold judgment process classifies the speed change rate parameter by setting multiple speed threshold intervals. When the speed change rate parameter is lower than the stationary threshold, it is marked as a stationary state; when the parameter is in the slow speed interval, it is marked as a slow movement state; when the parameter is in the fast speed interval, it is marked as a fast movement state. The speed state classification divides the movement state of the person into three basic speed categories: stationary, slow, and fast according to the numerical range of the speed change rate parameter. The duration statistics process analyzes the dwell time data in the dwell mode parameter to calculate the continuous length of time a person stays at the same location. When the dwell time exceeds the set threshold, it is determined to be a valid dwell behavior. The dwell state classification divides dwell behavior into three dwell state categories: short dwell, long dwell, and no dwell by statistically analyzing the dwell time distribution and dwell frequency distribution.

[0054] The combination matching process logically combines the speed state classification and the stop state classification. The stationary state combined with long-term stay forms stationary behavior, the slow state combined with the moving trajectory forms slow walking behavior, the fast state combined with continuous movement forms fast walking behavior, multiple people stopping at the same time forms gathering behavior, and abnormal speed changes or abnormal stop patterns form abnormal behavior. The five basic behavior states include stationary, slow walking, fast walking, gathering, and abnormal behavior. Each state has clear judgment conditions and characteristic descriptions. The probability calculation process counts the frequency of occurrence of various behavior states based on historical trajectory data and calculates the probability of occurrence of each behavior state under the current trajectory characteristic conditions. The behavior state probability distribution is adjusted to a probability distribution vector whose sum is equal to one through normalization processing.

[0055] The maximum probability determination process compares the size of each probability value in the probability distribution of the behavior state, and selects the behavior state with the largest probability value as the preliminary classification result of the current behavior. The behavior state classification result indicates the most likely behavior type of the person at the current moment. The preset behavior transfer rules define reasonable transfer paths between different behavior states. For example, the stationary state can only be transferred to the slow walking state, and cannot directly jump to the fast walking state. The state verification process checks whether the transfer relationship between the current behavior state classification result and the behavior state at the previous moment complies with the preset rules. If the transfer rules are violated, it is necessary to re-determine or adopt the behavior state with the second highest probability. The confirmed behavior state classification result is the final behavior classification after verification of the transfer rules.

[0056] The one-hot encoding algorithm converts categorical labels into numeric vectors, assigning a unique numeric position to each behavioral state. This numerical conversion process encodes each of the five behavioral states into a five-dimensional binary vector: the stationary state is encoded as a vector with the first bit set to one and all other bits zero, the slow walking state is encoded as a vector with the second bit set to one and all other bits zero, and so on. The behavioral encoding vector is a five-dimensional binary vector with only one position set to one and all other positions set to zero. This vector accurately represents the current behavioral state.

[0057] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0058] Perform feature fusion processing on the trajectory feature vector and the behavior coding vector to obtain a comprehensive feature input vector. Based on the comprehensive feature input vector, perform data input processing on the input layer of the multi-layer variable neural network model to obtain network input data.

[0059] Perform forward propagation calculation processing on the network input data to obtain the hidden layer feature output, and perform gradient calculation processing on the hidden layer feature output based on the trajectory tensor high-order backpropagation algorithm to obtain the network parameter gradient value;

[0060] The weight parameters of the multi-layer variable neural network model are updated and optimized according to the gradient values ​​of the network parameters to obtain the optimized network weights. The optimized network weights are subjected to convergence verification to obtain the trained multi-layer variable neural network model.

[0061] The real-time trajectory data of the park is input into the trained multi-layer variable neural network model for inference calculation and processing to obtain the predicted position coordinates. The deviation between the predicted position coordinates and the actual trajectory coordinates is compared to obtain the personnel position coordinate sequence and abnormal behavior alarm information.

[0062] Specifically, the feature fusion processing combines the trajectory feature vector and the behavior coding vector according to a preset connection method. The trajectory feature vector contains continuous numerical features such as movement direction, speed change, stop pattern and path curvature, and the behavior coding vector contains discrete behavior state features that have been one-hot encoded. The fusion process combines the two vectors into a longer comprehensive feature vector through a vector splicing operation. This vector contains both the physical motion features and behavioral semantic features of the trajectory. The data input processing uses the comprehensive feature input vector as the input data of the input layer of the multi-layer variable neural network model. The input layer receives the vector and distributes it to the first hidden layer of the network. The network input data is pre-processed by the input layer to form a standardized data format. Each input node corresponds to a feature dimension in the comprehensive feature vector.

[0063] The forward propagation computational process passes the network input data through the input layer to the first hidden layer. Each hidden layer node receives weighted data from the input layer and undergoes a nonlinear transformation through an activation function. A multi-layer variable neural network model contains multiple hidden layers. Data propagates sequentially between layers, with each layer extracting and transforming the input data, ultimately generating a prediction result at the output layer. The hidden layer feature output is the output value of each hidden layer node after being processed by the activation function. These output values ​​contain the abstract features learned by the network from the input data. The trajectory tensor high-order backpropagation algorithm is a gradient calculation method specifically designed for the characteristics of trajectory data. This algorithm considers the temporal continuity and spatial correlation constraints of trajectory data when calculating gradients. The gradient calculation process uses the backpropagation algorithm to calculate the partial derivatives of the loss function with respect to the weight parameters of each network layer. The gradient value of the network parameter indicates the direction and magnitude of the parameter adjustment.

[0064] The weight parameter update optimization process adjusts the connection weights of the neural network according to the calculated network parameter gradient values. The update formula adopts a variant algorithm of the gradient descent method. Each update adjusts the weight value in the direction of reducing the prediction error. The variable weight mechanism of the multi-layer variable neural network model allows different input features to have different degrees of influence on the network output, and the weight value will be dynamically adjusted according to the importance of the feature. The optimized network weights are parameter values ​​obtained after multiple iterative training. These weight values ​​can enable the network to achieve better performance in trajectory prediction tasks. The convergence verification process determines whether the network has reached a convergence state by monitoring the changing trend of the loss function during training. When the change amplitude of the loss function in multiple consecutive training cycles is less than the preset threshold, the network is considered to have converged. The trained multi-layer variable neural network model has the ability to accurately predict the trajectory of personnel in the park.

[0065] The inference calculation process inputs the trajectory data collected in real time by the park into the trained neural network model, and obtains the network's prediction of the future trajectory position through forward propagation calculation. The predicted position coordinates are the values ​​generated by the network output layer, which represent the spatial position of the person at the next moment predicted by the model. The deviation comparison process calculates the Euclidean distance between the predicted position coordinates and the coordinates of the person's actual arrival position. When the distance difference exceeds the preset abnormality threshold, the abnormal behavior detection mechanism is triggered. The personnel position coordinate sequence is formed by arranging the predicted coordinates of consecutive time points in chronological order. This sequence describes the movement trajectory path of the person. Abnormal behavior alarm information is generated based on the results of the deviation comparison. When an abnormal trajectory deviation is detected, alarm data containing the person's identity, abnormal location, and abnormal time is automatically generated.

[0066] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0067] Input the park's real-time trajectory data into the input layer of the trained multi-layer variable neural network model for data reception processing to obtain input layer data, and perform weighted product calculation on the input layer data to obtain the first hidden layer input data;

[0068] Performing nonlinear transformation processing on the first hidden layer input data based on the activation function to obtain the first hidden layer output data, and performing product operation processing on the first hidden layer output data and the variable weight to obtain the weighted first hidden layer data;

[0069] The weighted first hidden layer data is transferred to the second hidden layer for feature fusion processing to obtain the second hidden layer input data, and the second hidden layer input data is transformed by activation function processing to obtain the second hidden layer output data;

[0070] Inputting the output data of the second hidden layer into the third hidden layer for deep feature learning processing to obtain the output data of the third hidden layer, and performing output layer linear transformation processing on the output data of the third hidden layer to obtain the predicted position coordinates;

[0071] Based on the preset position deviation threshold, the difference between the predicted position coordinates and the actual trajectory coordinates is calculated to obtain the trajectory deviation value. According to the trajectory deviation value, the abnormal trajectory of the park personnel is judged and processed to obtain the abnormal behavior identification;

[0072] The predicted position coordinates are arranged and combined according to the time series to obtain the personnel position coordinate sequence, and the early warning signal is generated based on the abnormal behavior identification to obtain abnormal behavior alarm information.

[0073] Specifically, the data reception process receives the trajectory data collected in real time in the park through the input layer nodes. Each input node corresponds to a feature dimension after the trajectory feature vector and the behavior coding vector are fused. The input layer data is a normalized numerical vector. The weighted product calculation process multiplies each data value of the input layer with the corresponding connection weight one by one, and then sums all the product results to obtain the input value of each node in the first hidden layer. The input data of the first hidden layer is the result of the input layer data being transformed by the weight matrix. This data contains the network's preliminary weighted processing information on the input features. The activation function is a mathematical function used to introduce nonlinear transformations in neural networks. Commonly used activation functions include the ReLU function, the Sigmoid function, and the Tanh function. The nonlinear transformation process maps the linear input data of the first hidden layer to a nonlinear output through the activation function, so that the network has the ability to learn complex nonlinear relationships. The output data of the first hidden layer are numerical values ​​processed by the activation function. These numerical values ​​represent the abstract feature representations extracted by the network from the input trajectory features. Variable weights are the core feature of multi-layer variable neural network models. Unlike the fixed weights of traditional neural networks, variable weights dynamically adjust weight values ​​according to the feature importance of the input data. The product operation multiplies the output data of the first hidden layer by the corresponding variable weights. The weighted first hidden layer data highlights the influence of important features and suppresses the interference of unimportant features.

[0074] The feature fusion process passes the weighted first hidden layer data to the second hidden layer, which is responsible for fusing and combining different types of features to form a higher-level feature representation. The second hidden layer's input data undergoes weight matrix transformation and activation function processing, generating the second hidden layer's output data. This output data contains a fusion of trajectory and behavioral features. Deep feature learning is the primary function of the third hidden layer. Through deeper feature abstraction and combination, this layer learns a deep feature representation from the fused features that can accurately predict trajectory location. The output data of the third hidden layer is the high-level abstract features obtained by the network through multi-layer feature learning. These features contain the key information required for trajectory prediction. The output layer's linear transformation process maps the output data of the third hidden layer into specific position coordinate values ​​using a linear weight matrix. The predicted position coordinates consist of the predicted x and y coordinates, representing the network's predicted spatial location at the next moment.

[0075] The difference calculation process calculates the difference between the predicted location coordinates and the actual arrival location coordinates, typically using the Euclidean distance formula to calculate the straight-line distance between two points. The trajectory deviation value reflects the accuracy of the prediction. A preset location deviation threshold is a criterion set based on the campus environment and application requirements. When the trajectory deviation value exceeds this threshold, it indicates a significant discrepancy between the predicted result and the actual situation. The anomaly determination process compares the trajectory deviation value with a preset threshold. When the deviation value is greater than the threshold, it is considered abnormal; when it is less than the threshold, it is considered normal. The abnormal behavior flag is a Boolean flag that indicates whether the current trajectory is abnormal. The permutation and combination process arranges the predicted location coordinates of consecutive time points in chronological order to form a coordinate sequence that describes the person's movement trajectory. This coordinate sequence fully records the predicted movement path. The early warning signal process generates corresponding alert information based on the status of the abnormal behavior flag. When the abnormal behavior flag is true, detailed alert data including the abnormal location, abnormal time, and person identity is generated. The abnormal behavior alert information is pushed to relevant management personnel through the campus management platform.

[0076] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0077] Perform Euclidean distance calculation on the predicted position coordinates and the actual trajectory coordinates to obtain a position deviation distance value. When the position deviation distance value is greater than a preset deviation threshold, perform an abnormal marking process to obtain a trajectory abnormality mark;

[0078] Cross-border detection is performed based on the predicted location coordinates and the park boundary coordinates. When it is detected that the person's location exceeds the boundary of the park's authorized area, the outbound behavior is marked and an abnormal outbound behavior is obtained;

[0079] Perform logical OR operations based on the trajectory abnormality mark and the abnormal outbound behavior mark. When any abnormal mark exists, the abnormal behavior confirmation process is triggered to obtain a comprehensive abnormal behavior judgment result.

[0080] The comprehensive abnormal behavior judgment results are input into the early warning system. When the abnormal behavior judgment result is true, a high-level alarm signal is generated. When the judgment result is false, a normal state signal is generated to obtain graded abnormal behavior alarm information. At the same time, the predicted position coordinates are arranged and combined according to the time series to obtain a personnel position coordinate sequence.

[0081] Specifically, the Euclidean distance calculation process calculates the straight-line distance between the two points by extracting the x and y values ​​of the predicted position coordinates and the actual trajectory coordinates. The calculation formula is the square root of the sum of the squares of the difference between the two point coordinates. The position deviation distance value reflects the accuracy of the prediction. The preset deviation threshold is a distance standard pre-set according to the park environment and tracking accuracy requirements. The abnormality marking process compares the position deviation distance value with the preset deviation threshold. When the deviation distance value exceeds the threshold, the trajectory abnormality flag is set to a true value, otherwise it is set to a false value. The trajectory abnormality flag is a Boolean data flag used to record whether there is a significant deviation in the current trajectory prediction. The out-of-bounds detection process compares the predicted position coordinates with the pre-set boundary coordinates of the park. The park boundary coordinates define the spatial range of the allowed activities of personnel, including the boundary line coordinate points of each area. The detection algorithm determines whether the predicted position coordinates exceed the boundary range of the authorized area. When the x value or y value of the position coordinate exceeds the corresponding boundary limit, the out-of-bounds detection is triggered. Exit behavior flagging is a crucial component of campus security management. When a person's location is detected outside the boundaries of an authorized zone, the abnormal exit behavior flag is automatically set to true, indicating an unauthorized exit. The abnormal exit behavior flag records whether a person has violated regulations by leaving the designated area.

[0082] The logical OR operation is a type of Boolean logic operation. The result of the logical OR operation is true when either the trajectory anomaly flag or the abnormal exit behavior flag is true. It is false only when both flags are false. Abnormal behavior confirmation determines whether abnormal behavior exists based on the result of the logical OR operation. A true result triggers the abnormal behavior confirmation program, generating detailed abnormal behavior records. The comprehensive abnormal behavior determination result is a comprehensive assessment of multiple abnormal situations, which determines the level of subsequent warning processing and response measures.

[0083] The early warning system is a core component of campus security management, responsible for generating alarm information of corresponding levels based on the results of abnormal behavior determinations. When the comprehensive abnormal behavior determination result is true, the early warning system generates a high-level alarm signal containing the abnormality type, location, timestamp, and identity of the person, which is immediately pushed to the campus security management personnel. When the determination result is false, the early warning system generates a normal status signal, indicating that the current trajectory tracking is normal and there are no abnormalities. Graded abnormal behavior alarm information is divided into different alarm levels according to the severity of the abnormality, including general reminders, important warnings, and emergency alarms. Time series permutation and combination processing arranges the predicted position coordinates of consecutive time points in chronological order to form a complete description of the personnel movement trajectory. The personnel position coordinate sequence records the complete movement path information of the personnel within a specific time period.

[0084] The above describes the method for tracking personnel trajectory in a smart park based on data analysis in the embodiment of the present application. The following describes the system for tracking personnel trajectory in a smart park based on data analysis in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of a smart park personnel trajectory tracking system based on data analysis includes:

[0085] The acquisition module 201 is used to collect and process the location information of park personnel through multi-source sensors to obtain an original trajectory data set, and perform standardization processing on the original trajectory data set to obtain pre-processed trajectory data;

[0086] A construction module 202 is configured to map the pre-processed trajectory data into a three-dimensional tensor space for construction processing to obtain a trajectory spatiotemporal tensor, and perform feature extraction processing on the trajectory spatiotemporal tensor to obtain a trajectory feature vector;

[0087] The identification module 203 is used to identify the behavior state of the person according to the trajectory feature vector to obtain a behavior state classification result, and encode the behavior state classification result to obtain a behavior encoding vector;

[0088] The prediction module 204 is used to input the trajectory feature vector and the behavior coding vector into a multi-layer variable neural network model for training processing to obtain a trained multi-layer variable neural network model, and perform real-time prediction processing based on the multi-layer variable neural network model to obtain a personnel position coordinate sequence and abnormal behavior alarm information.

[0089] above Figure 2 The smart park personnel trajectory tracking system based on data analysis in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The smart park personnel trajectory tracking device based on data analysis in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0090] Reference Figure 3 In an embodiment of the present invention, a smart park personnel trajectory tracking device based on data analysis is also provided. The smart park personnel trajectory tracking device based on data analysis can be a server, and its internal structure can be as follows: Figure 3As shown. The smart park personnel trajectory tracking device based on data analysis includes a processor, memory, display screen, input device, network interface and database connected through a system bus. Among them, the computer-designed processor is used to provide computing and control capabilities. The memory of the smart park personnel trajectory tracking device based on data analysis includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the smart park personnel trajectory tracking device based on data analysis is used to store the corresponding data in this embodiment. The network interface of the smart park personnel trajectory tracking device based on data analysis is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0091] Those skilled in the art will understand that Figure 3 The structure shown in is only a block diagram of part of the structure related to the solution of the present invention, and does not constitute a limitation on the smart park personnel trajectory tracking device based on data analysis to which the solution of the present invention is applied.

[0092] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the smart park personnel trajectory tracking method based on data analysis.

[0093] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0094] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for enabling a smart park personnel trajectory tracking device based on data analysis (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk and other media that can store program code.

[0095] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for tracking personnel trajectories in a smart park based on data analysis, characterized in that: The method comprises: The location information of the park personnel is collected and processed by multi-source sensors to obtain an original trajectory data set, and the original trajectory data set is standardized to obtain pre-processed trajectory data; Mapping the pre-processed trajectory data to a three-dimensional tensor space for construction processing to obtain a trajectory spatiotemporal tensor, performing feature extraction processing on the trajectory spatiotemporal tensor to obtain a trajectory feature vector; Identify the behavior state of the person according to the trajectory feature vector to obtain a behavior state classification result, and encode the behavior state classification result to obtain a behavior encoding vector; The trajectory feature vector and the behavior coding vector are input into a multi-layer variable neural network model for training processing to obtain a trained multi-layer variable neural network model. Real-time prediction processing is performed based on the multi-layer variable neural network model to obtain a personnel position coordinate sequence and abnormal behavior alarm information.

2. The method for tracking personnel trajectories in a smart park based on data analysis according to claim 1 is characterized in that: The method includes collecting and processing the location information of park personnel through multi-source sensors to obtain an original trajectory data set, and performing standardization processing on the original trajectory data set to obtain pre-processed trajectory data, including: Positioning and scanning personnel in the park based on cameras, access control systems, and wireless probes to obtain multi-source location data, and performing time synchronization and correction processing on the multi-source location data to obtain a time-aligned location data set; Performing noise identification and elimination processing on the time-aligned position data set to obtain valid position data, and performing coordinate conversion processing on the valid position data according to the park coordinate system to obtain standard coordinate trajectory data; The standard coordinate trajectory data are grouped according to the personnel identity to obtain individual trajectory data groups, and the individual trajectory data groups are sorted in time series to obtain a continuous time trajectory sequence; The continuous time trajectory sequence is smoothed based on a sliding window algorithm to obtain smoothed trajectory data, and the smoothed trajectory data is resampled to obtain preprocessed trajectory data.

3. The method for tracking personnel trajectories in a smart park based on data analysis according to claim 1 is characterized in that: The mapping of the pre-processed trajectory data to a three-dimensional tensor space for construction processing to obtain a trajectory spatiotemporal tensor, and performing feature extraction processing on the trajectory spatiotemporal tensor to obtain a trajectory feature vector, including: Arranging the pre-processed trajectory data in three dimensions according to the park grid coordinates and time series to obtain space-time structured data, and re-dimensionalizing the space-time structured data based on a tensor construction algorithm to obtain a trajectory space-time tensor; Performing direction angle calculation processing on the trajectory coordinates of adjacent time points in the trajectory space-time tensor to obtain a moving direction vector, and performing speed calculation processing on the trajectory space-time tensor according to the position changes of consecutive time points to obtain a speed change rate parameter; Based on the dwell time distribution in the trajectory spatiotemporal tensor, the dwell frequency statistics of each area of ​​the park are processed to obtain dwell pattern parameters, and the movement direction vector is subjected to continuity analysis to obtain path curvature parameters; The moving direction vector, the speed change rate parameter, the dwell mode parameter, and the path curvature parameter are subjected to vector combination processing to obtain a trajectory feature vector.

4. The method for tracking personnel trajectories in a smart park based on data analysis according to claim 1 is characterized in that: The identifying process of the behavior state of the person according to the trajectory feature vector to obtain a behavior state classification result, and encoding the behavior state classification result to obtain a behavior encoding vector includes: Based on the speed change rate parameter in the trajectory feature vector, a threshold judgment process is performed on the moving speed of the personnel to obtain a speed state classification; and based on the stay mode parameter in the trajectory feature vector, a statistical process is performed on the duration of the personnel's stay behavior to obtain a stay state classification; The speed state classification and the stop state classification are combined and matched to obtain five basic behavior states: stationary, slow walking, fast walking, gathering, and abnormal behavior; and the five basic behavior states are subjected to probability calculation to obtain a behavior state probability distribution; Performing maximum probability determination processing on the current behavior of park personnel based on the behavior state probability distribution to obtain a behavior state classification result, and performing state verification processing on the behavior state classification result according to a preset behavior transfer rule to obtain a confirmed behavior state classification result; The confirmed behavior state classification result is input into a one-hot encoding algorithm for numerical conversion processing to obtain a behavior encoding vector.

5. The method for tracking personnel trajectories in a smart park based on data analysis according to claim 1 is characterized in that: The trajectory feature vector and the behavior coding vector are input into a multi-layer variable neural network model for training processing to obtain a trained multi-layer variable neural network model, and real-time prediction processing is performed based on the multi-layer variable neural network model to obtain a personnel position coordinate sequence and abnormal behavior alarm information, including: Performing feature fusion processing on the trajectory feature vector and the behavior coding vector to obtain a comprehensive feature input vector, and performing data input processing on the input layer of the multi-layer variable neural network model based on the comprehensive feature input vector to obtain network input data; Performing forward propagation calculation processing on the network input data to obtain hidden layer feature output, and performing gradient calculation processing on the hidden layer feature output based on the trajectory tensor high-order backpropagation algorithm to obtain the network parameter gradient value; performing an updating and optimization process on the weight parameters of the multi-layer variable neural network model according to the network parameter gradient values ​​to obtain optimized network weights, and performing a convergence verification process on the optimized network weights to obtain a trained multi-layer variable neural network model; The real-time trajectory data of the park is input into the trained multi-layer variable neural network model for inference calculation processing to obtain predicted position coordinates, and the deviation comparison processing is performed on the predicted position coordinates and the actual trajectory coordinates to obtain the personnel position coordinate sequence and abnormal behavior alarm information.

6. The method for tracking personnel trajectories in a smart park based on data analysis according to claim 5 is characterized in that: The real-time trajectory data of the park is input into the trained multi-layer variable neural network model for inference calculation processing to obtain predicted position coordinates, and the predicted position coordinates are compared with the actual trajectory coordinates to obtain a personnel position coordinate sequence and abnormal behavior alarm information, including: Inputting the park real-time trajectory data into the input layer of the trained multi-layer variable neural network model for data reception processing to obtain input layer data, and performing weighted product calculation processing on the input layer data to obtain first hidden layer input data; performing a nonlinear transformation on the first hidden layer input data based on an activation function to obtain first hidden layer output data, and performing a product operation on the first hidden layer output data and a variable weight to obtain weighted first hidden layer data; Passing the weighted first hidden layer data to the second hidden layer for feature fusion processing to obtain second hidden layer input data, and performing activation function transformation processing on the second hidden layer input data to obtain second hidden layer output data; Inputting the second hidden layer output data into the third hidden layer for deep feature learning processing to obtain third hidden layer output data, and performing output layer linear transformation processing on the third hidden layer output data to obtain predicted position coordinates; Based on a preset position deviation threshold, the predicted position coordinates and the actual trajectory coordinates are calculated to obtain a trajectory deviation value, and the trajectory of the park personnel is judged to be abnormal according to the trajectory deviation value to obtain an abnormal behavior identifier; The predicted position coordinates are arranged and combined according to a time series to obtain a personnel position coordinate sequence, and an early warning signal is generated based on the abnormal behavior identifier to obtain abnormal behavior alarm information.

7. The method for tracking personnel trajectories in a smart park based on data analysis according to claim 5 is characterized in that: The deviation comparison processing of the predicted position coordinates and the actual trajectory coordinates to obtain the personnel position coordinate sequence and abnormal behavior alarm information includes: Performing Euclidean distance calculation on the predicted position coordinates and the actual trajectory coordinates to obtain a position deviation distance value, and performing abnormal marking processing when the position deviation distance value is greater than a preset deviation threshold to obtain a trajectory abnormality flag; Performing out-of-bounds detection based on the predicted location coordinates and the park boundary coordinates, and marking the exit behavior when it is detected that the person's location exceeds the boundary of the park's authorized area to obtain an abnormal exit behavior identifier; Performing a logical OR operation based on the trajectory abnormality identifier and the abnormal outbound behavior identifier, triggering abnormal behavior confirmation processing when any abnormal identifier exists, and obtaining a comprehensive abnormal behavior determination result; The comprehensive abnormal behavior determination result is input into the early warning system. When the abnormal behavior determination result is true, a high-level alarm signal is generated. When the determination result is false, a normal state signal is generated to obtain graded abnormal behavior alarm information. At the same time, the predicted position coordinates are arranged and combined according to the time series to obtain a personnel position coordinate sequence.

8. A smart park personnel trajectory tracking system based on data analysis, characterized in that: For implementing the method for tracking personnel trajectories in a smart park based on data analysis according to any one of claims 1 to 7, the smart park personnel trajectories tracking system based on data analysis comprises: An acquisition module is used to collect and process the location information of park personnel through multi-source sensors to obtain an original trajectory data set, and to perform standardization processing on the original trajectory data set to obtain pre-processed trajectory data; A construction module, configured to map the pre-processed trajectory data into a three-dimensional tensor space for construction processing to obtain a trajectory spatiotemporal tensor, and perform feature extraction processing on the trajectory spatiotemporal tensor to obtain a trajectory feature vector; an identification module, configured to identify the behavior state of a person according to the trajectory feature vector to obtain a behavior state classification result, and encode the behavior state classification result to obtain a behavior encoding vector; The prediction module is used to input the trajectory feature vector and the behavior coding vector into a multi-layer variable neural network model for training processing to obtain a trained multi-layer variable neural network model, and perform real-time prediction processing based on the multi-layer variable neural network model to obtain a personnel position coordinate sequence and abnormal behavior alarm information.

9. A smart park personnel trajectory tracking device based on data analysis, characterized in that: It includes a memory and a processor, the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, it implements the smart park personnel trajectory tracking method based on data analysis as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the processor executes the method for tracking personnel trajectories in a smart park based on data analysis as described in any one of claims 1 to 7.

Images