Smart campus security data processing method based on big data

Abstract

The invention relates to the technical field of campus security management, in particular to an intelligent campus security data processing method based on big data, and the method comprises the steps: collecting video monitoring data and access control equipment data in real time, rapidly recognizing clear and high-frequency basic risks, triggering graded early warning to generate graded early warning signals, and for complex and fuzzy risks, carrying out early warning on the risk, and carrying out early warning on the risk. Performing secondary judgment by the campus risk model; the campus risk model carries out matching by calling historical features from a historical sample library, calculates the upgrading probability of the risk based on a matching result, and triggers graded early warning based on the upgrading probability to generate a graded early warning signal; and based on the intelligent decision model, automatically matching a linkage strategy according to the graded early warning signal. Therefore, the problems of campus security response lag, missed report and false report, insufficient complex risk identification precision, low early warning processing efficiency, resource waste and the like are solved.

Description

Technical Field

[0001] This application relates to the field of campus security management technology, and in particular to a smart campus security data processing method based on big data. Background Technology

[0002] With the deepening of educational informatization and the continuous expansion of campus scale, the traditional campus security model has gradually exposed its shortcomings in adapting to complex security needs, while the maturity of technologies such as big data and artificial intelligence has provided key support for the upgrading of smart campus security systems.

[0003] Traditional campus security relies heavily on manual monitoring and post-incident tracing. For sudden security incidents, it often only allows for post-incident review via video recordings, lacking real-time early warning capabilities. Manual monitoring is limited by manpower and attention spans, resulting in a false negative rate of over 30% for hidden risks such as wall climbing and unusual nighttime loitering. Furthermore, campus security involves a large amount of sensitive data, such as facial recognition information of faculty and students, behavioral trajectories, and dormitory entry and exit records. Directly sharing raw data during cross-campus collaborations or data collection by higher-level departments can easily lead to privacy leaks. For example, when building a unified security model, multi-campus universities often refuse to provide local abnormal behavior samples for fear of student identity information leakage, resulting in insufficient training data and limited recognition accuracy. Moreover, some risks on campus are complex and ambiguous, difficult to determine using single features, easily leading to insufficient accuracy in identifying complex risks. Traditional early warning systems often employ a "one-size-fits-all" approach, such as notifying all security personnel of all alerts, resulting in resource misallocation, low efficiency in early warning response, and resource waste. Summary of the Invention

[0004] This application provides a data processing method for smart campus security based on big data, in order to solve problems such as delayed response, missed or false alarms, insufficient accuracy in identifying complex risks, low efficiency in early warning and handling, and waste of resources in campus security.

[0005] The first aspect of this application provides a smart campus security data processing method based on big data. This method requires a smart campus security management system, which includes: an edge terminal, a campus risk model, and an early warning linkage module. The edge terminal includes an AI-powered smart camera and an access control device integrating multiple sensors. The smart camera, equipped with a lightweight YOLOv5 model, is installed in dormitory corridors, near windows in high-rise dormitories, around power distribution rooms, and on the playground to collect video surveillance data in real time. The access control device, integrating infrared sensors, a facial recognition module, and an RFID reader, is installed at the entrances and exits of teaching buildings, student dormitories, and dedicated laboratory passages to collect access control data. Based on the preset risk identification rules of the access control device and the preset behavior identification rules of the smart camera, clear and high-frequency basic risks are quickly identified, triggering tiered early warnings, generating tiered early warning signals, and adding the basic risk data to the historical database. For complex and ambiguous risks, the edge terminal encrypts the collected feature data and uploads it to the campus node, where the campus risk model performs a secondary judgment and adds the judgment result to the historical database. The campus risk model is used to match complex and ambiguous risks input by the edge terminal by calling historical features from the historical sample database. Based on the matching results, the campus risk model calculates the escalation probability of the complex and ambiguous risks, triggers a graded early warning based on the escalation probability, generates a graded early warning signal, and adds the complex and ambiguous risk data to the historical database. The campus risk model is built through federated learning and differential privacy collaborative operation. The construction process is as follows: each pilot campus selects 10,000 labeled samples from the historical database, performs desensitization and data transformation, and uploads them to the campus nodes. The campus-level platform trains a behavior recognition model based on the uploaded data and distributes it to the campus nodes through an encrypted channel. The campus nodes train the model using local data. After training, only the model parameters are extracted, encrypted using the RSA algorithm, and uploaded to the campus-level platform. The campus-level platform uses the "weighted average" method to aggregate the parameters to generate a global model and distributes it to each campus node, covering the local model. The early warning linkage module is based on an intelligent decision-making model and automatically matches linkage strategies according to the graded early warning signals. The big data-based smart campus security data processing method includes the following steps: S101. Real-time collection of video surveillance data and access control equipment data. Through the risk identification logic of access control equipment and the behavior recognition logic of cameras, it quickly identifies clear and high-frequency basic risks, triggers graded early warning, generates graded early warning signals, and adds basic risk data to the historical database. For complex and ambiguous risks, the collected feature data is uploaded to the campus node for secondary judgment by the campus risk model. S102. The campus risk model matches complex and ambiguous risks input from edge terminals by calling historical features from the historical sample database. Based on the matching results, the model calculates the escalation probability of complex and ambiguous risks, triggers graded early warning based on the escalation probability, generates graded early warning signals, and adds complex and ambiguous risk data to the historical database. S103. Based on the intelligent decision-making model, automatically match linkage strategies according to the graded early warning signals.

[0006] Optionally, the preset risk identification rules of the access control device include: when the infrared sensor detects that the first target and the second target pass through continuously within a first preset time, and the second target has not completed face recognition or card swiping, the risk is determined to be abnormal tailgating; when the face recognition result is matched with the access control permission database but the permission does not match, the risk is determined to be permission mismatch; when the same target fails to be recognized a preset number of times consecutively within a second preset time, the risk is determined to be suspicious probing.

[0007] Optionally, the process of comparing the facial recognition result with the access control permission database includes: Step 1: Constructing an access control permission database that stores the feature vectors and associated information of authorized personnel. After the camera captures a face, the facial information is converted into a string of feature vectors using a built-in feature extraction algorithm; Step 2: The access control device calculates the similarity between the real-time extracted facial feature vectors and the vectors in the permission database, outputs a matching score, and determines whether the match is successful based on the matching score; Step 3: If the feature match is successful, the permission information of the person is automatically associated to verify "whether they have permission to enter the current area" and "whether they are within the valid permission time"; Step 4: Based on the feature matching result and the permission verification result, a judgment result is derived, including feature matching and permission granted, feature matching but permission not matching, suspected feature matching, and feature not matching.

[0008] Optionally, the preset behavior recognition rules of the camera include: real-time detection of key points of the human skeleton in the image, excluding normal behavior, and determining that a person has fallen when the angle between the torso and the ground is less than or equal to a preset angle and lasts for a third preset time; analyzing dynamic changes in the image through frame difference method, and determining that an object is thrown from a height when it is detected moving rapidly in the air and its trajectory is free fall, combined with a timestamp; pre-marking restricted areas, and determining that a person is staying in a restricted area when the camera detects that a person has entered the area and stayed for more than a preset threshold; and counting the number of people in the area through target detection, and determining that a gathering conflict occurs when the number of people in the preset range exceeds a threshold and is accompanied by rapid limb waving.

[0009] Optionally, the campus risk model further includes an iterative optimization process, which includes repeating the following process weekly: campus nodes train the model using local data, inputting 2000 samples each time, adjusting the model weights through backpropagation, and after training, extracting only the model parameters, encrypting them using the RSA algorithm, and uploading them to the campus-level platform. The campus-level platform allocates weights according to the number of samples in each campus, generates a global model, and distributes it to each campus node, covering the local model.

[0010] Optionally, the campus risk model calculates the escalation probability of the complex and ambiguous risk by: extracting key features from real-time collected ambiguous risk data and historical samples to form a multi-dimensional input vector; training the historical sample database using machine learning algorithms to automatically calculate the weights of each feature; and multiplying the quantized values ​​of the real-time features by their corresponding weights and summing the results to obtain the total score. The calculation formula is as follows:

[0011] in, For the total score, For the first The basic weights of each feature For the first Quantized values ​​of real-time features, The number of features to be implemented; The probability of advancement is obtained by mapping the total score to a probability value between 0% and 100% using the sigmoid function. The calculation formula is as follows:

[0012] in, For risk probability, This is the total score.

[0013] Optionally, generating a tiered early warning signal based on the escalation probability includes: determining the relationship between the escalation probability and a preset threshold, wherein the preset threshold includes a first threshold, a second threshold, and a third threshold, wherein the first threshold is less than the second threshold, and the second threshold is less than the third threshold; determining a risk level based on the relationship between the escalation probability and the preset threshold, wherein the risk level includes a low-risk level, a medium-risk level, and a high-risk level; and determining a tiered early warning signal based on the risk level, wherein the tiered early warning signal includes a low-risk early warning signal, a medium-risk early warning signal, and a high-risk early warning signal; wherein a low-risk level triggers a low-risk early warning signal, a medium-risk level triggers a medium-risk early warning signal, and a high-risk level triggers a high-risk early warning signal.

[0014] Optionally, the automatic matching and linkage strategy based on the graded early warning signals includes: the graded early warning signals include low-risk early warning signals, medium-risk early warning signals, and high-risk early warning signals; the linkage strategy for low-risk early warning signals is that the system automatically performs minimal intervention operations; the linkage strategy for medium-risk early warning signals is that the system triggers a local response mechanism, expanding the linkage range to a 300-meter radius resource circle; the linkage strategy for medium-risk early warning signals is that the system activates an emergency response plan, linking all resources inside and outside the campus.

[0015] A second aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a smart campus security data processing method based on big data as described in the above embodiments.

[0016] A third aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement a smart campus security data processing method based on big data as described in the above embodiments.

[0017] Therefore, this application has the following beneficial effects: This application's embodiments, through the built-in basic rules and lightweight models on edge terminals, can directly judge the collected data, shortening the warning time for basic risks from minutes to seconds, while reducing the false alarm rate of manual judgment, thus solving the problems of delayed response, missed reporting, and false alarms in campus security. The campus risk model, through multi-dimensional feature matching and probability calculation, identifies fuzzy and complex risks, significantly improving the accuracy of complex risk identification and solving the problem of insufficient precision in complex risk identification. The intelligent decision-making model matches precise strategies according to the warning level, avoiding resource waste, and continuously improves the handling efficiency through a feedback mechanism, solving the problems of low warning handling efficiency and resource waste. Using a federated learning architecture and differential privacy, it integrates the data value of multiple campuses while complying with personal information protection regulations, resolving the contradiction between cross-campus data sharing and privacy protection.

[0018] This solved problems such as delayed response, missed or false alarms, insufficient accuracy in identifying complex risks, low efficiency in early warning and response, and waste of resources in campus security.

[0019] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0020] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart illustrating a smart campus security data processing method based on big data, according to an embodiment of this application. Figure 2 This is a schematic diagram of a smart campus security management system provided according to an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0021] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0022] The following description, with reference to the accompanying drawings, illustrates a data processing method for smart campus security based on big data, according to an embodiment of this application. Addressing the issues of delayed response, missed or false alarms, insufficient accuracy in identifying complex risks, low efficiency in early warning and response, and resource waste mentioned in the background art, this application provides a data processing method for smart campus security based on big data. By incorporating basic rules and lightweight models into edge terminals, the collected data can be directly evaluated, reducing the early warning time for basic risks from minutes to seconds, while simultaneously reducing the false alarm rate of manual judgment, thus solving the problems of delayed response, missed or false alarms in campus security. The campus risk model, through multi-dimensional feature matching and probability calculation, identifies fuzzy and complex risks, significantly improving the accuracy of complex risk identification and solving the problem of insufficient accuracy in complex risk identification. The intelligent decision-making model matches precise strategies according to the early warning level, avoiding resource waste, and continuously improves response efficiency through a feedback mechanism, solving the problems of low efficiency in early warning and response and resource waste. The use of a federated learning architecture and differential privacy integrates the data value of multiple campuses while complying with personal information protection regulations, resolving the conflict between cross-campus data sharing and privacy protection.

[0023] This solved problems such as delayed response, missed or false alarms, insufficient accuracy in identifying complex risks, low efficiency in early warning and response, and waste of resources in campus security.

[0024] Specifically, Figure 1 This is a flowchart illustrating a smart campus security data processing method based on big data, provided in an embodiment of this application.

[0025] like Figure 1 As shown, this big data-based smart campus security data processing method includes the following steps: In step S101, video surveillance data and access control device data are collected in real time. The risk identification logic of the access control device and the behavior recognition logic of the camera are used to quickly identify clear and high-frequency basic risks, trigger graded early warning, generate graded early warning signals, and add the basic risk data to the historical database. For complex and ambiguous risks, the collected feature data is uploaded to the campus node for secondary judgment by the campus risk model.

[0026] The video surveillance data is collected by smart cameras equipped with AI chips, while the access control equipment data is collected by access control devices that integrate infrared sensors, facial recognition modules, and RFID card readers. The preset risk identification rules for the access control devices include: when the infrared sensor detects two targets passing through consecutively within a first preset time period, and the second target fails to complete facial recognition or card swiping, the risk is determined to be abnormal tailgating; when the facial recognition result matches the access control permission database but the permissions do not match, the risk is determined to be permission mismatch; when the same target fails to be recognized a preset number of times consecutively within a second preset time period, the risk is determined to be suspicious probing.

[0027] For example, at 8:30 AM, in a student dormitory access control channel, student A opens the dormitory door using facial recognition. The infrared sensor is triggered for the first time and recorded as the first target passing through. Within 10 seconds, the infrared sensor is triggered again, detecting person B following student A into the channel and recorded as the second target passing through. However, B did not perform facial recognition or card swiping. The system judges this as abnormal tailgating, triggering a medium-risk warning and simultaneously pushing information to the dormitory manager. The dormitory manager verifies B's identity through channel monitoring, with the first preset time set to 15 seconds. In a dedicated laboratory channel, at 2:10 PM, person C performs facial recognition in the laboratory channel. The system extracts their facial features and compares them with the permission database. The feature matching rate reaches 92%, meeting the feature matching standard. Further verification of permissions reveals that C's identity information belongs to another laboratory member and is not authorized to enter the current laboratory. The system judges this as permission mismatch, triggering a medium-risk warning, keeping the access control closed, and simultaneously notifying the laboratory administrator to come and inquire. At the entrance of the teaching building, from 9:00 to 9:05 AM, with a second preset time of 5 minutes, Person D attempted facial recognition 4 times consecutively at the access control point. Of these, 3 attempts were preset, but all failed due to reasons such as "facial obstruction" and "angle deviation." Furthermore, Person D did not attempt to swipe their card. The system determined this to be a suspicious attempt, triggering a primary risk warning and alerting the security guard to pay attention to Person D.

[0028] Furthermore, the process of comparing the facial recognition results with the access control permission database includes: Step 1: Constructing an access control permission database that stores the feature vectors and associated information of authorized personnel. After the camera captures a face, the facial information is converted into a string of feature vectors using a built-in feature extraction algorithm; Step 2: The access control device calculates the similarity between the real-time extracted facial feature vectors and the vectors in the permission database, outputs a matching score, and determines whether a match is successful based on the matching score; Step 3: If the feature match is successful, the personnel's permission information is automatically associated to verify "whether they have permission to enter the current area" and "whether they are within the valid permission time"; Step 4: Based on the feature matching result and the permission verification result, a judgment result is derived, including feature matching and permission granted, feature matching but permission not applicable, suspected feature matching, and feature mismatch.

[0029] For example, the access control database for a chemistry lab at a university stores information on 10 authorized members of the lab: each person corresponds to a set of facial feature vectors, with associated information including "name, student ID / employee ID, and permitted entry time (weekdays 9:00-22:00)". Student E is an authorized member of the lab, student F is a member of the physics lab (not authorized by the chemistry lab), and external personnel G has no campus access control privileges.

[0030] The access control database has pre-stored the feature vector of student E. (Related permissions: Chemistry Lab, 9:00-22:00), Feature vector of student F (Associated permissions: Physics lab, no chemistry lab permissions) and other authorization information. At 10:00 AM, student E walks towards the chemistry lab corridor. The access control camera captures their face, and the built-in feature extraction algorithm converts their facial information into a real-time feature vector. Access control equipment will With the permission library , Calculate the similarity of each vector individually. and The similarity is 95% (higher than the preset matching threshold of 85%), the output matching score is 95, and the feature is judged to be "successfully matched". and The similarity is 60% (below the threshold), therefore a "feature mismatch" is determined. and The match was successful. The system automatically linked Student E's permission information and verified: "Does he / she have permission to enter the current area?": Student E's permissions include the chemistry lab, so it passed; "Is he / she within the valid permission time?": The current time is 10:00, which is within the 9:00-22:00 range, so it passed. Finally, Student E's features matched successfully and the permission verification passed, so the result was "features matched and permission passed," and the access control automatically opened.

[0031] Student F attempts to enter the chemistry lab; real-time feature vector. In the permission library The similarity was 93% (feature matching successful). After associating with access control information, it was found that student F did not have access to the chemistry lab. The judgment result was "feature matching but access not matching," and the access control remained closed, triggering a medium-risk warning. The facial feature vector of off-campus personnel G... With an authorized person in the permission database The similarity is 80% (between the "suspected threshold 70% and matching threshold 85%)", the judgment result is "feature suspected match", the system prompts "please align with the camera again", and notifies the security guard to pay attention. Facial feature vector of off-campus person G. The similarity with all authorized vectors is less than 70%, the judgment result is "feature mismatch", the access control refuses to open, and it is recorded as "unauthorized personnel attempted to enter".

[0032] Specifically, the camera's preset behavior recognition rules include: real-time detection of key points of the human skeleton in the image, excluding normal behavior, and determining that a person has fallen when the angle between the torso and the ground is less than or equal to a preset angle and lasts for a third preset time; analyzing dynamic changes in the image through frame difference method, and determining that an object is thrown from a height when it is detected moving rapidly in the air and its trajectory is free fall, combined with a timestamp; pre-marking restricted areas, and determining that a person is staying in a restricted area when the camera detects that they have entered the area and stayed for more than a preset threshold; and counting the number of people in an area through target detection, and determining that a gathering conflict occurs when the number of people in a preset range exceeds a threshold and is accompanied by rapid limb waving.

[0033] For example, at 11:30 PM, student H slipped and fell in the dormitory corridor due to the wet floor. The angle between his torso and the ground instantly became 30° (the preset angle is set to 45°), and this posture lasted for 10 seconds (the third preset time is set to 5 seconds). The system excluded normal behaviors such as "squatting to tie shoelaces" and judged it as "person falling to the ground," triggering a medium-risk warning and notifying the dormitory manager to come to the scene immediately. At 3:00 PM, an empty beverage bottle fell from the window of a high-rise dormitory. The camera detected that the object was moving quickly in the air, with a parabolic trajectory. Based on the timestamp, the falling time was calculated to be about 2 seconds. The system judged it as "object thrown from a height," triggering a high-risk warning. It linked all cameras in the building facing the square to trace the source of the object and notified security personnel to evacuate the crowd below. At 12:15 PM, an outsider, I, curiously approached the power distribution room, entered the restricted area, and stood there for 1 minute. The preset threshold was set to 30 seconds. The system detected that the person was continuously in the restricted area and judged it as "staying in the restricted area," triggering a medium-risk warning and notifying security patrol personnel to persuade him to leave. At 6:00 PM, several students got into a scuffle while playing ball. The camera detected that there were 8 people in the area (the threshold was set to 5), and many of them were making rapid arm swinging and body collisions. The system judged this as a "gathering conflict", triggered a medium-risk warning, and notified the playground security guards to intervene and mediate immediately.

[0034] Understandably, by using built-in basic rules and lightweight models on edge terminals, the collected data can be directly evaluated, reducing the warning time for basic risks from minutes to seconds, while also reducing the false alarm rate of manual judgment. This solves the problems of delayed response, missed reports, and false alarms in campus security. The complete process of comparing facial recognition results with the access control database confirms identity through feature vectors and verifies access control area permissions, quickly identifying risky behaviors in special scenarios and improving warning efficiency. The camera's preset behavior recognition rules target high-frequency risk scenarios on campus and combine computer vision technology to achieve automated judgment, making up for the lack of real-time performance of manual monitoring and providing accurate basis for front-end risk filtering.

[0035] In step S102, the campus risk model matches the complex and ambiguous risks input by the edge terminal by calling historical features from the historical sample database. Based on the matching results, the model calculates the escalation probability of the complex and ambiguous risks, triggers a graded early warning based on the escalation probability, generates a graded early warning signal, and adds the complex and ambiguous risk data to the historical database.

[0036] The campus risk model is built through federated learning and differential privacy collaborative operation. The construction process is as follows: each pilot campus selects 10,000 labeled samples from the historical database, performs anonymization and data transformation, and uploads them to the campus node. The campus-level platform trains the behavior recognition model based on the uploaded data and distributes it to the campus node through an encrypted channel. The campus node uses local data to train the model. After training, only the model parameters are extracted, encrypted with the RSA algorithm, and uploaded to the campus-level platform. The campus-level platform uses the "weighted average" method to aggregate the parameters to generate a global model and distributes it to each campus node to cover the local model.

[0037] Furthermore, the campus risk model also includes an iterative optimization process, which includes: The following process is repeated weekly: Campus nodes train models using local data, with 2000 samples input for each iteration. Model weights are adjusted through backpropagation. After training, only the model parameters are extracted, encrypted using the RSA algorithm, and uploaded to the campus-level platform. The campus-level platform allocates weights according to the number of samples from each campus, generates a global model, and distributes it to each campus node, overriding the local model.

[0038] Understandably, the campus risk model identifies ambiguous and complex risks through multi-dimensional feature matching and probability calculation, significantly improving the accuracy of complex risk identification and solving the problem of insufficient precision in complex risk identification. By using a federated learning architecture and differential privacy, it integrates the value of data from multiple campuses while complying with personal information protection regulations, thus resolving the conflict between cross-campus data sharing and privacy protection.

[0039] Furthermore, the campus risk model calculates the escalation probability of complex and ambiguous risks by: extracting key features from real-time collected ambiguous risk data and historical samples to form a multi-dimensional input vector; training the historical sample database using machine learning algorithms to automatically calculate the weights of each feature; and multiplying the quantized values ​​of the real-time features by their corresponding weights and summing the results to obtain the total score. The calculation formula is as follows:

[0040] in, For the total score, For the first The basic weights of each feature For the first Quantized values ​​of real-time features, The number of features to be implemented; The probability of advancement is obtained by mapping the total score to a probability value between 0% and 100% using the sigmoid function. The calculation formula is as follows:

[0041] in, For risk probability, This is the total score.

[0042] For example, the probability of a conflict arising from a gathering of 4 people is calculated by extracting key features from real-time collected fuzzy risk data and historical samples to form a multi-dimensional input vector. Taking the scenario of a gathering of 4 people as an example, the feature dimensions include: the magnitude of physical actions (e.g., "slight pushing = 1 point, punching = 3 points"), the intensity of verbal conflict (judged by sound recognition, e.g., "no arguing = 0 points, loud arguing = 2 points"), the type of area (e.g., "library = 1, playground = 0.5, quiet areas have a higher risk of conflict"), the time (e.g., "nighttime = 1.2, daytime = 0.8"), the relationship between people (e.g., "the percentage of unfamiliar faces = 0.3"), and the escalation rate of similar scenarios in the area over the past 3 months (e.g., "15% of gatherings of 4 people result in conflict").

[0043] The model is trained on a historical sample database using machine learning algorithms, automatically calculating the weights of each feature. Assuming the model learns from 100 historical cases that: "Punching gesture" has the strongest correlation with the upgrade result, with a weight of 0.3; "Library area" has a weight of 0.2; "Nighttime" has a weight of 0.15; and the total weight of other features is 0.35. A higher weight value indicates a greater impact of that feature on the upgrade probability. Total Score Upgrade probability That is, the upgrade probability is 77.7%.

[0044] Furthermore, the generation of tiered early warning signals based on escalation probability triggering tiered early warning includes: The relationship between the escalation probability and preset thresholds is determined. These preset thresholds include a first threshold, a second threshold, and a third threshold, where the first threshold is less than the second threshold, and the second threshold is less than the third threshold. Based on this relationship, risk levels are determined, including low-risk, medium-risk, and high-risk levels. Tiered early warning signals are then determined according to the risk levels, including low-risk, medium-risk, and high-risk warning signals. Specifically, a low-risk level triggers a low-risk warning signal, a medium-risk level triggers a medium-risk warning signal, and a high-risk level triggers a high-risk warning signal.

[0045] Understandably, tiered early warning systems can transform vague possibilities into specific event probabilities and impact levels, thus refining the risk level and facilitating rapid risk identification. Tiered signals allow relevant personnel to quickly assess the urgency of events, avoiding wasting time on low-priority events and ensuring that high-risk events are handled first. After tiering, resources can be concentrated on high-risk events, improving resource utilization.

[0046] In step S103, based on the intelligent decision-making model, linkage strategies are automatically matched according to the hierarchical early warning signals.

[0047] The automatic matching and linkage strategy based on the graded early warning signals includes: graded early warning signals include low-risk, medium-risk, and high-risk early warning signals. The linkage strategy for low-risk early warning signals is that the system automatically performs minimal intervention operations; the linkage strategy for medium-risk early warning signals is that the system triggers a local response mechanism, expanding the linkage range to a 300-meter radius resource circle; and the linkage strategy for high-risk early warning signals is that the system activates the emergency response plan, linking all resources inside and outside the campus.

[0048] Specifically, the system automatically performs minimal intervention operations including: controlling the camera in the area to increase the frame rate, switching to "key recording mode," and extending the storage period to 7 days, without triggering audible and visual alarms; simultaneously maintaining normal access control without restrictions; only pushing low-risk alerts to the mobile apps of on-duty security personnel in the area, including location, current probability, and instructions to remotely check every 15 minutes, without requiring on-site handling; automatically generating linkage operation logs and synchronizing them to campus nodes for evidence storage, facilitating later traceability. The system triggers a local response mechanism, expanding the linkage range to a 300-meter radius resource circle, including: switching the access control system in the area to exit-only mode, activating the camera to select and continuously focus on the disputed person, and automatically lowering the background music volume if there is broadcasting equipment in the area; sending medium-risk dispatch instructions to patrolling security personnel within 300 meters via walkie-talkie, and simultaneously pushing warning information to the counselors of the departments to which the area belongs, with suggestions on the identity of suitable personnel to assist; updating on-site data to the campus platform every 30 seconds, dynamically updating the upgrade probability. The system activated its emergency response plan, coordinating all resources inside and outside the campus, including: turning on all emergency lighting in the affected area and within a 50-meter radius, automatically unlocking fire exit doors, turning all cameras towards the area to form 360° surveillance without blind spots, switching the access control system to emergency lock status across the entire area, and continuously broadcasting evacuation instructions through the campus broadcasting system; sending a full assembly order to all security personnel through the emergency command center, automatically dialing school leaders and the local police station, and pushing an emergency alarm to the entire faculty and student APP, along with an evacuation route map.

[0049] Understandably, matching precise strategies to early warning levels avoids resource waste, while continuously improving response efficiency through feedback mechanisms solves the problems of low efficiency and resource waste in early warning response. Low-risk minimal intervention operations do not require manpower, saving core resources such as school security teams and medical personnel. The 300-meter radius resource circle for medium-risk events avoids mobilizing the entire school's resources for localized incidents. High-risk emergency handling only activates full resource coordination in extreme cases, ensuring that limited emergency resources are used for major emergencies and avoiding cost pressures caused by excessive reserves or mobilization in daily operations.

[0050] Optionally, in some embodiments, the scenario is as follows: Behind a university dormitory building is a green belt, which is a restricted area (such as a construction area or a dangerous area), and students are usually prohibited from entering. At the same time, there is a risk of individual students throwing objects from the upper floors of the dormitory building. One night at 11:15 PM, a student, A, became emotionally agitated, carried an empty beverage bottle to the area, and appeared to be throwing it upwards.

[0051] Smart cameras are deployed in the dormitory corridors and rear areas to collect video streams in real time. The cameras incorporate a lightweight YOLOv5 model for behavior recognition. The green area has been pre-marked as a "restricted zone." When person A enters the restricted zone and remains there for 60 seconds (exceeding the preset threshold of 30 seconds), the "restricted zone stay" risk is triggered. Simultaneously, frame difference analysis reveals that person A is holding an object moving rapidly upwards. The trajectory does not conform to free fall, but the movement is abnormal and ambiguous, making it impossible to definitively classify it as throwing an object from a height or a normal action. The access control system did not detect any anomalies. Because "restricted zone stay" is a clear, high-frequency basic risk, a low-risk warning signal is immediately triggered, and basic risk data is recorded in the historical database. "Suspected throwing action" is a complex, ambiguous risk. The video feature data for that period (such as movement trajectory, speed, direction, and key points of human posture) is encrypted and uploaded to the campus node for secondary analysis by the campus risk model.

[0052] The campus risk model receives "suspected object throwing" feature data uploaded by edge terminals. The model retrieves similar cases from the historical sample database for matching. For example, there are 100 "high-altitude object throwing" incidents in the historical data, of which 80% occurred between 22:00 and 24:00; 70% of the throwing behavior was preceded by "long-term loitering in the restricted area"; 60% of the perpetrators had "emotional agitation" or "limb waving" characteristics; and 50% were related to "recent psychological abnormality records".

[0053] The model extracts a multi-dimensional feature vector of the current event: Feature 1: Time is 23:15 (nighttime, high weight); Feature 2: Duration is 60 seconds (exceeds threshold, medium weight); Feature 3: Movement trajectory shows an upward throwing trend (not free fall, medium weight); Feature 4: No other people gathered (excluding group conflict, low weight); Feature 5: Student A has two records of returning home late in the past week. Assume the weights and quantification values ​​are as follows: Time weight Quantization value ; stay time , ; movement trajectory , Recent behavior , ;other , Total Score Upgrade probability That is, the upgrade probability is 72%.

[0054] Thresholds are set: the first threshold is 30%, the second threshold is 60%, and the third threshold is 85%. If 60% < 72% < 85%, it is judged as a medium risk level, and a medium risk warning signal is generated. The model will judge "suspected high-altitude littering + emotional behavior" as medium risk and update it to the historical database.

[0055] Upon receiving a "medium-risk warning signal," the early warning and linkage module automatically matches a linkage strategy based on a preset intelligent decision-making model: triggering a local response mechanism, expanding the linkage range to a 300-meter radius resource circle, and automatically executing the linkage operation of notifying the nearest security patrol personnel, retrieving surrounding cameras (playground, corridors) for linkage tracking, continuously monitoring the student's movements, pushing information to the dormitory administrator to remind them to pay attention to the dynamics of the student's dormitory floor, and recording the entire process of the incident for subsequent psychological counseling or safety education.

[0056] Security personnel arrived at the scene within 5 minutes and found that student A was in a low mood and had not thrown the empty bottle in his hand. After being comforted, A said that he just wanted some peace and quiet and had no malicious intent. The incident was marked as a "false alarm", but the system recorded it as a "successful warning for medium risk" for subsequent model optimization.

[0057] The incident was included in the local historical database. A week later, the campus node used 2,000 new samples, including this case, to train the local model. The model parameters were encrypted and uploaded to the campus platform to participate in federated learning aggregation and global model updates. In the future, the accuracy of identifying behaviors such as "nighttime loitering + upward movement" will be improved.

[0058] In this embodiment, "restricted areas" are quickly identified by cameras, with local response reducing latency; "suspected objects" that cannot be determined are uploaded to the central model for secondary analysis, improving the accuracy of identifying complex and ambiguous risks; upgrading probabilities by combining historical data enables intelligent grading, allowing for reasonable resource allocation and avoiding waste; multi-campus data collaborative modeling protects student privacy; and event feedback is used for model iteration, improving the model's accuracy and speed in identification.

[0059] The smart campus security data processing method based on big data proposed in this application, through the built-in basic rules and lightweight models of the edge terminal, can directly judge the collected data, shortening the warning time of basic risks from minutes to seconds, while reducing the false alarm rate of manual judgment, and solving the problems of delayed response, missed reporting, and false alarms in campus security. The campus risk model identifies fuzzy and complex risks through multi-dimensional feature matching and probability calculation, significantly improving the accuracy of complex risk identification and solving the problem of insufficient accuracy in complex risk identification. The intelligent decision-making model matches precise strategies according to the warning level, avoiding resource waste, and continuously improves the handling efficiency through the feedback mechanism, solving the problems of low warning handling efficiency and resource waste. The use of federated learning architecture and differential privacy integrates the data value of multiple campuses and complies with personal information protection regulations, solving the contradiction between cross-campus data sharing and privacy protection.

[0060] This solved problems such as delayed response, missed or false alarms, insufficient accuracy in identifying complex risks, low efficiency in early warning and response, and waste of resources in campus security.

[0061] Next, referring to the accompanying drawings, a smart campus security management system based on an embodiment of this application is described.

[0062] Figure 2 This is a block diagram of a smart campus security management system according to an embodiment of this application.

[0063] like Figure 2 As shown, the smart campus security management system 10 includes: an edge terminal 100, a campus risk model 200, and an early warning linkage module 300.

[0064] The edge terminal 100 includes an AI-powered smart camera and an access control device integrating multiple sensors. The smart camera, equipped with a lightweight YOLOv5 model, is installed in dormitory corridors, near windows in high-rise dormitories, around power distribution rooms, and on the playground to collect video surveillance data in real time. The access control device, integrating infrared sensors, a face recognition module, and an RFID reader, is installed at the entrances and exits of teaching buildings, student dormitories, and dedicated laboratory passages to collect access control data. Through the preset risk identification rules of the access control device and the preset behavior identification rules of the smart camera, clear and high-frequency basic risks are quickly identified, triggering graded early warnings, generating graded early warning signals, and adding the basic risk data to the historical database. For complex and ambiguous risks, the edge terminal encrypts the collected feature data and uploads it to the campus node, where the campus risk model makes a secondary judgment and adds the judgment result to the historical database. The Campus Risk Model 200 is used to match complex and ambiguous risks input from edge terminals by calling historical features from the historical sample database. Based on the matching results, the Campus Risk Model 200 calculates the escalation probability of complex and ambiguous risks, triggers tiered early warning based on the escalation probability, generates tiered early warning signals, and adds complex and ambiguous risk data to the historical database. The Campus Risk Model is built through federated learning and differential privacy collaborative operation. The construction process is as follows: each pilot campus selects 10,000 labeled samples from the historical database, performs desensitization and data transformation, and uploads them to the campus nodes. The campus-level platform trains the behavior recognition model based on the uploaded data and distributes it to the campus nodes through an encrypted channel. The campus nodes use local data to train the model. After training, only the model parameters are extracted, encrypted with the RSA algorithm, and uploaded to the campus-level platform. The campus-level platform uses the "weighted average" method to aggregate the parameters to generate a global model and distributes it to each campus node to cover the local model. The early warning linkage module 300 is based on an intelligent decision-making model and automatically matches linkage strategies according to the graded early warning signals.

[0065] It should be noted that the foregoing explanation of the embodiment of the smart campus security data processing method based on big data also applies to the smart campus security management system of this embodiment, and will not be repeated here.

[0066] The smart campus security management system proposed in this application, through the built-in basic rules and lightweight models on the edge terminal, can directly judge the collected data, shortening the warning time of basic risks from minutes to seconds, while reducing the false alarm rate of manual judgment, thus solving the problems of delayed response, missed reporting, and false alarms in campus security. The campus risk model, through multi-dimensional feature matching and probability calculation, identifies fuzzy and complex risks, significantly improving the accuracy of complex risk identification and solving the problem of insufficient accuracy in complex risk identification. The intelligent decision-making model matches precise strategies according to the warning level, avoiding resource waste, and continuously improves the handling efficiency through a feedback mechanism, thus solving the problems of low warning handling efficiency and resource waste. The use of federated learning architecture and differential privacy integrates the data value of multiple campuses while complying with personal information protection regulations, thus solving the contradiction between cross-campus data sharing and privacy protection.

[0067] This solved problems such as delayed response, missed or false alarms, insufficient accuracy in identifying complex risks, low efficiency in early warning and response, and waste of resources in campus security.

[0068] Figure 3 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: The memory 301, the processor 302, and the computer program stored on the memory 301 and capable of running on the processor 302.

[0069] When the processor 302 executes the program, it implements the smart campus security data processing method based on big data provided in the above embodiments.

[0070] Furthermore, electronic devices also include: Communication interface 303 is used for communication between memory 301 and processor 302.

[0071] The memory 301 is used to store computer programs that can run on the processor 302.

[0072] The memory 301 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0073] If the memory 301, processor 302, and communication interface 303 are implemented independently, then the communication interface 303, memory 301, and processor 302 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 3 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0074] Optionally, in a specific implementation, if the memory 301, processor 302, and communication interface 303 are integrated on a single chip, then the memory 301, processor 302, and communication interface 303 can communicate with each other through an internal interface.

[0075] Processor 302 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0076] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described smart campus security data processing method based on big data.

[0077] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0078] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0079] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0080] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0081] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

Claims

1. A smart campus security data processing method based on big data, characterized in that, The smart campus security data processing method requires a smart campus security management system during its implementation. This system includes: edge terminals, a campus risk model, and an early warning and linkage module. The edge terminal includes an AI-powered smart camera and an access control device integrating multiple sensors. The smart camera, equipped with a lightweight YOLOv5 model, is installed in dormitory corridors, near windows in high-rise dormitories, around power distribution rooms, and on the playground to collect video surveillance data in real time. The access control device, integrating infrared sensors, a facial recognition module, and an RFID reader, is installed at the entrances and exits of teaching buildings, student dormitories, and dedicated laboratory passages to collect access control data. Based on the preset risk identification rules of the access control device and the preset behavior identification rules of the smart camera, clear and high-frequency basic risks are quickly identified, triggering tiered early warnings, generating tiered early warning signals, and adding the basic risk data to the historical database. For complex and ambiguous risks, the edge terminal encrypts the collected feature data and uploads it to the campus node, where the campus risk model performs a secondary judgment and adds the judgment result to the historical database. The campus risk model is used to match complex and ambiguous risks input by the edge terminal by calling historical features from the historical sample database. Based on the matching results, the campus risk model calculates the escalation probability of the complex and ambiguous risks, triggers a graded early warning based on the escalation probability, generates a graded early warning signal, and adds the complex and ambiguous risk data to the historical database. The campus risk model is built through federated learning and differential privacy collaborative operation. The construction process is as follows: each pilot campus selects 10,000 labeled samples from the historical database, performs de-identification and data transformation, and uploads them to the campus nodes. The campus-level platform trains a behavior recognition model based on the uploaded data and distributes it to the campus nodes through an encrypted channel. The campus nodes train the model using local data. After training, only the model parameters are extracted, encrypted with the RSA algorithm, and uploaded to the campus-level platform. The campus-level platform uses the "weighted average" method to aggregate the parameters to generate a global model and distributes it to each campus node, covering the local model. The early warning linkage module is based on an intelligent decision-making model and automatically matches linkage strategies according to the graded early warning signals. The big data-based smart campus security data processing method includes the following steps: S101. Real-time collection of video surveillance data and access control equipment data. Through the risk identification logic of access control equipment and the behavior recognition logic of cameras, it quickly identifies clear and high-frequency basic risks, triggers graded early warning, generates graded early warning signals, and adds basic risk data to the historical database. For complex and ambiguous risks, the collected feature data is uploaded to the campus node for secondary judgment by the campus risk model. S102. The campus risk model matches complex and ambiguous risks input from edge terminals by calling historical features from the historical sample database. Based on the matching results, the model calculates the escalation probability of complex and ambiguous risks, triggers graded early warning based on the escalation probability, generates graded early warning signals, and adds complex and ambiguous risk data to the historical database. S103. Based on the intelligent decision-making model, automatically match linkage strategies according to the graded early warning signals.

2. The smart campus security data processing method based on big data according to claim 1, characterized in that, The preset risk identification rules of the access control device include: When the infrared sensor detects that the first target and the second target pass through continuously within a first preset time, and the second target has not completed face recognition or card swiping, the risk is determined to be abnormal tailgating; When the facial recognition result matches the access control permission database but the permissions do not match, the risk is determined to be permission mismatch; If the same target fails to be identified a preset number of times within a second preset time period, the risk is judged as a suspicious attempt.

3. The smart campus security data processing method based on big data according to claim 2, characterized in that, The process of comparing facial recognition results with the access control database includes: Step 1: Construct an access control permission library that stores the feature vectors and associated information of authorized personnel. After the camera captures a face, the facial information is converted into a string of feature vectors using the built-in feature extraction algorithm. Step 2: The access control device will calculate the similarity between the real-time extracted facial feature vector and the vector in the permission database, output a matching score, and determine whether the match is successful based on the matching score; Step 3: If the feature match is successful, automatically associate the person's permission information and verify "whether they have permission to enter the current area" and "whether they are within the valid permission time". Step 4: Based on the feature matching results and permission verification results, a judgment result is obtained. The judgment result includes feature matching and permission granted, feature matching but permission not met, suspected feature matching, and feature not matching.

4. The smart campus security data processing method based on big data according to claim 1, characterized in that, The preset behavior recognition rules of the camera include: Real-time detection of key points of the human skeleton in the image, excluding normal behavior, and determining that a person has fallen to the ground when the angle between the torso and the ground is less than or equal to a preset angle and lasts for a third preset time. By analyzing the dynamic changes in the image using the frame difference method, when an object is detected moving rapidly in the air and its trajectory is in free fall, it is determined to be a high-altitude object thrown by combining the timestamp. Restricted areas are marked in advance. When the camera detects that a person has entered the area and stayed there for more than a preset threshold, it is determined to be a restricted area. By detecting the number of people within a target area, a conflict is identified when the number of people within a preset range exceeds a threshold and is accompanied by rapid limb waving.

5. The smart campus security data processing method based on big data according to claim 1, characterized in that, The campus risk model also includes an iterative optimization process, which includes: The following process is repeated weekly: Campus nodes train models using local data, with 2000 samples input for each iteration. Model weights are adjusted through backpropagation. After training, only the model parameters are extracted, encrypted using the RSA algorithm, and uploaded to the campus-level platform. The campus-level platform allocates weights according to the number of samples from each campus, generates a global model, and distributes it to each campus node, overriding the local model.

6. The smart campus security data processing method based on big data according to claim 1, characterized in that, The campus risk model calculates the escalation probability of the complex and ambiguous risks as follows: Key features are extracted from real-time collected fuzzy risk data and historical samples to form a multi-dimensional input vector; The historical sample database is trained using machine learning algorithms to automatically calculate the weights of each feature; The total score is obtained by multiplying the quantized values ​​of the real-time features by their corresponding weights and then summing the results. The calculation formula is as follows: ； in, For the total score, For the first The basic weights of each feature For the first Quantized values ​​of real-time features, The number of features to be implemented; The probability of advancement is obtained by mapping the total score to a probability value between 0% and 100% using the sigmoid function. The calculation formula is as follows: ； in, For risk probability, This is the total score.

7. The smart campus security data processing method based on big data according to claim 1, characterized in that, Generating tiered early warning signals based on the escalation probability trigger includes: Determine the relationship between the upgrade probability and a preset threshold, wherein the preset threshold includes a first threshold, a second threshold, and a third threshold, wherein the first threshold is less than the second threshold, and the second threshold is less than the third threshold; Based on the relationship between the upgrade probability and the preset threshold, the risk level is determined, including low risk level, medium risk level and high risk level; A tiered early warning signal is determined based on the risk level, and the tiered early warning signal includes a low-risk early warning signal, a medium-risk early warning signal, and a high-risk early warning signal; Among them, a low-risk level triggers a low-risk warning signal, a medium-risk level triggers a medium-risk warning signal, and a high-risk level triggers a high-risk warning signal.

8. The smart campus security data processing method based on big data according to claim 1, characterized in that, The automatic matching and linkage strategy based on the aforementioned graded early warning signals includes: The tiered early warning signals include low-risk, medium-risk, and high-risk early warning signals. The linkage strategy for the low-risk early warning signal is that the system automatically performs minimal intervention operations. The linkage strategy for the medium-risk early warning signal is a system-triggered local response mechanism, with the linkage range expanded to a resource circle with a radius of 300 meters; The linkage strategy for the high-risk early warning signal is to activate the emergency response plan of the system and link all resources inside and outside the campus.

9. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a smart campus security data processing method based on big data as described in any one of claims 1-8.

10. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, The program or instructions are executed by the processor to implement a smart campus security data processing method based on big data as described in any one of claims 1-8.

Images