Intelligent campus safety monitoring and early warning system and method thereof

Abstract

The invention relates to the technical field of campus safety, in particular to an intelligent campus safety monitoring and early warning system and method, and an image acquisition module is responsible for acquiring and transmitting campus monitoring images to a server. The video preprocessing module receives the image, then carries out noise reduction and color correction, and then sends the image to the video analysis module. And the video analysis module analyzes the image by using deep learning, identifies an abnormal event and judges and alarms. The abnormal behavior database stores related data. And the abnormal behavior analysis module performs feature extraction and clustering, identifies potential anomalies and corrects prediction data. And the abnormal behavior prediction module predicts potential abnormity and judges and alarms. And the abnormal behavior feedback module generates a report and early warning information. The alarm module receives the alarm signal and gives an alarm. And the monitoring and early warning server uniformly manages and processes information and sends early warning to the feedback module, so that the accuracy and the real-time performance of abnormal event identification are greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of campus safety technology, in particular to an intelligent campus safety monitoring and early warning system and method thereof. Background Art

[0002] As campus safety issues are increasingly attracting social attention, traditional campus safety monitoring systems can no longer meet the needs of the current complex and ever-changing security situation. The existing campus safety monitoring systems mainly have the following problems:

[0003] First, most existing systems still rely on manual monitoring, which not only consumes a lot of human resources, but is also easily affected by human factors, resulting in unstable monitoring results. Even if some automation technologies are adopted, their intelligence level is low, and they can often only identify simple abnormal situations, and lack effective early warning capabilities for complex security threats.

[0004] Second, existing systems usually use a single data source for analysis, such as relying solely on video surveillance data. This approach ignores the complexity of the campus environment and fails to fully grasp the security situation. For example, potential dangers caused by certain environmental factors, such as harmful gas leaks or smoke in the early stages of a fire, cannot be detected through video alone.

[0005] Furthermore, most current campus security systems use a centralized data processing architecture, and all data needs to be transmitted to a central server for processing. This approach not only increases network bandwidth pressure, but also prolongs system response time, making it difficult to meet the needs of real-time monitoring and rapid response.

[0006] In addition, existing systems have limited capabilities in abnormal behavior analysis and prediction. They usually adopt fixed rules or simple statistical models, which are difficult to adapt to changing security threats. This makes the system prone to false positives or false negatives, reducing the efficiency and reliability of security management.

[0007] Finally, the existing campus security system often lacks effective early warning and emergency response mechanisms. Even if an abnormal situation is detected, it is difficult to quickly and accurately assess the scope of impact and formulate appropriate response strategies, resulting in untimely and inadequate handling of security incidents. Summary of the invention

[0008] In view of the above problems, the present invention proposes an intelligent campus safety monitoring and early warning system and method thereof. The system integrates multiple functional modules to achieve full-process intelligent management from data collection, preprocessing, analysis, prediction to early warning.

[0009] The present invention proposes an intelligent campus safety monitoring and early warning system, comprising:

[0010] Image acquisition module for:

[0011] Collect images of the campus surveillance area;

[0012] Send the collected images to the monitoring and warning server;

[0013] The video preprocessing module is connected to the image acquisition module for:

[0014] Receiving the image sent by the image acquisition module;

[0015] Performing noise reduction filtering and color correction on the received image;

[0016] Send the preprocessed image to the video analysis module;

[0017] The video analysis module is connected to the video preprocessing module for:

[0018] Receiving the preprocessed image sent by the video preprocessing module;

[0019] The received images are analyzed based on the deep learning monitoring model;

[0020] Identify unusual events on campus;

[0021] Determine whether to trigger an alarm;

[0022] Abnormal behavior database module, used for:

[0023] Storing abnormal behavior detection data;

[0024] Storing abnormal behavior prediction data;

[0025] An abnormal behavior analysis module, which is in communication with the abnormal behavior database module and the video analysis module, is used to:

[0026] Receiving abnormal event information identified by the video analysis module;

[0027] Perform feature extraction and clustering on the received abnormal event information;

[0028] Identify potentially anomalous behavior;

[0029] Incremental learning and correction of the data input to the abnormal behavior prediction module;

[0030] The abnormal behavior prediction module is in communication with the abnormal behavior analysis module and is used to:

[0031] Receiving potential abnormal behavior information sent by the abnormal behavior analysis module;

[0032] Predict potential abnormal behaviors;

[0033] Determine whether to trigger an alarm;

[0034] An abnormal behavior feedback module is in communication with the abnormal behavior analysis module and the abnormal behavior prediction module, and is used to:

[0035] Receiving analysis results of the abnormal behavior analysis module and the abnormal behavior prediction module;

[0036] Generate abnormal behavior reports and early warning information;

[0037] An alarm module is communicatively connected with the video analysis module, the abnormal behavior prediction module and the abnormal behavior feedback module, and is used to:

[0038] receiving a signal that triggers an alarm;

[0039] Send out alarm information;

[0040] A monitoring and early warning server is connected in communication with the image acquisition module, the abnormal behavior analysis module, the abnormal behavior prediction module, the abnormal behavior feedback module and the alarm module, and is used to:

[0041] Receive information from each module;

[0042] Centralized management and processing of abnormal alarm information;

[0043] Send warning information to the abnormal behavior feedback module.

[0044] Preferably, the video analysis module comprises:

[0045] Convolutional Neural Network units for:

[0046] Extracting features of the preprocessed image;

[0047] A recurrent neural network unit, in communication with the convolutional neural network unit, for:

[0048] Receiving image features extracted by the convolutional neural network unit;

[0049] Establishing the temporal relationship between consecutive video frames;

[0050] A fully connected layer network unit is communicatively connected to the recurrent neural network unit and is used to:

[0051] Combined with feature selection algorithm, calculate the anomaly score;

[0052] A softmax judgment unit is connected to the fully connected layer network unit for:

[0053] Making a final judgment based on the abnormality score;

[0054] Output abnormal event recognition results.

[0055] As a preference, it also includes:

[0056] Sensor modules for:

[0057] Collect environmental parameters on campus, including temperature, humidity, wind speed, carbon dioxide concentration, sound, smoke concentration, and dust concentration;

[0058] Sending the collected environmental parameters to the abnormal behavior analysis module;

[0059] Wherein, the abnormal behavior analysis module is also used for:

[0060] Receiving environmental parameters sent by the sensor module;

[0061] A comprehensive abnormal behavior analysis is performed by combining the environmental parameters and the video analysis results.

[0062] Preferably, the abnormal behavior analysis module comprises:

[0063] Multi-stream convolutional neural network units for:

[0064] Perform multi-scale feature extraction on the received video data;

[0065] A long short-term memory network unit is communicatively connected to the multi-stream convolutional neural network unit, and is used to:

[0066] Modeling video data sequences;

[0067] Statistical analysis unit for:

[0068] Statistical modeling of normal and abnormal behavior;

[0069] Deep Uncertain Clustering Unit, used for:

[0070] Perform cluster analysis on potential abnormal behaviors;

[0071] Among them, the abnormal behavior analysis module realizes comprehensive analysis and identification of abnormal behavior by fusing the analysis results of the multi-stream convolutional neural network unit, the long short-term memory network unit, the statistical analysis unit and the deep uncertain clustering unit.

[0072] Preferably, the abnormal behavior prediction module comprises:

[0073] Feature extractors for:

[0074] Receiving potential abnormal behavior information from the abnormal behavior analysis module;

[0075] Extract features of potential abnormal behavior;

[0076] A Gaussian distribution model unit is communicatively connected to the feature extractor and is used to:

[0077] Establishing a Gaussian probability distribution model based on the extracted features;

[0078] A maximum likelihood optimization unit is communicatively connected to the Gaussian distribution model unit and is used to:

[0079] Transform the abnormal behavior prediction problem into a maximum likelihood optimization problem;

[0080] Solve the maximum likelihood optimization problem;

[0081] A prediction result output unit is communicatively connected to the maximum likelihood optimization unit and is used to:

[0082] Output abnormal behavior prediction results based on maximum likelihood optimization results;

[0083] Determine whether to trigger an alarm.

[0084] As a preference, it also includes:

[0085] The edge computing module is connected to the image acquisition module for:

[0086] Conduct preliminary abnormal behavior judgment in the smart camera;

[0087] When potential abnormal behavior is detected, trigger the sending of a complete video stream to the monitoring and warning server;

[0088] Among them, the edge computing module reduces the amount of data transmission and improves the real-time performance of the system by performing preliminary processing at the front end.

[0089] Preferably, the monitoring and early warning server further includes:

[0090] Multi-level early warning unit for:

[0091] Determining a warning level based on output results of the abnormal behavior analysis module and the abnormal behavior prediction module;

[0092] Select the corresponding warning method according to the warning level;

[0093] Personnel management unit for:

[0094] Maintain campus personnel information database;

[0095] Combined with the abnormal behavior analysis results, assess the scope of people who may be affected;

[0096] Emergency response unit for:

[0097] Automatically generate emergency response plans based on the warning level and the range of people that may be affected;

[0098] Push emergency response instructions to relevant personnel.

[0099] As a preference, it also includes:

[0100] The security management platform is connected to the monitoring and early warning server in communication, and includes:

[0101] Mobile applications for:

[0102] Receiving warning information pushed by the monitoring and warning server;

[0103] Display real-time campus safety status;

[0104] Provides rapid reporting of abnormal events;

[0105] Web management interface for:

[0106] Provide visual display of campus safety data;

[0107] Support security policy configuration and system parameter adjustment;

[0108] Generate a safety analysis report.

[0109] As a preference, it also includes:

[0110] The adaptive optimization module is connected to the abnormal behavior analysis module, the abnormal behavior prediction module and the monitoring and early warning server for:

[0111] Collect system operation data and user feedback information;

[0112] Based on the collected data, incremental learning and parameter adjustment are performed on the abnormal behavior analysis model;

[0113] Dynamically optimize warning rules and thresholds;

[0114] Generate system performance evaluation report to provide basis for further optimization.

[0115] The method based on the intelligent campus safety monitoring and early warning system comprises the following steps:

[0116] S1. Collect images of the campus monitoring area through the image acquisition module;

[0117] S2. The video preprocessing module performs noise reduction filtering and color correction on the captured image;

[0118] S3. The video analysis module analyzes the preprocessed images based on a deep learning monitoring model to identify abnormal events on campus;

[0119] S4. The abnormal behavior analysis module extracts and clusters the identified abnormal events to identify potential abnormal behaviors;

[0120] S5. The abnormal behavior prediction module predicts potential abnormal behavior and determines whether to trigger an alarm;

[0121] S6. The abnormal behavior feedback module generates abnormal behavior reports and warning information based on the analysis and prediction results;

[0122] S7. When an alarm is required, the alarm module issues an alarm message;

[0123] S8. The monitoring warning server receives information from each module, uniformly manages and processes abnormal alarm information, and sends warning information to the abnormal behavior feedback module;

[0124] S9. The adaptive optimization module dynamically optimizes the models and parameters of each module based on system operation data and user feedback;

[0125] Wherein, step S4 also includes:

[0126] Receiving environmental parameters collected by the sensor module;

[0127] Combine environmental parameters and video analysis results to conduct comprehensive abnormal behavior analysis;

[0128] Steps S3 and S4 also include:

[0129] The edge computing module performs preliminary abnormal behavior judgment in the smart camera, and when potential abnormal behavior is detected, triggers the sending of a complete video stream to the monitoring and early warning server;

[0130] Step S8 also includes:

[0131] Determine the warning level and select the warning method through the multi-level warning unit;

[0132] Assess the scope of personnel that may be affected through the personnel management unit;

[0133] Generate an emergency response plan and push an emergency response instruction through the emergency response unit;

[0134] The method further includes:

[0135] The mobile application and web management interface of the security management platform can realize rapid reporting of abnormal events, visualization of security data and system management functions.

[0136] Compared with the prior art, the present invention has the following beneficial effects:

[0137] First, the present invention uses deep learning-based video analysis technology, which greatly improves the accuracy and real-time performance of abnormal event recognition. The system can automatically learn and adapt to different types of abnormal behavior patterns, greatly reducing the burden of manual monitoring while improving the comprehensiveness and continuity of monitoring.

[0138] Secondly, the present invention innovatively integrates multi-source data, including video data and environmental sensor data. This multimodal data fusion method enables the system to perceive the campus environment more comprehensively, effectively improving the accuracy and reliability of abnormal event detection. For example, the system can simultaneously analyze suspicious behaviors in the video and abnormal changes in environmental parameters, thereby identifying potential security threats earlier.

[0139] Furthermore, the present invention introduces edge computing technology to perform preliminary abnormal behavior judgment on the smart camera side. This distributed processing architecture significantly reduces the amount of data transmission and improves the real-time performance of the system. At the same time, it also enhances the scalability of the system, enabling it to better cope with the monitoring needs of large-scale campuses.

[0140] In addition, the present invention uses advanced machine learning algorithms in abnormal behavior analysis and prediction. The system can not only identify current abnormal behavior, but also predict potential security risks. This predictive analysis greatly improves the initiative and foresight of campus security management.

[0141] Finally, the present invention has built a complete set of multi-level early warning and emergency response mechanisms. The system can automatically determine the early warning level according to the severity of the abnormal event, assess the scope of impact, and generate a corresponding emergency response plan. This greatly improves the efficiency and accuracy of security incident handling.

[0142] In general, the intelligent campus safety monitoring and early warning system of the present invention realizes the intelligent, precise and active campus safety management by integrating advanced artificial intelligence technology, multi-source data fusion, edge computing and intelligent early warning mechanism. It can not only effectively improve the efficiency of discovering and handling safety hazards, but also provide campus managers with more comprehensive and in-depth safety situation analysis, thus providing strong technical support for building a harmonious and safe campus environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0143] Figure 1 The figure is an overall block diagram of the system of the present invention.

[0144] Figure 2 It is a logic block diagram of the video analysis module of the present invention.

[0145] Figure 3 It is a logic block diagram of the abnormal behavior analysis module of the present invention.

[0146] Figure 4 It is a logic block diagram of the abnormal behavior prediction module of the present invention.

[0147] Figure 5 This is a logical block diagram of the monitoring and early warning server of the present invention.

[0148] Figure 6 This is a logical block diagram of the security management platform of the present invention. DETAILED DESCRIPTION

[0149] Please refer to Figure 1-6 The present invention discloses an intelligent campus safety monitoring and early warning system and method thereof. The system comprises an image acquisition module 1, a video preprocessing module 2, a video analysis module 3, an abnormal behavior database module 4, an abnormal behavior analysis module 6, an abnormal behavior prediction module 5, an abnormal behavior feedback module 7, an alarm module 8 and a monitoring and early warning server 9. These modules work together to construct a comprehensive campus safety monitoring and early warning system.

[0150] The image acquisition module 1 is used to acquire images of the campus monitoring area and send the acquired images to the monitoring and warning server 9. Preferably, the image acquisition module 1 uses a high-definition camera with a resolution of 1920x1080 or higher to ensure image quality. In one embodiment of the present invention, the image acquisition module 1 can also acquire infrared images to enhance nighttime monitoring capabilities.

[0151] The video preprocessing module 2 is connected to the image acquisition module 1 for receiving the image sent by the image acquisition module 1, performing noise reduction filtering and color correction on the received image, and then sending the preprocessed image to the video analysis module 3. The present invention adopts Gaussian filtering for noise reduction, wherein the size of the Gaussian kernel is usually selected as 3x3 or 5x5, and the standard deviation σ can be adjusted according to the noise level, generally between 0.5 and 2. The color correction adopts a white balance algorithm to ensure the accuracy of the image color.

[0152] The video analysis module 3 is connected to the video preprocessing module 2 for receiving the preprocessed image sent by the video preprocessing module 2. The module analyzes the received image based on the monitoring model of deep learning, identifies abnormal events on campus, and determines whether to trigger an alarm. The video analysis module 3 of the present invention adopts an innovative deep learning architecture, combining a convolutional neural network (CNN) and a long short-term memory network (LSTM). Among them, CNN is used to extract the spatial features of the image, and LSTM is used to capture the temporal dependency of the video sequence. Specifically, the CNN part uses ResNet-50 as the backbone network, and the number of LSTM units is set to 256. The model training uses the Adam optimizer, the initial value of the learning rate is set to 0.001, and it decays by 10 times every 50 epochs.

[0153] The abnormal behavior database module 4 is used to store abnormal behavior detection data and abnormal behavior prediction data. This module adopts a distributed database architecture to improve the efficiency of data storage and retrieval. In one embodiment of the present invention, the abnormal behavior database module 4 uses Apache Cassandra as the underlying database and cooperates with Elasticsearch to achieve efficient full-text retrieval.

[0154] The abnormal behavior analysis module 6 is in communication with the abnormal behavior database module 4 and the video analysis module 3. The module receives the abnormal event information identified by the video analysis module 3, performs feature extraction and clustering on the received abnormal event information, identifies potential abnormal behaviors, and performs incremental learning and correction on the data input by the abnormal behavior prediction module 5. The present invention adopts an innovative multimodal fusion method in abnormal behavior analysis, combining visual features and contextual information. The feature extraction adopts a deep convolutional autoencoder, and the clustering algorithm adopts an improved DBSCAN algorithm, in which the ε parameter is adaptively adjusted and MinPts is set to 5.

[0155] The abnormal behavior prediction module 5 is connected to the abnormal behavior analysis module 6 for receiving the potential abnormal behavior information sent by the abnormal behavior analysis module 6, predicting the potential abnormal behavior, and determining whether to trigger an alarm. The abnormal behavior prediction of the present invention adopts a sequence-to-sequence (Seq2Seq) model based on the attention mechanism. Both the encoder and the decoder use bidirectional LSTM, and the hidden layer dimension is 512. The attention mechanism uses Bahdanau attention, and the attention weight is normalized by the softmax function. The prediction threshold is dynamically adjusted according to the actual application scenario, and is initially set to 0.7.

[0156] The abnormal behavior feedback module 7 is in communication with the abnormal behavior analysis module 6 and the abnormal behavior prediction module 5, and is used to receive the analysis results of the abnormal behavior analysis module 6 and the abnormal behavior prediction module 5, and form an abnormal behavior report and warning information. This module uses natural language generation (NLG) technology to convert the analysis results into an easy-to-understand text report.

[0157] The alarm module 8 is connected to the video analysis module 3, the abnormal behavior prediction module 5 and the abnormal behavior feedback module 7 for receiving the alarm triggering signal and issuing an alarm message. The alarm module 8 of the present invention supports multiple alarm modes, including sound and light alarm, SMS notification and APP push. The alarm level is divided into three levels, corresponding to mild, moderate and severe abnormal conditions, respectively, and each level adopts a different alarm strategy.

[0158] The monitoring and early warning server 9 is connected to the image acquisition module 1, the abnormal behavior analysis module 6, the abnormal behavior prediction module 5, the abnormal behavior feedback module 7 and the alarm module 8 for receiving information from each module, uniformly managing and processing abnormal alarm information, and sending early warning information to the abnormal behavior feedback module 7. The monitoring and early warning server 9 adopts a distributed architecture to ensure high availability and scalability of the system.

[0159] In a preferred embodiment of the present invention, the video analysis module 3 includes a convolutional neural network unit 31, a recurrent neural network unit 32, a fully connected layer network unit 33 and a softmax judgment unit 34. The convolutional neural network unit 31 is used to extract features of the preprocessed image. The recurrent neural network unit 32 is communicatively connected to the convolutional neural network unit 31, and is used to receive the image features extracted by the convolutional neural network unit 31, and establish a temporal relationship between consecutive video frames. The fully connected layer network unit 33 is communicatively connected to the recurrent neural network unit 32, and is used to calculate an anomaly score in combination with a feature selection algorithm. The softmax judgment unit 34 is communicatively connected to the fully connected layer network unit 33, and is used to make a final judgment based on the anomaly score, and output an abnormal event recognition result.

[0160] The system of the present invention also includes a sensor module 10, which is used to collect environmental parameters on campus, including temperature, humidity, wind speed, carbon dioxide concentration, sound, smoke concentration and dust concentration, and send the collected environmental parameters to the abnormal behavior analysis module 6. The abnormal behavior analysis module 6 is also used to receive the environmental parameters sent by the sensor module 10, and perform a comprehensive abnormal behavior analysis in combination with the environmental parameters and the video analysis results. This multi-source data fusion method significantly improves the accuracy of abnormal behavior identification. For example, when an abnormal sound is detected, if the smoke concentration suddenly increases, the system will more quickly determine that a fire may occur, thereby triggering an alarm in time.

[0161] The present invention realizes accurate identification and prediction of campus safety incidents through deep learning technology and multi-source data fusion. The modular design of the system makes it have good scalability and adaptability, and can adapt to campus environments of different sizes and types. In addition, the adaptive learning mechanism of the present invention enables the system to continuously optimize its performance, improve the recognition rate of abnormal events, and reduce the false alarm rate. This intelligent campus safety monitoring and early warning system provides a powerful tool for campus managers, which helps to timely discover and deal with potential safety hazards, thereby creating a safer learning and living environment for teachers and students. In a preferred embodiment of the present invention, the abnormal behavior analysis module 6 includes a multi-stream convolutional neural network unit 61, a long short-term memory network unit 62, a statistical analysis unit 63 and a deep uncertain clustering unit 64. These units work together to achieve comprehensive analysis and identification of abnormal behavior.

[0162] The multi-stream convolutional neural network unit 61 is used to extract multi-scale features from the received video data. The present invention adopts a three-stream architecture, corresponding to the spatial stream, the temporal stream and the optical stream. The spatial stream uses ResNe t-101 as the backbone network, the temporal stream uses a 3D convolutional network, and the optical stream is calculated using the TVL1 algorithm. This multi-stream architecture can effectively capture the spatial-temporal features in the video and improve the accuracy of abnormal behavior recognition.

[0163] The long short-term memory network unit 62 is connected to the multi-stream convolutional neural network unit 61 for modeling the video data sequence. This unit adopts a bidirectional LSTM structure, and the hidden layer dimension is set to 512. Preferably, the present invention also introduces an attention mechanism to capture important information of key time steps. The attention weight is calculated by the following formula:

[0164]

[0165] Among them, h t represents the hidden state of the current time step, h s It represents the summary representation of the sequence, and the score function adopts the additive attention mechanism.

[0166] The statistical analysis unit 63 is used to perform statistical modeling on normal behavior and abnormal behavior. This unit uses a Gaussian mixture model (GMM) to describe the normal behavior pattern, and the abnormal behavior is regarded as a sample that deviates from the normal pattern. The number of components of the GMM is automatically selected by the Bayesian Information Criterion (BIC), usually between 3 and 8. Preferably, the present invention also introduces an online learning mechanism so that the model can adapt to the dynamic changes of the behavior pattern.

[0167] The deep uncertainty clustering unit 64 is used to perform cluster analysis on potential abnormal behaviors. This unit adopts an innovative deep uncertainty clustering algorithm that combines the representation power of deep learning and the uncertainty estimation of Bayesian inference. Specifically, the present invention uses a variational autoencoder (VAE) as a feature extractor and then applies Gaussian mixture variational inference (GMVAE) in the latent space for clustering. The number of clusters K is automatically determined by the variational lower bound (ELBO), the initial value is set to 10, and it is dynamically adjusted during the training process.

[0168] The abnormal behavior analysis module 6 realizes comprehensive analysis and identification of abnormal behaviors by fusing the analysis results of the multi-stream convolutional neural network unit 61, the long short-term memory network unit 62, the statistical analysis unit 63 and the deep uncertainty clustering unit 64. The fusion strategy adopts a weighted voting mechanism, and the weights are determined by cross-validation. This multi-modal, multi-algorithm fusion method significantly improves the robustness and generalization ability of the system.

[0169] In another embodiment of the present invention, the abnormal behavior prediction module 5 includes a feature extractor 51, a Gaussian distribution model unit 52, a maximum likelihood optimization unit 53 and a prediction result output unit 54. The coordinated work of these units realizes accurate prediction of potential abnormal behaviors.

[0170] The feature extractor 51 is used to receive potential abnormal behavior information from the abnormal behavior analysis module 6 and extract the features of the potential abnormal behavior. The present invention adopts a feature extraction method based on a graph convolutional network (GCN), which can effectively capture the correlation between abnormal behaviors. The number of layers of GCN is set to 3, and the output dimensions of each layer are 64, 32 and 16 respectively.

[0171] The Gaussian distribution model unit 52 is in communication with the feature extractor 51 and is used to establish a Gaussian probability distribution model based on the extracted features. This unit uses multivariate Gaussian distribution to describe the distribution of abnormal behavior features, and its probability density function is as follows:

[0172]

[0173] Among them, x is the d-dimensional feature vector, μ is the mean vector, and Σ is the covariance matrix.

[0174] The maximum likelihood optimization unit 53 is connected to the Gaussian distribution model unit 52 for converting the abnormal behavior prediction problem into a maximum likelihood optimization problem and solving it. This unit uses the stochastic gradient descent (SGD) algorithm for optimization, the initial value of the learning rate is set to 0.01, and the cosine annealing strategy is used for adjustment. Preferably, the present invention also introduces a regularization term to prevent overfitting, and the regularization coefficient λ is determined by cross-validation, usually between 0.001 and 0.1.

[0175] The prediction result output unit 54 is connected to the maximum likelihood optimization unit 53 for outputting the abnormal behavior prediction result based on the maximum likelihood optimization result and determining whether to trigger an alarm. This unit uses a threshold method for determination, and the threshold θ is initially set to 0.8, which can be dynamically adjusted according to the actual application scenario. When the predicted probability exceeds the threshold, the system will trigger an alarm.

[0176] The system of the present invention also includes an edge computing module 11, which is in communication with the image acquisition module 1. The edge computing module 11 is used to make preliminary abnormal behavior judgments in the smart camera, and when potential abnormal behavior is detected, trigger the sending of a complete video stream to the monitoring and warning server 9. This edge computing architecture greatly reduces the amount of data transmission and improves the real-time performance of the system.

[0177] Preferably, the edge computing module 11 of the present invention uses a lightweight convolutional neural network MobileNetV2 as a feature extractor, and cooperates with a simplified version of LSTM for time series modeling. Model quantization uses 8-bit fixed-point to adapt to the computing power of edge devices. The initial abnormal behavior judgment adopts the sliding window method, with the window size set to 5 seconds and the step size of 1 second. When three consecutive windows detect anomalies, the transmission of the complete video stream will be triggered to reduce false alarms.

[0178] The intelligent campus safety monitoring and early warning system of the present invention realizes accurate identification and early warning of abnormal behaviors on campus through multi-level and multi-modal analysis and prediction methods. The modular design and edge computing architecture of the system not only improves the performance, but also enhances the scalability and adaptability of the system. This innovative technical solution provides strong support for campus safety management, helps to timely discover and prevent various safety hazards, and creates a safer and more harmonious campus environment for teachers and students. In a preferred embodiment of the present invention, the monitoring and early warning server 9 also includes a multi-level early warning unit 91, a personnel management unit 92 and an emergency response unit 93. The collaborative work of these units enables the system of the present invention to respond to various campus safety incidents more intelligently and efficiently.

[0179] The multi-level warning unit 91 is used to determine the warning level based on the output results of the abnormal behavior analysis module 6 and the abnormal behavior prediction module 5, and select the corresponding warning method according to the warning level. The present invention adopts an innovative multi-level warning mechanism, which divides the warning level into four levels: reminder, warning, serious and emergency. Each level corresponds to a different warning method and processing flow.

[0180] Preferably, the multi-level warning unit 91 of the present invention uses a fuzzy logic controller to determine the warning level. The fuzzy rule base is constructed based on expert knowledge, and the input variables include the type, duration and impact range of abnormal behavior. For example, a typical fuzzy rule may be:

[0181] IF the abnormal behavior type IS suspicious person AND the duration IS long AND the impact scope IS medium THEN the warning level IS warning.

[0182] The fuzzy reasoning uses the Mamdani method, and the defuzzification method uses the centroid method. This early warning mechanism based on fuzzy logic can better handle the uncertainty and ambiguity in actual situations and improve the accuracy and rationality of early warning.

[0183] The personnel management unit 92 is used to maintain the campus personnel information database and evaluate the scope of personnel that may be affected in combination with the abnormal behavior analysis results. This unit uses the graph database Neo4j to store and manage the personnel relationship network to achieve rapid association analysis. When abnormal behavior is detected, the system can quickly locate the personnel that may be affected and generate an impact range report.

[0184] Preferably, the personnel management unit 92 of the present invention also introduces a reasoning mechanism based on a knowledge graph. The knowledge graph contains entities such as personnel, locations, time and events, as well as various relationships between them. By reasoning on the knowledge graph, the system can more comprehensively evaluate the potential impact of abnormal events. For example, if a suspicious person is detected in a certain teaching building, the system will not only consider the people currently in the building, but also evaluate the possible impact on the people who are about to arrive in the building (based on the course schedule information).

[0185] The emergency response unit 93 is used to automatically generate an emergency response plan based on the warning level and the scope of people who may be affected, and push emergency response instructions to relevant personnel. This unit uses a case-based reasoning (CBR) method to generate an emergency response plan. The CBR system maintains a case library containing historical emergency events and their handling plans. When a new abnormal event occurs, the system will retrieve the most similar historical cases, make appropriate adjustments based on the current situation, and generate a new emergency response plan.

[0186] Preferably, the emergency response unit 93 of the present invention also integrates a reinforcement learning model for continuously optimizing the emergency response strategy. The system records the process and results of each emergency response and uses these data to train the reinforcement learning model. The model adopts a deep Q network (DQN) architecture, and the state space includes abnormal event characteristics, information on affected personnel, and current environmental conditions, and the action space includes various possible emergency response measures. The reward function design takes into account factors such as response time, impact range control, and resource utilization efficiency. Through continuous learning and optimization, the system can gradually improve the effect of emergency response.

[0187] In another embodiment of the present invention, the system further includes a security management platform 12, which is in communication with the monitoring and warning server 9. The security management platform 12 includes a mobile application 121 and a Web management interface 122, providing a comprehensive and convenient management tool for campus security managers.

[0188] The mobile application 121 is used to receive the warning information pushed by the monitoring warning server 9, display the real-time campus safety status, and provide a quick report function for abnormal events. The mobile application of the present invention is developed using the React Native framework to ensure a consistent experience on iOS and Android platforms. The application interface design follows the Material Design specification and provides an intuitive and easy-to-operate user interface.

[0189] Preferably, the mobile application 121 of the present invention integrates augmented reality (AR) technology to provide security managers with more intuitive on-site information. For example, when managers arrive at the scene of an abnormal event, they can use the AR function to view the location and coverage of surrounding cameras in real time, quickly locate possible evidence locations, and view key information such as surrounding evacuation routes.

[0190] The web management interface 122 is used to provide a visual display of campus security data, support security policy configuration and system parameter adjustment, and generate security analysis reports. The web management interface of the present invention is built using the Vue.js framework, and the backend API is implemented using the Django REST framework. Data visualization uses the ECharts library to provide a variety of chart types and interactive functions.

[0191] Preferably, the web management interface 122 of the present invention also integrates an anomaly detection algorithm based on machine learning for analyzing long-term security data trends. The system uses an LSTM autoencoder model to detect abnormal patterns in time series data. The model training uses security event data from the past year, and the input features include event type, frequency, duration, etc. By comparing reconstruction errors, the system can identify potential abnormal trends and provide managers with a basis for long-term security strategy formulation.

[0192] The present invention also includes an adaptive optimization module 13, which is in communication with the abnormal behavior analysis module 6, the abnormal behavior prediction module 5 and the monitoring and early warning server 9. The adaptive optimization module 13 is used to collect system operation data and user feedback information, perform incremental learning and parameter adjustment on the abnormal behavior analysis model based on the collected data, dynamically optimize the early warning rules and thresholds, and generate a system performance evaluation report to provide a basis for further optimization.

[0193] Preferably, the adaptive optimization module 13 of the present invention adopts a federated learning framework, so that the system can use data from multiple campuses for model optimization while protecting privacy. Each campus trains the model locally, and then sends the model update (rather than the original data) to the central server for aggregation. The aggregation algorithm adopts FedAvg, which takes into account the data quality and quantity of different campuses, ensuring fair and effective model updates.

[0194] In addition, the adaptive optimization module 13 of the present invention also introduces an active learning mechanism to improve the efficiency of data annotation. The system automatically selects the most informative samples for manual annotation, and the annotation strategy adopts a method combining uncertainty sampling and diversity sampling. This method greatly reduces the amount of annotation data required while ensuring the continuous improvement of model performance.

[0195] Finally, the present invention also proposes a method based on the above intelligent campus safety monitoring and early warning system. The method includes steps such as image acquisition, video preprocessing, video analysis, abnormal behavior analysis, abnormal behavior prediction, abnormal behavior feedback, alarm, monitoring and early warning management, and adaptive optimization. The organic combination of these steps forms a complete campus safety monitoring and early warning process.

[0196] In particular, in the abnormal behavior analysis step, the method also combines the environmental parameter data from the sensor module 10 to achieve fusion analysis of multi-source data. In the video analysis and abnormal behavior analysis steps, the method uses the edge computing module 11 to perform preliminary abnormal behavior judgment in the smart camera, greatly improving the real-time performance of the system. In the monitoring and early warning management step, the method determines the early warning level and selects the early warning method through the multi-level early warning unit 91, evaluates the range of personnel that may be affected through the personnel management unit 92, generates an emergency response plan and pushes emergency response instructions through the emergency response unit 93, and achieves comprehensive and intelligent security management.

[0197] In addition, the method also includes realizing rapid reporting of abnormal events, visualization of security data and system management functions through the mobile application 121 and Web management interface 122 of the security management platform 12, providing convenient and powerful tools for security managers.

[0198] Through the above innovative technical solutions, the intelligent campus security monitoring and early warning system and its method of the present invention realizes all-round and intelligent management of campus security. The adaptability and scalability of the system enable it to continuously optimize performance and adapt to the characteristics and needs of different campuses. This not only improves the efficiency and accuracy of campus security management, but also provides strong technical support for building a harmonious and safe campus environment.

[0199] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. Intelligent campus safety monitoring and early warning system, characterized by: include: Image acquisition module for: Collect images of the campus surveillance area; Send the collected images to the monitoring and warning server; The video preprocessing module is connected to the image acquisition module for: Receiving the image sent by the image acquisition module; Performing noise reduction filtering and color correction on the received image; Send the preprocessed image to the video analysis module; The video analysis module is connected to the video preprocessing module for: Receiving the preprocessed image sent by the video preprocessing module; The received images are analyzed based on the deep learning monitoring model; Identify unusual events on campus; Determine whether to trigger an alarm; Abnormal behavior database module, used for: Storing abnormal behavior detection data; Storing abnormal behavior prediction data; An abnormal behavior analysis module, which is in communication with the abnormal behavior database module and the video analysis module, is used to: Receiving abnormal event information identified by the video analysis module; Perform feature extraction and clustering on the received abnormal event information; Identify potentially anomalous behavior; Incremental learning and correction of the data input to the abnormal behavior prediction module; The abnormal behavior prediction module is in communication with the abnormal behavior analysis module and is used to: Receiving potential abnormal behavior information sent by the abnormal behavior analysis module; Predict potential abnormal behaviors; Determine whether to trigger an alarm; An abnormal behavior feedback module is in communication with the abnormal behavior analysis module and the abnormal behavior prediction module, and is used to: Receiving analysis results of the abnormal behavior analysis module and the abnormal behavior prediction module; Generate abnormal behavior reports and early warning information; an alarm module, which is in communication with the video analysis module, the abnormal behavior prediction module and the abnormal behavior feedback module, and is used to: receiving a signal that triggers an alarm; Send out alarm information; A monitoring and early warning server is connected in communication with the image acquisition module, the abnormal behavior analysis module, the abnormal behavior prediction module, the abnormal behavior feedback module and the alarm module, and is used to: Receive information from each module; Centralized management and processing of abnormal alarm information; Send warning information to the abnormal behavior feedback module.

2. The intelligent campus safety monitoring and early warning system according to claim 1 is characterized in that: The video analysis module comprises: Convolutional Neural Network units for: Extracting features of the preprocessed image; A recurrent neural network unit, in communication with the convolutional neural network unit, for: Receiving image features extracted by the convolutional neural network unit; Establishing the temporal relationship between consecutive video frames; A fully connected layer network unit is communicatively connected to the recurrent neural network unit and is used to: Combined with feature selection algorithm, calculate the anomaly score; A softmax judgment unit is connected to the fully connected layer network unit for: Making a final judgment based on the abnormality score; Output abnormal event recognition results.

3. The intelligent campus safety monitoring and early warning system according to claim 1 is characterized in that: Also includes: Sensor modules for: Collect environmental parameters on campus, including temperature, humidity, wind speed, carbon dioxide concentration, sound, smoke concentration, and dust concentration; Sending the collected environmental parameters to the abnormal behavior analysis module; Wherein, the abnormal behavior analysis module is also used for: Receiving environmental parameters sent by the sensor module; A comprehensive abnormal behavior analysis is performed by combining the environmental parameters and the video analysis results.

4. The intelligent campus safety monitoring and early warning system according to claim 1 is characterized in that: The abnormal behavior analysis module includes: Multi-stream convolutional neural network units for: Perform multi-scale feature extraction on the received video data; A long short-term memory network unit is communicatively connected to the multi-stream convolutional neural network unit, and is used to: Modeling video data sequences; Statistical Analysis Unit for: Statistical modeling of normal and abnormal behavior; Deep Uncertain Clustering Unit, used for: Perform cluster analysis on potential abnormal behaviors; Among them, the abnormal behavior analysis module realizes comprehensive analysis and identification of abnormal behavior by fusing the analysis results of the multi-stream convolutional neural network unit, the long short-term memory network unit, the statistical analysis unit and the deep uncertain clustering unit.

5. The intelligent campus safety monitoring and early warning system according to claim 1 is characterized in that: The abnormal behavior prediction module includes: Feature extractors for: Receiving potential abnormal behavior information from the abnormal behavior analysis module; Extract features of potential abnormal behavior; A Gaussian distribution model unit is communicatively connected to the feature extractor and is used to: Establishing a Gaussian probability distribution model based on the extracted features; A maximum likelihood optimization unit is communicatively connected to the Gaussian distribution model unit and is used to: Transform the abnormal behavior prediction problem into a maximum likelihood optimization problem; Solve the maximum likelihood optimization problem; A prediction result output unit is communicatively connected to the maximum likelihood optimization unit and is used to: Output abnormal behavior prediction results based on maximum likelihood optimization results; Determine whether to trigger an alarm.

6. The intelligent campus safety monitoring and early warning system according to claim 1 is characterized in that: Also includes: The edge computing module is connected to the image acquisition module for: Conduct preliminary abnormal behavior judgment in the smart camera; When potential abnormal behavior is detected, trigger the sending of a complete video stream to the monitoring and warning server; Among them, the edge computing module reduces the amount of data transmission and improves the real-time performance of the system by performing preliminary processing at the front end.

7. The intelligent campus safety monitoring and early warning system according to claim 1 is characterized in that: The monitoring and early warning server also includes: Multi-level early warning unit for: Determining a warning level based on output results of the abnormal behavior analysis module and the abnormal behavior prediction module; Select the corresponding warning method according to the warning level; Personnel management unit for: Maintain campus personnel information database; Combined with the abnormal behavior analysis results, assess the scope of people who may be affected; Emergency response unit for: Automatically generate emergency response plans based on the warning level and the range of people that may be affected; Push emergency response instructions to relevant personnel.

8. The intelligent campus safety monitoring and early warning system according to claim 1 is characterized in that: Also includes: The security management platform is connected to the monitoring and early warning server in communication, and includes: Mobile applications for: Receiving warning information pushed by the monitoring and warning server; Display real-time campus safety status; Provides rapid reporting of abnormal events; Web management interface for: Provide visual display of campus safety data; Support security policy configuration and system parameter adjustment; Generate a safety analysis report.

9. The intelligent campus safety monitoring and early warning system according to claim 1 is characterized in that: Also includes: The adaptive optimization module is connected to the abnormal behavior analysis module, the abnormal behavior prediction module and the monitoring and early warning server for: Collect system operation data and user feedback information; Based on the collected data, incremental learning and parameter adjustment are performed on the abnormal behavior analysis model; Dynamically optimize warning rules and thresholds; Generate system performance evaluation report to provide basis for further optimization.

10. A method based on the intelligent campus safety monitoring and early warning system according to any one of claims 1 to 9, characterized in that: The following steps are involved: S1. Collect images of the campus monitoring area through the image acquisition module; S2. The video preprocessing module performs noise reduction filtering and color correction on the captured image; S3. The video analysis module analyzes the preprocessed images based on a deep learning monitoring model to identify abnormal events on campus; S4. The abnormal behavior analysis module extracts and clusters the identified abnormal events to identify potential abnormal behaviors; S5. The abnormal behavior prediction module predicts potential abnormal behavior and determines whether to trigger an alarm; S6. The abnormal behavior feedback module generates abnormal behavior reports and warning information based on the analysis and prediction results; S7. When an alarm is required, the alarm module issues an alarm message; S8. The monitoring warning server receives information from each module, uniformly manages and processes abnormal alarm information, and sends warning information to the abnormal behavior feedback module; S9. The adaptive optimization module dynamically optimizes the models and parameters of each module based on system operation data and user feedback; Wherein, step S4 also includes: Receiving environmental parameters collected by the sensor module; Combine environmental parameters and video analysis results to conduct comprehensive abnormal behavior analysis; Steps S3 and S4 also include: The edge computing module performs preliminary abnormal behavior judgment in the smart camera, and when potential abnormal behavior is detected, triggers the sending of a complete video stream to the monitoring and early warning server; Step S8 also includes: Determine the warning level and select the warning method through the multi-level warning unit; Assess the scope of personnel that may be affected through the personnel management unit; Generate an emergency response plan and push an emergency response instruction through the emergency response unit; The method further includes: The mobile application and web management interface of the security management platform can realize rapid reporting of abnormal events, visualization of security data and system management functions.

Images