An internet-based campus security management system and method

Abstract

The application belongs to the technical field of campus safety, and discloses a campus safety management system and method based on the Internet, which comprises a data acquisition module for acquiring campus safety monitoring data; a data processing module for pre-processing the campus safety monitoring data to obtain a campus safety comprehensive feature data set; a threat prediction module for training a campus safety threat prediction model according to the campus safety comprehensive feature data set, predicting a campus safety threat factor based on the campus safety threat prediction model, and judging whether the campus safety is threatened according to the predicted campus safety threat factor; a threat diagnosis module for acquiring campus safety threat data if the campus safety is threatened, inputting the campus safety threat data into a trained safety threat diagnosis model, and predicting a campus safety threat type; and an emergency response module for formulating a corresponding emergency plan according to the campus safety threat type, so that the intelligent level of management is improved.

Description

Technical Field

[0001] The present invention relates to the field of campus safety technology, and more particularly to an Internet-based campus safety management system and method. Background Art

[0002] Patent announcement number CN115311824A discloses an Internet-based campus safety management system and management method, including a security monitoring module for connecting to a network through a wireless network and pre-set monitoring equipment, and monitoring the campus according to the monitoring equipment to obtain monitoring data; a security judgment module for judging campus accidents based on the monitoring data and obtaining judgment results; an alarm processing module for executing a linkage alarm operation when the judgment result shows that a campus accident has occurred. The campus monitoring network constructed by the present invention can quickly transmit abnormal campus situations, thereby saving a lot of alarm time. The set joint alarm can not only save alarm time, but also ensure sufficient and rapid police dispatch, thereby more comprehensively ensuring campus safety.

[0003] The existing campus safety management systems and methods have the following main problems:

[0004] Failure to identify students' abnormal behavior patterns, such as staying in inactive areas for a long time or moving abnormally frequently, will make it difficult to detect potential safety risks in a timely manner; the lack of quantitative analysis of state transition probabilities will make it impossible to accurately reveal the trend of dynamic changes in students' behavior, and thus may miss the opportunity to prevent safety accidents; the inability to detect irregular state transitions in students' behavior, such as moving directly from the teaching building to an unsafe area, will increase safety hazards; the system will find it difficult to effectively screen abnormal trajectories, resulting in limited abnormal behavior detection capabilities of the system; the inability to automatically adapt to the trajectory patterns of different students will reduce the accuracy and versatility of the analysis results; campus management efficiency is low, and it is difficult to formulate effective management measures for different behavior patterns; the failure to timely detect and intervene in potential safety hazards will directly affect students' safety and may lead to safety accidents;

[0005] Without the support of dynamic adjustment factors, the trajectory alignment algorithm may not be able to accurately respond to the differences in time span and shape of students' behavioral trajectories, resulting in inaccurate alignment results, which will directly affect the subsequent behavioral pattern analysis and safety hazard identification; without an automated adjustment mechanism, the system may require frequent manual intervention and manual parameter adjustment to cope with changes in trajectory data, which not only increases management costs, but may also introduce human errors and reduce the overall performance of the system; the lack of dynamic limiting factors limits the system's ability to automatically optimize the trajectory alignment algorithm according to changes in student behavior data, reducing the system's intelligence level; inaccurate trajectory alignment may affect the ability to identify safety hazards.

[0006] In view of this, the present invention proposes an Internet-based campus security management system and method to solve the above problems. Summary of the Invention

[0007] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solution: an Internet-based campus safety management system, comprising:

[0008] Data acquisition module, used to obtain campus security monitoring data;

[0009] The data processing module is used to pre-process the campus security monitoring data to obtain a comprehensive campus security feature data set;

[0010] The threat prediction module is used to train a campus security threat prediction model based on the campus security comprehensive feature dataset, predict campus security threat factors based on the campus security threat prediction model, and determine whether there is a threat to campus security based on the predicted campus security threat factors;

[0011] The threat diagnosis module obtains campus security threat data if there is a threat to campus security, inputs the campus security threat data into the trained security threat diagnosis model, and predicts the type of campus security threat;

[0012] The emergency response module is used to formulate corresponding emergency plans according to the type of campus security threat and generate different alarm information; send alarm information to relevant personnel through the campus security management terminal and automatically activate the emergency plan;

[0013] The modules are connected via wired and / or wireless means to achieve data transmission between modules.

[0014] Furthermore, the campus safety monitoring data includes student behavior data, campus environment data and campus video data; student behavior data includes student movement trajectory data and violent conflict data; campus environment data includes building structure status data, power equipment status data and fire equipment status data; campus video data includes campus public area videos, campus activity videos and individual behavior monitoring videos.

[0015] Furthermore, the method of preprocessing the campus safety monitoring data to obtain a campus safety comprehensive feature data set includes:

[0016] Trajectory data mining and Markov chain are used to extract features from student behavior data to obtain a student behavior feature dataset. The K-Means clustering algorithm is used to identify and eliminate outliers in the campus environment data and campus video data included in the campus security monitoring data to obtain a campus environment feature dataset and a campus video feature dataset.

[0017] Perform standard deviation normalization on the student behavior feature dataset, campus environment feature dataset, and campus video feature dataset, converting them into standard normal distribution to obtain the normalized student behavior feature dataset, campus environment feature dataset, and campus video feature dataset;

[0018] The normalized student behavior feature dataset, campus environment feature dataset and campus video feature dataset are weighted fused to obtain a comprehensive feature dataset.

[0019] Furthermore, the method of extracting features from student behavior data using trajectory data mining and Markov chains to obtain a student behavior feature dataset includes:

[0020] S41. Preset the student's movement trajectory data as a set of spatial position coordinate points in a time series: ;in, A trajectory for students to move; For students in The position coordinates of each time point; For students in The position coordinates of each time point; For students in The horizontal coordinate of the position of each time point; For students in The vertical coordinate of the position at each time point; For students in The timestamp of the position coordinate record of each time point; The total number of time points in the student's movement trajectory data; is the index of the time point in the student's movement trajectory data, The value range is from 1 to ;

[0021] S42, preset another trajectory of student movement is , the similarity between the two student trajectories is calculated using the Frechet distance metric formula; the Frechet distance metric formula is: ;in, Move tracks for two students and The Frechet distance between is the alignment path function; is the set of all alignment paths; Indicates that all alignment paths are confirmed; Indicates the supremum operation for all time points; is the Euclidean distance; Mobile tracks for students At the time point The location coordinates of

[0022] S43. Dynamically design the alignment path function by introducing the total number of time points in the student's movement trajectory data; the alignment path function is: ;in, Represents student movement trajectory In the The location at a point in time; Represents student movement trajectory In the The location at a point in time; Adjustment factor for alignment paths; is the normalization factor; the alignment path adjustment factor Perform dynamic constraints and automatically adjust the value of the constraint alignment path adjustment factor;

[0023] S44. Group all students’ movement trajectories using the DBSCAN clustering algorithm and divide the student movement trajectories into different behavior patterns;

[0024] S45. Based on trajectory data mining, we use Markov chains to model different behavior patterns, define the state space of student behavior, and divide the behavior of students at different locations into different states. The state space of student behavior is: ;in, For students a behavioral pattern; An index of behavioral patterns; is the total number of behavior patterns;

[0025] S46. Define the state sequence in the student behavior trajectory as , then the state transfer matrix is ​​expressed as: ;in, represents the state transition matrix; Indicates the slave state Transfer to state probability; Represents the number of rows and columns in the state transition matrix; and The index representing the state;

[0026] S47, using Markov chain, extracting student behavior feature data set, student behavior features include state transition probability and residence time; preset from state Transfer to state The number of times is , then the state transition probability is expressed as: ;in, Indicates the slave state Transfer to state probability; To indicate status Transfer to state the number of times; Indicates the slave state The total number of transitions to all states;

[0027] For each state , define the dwell time The length of time the student stays in the state is calculated using the residence time evaluation formula. The residence time evaluation formula is: ;in, Status duration of stay; A specific point in time in the student's movement trajectory; is the indicator function; For the time point When, the student's status;

[0028] S48. Combine the extracted student behavior features to obtain a student behavior feature dataset.

[0029] Furthermore, the alignment path adjustment factor Methods for dynamic limiting include:

[0030] Adjust the alignment path factor using the adjustment factor constraint formula Perform dynamic restriction and adjust the factor restriction formula as follows: ;in, To control the alignment path adjustment factor Constant factors of maximum values; Adjustment factor of alignment path to control trajectory time span variation The weighting factor of the impact degree; Adjustment factor for alignment paths to control trajectory shape differences The weighting factor of the impact degree; The change in the total number of time points in the student's movement trajectory data.

[0031] Furthermore, the training method of the campus security threat prediction model includes:

[0032] The dataset is divided into training, validation, and test sets to train the model and evaluate its performance. A hazard probability prediction model is constructed, with the sample set being a subset of the dataset. Each sample set includes a historical campus safety comprehensive feature dataset and corresponding campus safety threat factors.

[0033] The campus security threat prediction model includes an input layer, a hidden layer, and an output layer; the hidden layer uses a ReLU activation function; the input layer of the model is used to input a historical campus security comprehensive feature dataset; the output layer of the model is used to output campus security threat factors; the campus security threat prediction model is a multimodal neural network model;

[0034] The mean squared error (MSE) is used as the loss function to measure the error between the model's predicted value and the actual value. The hazard probability prediction model is trained using the training set, and the model parameters are updated using the backpropagation algorithm and gradient descent method to minimize the loss function. The performance of the campus safety threat prediction model is evaluated by calculating the coefficient of determination using the validation set, and the model's hyperparameters are tuned.

[0035] The Adam optimization algorithm is selected as the optimizer, and the model's hyperparameters are adjusted until the performance no longer improves or the preset number of iterations is reached. The test set is used to evaluate the model's performance in the prediction task, and the trained campus security threat prediction model is used to predict the current campus security comprehensive feature dataset to obtain the campus security threat factor.

[0036] Furthermore, the method for determining whether there is a threat to campus security based on the predicted campus security threat factors includes:

[0037] Preset a campus security threat factor threshold, and compare the predicted campus security threat factor with the preset campus security threat factor threshold;

[0038] If the predicted campus security threat factor is greater than or equal to the preset campus security threat factor threshold, it is determined that there is a threat to campus security;

[0039] If the predicted campus security threat factor is less than the preset campus security threat factor threshold, it is determined that there is no threat to campus security.

[0040] Furthermore, the training method of the security threat diagnosis model includes:

[0041] The dataset is divided into training set, validation set and test set to train the model and evaluate the model performance; the sample set is a subset of the dataset, and each sample contains historical campus security threat data and the corresponding campus security threat type;

[0042] A security threat diagnosis model is constructed, comprising an input layer, an encoding layer, a multimodal fusion layer, a decoding layer, and an output layer. The input layer is used to input historical campus security threat data, and the output layer is used to output campus security threat types. The output layer is configured with neurons equal to the number of campus security threat types, with each neuron corresponding to the predicted probability of a campus security threat type. The softmax function is used as the activation function. The security threat diagnosis model is a multimodal Transformer model.

[0043] The multi-classification cross-entropy is used as the loss function of the model to measure the difference between the predicted value and the actual value of the model; the security threat diagnosis model is trained using the training set, the model parameters are updated through the back propagation algorithm to minimize the loss function; the mini-batch gradient descent method is used, and the model is updated once for each batch; the performance of the security threat diagnosis model is evaluated by calculating the recall rate index using the validation set;

[0044] The SGD optimization algorithm is selected as the optimizer, the model is optimized according to the performance feedback of the validation set, the model parameters are adjusted until the performance no longer improves or the preset stopping condition is reached; the performance of the model in the prediction task is evaluated using the test set, and the trained security threat diagnosis model is used to predict the current campus security threat data to obtain the campus security threat type.

[0045] Further, the method of formulating corresponding emergency plans and generating different alarm information according to the campus security threat type comprises:

[0046] The campus security threat type includes violent incidents, self-harm behavior, fire safety threats, campus outsiders, network security threats, public safety events, traffic safety accidents, sudden public health events, and terrorist activities on campus;

[0047] For violent incidents, start campus security patrol, alarm to security or police station; for self-harm behavior, start psychological counseling center intervention, notify mental health teacher; for fire safety threat, start fire fighting plan, broadcast emergency evacuation instruction and dial fire telephone; for campus outsiders, track outsiders through camera, set up checkpoints, strengthen patrol, and strengthen personnel identity verification at school gate; for network security threat, monitor network abnormal behavior through firewall, and timely discover and handle network attack;

[0048] For public safety events, security personnel cooperate with on-site responsible person for persuasion and dissuasion, if violent or other dangerous behavior occurs, report to relevant law enforcement agencies immediately for intervention, and coordinate police forces for dispersal; for traffic safety accidents, analyze accident causes through camera, dispatch traffic police and emergency personnel; for sudden public health events, immediately start public health emergency plan, notify school medical room and health department for on-site inspection, block epidemic transmission source, and perform isolation treatment; for terrorist activities on campus, notify police and special police for support, implement blockade and conduct rigorous search; according to different campus security threat types, different alarm information is generated to indicate different alarm levels for relevant personnel.

[0049] A campus security management method based on the Internet comprises:

[0050] S1. Obtain campus security monitoring data;

[0051] S2. Preprocess the campus security monitoring data to obtain a comprehensive campus security feature dataset;

[0052] S3. Train a hazard probability prediction model based on the campus safety comprehensive feature dataset, predict campus safety threat factors based on the hazard probability prediction model, and determine whether there is a threat to campus safety based on the predicted campus safety threat factors;

[0053] S4. If there is a threat to campus security, obtain campus security threat data, input the campus security threat data into the trained security threat diagnosis model, and predict the type of campus security threat;

[0054] S5. Develop corresponding emergency plans based on the type of campus security threat and generate different alarm information; send alarm information to relevant personnel through the campus security management terminal and automatically activate the emergency plan.

[0055] The technical effects and advantages of the Internet-based campus security management system and method of the present invention are as follows:

[0056] Through trajectory data mining and clustering algorithms, the present invention can identify students' abnormal behavior patterns (such as staying in inactive areas for a long time or abnormally frequent movements), thereby improving the ability to discover potential risks; quantify the rules of students' transitions from one behavior pattern (such as location or activity) to another through state transition probability, revealing the trend of dynamic changes in behavior; evaluate the activity intensity of students in specific locations or states through residence time, providing a basis for analyzing students' learning and life behaviors; use Markov chains to model behavior patterns, and discover irregular state transitions in students' behavior (such as moving directly from a teaching building to an unsafe area), which helps to quickly identify safe areas. The Frechet distance, a measure of trajectory similarity, can effectively screen out abnormal trajectories and further enhance the system's abnormal behavior detection capabilities. By dynamically designing the alignment path function and dynamically limiting the adjustment factors, the system can automatically adapt to the trajectory patterns of different students, ensuring the accuracy and universality of the analysis results. The clustering algorithm groups student behaviors and identifies different behavioral patterns (such as common commuting trajectories and group activity trajectories). It can analyze both individual and group behaviors, meeting the diverse needs of campus management. By extracting and analyzing student behavioral characteristics, potential safety hazards can be promptly discovered and intervened, providing a safer campus environment for students.

[0057] Dynamically limiting the alignment path adjustment factor can optimize the accuracy of trajectory alignment. During trajectory comparison, students' behavioral trajectories may differ in time span and shape. Using adjustment factors to fine-tune alignment ensures the accuracy of the alignment path, thereby better capturing behavioral patterns. By introducing factors influencing trajectory time span changes and trajectory shape differences, this method can adapt to the temporal and spatial differences between different trajectories. For example, when students' activities occur in different time periods, the dynamic limiting factor can be flexibly adjusted to ensure the rationality and stability of trajectory alignment. This reduces the need for manual intervention and parameter adjustment, and the system automatically optimizes the trajectory alignment algorithm based on changes in student behavioral data, improving the degree of automation and system intelligence. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 This is a schematic structural diagram of an Internet-based campus security management system of the present invention;

[0059] Figure 2 The present invention is a flow chart of an Internet-based campus safety management method. DETAILED DESCRIPTION

[0060] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0061] Example 1

[0062] See also Figure 1 As shown, this embodiment provides an Internet-based campus security management system, including:

[0063] Data acquisition module, used to obtain campus security monitoring data;

[0064] The data processing module is used to pre-process the campus security monitoring data to obtain a comprehensive campus security feature data set;

[0065] The threat prediction module is used to train a campus security threat prediction model based on the campus security comprehensive feature dataset, predict campus security threat factors based on the campus security threat prediction model, and determine whether there is a threat to campus security based on the predicted campus security threat factors;

[0066] The threat diagnosis module obtains campus security threat data if there is a threat to campus security, inputs the campus security threat data into the trained security threat diagnosis model, and predicts the type of campus security threat;

[0067] The emergency response module is used to formulate corresponding emergency plans according to the type of campus security threat and generate different alarm information; send alarm information to relevant personnel through the campus security management terminal and automatically activate the emergency plan;

[0068] The modules are connected via wired and / or wireless means to achieve data transmission between modules.

[0069] Campus safety monitoring data includes student behavior data, campus environment data, and campus video data; student behavior data includes student movement trajectory data and violent conflict data; campus environment data includes building structure status data, power equipment status data, and fire equipment status data; campus video data includes campus public area videos, campus activity videos, and individual behavior monitoring videos.

[0070] The method of preprocessing campus security monitoring data to obtain a comprehensive campus security feature dataset includes:

[0071] Trajectory data mining and Markov chain are used to extract features from student behavior data to obtain a student behavior feature dataset. The K-Means clustering algorithm is used to identify and eliminate outliers in the campus environment data and campus video data included in the campus security monitoring data to obtain a campus environment feature dataset and a campus video feature dataset.

[0072] Perform standard deviation normalization on the student behavior feature dataset, campus environment feature dataset, and campus video feature dataset, converting them into standard normal distribution to obtain the normalized student behavior feature dataset, campus environment feature dataset, and campus video feature dataset;

[0073] The normalized student behavior feature dataset, campus environment feature dataset and campus video feature dataset are weighted fused to obtain a comprehensive feature dataset.

[0074] Methods for extracting features from student behavior data using trajectory data mining and Markov chains to obtain student behavior feature datasets include:

[0075] S41. Preset the student's movement trajectory data as a set of spatial position coordinate points in a time series: ;in, A trajectory for students to move; For students in The position coordinates of each time point; For students in The position coordinates of each time point; For students in The horizontal coordinate of the position of each time point; For students in The vertical coordinate of the position at each time point; For students in The timestamp of the position coordinate record of each time point; The total number of time points in the student's movement trajectory data; is the index of the time point in the student's movement trajectory data, The value range is from 1 to ;

[0076] S42, preset another trajectory of student movement is , the similarity of the two students' movement trajectories is calculated using the Frechet distance measurement formula; by calculating the trajectory similarity, students with similar behavior patterns can be classified together to find out whether their behavior has regularity or detect whether there is abnormal behavior; the Frechet distance measurement formula is: ;in, Move tracks for two students and The Frechet distance between them is used to measure the shape difference of two students' movement trajectories. It measures the shape similarity of two students' movement trajectories. The smaller the distance, the higher the similarity of the trajectories. is the alignment path function, indicating the time point A function to align the corresponding points of two student movement trajectories; is the set of all alignment paths; It means to perform the next determination operation on all alignment paths, that is, to find an optimal alignment path so that the maximum spatial distance between the two students' movement trajectories is minimized; Indicates that the supremum operation is performed on all time points, that is, the interval is found All time points The maximum value of is the Euclidean distance, used to calculate the alignment path function and trajectory At the time point distance; Mobile tracks for students At the time point The location coordinates of

[0077] S43. Dynamically design the alignment path function by introducing the total number of time points in the student movement trajectory data to improve the alignment capability of complex student movement trajectories; the alignment path function is: ;in, Represents student movement trajectory In the The location at a point in time; Represents student movement trajectory In the The location at a point in time; Adjust the alignment path factor to control the student's movement trajectory and student movement trajectories The proportion of influence on the alignment path; is the normalization factor used to index the time points Convert to a value in the range [0,1] to align between two tracks; adjust the alignment path factor Perform dynamic constraints and automatically adjust the value of the constraint alignment path adjustment factor;

[0078] When the time When it approaches 0, the normalization factor The proportion will be very small, and the student movement trajectory The position has a greater impact; when the time point Approaching When the normalization factor The ratio will be close to 1, at this time the student movement trajectory The position has a greater impact;

[0079] For example, there are two student movement trajectories and , respectively represent the positions of student 1 and student 2 at different time points;

[0080] Student 1's movement trajectory Contains 5 time points ;

[0081] Student 2's movement trajectory Contains 5 time points ;

[0082] Alignment path adjustment factor 0.5, indicating the student's movement trajectory and The same effect on the alignment path, the normalization factor at the first time point , the normalization factor at the second time point , the normalization factor at the third time point , the normalization factor at the 4th time point , the normalization factor at the 5th time point ;

[0083] Compute the aligned trajectory for each time point:

[0084] ;

[0085] ;

[0086] ;

[0087] ;

[0088] ;

[0089] The coordinate points corresponding to the final alignment path are: , , ; ; ;

[0090] S44. All student movement trajectories are grouped using the DBSCAN clustering algorithm, and student movement trajectories are divided into different behavior patterns (such as attending classes, eating, resting, gathering activities, etc.), helping campus administrators understand students' daily activity patterns; abnormal trajectories that deviate from normal behavior patterns (such as frequently wandering in restricted areas, staying in inactive areas for a long time, etc.) are detected to provide early warning for campus safety.

[0091] S45. Based on trajectory data mining, we use Markov chains to model different behavior patterns, define the state space of student behavior, and divide the behavior of students at different locations into different states. The state space of student behavior is: ;in, For students a behavioral pattern; An index of behavioral patterns; is the total number of behavior patterns;

[0092] S46. Define the state sequence in the student behavior trajectory as , then the state transfer matrix is ​​expressed as: ;in, represents the state transition matrix; Indicates the slave state Transfer to state probability; Represents the number of rows and columns in the state transition matrix, that is, the state sequence The total number of states in the and The index representing the state;

[0093] S47, using Markov chain, extracting student behavior feature data set, student behavior features include state transition probability and residence time; preset from state Transfer to state The number of times is , then the state transition probability is expressed as: ;in, Indicates the slave state Transfer to state probability; To indicate status Transfer to state the number of times; Indicates the slave state The total number of transitions to all states; the total number of behavior patterns Less than state sequence The total number of states ,because a behavior pattern may contain multiple states; a behavior pattern may be extracted from the combination or analysis of multiple states, such as "learning mode" and "leisure mode";

[0094] For each state , define the dwell time The length of time the student stays in the state is calculated using the residence time evaluation formula. The residence time evaluation formula is: ;in, Status The duration of stay of students in a certain state the total length of stay; A specific point in time in the student's movement trajectory; Is an indicator function, used to determine the time point Is the student in the state ,when hour, Take 1, otherwise 0; For the time point When, the student's status;

[0095] S48. Combine the extracted student behavior features to obtain a student behavior feature dataset.

[0096] Adjustment factor for alignment paths Methods for dynamic limiting include:

[0097] Adjust the alignment path factor using the adjustment factor constraint formula Perform dynamic restriction and adjust the factor restriction formula as follows: ;in, To control the alignment path adjustment factor Constant factors of maximum values; Adjustment factor of alignment path to control trajectory time span variation The weighting factor of the impact degree; Adjustment factor for alignment paths to control trajectory shape differences The weighting factor of the impact degree; The change in the total number of time points in the student's movement trajectory data, that is, the difference in the number of time points in the movement trajectories of two students;

[0098] The number of time points of the two student movement trajectories may be different. If the number of time points of the two student movement trajectories is very different, using the same alignment strategy may lead to errors, because the time span difference may affect the alignment of the trajectories. In order to avoid over-reliance on trajectories with a large number of time points, the adjustment factor limit formula is passed (difference in the number of time points) to adjust the factor Specifically, the greater the difference in the number of time points, the greater the factor The adjustment factor constraint formula can reduce the alignment weight when the track shapes differ greatly, thus avoiding unreasonable alignment results. This ensures that when the shapes of the tracks differ greatly, the tracks will not be overly aligned by linear interpolation, making the alignment process more flexible and reasonable.

[0099] For example, student movement trajectory Number of time points , student movement trajectory Number of time points ; Two student movement trajectories and Frechet distance between is 5, which is the change in the total number of time points in the student's movement trajectory data , total time points ,so: , controls the alignment path adjustment factor Constant factor of maximum value 1; Controls the adjustment factor of the alignment path due to the change in trajectory time span Weight factor of influence 1; control the trajectory shape difference to the alignment path adjustment factor Weight factor of influence is 0.2; then the alignment path adjustment factor after dynamic restriction is ;

[0100] Generally, the number of time points in a trajectory has a greater impact on the alignment process, especially when the number of time points in two trajectories differs greatly. The difference in the number of time points directly affects the temporal resolution of the trajectory and may cause a large deviation in the alignment path. Therefore, if the difference in the number of time points is large, a larger weight factor should be given. Generally speaking, setting 1 is more appropriate;

[0101] The shape difference of the trajectory reflects the difference in the spatial distribution of the two trajectories. Generally, the shape difference of the trajectory plays a small role in the alignment because the adjustment of spatial difference usually depends on the influence of temporal difference. However, if the shape difference of the trajectory is large, it may have a greater impact on the alignment, especially when the time points of the trajectories are very close. The weight factor for the shape difference is generally small, such as 0.2 or less, to avoid excessively affecting the alignment process;

[0102] In order to obtain reasonable alignment results, multiple experiments are required. For example, if the experimental data shows that the difference in the number of time points is much greater than the effect of the shape difference on the alignment results, then the weight factor of the number of time points can be increased. to make its impact more prominent.

[0103] The training methods for the campus security threat prediction model include:

[0104] The dataset is divided into training, validation, and test sets to train the model and evaluate its performance. A hazard probability prediction model is constructed, with the sample set being a subset of the dataset. Each sample set includes a historical campus safety comprehensive feature dataset and corresponding campus safety threat factors.

[0105] The campus security threat prediction model consists of an input layer, a hidden layer, and an output layer. The hidden layer uses the ReLU activation function. The input layer of the model is used to input a historical campus security comprehensive feature dataset. The output layer of the model is used to output campus security threat factors. The campus security threat prediction model is a multimodal neural network model.

[0106] The mean squared error (MSE) is used as the loss function to measure the error between the model's predicted value and the actual value. The hazard probability prediction model is trained using the training set, and the model parameters are updated using the backpropagation algorithm and gradient descent method to minimize the loss function. The performance of the campus safety threat prediction model is evaluated by calculating the coefficient of determination using the validation set, and the model's hyperparameters are tuned.

[0107] The Adam optimization algorithm is selected as the optimizer, and the model's hyperparameters are adjusted until the performance no longer improves or the preset number of iterations is reached. The test set is used to evaluate the model's performance in the prediction task, and the trained campus security threat prediction model is used to predict the current campus security comprehensive feature dataset to obtain the campus security threat factor.

[0108] Based on the predicted campus security threat factors, methods for determining whether campus security is threatened include:

[0109] Preset a campus security threat factor threshold, and compare the predicted campus security threat factor with the preset campus security threat factor threshold;

[0110] If the predicted campus security threat factor is greater than or equal to the preset campus security threat factor threshold, it is determined that there is a threat to campus security;

[0111] If the predicted campus security threat factor is less than the preset campus security threat factor threshold, it is determined that there is no threat to campus security.

[0112] The training methods for the security threat diagnosis model include:

[0113] The dataset is divided into training set, validation set and test set to train the model and evaluate the model performance; the sample set is a subset of the dataset, and each sample contains historical campus security threat data and the corresponding campus security threat type;

[0114] A security threat diagnosis model was constructed, consisting of an input layer, an encoding layer, a multimodal fusion layer, a decoding layer, and an output layer. The input layer was used to input historical campus security threat data, and the output layer was used to output the campus security threat types. The output layer was configured with neurons equal to the number of campus security threat types, with each neuron corresponding to the predicted probability of a campus security threat type. The softmax function was used as the activation function. The security threat diagnosis model was a multimodal Transformer model.

[0115] Multi-class cross entropy is used as the model's loss function to measure the difference between the model's predicted value and the actual value. The security threat diagnosis model is trained using the training set, and the model parameters are updated using the backpropagation algorithm to minimize the loss function. The model is updated once per batch using the mini-batch gradient descent method. The performance of the security threat diagnosis model is evaluated by calculating the recall metric using the validation set.

[0116] The SGD optimization algorithm is selected as the optimizer, and the model is tuned according to the performance feedback of the validation set. The model parameters are adjusted until the performance no longer improves or the preset stopping condition is reached. The test set is used to evaluate the performance of the model in the prediction task, and the trained security threat diagnosis model is used to predict the current campus security threat data to obtain the campus security threat type.

[0117] Develop corresponding emergency plans based on the type of campus security threat. The methods for generating different alarm information include:

[0118] Campus safety threat types include violent incidents, self-harm behaviors, fire safety threats, threats from outsiders on campus, cybersecurity threats, public safety incidents, traffic accidents, public health emergencies, and terrorist activities on campus;

[0119] For violent incidents, we will initiate on-campus security patrols and alert the security guards or police. For self-harm, we will activate the psychological counseling center for intervention and notify the mental health counselor. For fire safety threats, we will activate the fire emergency plan, broadcast emergency evacuation instructions, and call the fire department. For threats from outsiders on campus, we will track outsiders through cameras, have security personnel set up checkpoints and increase patrols, and strengthen identity verification at the school gate. For network security threats, we will monitor network abnormalities through firewalls to detect and address network attacks in a timely manner.

[0120] For public safety incidents, security personnel will cooperate with on-site personnel to provide guidance and dissuasion. If violence or other dangerous behavior occurs, they will immediately report it to the relevant law enforcement agencies for intervention and coordinate police forces to disperse the situation. For traffic safety accidents, they will analyze the cause of the accident through cameras, dispatch traffic police and emergency personnel, maintain smooth traffic flow and provide timely treatment for the injured. For public health emergencies, they will immediately activate the public health emergency plan, notify the school clinic and health department to conduct on-site inspections, block the source of the epidemic, and isolate and treat the incident. For terrorist activities on campus, they will notify the police and special police for support, implement blockades and conduct strict searches to ensure the safe evacuation of teachers and students on campus and protect key facilities.

[0121] Generate different alarm messages based on different campus security threat types and indicate different alarm levels for relevant personnel;

[0122] For example, create a set of standardized alert message templates for each threat type, including the following:

[0123] Threat Type: Clearly indicate the type of incident (e.g., violent incident, fire, safety incident, etc.).

[0124] Threat Level: Defines the alert level based on the urgency of the threat (e.g., Emergency, Warning, Information).

[0125] Time and place: The time and specific location (such as building, area, class, etc.) when the incident occurred.

[0126] Event Description: Briefly describe the occurrence and nature of the threat event.

[0127] Emergency measures: The emergency plans, measures or actions recommended for this type of threat.

[0128] Alarm Level: Indicates the alarm level to systems and personnel to determine response speed and staffing (e.g., red, orange, yellow).

[0129] Example of alarm information content:

[0130] Violence:

[0131] Threat Type: Violence;

[0132] Threat Level: Urgent;

[0133] Time and place: January 20, 2025, Teaching Building A101;

[0134] Brief description of the incident: Students were found fighting in A101 of the teaching building;

[0135] Emergency measures: Initiate on-campus security patrols, call the security or police station, and conduct on-site intervention;

[0136] Alarm level: red.

[0137] Self-harm behaviors:

[0138] Threat Type: Self-harm;

[0139] Threat Level: Urgent;

[0140] Time and place: January 20, 2025, Student Dormitory Building B;

[0141] Brief description of the incident: A student was found to be engaging in self-harming behavior;

[0142] Emergency measures: notify the psychological counseling center and immediately notify the mental health teacher to intervene;

[0143] Alarm level: orange.

[0144] Fire safety threats:

[0145] Threat type: Fire safety threat;

[0146] Threat Level: Urgent;

[0147] Time and place: January 20, 2025, Teaching Building C;

[0148] Brief description of the incident: A fire broke out in teaching building C, and smoke filled the building;

[0149] Emergency measures: activate the fire emergency plan, broadcast emergency evacuation instructions, call the fire department, and evacuate people in a timely manner;

[0150] Alarm level: red.

[0151] Cybersecurity threats:

[0152] Threat type: Cyber ​​attack;

[0153] Threat Level: Warning;

[0154] Time and place: January 20, 2025, campus network;

[0155] Brief description of the incident: It was discovered that the campus network was under a DDoS attack.

[0156] Emergency measures: Isolate the attacked network through firewall monitoring, immediately fix the vulnerability, and notify the IT department;

[0157] Alarm level: yellow.

[0158] Public safety incidents:

[0159] Threat Type: Public safety incident;

[0160] Threat Level: Warning;

[0161] Time and location: January 20, 2025, Campus Square;

[0162] Brief description of the incident: A large-scale demonstration was observed in the campus square.

[0163] Emergency measures: Security personnel will conduct mediation. If the violence escalates, call the police.

[0164] Alarm level: orange.

[0165] Traffic safety accidents:

[0166] Threat type: traffic accident;

[0167] Threat Level: Warning;

[0168] Time and location: January 20, 2025, at the east gate of the campus;

[0169] Brief description of the incident: A traffic accident occurred, resulting in traffic congestion on campus.

[0170] Emergency measures: dispatch traffic police and ambulances, direct traffic, and treat the injured.

[0171] Alarm level: yellow.

[0172] Public Health Emergencies:

[0173] Threat Type: Public Health Emergency;

[0174] Threat Level: Urgent;

[0175] Time and place: January 20, 2025, student dormitory area;

[0176] Brief description of the incident: A food poisoning incident occurred, with several students experiencing vomiting and abdominal pain;

[0177] Emergency measures: Activate the public health emergency plan, immediately notify the medical office to handle the situation, seal off the source of contamination and isolate the victims;

[0178] Alarm level: red.

[0179] Terrorist activities on campus:

[0180] Threat Type: Terrorism;

[0181] Threat Level: Urgent;

[0182] Date and Location: January 20, 2025, Campus Administration Building;

[0183] Brief description of the incident: A report was received of a suspicious person carrying suspicious items on campus;

[0184] Emergency measures: Notify police and special police for support, implement blockade and search, and ensure safe evacuation;

[0185] Alarm level: red.

[0186] The preset campus security threat factor threshold is set by the staff. Different campus security threat factors are obtained through the campus security management terminal, and the average value of multiple campus security threat factors is taken as the preset campus security threat factor threshold.

[0187] In this embodiment, through trajectory data mining and clustering algorithms, students' abnormal behavior patterns (such as staying in an inactive area for a long time or abnormally frequent movement) can be identified, thereby improving the ability to discover potential risks; the law of students' transition from one behavior pattern (such as location or activity) to another behavior pattern is quantified through state transition probability, revealing the trend of dynamic changes in behavior; the intensity of students' activities in a specific location or state is evaluated through residence time, providing a basis for analyzing students' learning and life behaviors; using Markov chains to model behavior patterns, it is possible to discover irregular state transitions in students' behavior (such as moving directly from a teaching building to an unsafe area), which helps to quickly identify safe areas. The Frechet distance, a measure of trajectory similarity, can effectively screen out abnormal trajectories and further enhance the system's abnormal behavior detection capabilities. By dynamically designing the alignment path function and dynamically limiting the adjustment factors, the system can automatically adapt to the trajectory patterns of different students, ensuring the accuracy and universality of the analysis results. The clustering algorithm groups student behaviors and identifies different behavioral patterns (such as common commuting trajectories and group activity trajectories). It can analyze both individual and group behaviors, meeting the diverse needs of campus management. By extracting and analyzing student behavioral characteristics, potential safety hazards can be promptly discovered and intervened, providing a safer campus environment for students.

[0188] Dynamically limiting the alignment path adjustment factor can optimize the accuracy of trajectory alignment. During trajectory comparison, students' behavioral trajectories may differ in time span and shape. Using adjustment factors to fine-tune alignment ensures the accuracy of the alignment path, thereby better capturing behavioral patterns. By introducing factors influencing trajectory time span changes and trajectory shape differences, this method can adapt to the temporal and spatial differences between different trajectories. For example, when students' activities occur in different time periods, the dynamic limiting factor can be flexibly adjusted to ensure the rationality and stability of trajectory alignment. This reduces the need for manual intervention and parameter adjustment, and the system automatically optimizes the trajectory alignment algorithm based on changes in student behavioral data, improving the degree of automation and system intelligence.

[0189] Example 2

[0190] See also Figure 2 As shown, for the parts not described in detail in this embodiment, please refer to the description of Example 1. A campus safety management method based on the Internet is provided, including:

[0191] S1. Obtain campus security monitoring data;

[0192] S2. Preprocess the campus security monitoring data to obtain a comprehensive campus security feature dataset;

[0193] S3. Train a hazard probability prediction model based on the campus safety comprehensive feature dataset, predict campus safety threat factors based on the hazard probability prediction model, and determine whether there is a threat to campus safety based on the predicted campus safety threat factors;

[0194] S4. If there is a threat to campus security, obtain campus security threat data, input the campus security threat data into the trained security threat diagnosis model, and predict the type of campus security threat;

[0195] S5. Develop corresponding emergency plans based on the type of campus security threat and generate different alarm information; send alarm information to relevant personnel through the campus security management terminal and automatically activate the emergency plan.

[0196] Since the electronic device introduced in this embodiment is an electronic device used for implementing an Internet-based campus security management system and method in the embodiments of the present application, based on the Internet-based campus security management system and method introduced in the embodiments of the present application, those skilled in the art can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiments of the present application is not described in detail here. As long as those skilled in the art implement the electronic device used for an Internet-based campus security management system and method in the embodiments of the present application, they all fall within the scope protected by this application.

[0197] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters and thresholds in the formulas are set by technicians in this field according to actual conditions.

[0198] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the principles of the present invention are within the scope of protection of the present invention. It should be noted that for users of ordinary skill in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A campus safety management system based on the Internet, characterized in that: include: Data acquisition module, used to obtain campus security monitoring data; The data processing module is used to pre-process the campus security monitoring data to obtain a comprehensive campus security feature data set; Campus safety monitoring data includes student behavior data, campus environment data, and campus video data; Trajectory data mining and Markov chain are used to extract features from student behavior data to obtain a student behavior feature dataset; specifically, The student movement trajectory data is preset as a set of spatial position coordinate points in a time series. is a track for students to move, and another track for students to move is preset as , the similarity between the two student trajectories is calculated using the Frechet distance metric formula; the Frechet distance metric formula is: ;in, Move tracks for two students and The Frechet distance between is the alignment path function; is the set of all alignment paths; Indicates the infimum operation on all alignment paths; Indicates the supremum operation for all time points; is the Euclidean distance; Mobile tracks for students At the time point The location coordinates of The alignment path function is dynamically designed by introducing the total number of time points in the student movement trajectory data; the alignment path function is: ;in, Represents student movement trajectory In the The location at a point in time; Represents student movement trajectory In the The location at a point in time; Adjustment factor for alignment paths; is the normalization factor; the alignment path adjustment factor Perform dynamic constraints and automatically adjust the value of the constraint alignment path adjustment factor; All student movement trajectories are grouped using the DBSCAN clustering algorithm and classified into different behavior patterns. Based on trajectory data mining, different behavior patterns are modeled using Markov chains, the state space of student behavior is defined, and the behavior of students at different locations is divided into different states. Using Markov chains, we extract student behavior feature datasets, which include state transition probabilities and residence times. Combine the extracted student behavior features to obtain the student behavior feature dataset; The threat prediction module is used to train a campus security threat prediction model based on the campus security comprehensive feature dataset, predict campus security threat factors based on the campus security threat prediction model, and determine whether there is a threat to campus security based on the predicted campus security threat factors; The threat diagnosis module obtains campus security threat data if there is a threat to campus security, inputs the campus security threat data into the trained security threat diagnosis model, and predicts the type of campus security threat; The emergency response module is used to formulate corresponding emergency plans according to the type of campus security threat and generate different alarm information; send alarm information to relevant personnel through the campus security management terminal and automatically activate the emergency plan; The modules are connected via wired and / or wireless means to achieve data transmission between modules.

2. The Internet-based campus safety management system according to claim 1, characterized in that: The student behavior data includes student movement trajectory data and violent conflict data; the campus environment data includes building structure status data, power equipment status data and fire equipment status data; the campus video data includes campus public area videos, campus activity videos and individual behavior monitoring videos.

3. The Internet-based campus safety management system according to claim 2, characterized in that: The method of preprocessing the campus safety monitoring data to obtain a campus safety comprehensive feature data set also includes: The K-Means clustering algorithm is used to identify and eliminate outliers in the campus environment data and campus video data included in the campus security monitoring data, and obtain the campus environment feature data set and the campus video feature data set; Perform standard deviation normalization on the student behavior feature dataset, campus environment feature dataset, and campus video feature dataset, converting them into standard normal distribution, and obtaining the normalized student behavior feature dataset, campus environment feature dataset, and campus video feature dataset; The normalized student behavior feature dataset, campus environment feature dataset and campus video feature dataset are weighted fused to obtain a comprehensive campus safety feature dataset.

4. The Internet-based campus safety management system according to claim 3 is characterized in that: The method of extracting features from student behavior data using trajectory data mining and Markov chains to obtain a student behavior feature dataset also includes: Trajectory ;in, For students in The position coordinates of each time point; For students in The position coordinates of each time point; For students in The horizontal coordinate of the position of each time point; For students in The vertical coordinate of the position at each time point; For students in The timestamp of the position coordinate record of each time point; The total number of time points in the student's movement trajectory data; is the index of the time point in the student's movement trajectory data, The value range is from 1 to ; The state space of student behavior is: ;in, For students a behavioral pattern; An index of behavioral patterns; is the total number of behavior patterns; Define the state sequence in the student behavior trajectory as , then the state transfer matrix is ​​expressed as: ;in, represents the state transition matrix; Indicates the slave state Transfer to state probability; Represents the number of rows and columns in the state transition matrix; and The index representing the state; Preset from state Transfer to state The number of times is , then the state transition probability is expressed as: ;in, Indicates the slave state Transfer to state probability; To indicate status Transfer to state the number of times; Indicates the slave state The total number of transitions to all states; For each state , define the dwell time The length of time the student stays in the state is calculated using the residence time evaluation formula. The residence time evaluation formula is: ;in, Status duration of stay; A specific point in time in the student's movement trajectory; is the indicator function; For the time point When, the student's status.

5. The Internet-based campus safety management system according to claim 4, characterized in that: The alignment path adjustment factor Methods for dynamic limiting include: Adjust the alignment path factor using the adjustment factor constraint formula Perform dynamic restriction and adjust the factor restriction formula as follows: ;in, To control the alignment path adjustment factor Constant factors of maximum values; Adjustment factor of alignment path to control trajectory time span variation The weighting factor of the impact degree; Adjustment factor for alignment paths to control trajectory shape differences The weighting factor of the impact degree; The change in the total number of time points in the student's movement trajectory data.

6. The Internet-based campus safety management system according to claim 5, characterized in that: The training method of the campus security threat prediction model includes: The dataset is divided into training, validation, and test sets to train the model and evaluate its performance. A hazard probability prediction model is constructed, with the sample set being a subset of the dataset. Each sample set includes a historical campus safety comprehensive feature dataset and corresponding campus safety threat factors. The campus security threat prediction model includes an input layer, a hidden layer, and an output layer; the hidden layer uses a ReLU activation function; the input layer of the model is used to input a historical campus security comprehensive feature dataset; the output layer of the model is used to output campus security threat factors; the campus security threat prediction model is a multimodal neural network model; The mean squared error (MSE) is used as the loss function to measure the error between the model's predicted value and the actual value. The hazard probability prediction model is trained using the training set, and the model parameters are updated using the backpropagation algorithm and gradient descent method to minimize the loss function. The performance of the campus safety threat prediction model is evaluated by calculating the coefficient of determination using the validation set, and the model's hyperparameters are tuned. The Adam optimization algorithm is selected as the optimizer, and the model's hyperparameters are adjusted until the performance no longer improves or the preset number of iterations is reached. The test set is used to evaluate the model's performance in the prediction task, and the trained campus security threat prediction model is used to predict the current campus security comprehensive feature dataset to obtain the campus security threat factor.

7. The Internet-based campus safety management system according to claim 6, characterized in that: The method for determining whether campus security is threatened based on the predicted campus security threat factors includes: Preset a campus security threat factor threshold, and compare the predicted campus security threat factor with the preset campus security threat factor threshold; If the predicted campus security threat factor is greater than or equal to the preset campus security threat factor threshold, it is determined that there is a threat to campus security; If the predicted campus security threat factor is less than the preset campus security threat factor threshold, it is determined that there is no threat to campus security.

8. The Internet-based campus safety management system according to claim 7, characterized in that: The training method of the security threat diagnosis model includes: The dataset is divided into training set, validation set and test set to train the model and evaluate the model performance; the sample set is a subset of the dataset, and each sample contains historical campus security threat data and the corresponding campus security threat type; A security threat diagnosis model is constructed, comprising an input layer, an encoding layer, a multimodal fusion layer, a decoding layer, and an output layer. The input layer is used to input historical campus security threat data, and the output layer is used to output campus security threat types. The output layer is configured with neurons equal to the number of campus security threat types, with each neuron corresponding to the predicted probability of a campus security threat type. The softmax function is used as the activation function. The security threat diagnosis model is a multimodal Transformer model. Multi-class cross entropy is used as the model's loss function to measure the difference between the model's predicted value and the actual value. The security threat diagnosis model is trained using the training set, and the model parameters are updated using the backpropagation algorithm to minimize the loss function. The model is updated once per batch using the mini-batch gradient descent method. The performance of the security threat diagnosis model is evaluated by calculating the recall metric using the validation set. The SGD optimization algorithm is selected as the optimizer, and the model is tuned according to the performance feedback of the validation set. The model parameters are adjusted until the performance no longer improves or the preset stopping condition is reached. The test set is used to evaluate the performance of the model in the prediction task, and the trained security threat diagnosis model is used to predict the current campus security threat data to obtain the campus security threat type.

9. The Internet-based campus safety management system according to claim 8, characterized in that: The method of formulating corresponding emergency plans and generating different alarm information according to the type of campus security threat includes: The types of campus safety threats include violent incidents, self-harm, fire safety threats, threats from outsiders on campus, cybersecurity threats, public safety incidents, traffic accidents, public health emergencies, and terrorist activities on campus; For violent incidents, we will initiate on-campus security patrols and alert the security guards or police. For self-harm, we will activate the psychological counseling center for intervention and notify the mental health counselor. For fire safety threats, we will activate the fire emergency plan, broadcast emergency evacuation instructions, and call the fire department. For threats from outsiders on campus, we will track outsiders through cameras, have security personnel set up checkpoints and increase patrols, and strengthen identity verification at the school gate. For network security threats, we will monitor network abnormalities through firewalls to detect and address network attacks in a timely manner. For public safety incidents, security personnel will cooperate with on-site persons in charge to conduct guidance and dissuasion. If violence or other dangerous behaviors occur, they will immediately report to the relevant law enforcement agencies for intervention and coordinate police forces to disperse them. For traffic safety accidents, the cause of the accident will be analyzed through cameras, and traffic police and emergency personnel will be dispatched. For public health emergencies, the public health emergency plan will be immediately activated, and the school clinic and health department will be notified to conduct on-site inspections, block the source of the epidemic, and isolate them. For terrorist activities on campus, the police and special police will be notified for support, and blockades will be implemented and strict searches will be conducted. Different alarm information will be generated according to different types of campus security threats, indicating different alarm levels for relevant personnel.

10. An Internet-based campus safety management method, applied to the Internet-based campus safety management system according to any one of claims 1 to 9, characterized in that: include: S1. Obtain campus security monitoring data; S2. Preprocess the campus security monitoring data to obtain a comprehensive campus security feature dataset; S3. Train a hazard probability prediction model based on the campus safety comprehensive feature dataset, predict campus safety threat factors based on the hazard probability prediction model, and determine whether there is a threat to campus safety based on the predicted campus safety threat factors; S4. If there is a threat to campus security, obtain campus security threat data, input the campus security threat data into the trained security threat diagnosis model, and predict the type of campus security threat; S5. Develop corresponding emergency plans based on the type of campus security threat and generate different alarm information; send alarm information to relevant personnel through the campus security management terminal and automatically activate the emergency plan.

Images