An artificial intelligence-based university campus abnormal behavior real-time monitoring and early warning system

Abstract

This invention belongs to the field of smart campus security technology and discloses an artificial intelligence-based real-time monitoring and early warning system for abnormal behavior on university campuses. The system includes: a perception layer: deploying AI cameras, environmental sensors, audio equipment, and a positioning system to cover key areas such as campus gates, walls, dormitories, and laboratories, collecting multimodal data including video, environmental, audio, and personnel positioning; an edge computing layer: using edge computing devices to process the real-time data collected by the perception layer locally, including video analysis, behavior recognition, and data preprocessing, achieving low-latency data processing; a cloud analysis layer: storing historical data, deploying and continuously training and optimizing deep learning models, supporting cross-campus risk model sharing and big data analysis; and an application layer: integrating a security command center screen, a mobile APP, and an emergency response system to achieve event visualization, real-time early warning push, and cross-departmental collaborative handling. This invention achieves dual monitoring of physical safety and mental health, covering all campus security needs; multimodal data fusion and deep learning algorithms improve recognition accuracy, reduce false alarm rates, and ensure the reliability of early warnings.

Description

Technical Field

[0001] This invention belongs to, but is not limited to, the field of smart campus security technology, and particularly relates to a real-time monitoring and early warning system for abnormal behavior on university campuses based on artificial intelligence. Background Technology

[0002] In the context of current technology, with the development of information technology and intelligent security technology, campus security and prevention are gradually shifting from the traditional human-based model to a "video surveillance + information management" model. As densely populated and complex public spaces, universities have long relied on technologies such as video surveillance systems, access control systems, and alarm systems for security management. These systems enable the detection and handling of abnormal events through real-time monitoring, video recording, and post-event tracing of key areas. This type of technology system, with video surveillance at its core, achieves unified supervision of campus entrances and exits, public areas, and key locations through network transmission and central platform management. This, to some extent, compensates for the low efficiency and limited coverage of manual patrols and enhances campus security capabilities.

[0003] Building upon this foundation, further developments in smart campus security systems have gradually incorporated the Internet of Things (IoT) and cloud platform architectures. These systems collect data through various devices such as cameras, positioning devices, and environmental monitoring modules. The data is then aggregated and processed by the main controller before being uploaded to a cloud server for storage and analysis. Simultaneously, information display and alarm management are achieved through a backend management system and mobile terminals. Such systems typically include modules such as data acquisition terminals, communication networks, cloud storage and analysis platforms, and application terminals, forming a relatively complete technical system with functions including video surveillance, environmental monitoring, personnel positioning, and alarm management ([Google Patent][2]).

[0004] However, the aforementioned existing technologies still primarily revolve around a security approach dominated by video surveillance. Their data processing methods are based on single or weakly integrated information sources, which limits their practical application. First, while these systems can achieve video information collection and basic analysis, their capabilities in multimodal data collaborative processing are limited. They typically only focus on utilizing video data, lacking effective fusion mechanisms for audio, environmental, and other data, resulting in insufficient ability to identify complex abnormal behaviors. Second, existing systems largely rely on a central platform for data processing and analysis, requiring centralized data transmission to the cloud for processing, which easily introduces latency issues and is difficult to meet the high real-time requirements of campus security scenarios. Third, existing technologies focus on the detection and recording of already occurred or overt behaviors, and their early warning mechanisms are mostly based on rules or simple threshold triggers, lacking the ability to continuously analyze the evolution trend of behaviors. Therefore, they are difficult to detect and warn of highly concealed or progressive abnormal behaviors in a timely manner.

[0005] Furthermore, existing technologies often emphasize functional integration in system design but lack a multi-layered collaborative processing mechanism at the overall architecture level. Modules are loosely connected, and data and control flows do not form a closed-loop linkage, limiting the accuracy and stability of identification results. Simultaneously, in the process of data use, existing systems primarily rely on centralized storage and retrieval of personnel identity and behavioral data, lacking privacy protection mechanisms deeply integrated with the data processing workflow, posing certain compliance risks. Moreover, existing technologies typically do not incorporate implicit risk factors such as students' psychological state into the monitoring system; system functions remain mainly focused on physical security, failing to meet the actual needs of collaborative management of campus safety and mental health.

[0006] In summary, while existing technologies have established a campus security system centered on video surveillance and cloud platforms, significant shortcomings remain in areas such as multimodal data fusion capabilities, real-time processing capabilities, system collaboration mechanisms, and the depth of risk warning. These limitations hinder the comprehensive, accurate, and real-time monitoring of abnormal behavior in complex campus scenarios. Therefore, developing a campus abnormal behavior monitoring and early warning technology solution that leverages existing technologies to achieve multi-source data collaborative processing, improves real-time performance and accuracy, and ensures security and compliance is a critical technical challenge that urgently needs to be addressed in this field. Summary of the Invention

[0007] To address the problems existing in the prior art, this invention provides a real-time monitoring and early warning system for abnormal behavior on university campuses based on artificial intelligence.

[0008] This invention is implemented as follows: a real-time monitoring and early warning system for abnormal behavior on university campuses based on artificial intelligence, characterized in that the system includes:

[0009] Perception layer: Deploy AI cameras, environmental sensors, audio equipment and positioning systems to cover key areas such as campus gates, walls, dormitories, and laboratories, and collect multimodal data including video, environmental, audio and personnel positioning;

[0010] Edge computing layer: Employs edge computing devices to process real-time data collected by the perception layer locally, including video analysis, behavior recognition, and data preprocessing, to achieve low-latency data processing;

[0011] Cloud-based analytics layer: Stores historical data, deploys and continuously trains and optimizes deep learning models, and supports cross-school risk model sharing and big data analysis;

[0012] Application layer: Integrates the security command center's large screen, mobile APP, and emergency response system to achieve event visualization, real-time early warning push, and cross-departmental collaborative handling.

[0013] Furthermore, in the perception layer: the AI ​​camera supports facial recognition, behavior analysis, and object recognition functions; the environmental sensors include smoke sensors, temperature and humidity sensors, and water level sensors; the audio equipment can identify abnormal audio such as screams and glass breaking sounds; and the positioning system uses electronic student ID cards or campus cards to locate personnel.

[0014] Furthermore, the edge computing layer uses NVIDIA Jetson Orin devices, supports 60 FPS four-channel video parallel processing, data processing response latency of less than 50ms, and integrates low-light enhancement algorithms to improve nighttime data acquisition and recognition effects.

[0015] Furthermore, the deep learning models deployed in the cloud analysis layer include: the object detection model YOLOv8, the pose estimation model OpenPose, the behavior analysis model 3D CNN, the time series analysis model LSTM, and the Transformer, which are used to realize abnormal behavior recognition, psychological anomaly early warning, and multimodal data fusion analysis.

[0016] Furthermore, the core functions of this system include:

[0017] Abnormal Behavior Recognition: Through target detection, pose estimation, and behavior analysis algorithms, it identifies physical safety anomalies such as fighting, climbing walls, falls and unconsciousness, abnormal gatherings, failure to wear protective equipment in the laboratory, and illegal parking / speeding of vehicles, with an accuracy rate of ≥95%.

[0018] Multimodal fusion early warning: Combining cross-analysis of video, audio, and environmental data to form a linked early warning logic, reducing the false alarm rate to <1%;

[0019] Early warning of psychological abnormalities: An "isolation index" is constructed based on behavioral data such as students' daily routines and social patterns. A time-series analysis model is used to identify psychological abnormalities such as social withdrawal and emotional breakdown, with a false alarm rate of ≤2%.

[0020] Privacy protection: Facial desensitization, data de-identification, and anonymization are employed. Regular video recordings are looped for 7 days, and only event-related data fragments are stored, which complies with the requirements of the Personal Information Protection Law.

[0021] Furthermore, the application layer's early warning mechanism adopts a hierarchical design: Category A direct threat events trigger SMS notifications + security command center large screen pop-up warnings; Category B violations are alerted via mobile APP push notifications, while also supporting one-click alarm function, which can link with emergency departments such as fire and medical services, automatically plan escape routes and guide evacuation.

[0022] Furthermore, the cloud analytics layer supports an open platform architecture, enabling data integration and cross-campus collaboration with campus access control systems, attendance systems, and parent apps. The edge computing layer achieves low-latency data transmission with the cloud analytics layer via a 5G private network, with video transmission latency <50ms.

[0023] Furthermore, the system also includes a false alarm rate optimization module: by continuously sampling and correcting labels for 40 hours, and combining multimodal data cross-validation, the warning results are filtered and optimized. At the same time, an access control mechanism is set up to restrict the scope of data use, and compliance is ensured by privacy notices and label posting.

[0024] Another objective of this invention is to provide an AI-based real-time monitoring and early warning method for abnormal behavior on university campuses, implementing the aforementioned AI-based real-time monitoring and early warning system for abnormal behavior on university campuses. This method includes:

[0025] S1: Utilize the perception layer to collect multimodal data including video, environment, audio, and personnel positioning;

[0026] S2: Utilize the edge computing layer to process real-time data collected by the perception layer locally, including video analysis, behavior recognition, and data preprocessing, to achieve low-latency data processing;

[0027] S3: Utilizes a cloud-based analytics layer to store historical data, deploys and continuously trains and optimizes deep learning models, and supports cross-school risk model sharing and big data analysis;

[0028] S4: By integrating the security command center's large screen, mobile app, and emergency response system at the application layer, it enables event visualization, real-time early warning push notifications, and cross-departmental collaborative handling.

[0029] Another objective of this invention is to provide an information data processing terminal, which is used to implement the artificial intelligence-based real-time monitoring and early warning system for abnormal behavior on university campuses.

[0030] Based on the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solution to be protected by this invention are as follows:

[0031] The system of this invention, through multi-technology integration and a closed-loop design throughout the entire process, has the following advantages: 1. It achieves dual monitoring of physical safety and mental health, covering all campus safety needs across all scenarios; 2. Multimodal data fusion and deep learning algorithms improve recognition accuracy, reduce false alarm rates, and ensure reliable early warnings; 3. The combination of edge computing and 5G technology meets the requirements for real-time monitoring and low-latency response; 4. A robust privacy protection mechanism ensures data compliance and reduces the risk of privacy leaks; 5. The open platform architecture supports multi-system integration, improving the efficiency of campus management collaboration. This system can effectively reduce the incidence of campus safety incidents, improve management efficiency and parental satisfaction, and provide core support for the construction of smart campuses. Attached Figure Description

[0032] Figure 1 This is a structural diagram of a real-time monitoring and early warning system for abnormal behavior on a university campus based on artificial intelligence, provided in an embodiment of the present invention.

[0033] Figure 2 This is a schematic diagram of the perception layer function provided in an embodiment of the present invention;

[0034] Figure 3 This is a schematic diagram of the core functions provided in the embodiments of the present invention;

[0035] Figure 4 This is a flowchart of an artificial intelligence-based real-time monitoring and early warning method for abnormal behavior on university campuses, provided in an embodiment of the present invention.

[0036] In the diagram: 1. Perception layer; 2. Edge computing layer; 3. Cloud analytics layer; 4. Application layer; 5. AI camera; 6. Environmental sensor; 7. Audio device; 8. Positioning system. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0038] This invention provides an artificial intelligence-based real-time monitoring and early warning system for abnormal behavior on university campuses. The system includes:

[0039] Perception Layer 1: Deploy AI cameras, environmental sensors, audio equipment and positioning systems to cover key areas such as campus gates, walls, dormitories, and laboratories, and collect multimodal data including video, environmental, audio and personnel positioning;

[0040] Edge computing layer 2: Employs edge computing devices to process real-time data collected by the perception layer locally, including video analysis, behavior recognition, and data preprocessing, to achieve low-latency data processing;

[0041] Cloud analytics layer 3: Stores historical data, deploys and continuously trains and optimizes deep learning models, and supports cross-school risk model sharing and big data analysis;

[0042] Application Layer 4: Integrates the security command center's large screen, mobile APP, and emergency response system to achieve event visualization, real-time early warning push, and cross-departmental collaborative handling.

[0043] like Figure 2 As shown, in the perception layer: the AI ​​camera supports facial recognition, behavior analysis, and object recognition functions; the environmental sensors include smoke sensors, temperature and humidity sensors, and water level sensors; the audio device can identify abnormal audio such as screams and glass breaking sounds; and the positioning system uses electronic student ID cards or campus cards to locate personnel.

[0044] The edge computing layer uses NVIDIA Jetson Orin devices, supports 60 FPS four-channel video parallel processing, data processing response latency of less than 50ms, and integrates low light enhancement algorithms to improve nighttime data acquisition and recognition effects.

[0045] The deep learning models deployed in the cloud-based analytics layer include: YOLOv8 object detection model, OpenPose pose estimation model, 3D CNN behavior analysis model, LSTM time series analysis model, and Transformer, which are used to realize abnormal behavior recognition, psychological anomaly early warning, and multimodal data fusion analysis.

[0046] The AI-based real-time monitoring and early warning system for abnormal behavior on university campuses, as described in this invention, works by establishing an integrated closed-loop collaborative mechanism among the perception layer 1, edge computing layer 2, cloud analysis layer 3, and application layer 4. This mechanism enables seamless linkage throughout the entire process, from data acquisition and real-time analysis to cross-modal fusion, hierarchical early warning, and continuous optimization. This system is not a simple assembly of modules; rather, it constructs a unified operating logic through spatiotemporal synchronization, model collaboration, and feedback adaptive mechanisms. This creates strong coupling between the layers in data flow and control flow, resulting in significantly superior overall performance compared to simple combinations of existing technologies.

[0047] The perception layer 1 continuously collects video streams, environmental parameters, audio signals, and personnel location information in key areas of the campus. It then aligns the multi-source data using a unified timestamp synchronization mechanism, ensuring a strict correspondence between different modalities in the time dimension and guaranteeing the effectiveness of subsequent fusion analysis from the source. The AI ​​cameras, environmental sensors, audio devices, and positioning system do not operate independently; instead, they are initially associated and labeled according to preset multimodal association rules during the data acquisition phase. For example, video behavior segments within the same time window are bound to corresponding audio events and spatial locations, forming structured multimodal data units.

[0048] After receiving the structured multimodal data units, edge computing layer 2 performs real-time processing based on a pre-built lightweight inference model. Video data is processed by cascaded inference using a target detection model and a pose estimation model to extract behavioral features; audio data is analyzed using spectral feature extraction and an abnormal voiceprint recognition model; and environmental data is screened for anomalies using a dynamic threshold correction algorithm. In this process, edge computing layer 2 does not perform independent identification but completes the first round of association judgment locally through a cross-modal rapid fusion mechanism. For example, when rapid human movement is detected, it matches scream signals and abnormal positional changes to generate a preliminary set of candidate anomalies, thereby reducing the probability of false alarms propagating to the cloud at the source.

[0049] Furthermore, edge computing layer 2 uploads the screened candidate events and compressed key feature data to cloud analysis layer 3. Data transmission is synchronized with low latency via a 5G private network and employs an event-driven triggering mechanism, uploading only when the multimodal association threshold condition is met. This avoids the bandwidth consumption and response latency issues caused by massive redundant data transmission in traditional systems. Cloud analysis layer 3 performs deep fusion processing on the received data, constructing a unified multimodal inference link through target detection models, pose estimation models, behavior analysis models, and time series analysis models. The models do not operate independently but progress step-by-step in the order of "spatial behavior features—temporal evolution features—group relationship features," achieving multi-dimensional judgment of abnormal behavior.

[0050] In this multi-model collaborative process, the system achieves information exchange between models by sharing a feature vector space. For example, the pose estimation result is used as the input constraint of the behavior analysis model, while the output of the time-series model is used as the dynamic weight of the anomaly score, thus forming a cross-model feedback adjustment mechanism, which makes the recognition results continuous and stable. This mechanism effectively avoids the judgment conflict problem caused by the isolated operation of each model in the prior art, and improves the recognition accuracy and robustness at the system level.

[0051] After anomaly detection, the cloud analytics layer 3 transmits the results to the application layer 4. The application layer 4 executes a tiered response strategy based on event type, risk level, and impact scope. Category A events trigger a high-priority alarm directly on the security command center and across multiple terminals, while Category B events are alerted via mobile devices and their behavior is recorded. During the response process, the application layer 4 does not simply receive results; instead, it sends feedback information back to the cloud analytics layer 3 as labeled data for model retraining and parameter optimization, thus constructing a continuous learning loop.

[0052] Furthermore, this system employs a dynamic self-correction mechanism through a false alarm rate optimization module. This module performs statistical analysis on the model output based on continuously sampled data and incorporates multimodal consistency verification rules to conduct secondary screening of abnormal events. For example, when video detection results are inconsistent with audio and environmental data, the system automatically lowers the abnormal score of the event or marks it as pending confirmation, thereby suppressing the accumulation of false alarms at the system level. This mechanism, together with the initial screening mechanism at the edge computing layer and the multi-model fusion mechanism in the cloud, forms a three-level filtering structure, significantly reducing the overall false alarm rate.

[0053] Regarding privacy protection, the system embeds an anonymization mechanism into the data flow transmission path, performing feature processing on facial data at the edge computing layer, retaining only the necessary recognition vectors, and employing a de-identification storage strategy in the cloud to ensure that data remains under control throughout its flow at each layer. This privacy protection mechanism is not an add-on feature but is deeply integrated into the data processing workflow, ensuring compliance at the system architecture level.

[0054] This invention achieves holistic linkage between the perception layer 1, edge computing layer 2, cloud analysis layer 3, and application layer 4 in terms of time synchronization, feature sharing, model collaboration, and feedback optimization. This results in a highly coupled system structure and a continuous closed loop in functionality, significantly different from the simple splicing or parallel combination of modules in existing technologies. This holistic collaborative mechanism not only achieves synergistic improvements in real-time performance, accuracy, and false alarm control, but also enables continuous evolution of system performance through multi-level data filtering and dynamic optimization. Consequently, the technical effects achieved by this invention cannot be derived from simple combinations of existing technologies.

[0055] like Figure 3 As shown, the core functions of this system include:

[0056] Abnormal Behavior Recognition: Through target detection, pose estimation, and behavior analysis algorithms, it identifies physical safety anomalies such as fighting, climbing walls, falls and unconsciousness, abnormal gatherings, failure to wear protective equipment in the laboratory, and illegal parking / speeding of vehicles, with an accuracy rate of ≥95%.

[0057] Multimodal fusion early warning: Combining cross-analysis of video, audio, and environmental data to form a linked early warning logic, reducing the false alarm rate to <1%;

[0058] Early warning of psychological abnormalities: An "isolation index" is constructed based on behavioral data such as students' daily routines and social patterns. A time-series analysis model is used to identify psychological abnormalities such as social withdrawal and emotional breakdown, with a false alarm rate of ≤2%.

[0059] Privacy protection: Facial desensitization, data de-identification, and anonymization are employed. Regular video recordings are looped for 7 days, and only event-related data fragments are stored, which complies with the requirements of the Personal Information Protection Law.

[0060] The application layer's early warning mechanism adopts a hierarchical design: Category A direct threat events trigger SMS notifications + security command center large screen pop-up warnings; Category B violations are pushed to the mobile APP for reminders, and also support one-click alarm function, which can link with emergency departments such as fire and medical services, automatically plan escape routes and guide evacuation.

[0061] The cloud analytics layer supports an open platform architecture, enabling data integration and cross-campus collaboration with campus access control systems, attendance systems, and parent apps. The edge computing layer achieves low-latency data transmission with the cloud analytics layer via a 5G private network, with video transmission latency <50ms.

[0062] The system also includes a false alarm rate optimization module: by continuously sampling and correcting labels for 40 hours, and combining multimodal data cross-validation, the warning results are filtered and optimized. At the same time, an access control mechanism is set up to restrict the scope of data use, and compliance is ensured by privacy notices and label posting.

[0063] like Figure 4 As shown in the figure, this invention provides a method for real-time monitoring and early warning of abnormal behavior on university campuses using artificial intelligence, implementing the aforementioned artificial intelligence-based real-time monitoring and early warning system for abnormal behavior on university campuses. The method includes:

[0064] S1: Utilize the perception layer to collect multimodal data including video, environment, audio, and personnel positioning;

[0065] S2: Utilize the edge computing layer to process real-time data collected by the perception layer locally, including video analysis, behavior recognition, and data preprocessing, to achieve low-latency data processing;

[0066] S3: Utilizes a cloud-based analytics layer to store historical data, deploys and continuously trains and optimizes deep learning models, and supports cross-school risk model sharing and big data analysis;

[0067] S4: By integrating the security command center's large screen, mobile app, and emergency response system at the application layer, it enables event visualization, real-time early warning push notifications, and cross-departmental collaborative handling.

[0068] The real-time monitoring and early warning method for abnormal behavior on university campuses using artificial intelligence, as described in this embodiment of the invention, works by using an information data processing terminal as the core control unit to drive the perception layer 1, edge computing layer 2, cloud analysis layer 3, and application layer 4 to form a continuous data flow and control flow collaborative operation mechanism, thereby achieving real-time identification and hierarchical early warning of abnormal behavior.

[0069] In step S1, the information data processing terminal accesses the perception layer 1 through the communication interface, uniformly receives the multimodal data collected by the AI ​​camera, environmental sensor, audio device and positioning system, and performs alignment processing on video frames, audio signals, environmental parameters and location information based on the time synchronization mechanism to generate a multimodal data sequence with a unified time identifier, providing a consistent data foundation for subsequent processing.

[0070] In step S2, the information data processing terminal schedules the edge computing layer 2 to perform local real-time processing on the aforementioned multimodal data sequences. Specifically, video data undergoes behavioral feature extraction using target detection and pose estimation algorithms, audio data undergoes acoustic feature extraction using an abnormal voiceprint recognition algorithm, and environmental data undergoes anomaly screening using a dynamic threshold model. Data cleaning, compression, and feature encoding are also performed at the edge. Simultaneously, the edge computing layer 2 performs preliminary fusion judgment on various features based on cross-modal association rules, generates anomaly candidate events, and achieves rapid response through a low-latency mechanism.

[0071] In step S3, the information data processing terminal uploads the filtered key feature data and abnormal candidate events to the cloud analysis layer 3. The cloud analysis layer 3 performs in-depth fusion analysis on the data, constructing a multi-layer inference chain through target detection models, behavior analysis models, and time series models to make precise judgments on abnormal behaviors. At the same time, the cloud analysis layer 3 continuously stores and trains historical data, improving recognition accuracy and generalization ability through model parameter updates and cross-school data sharing mechanisms.

[0072] In step S4, the information data processing terminal transmits the cloud analysis results to application layer 4. Application layer 4 triggers corresponding early warning strategies based on the event risk level, displays the results visually on the security command center's large screen, pushes them in real time via a mobile app, and enables cross-departmental collaborative handling through the emergency response system. Simultaneously, application layer 4 sends the handling feedback information back to cloud analysis layer 3 for model optimization and false alarm correction, forming a closed-loop feedback mechanism.

[0073] Through the coordinated operation of the above steps, this method forms a continuous closed loop between data acquisition, edge processing, cloud analysis and application response, enabling unified scheduling and dynamic optimization of information data processing terminals across all layers. This improves the real-time performance, accuracy and system stability of abnormal behavior identification as a whole, which is significantly different from the independent implementation of each processing step in existing technologies.

[0074] It should be noted that embodiments of the present invention can be implemented in hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above-described hardware circuitry and software, such as firmware.

[0075] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An artificial intelligence-based real-time monitoring and early warning system for abnormal behavior on a university campus, characterized in that, The system includes: Perception layer: Deploy AI cameras, environmental sensors, audio equipment and positioning systems to cover key areas such as campus gates, walls, dormitories, and laboratories, and collect multimodal data including video, environmental, audio and personnel positioning; Edge computing layer: Employs edge computing devices to process real-time data collected by the perception layer locally, including video analysis, behavior recognition, and data preprocessing, to achieve low-latency data processing; Cloud-based analytics layer: Stores historical data, deploys and continuously trains and optimizes deep learning models, and supports cross-school risk model sharing and big data analysis; Application layer: Integrates the security command center's large screen, mobile APP, and emergency response system to achieve event visualization, real-time early warning push, and cross-departmental collaborative handling.

2. The real-time monitoring and early warning system for abnormal behavior on a university campus based on artificial intelligence according to claim 1, characterized in that, In the perception layer: the AI ​​camera supports facial recognition, behavior analysis, and object recognition functions; the environmental sensors include smoke sensors, temperature and humidity sensors, and water level sensors; the audio equipment can identify abnormal audio such as screams and glass breaking sounds; and the positioning system uses electronic student ID cards or campus cards to locate personnel. 3.The real-time monitoring and early warning system for abnormal behavior on a university campus based on artificial intelligence according to claim 1, characterized in that, The edge computing layer uses NVIDIA Jetson Orin devices, supports 60 FPS four-channel video parallel processing, data processing response latency of less than 50ms, and integrates low light enhancement algorithms to improve nighttime data acquisition and recognition effects.

4. The real-time monitoring and early warning system for abnormal behavior on a university campus based on artificial intelligence according to claim 1, characterized in that, The deep learning models deployed in the cloud-based analytics layer include: YOLOv8 object detection model, OpenPose pose estimation model, 3D CNN behavior analysis model, LSTM time series analysis model, and Transformer, which are used to realize abnormal behavior recognition, psychological anomaly early warning, and multimodal data fusion analysis. 5.The real-time monitoring and early warning system for abnormal behavior on a university campus based on artificial intelligence according to claim 1, characterized in that, The core functions of the system include: Abnormal Behavior Recognition: Through target detection, pose estimation, and behavior analysis algorithms, it identifies physical safety anomalies such as fighting, climbing walls, falls and unconsciousness, abnormal gatherings, failure to wear protective equipment in the laboratory, and illegal parking / speeding of vehicles, with an accuracy rate of ≥95%. Multimodal fusion early warning: Combining cross-analysis of video, audio, and environmental data to form a linked early warning logic, reducing the false alarm rate to <1%; Early warning of psychological abnormalities: An "isolation index" is constructed based on behavioral data such as students' daily routines and social patterns. A time-series analysis model is used to identify psychological abnormalities such as social withdrawal and emotional breakdown, with a false alarm rate of ≤2%. Privacy protection: Facial desensitization, data de-identification, and anonymization are employed. Regular video recordings are looped for 7 days, and only event-related data fragments are stored, which complies with the requirements of the Personal Information Protection Law.

6. The real-time monitoring and early warning system for abnormal behavior on university campuses based on artificial intelligence as described in claim 1, characterized in that, The application layer's early warning mechanism adopts a hierarchical design: Category A direct threat events trigger SMS notifications + security command center large screen pop-up warnings; Category B violations are pushed to the mobile APP for reminders, and also support one-click alarm function, which can link with emergency departments such as fire and medical services, automatically plan escape routes and guide evacuation.

7. The real-time monitoring and early warning system for abnormal behavior on university campuses based on artificial intelligence as described in claim 1, characterized in that, The cloud analytics layer supports an open platform architecture, enabling data integration and cross-campus collaboration with campus access control systems, attendance systems, and parent apps. The edge computing layer achieves low-latency data transmission with the cloud analytics layer via a 5G private network, with video transmission latency <50ms.

8. The real-time monitoring and early warning system for abnormal behavior on university campuses based on artificial intelligence as described in claim 1, characterized in that, The system also includes a false alarm rate optimization module: by continuously sampling and correcting labels for 40 hours, and combining multimodal data cross-validation, the warning results are filtered and optimized. At the same time, an access control mechanism is set up to restrict the scope of data use, and compliance is ensured by privacy notices and label posting.

9. A method for real-time monitoring and early warning of abnormal behavior on a university campus using artificial intelligence, implementing the artificial intelligence-based real-time monitoring and early warning system for abnormal behavior on a university campus as described in any one of claims 1-8, characterized in that, The method includes: S1: Utilize the perception layer to collect multimodal data including video, environment, audio, and personnel positioning; S2: Utilize the edge computing layer to process real-time data collected by the perception layer locally, including video analysis, behavior recognition, and data preprocessing, to achieve low-latency data processing; S3: Utilizes a cloud-based analytics layer to store historical data, deploys and continuously trains and optimizes deep learning models, and supports cross-school risk model sharing and big data analysis; S4: By integrating the security command center's large screen, mobile app, and emergency response system at the application layer, it enables event visualization, real-time early warning push notifications, and cross-departmental collaborative handling.

10. An information data processing terminal, characterized in that, The information data processing terminal is used to implement the artificial intelligence-based real-time monitoring and early warning system for abnormal behavior on university campuses as described in claim 9.

Images