Community safety environment supervision system based on artificial intelligence

Abstract

The invention, which relates to the technical field of community safety supervision, discloses an artificial intelligence-based community safety environment supervision system comprising a data acquisition module, a data processing and analysis module, an intelligent decision module, an early warning response module and a system management module. The data acquisition module is used as a sensing layer of the system; the data processing and analysis module specifically comprises a feature extraction unit and a behavior recognition unit; the intelligent decision module is used for receiving the risk assessment result output by the data processing and analysis module; and the early warning response module generates the scheme according to the intelligent decision module. According to the community safety environment supervision system based on artificial intelligence, intelligent supervision of a community safety environment is realized through a complete closed loop of data acquisition, data processing, intelligent decision making, early warning response and system management; all the modules are in close cooperation, full-process automation from data collection to emergency response is ensured, and the efficiency and accuracy of community safety management are greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of community safety supervision, and in particular to a community safety environment supervision system based on artificial intelligence. Background Art

[0002] With the acceleration of urbanization, community safety issues have become increasingly prominent, and traditional security systems can no longer meet the needs of modern community safety management. Current community safety supervision mainly relies on fixed camera monitoring and manual patrols, which have many blind spots, slow response speeds, and high labor costs. In recent years, artificial intelligence technology has made significant progress in image recognition, behavior analysis, and anomaly detection, providing new technical support for intelligent security systems.

[0003] Existing community security systems typically rely on simple motion detection or facial recognition technologies, lacking the ability to comprehensively analyze complex scenarios. Environmental safety factors such as fire hazards and equipment failures often require separate monitoring systems, making unified management difficult. Furthermore, traditional systems offer limited early warning and response mechanisms for emergencies, failing to dynamically adjust response strategies based on the severity of the incident. With the development of the Internet of Things (IoT), various sensor devices are widely used in communities, generating large amounts of heterogeneous data, but existing systems lack the ability to effectively integrate and analyze this data. Community safety involves many factors such as public area monitoring, equipment status monitoring, personnel behavior analysis, and environmental parameter collection. It requires an integrated system that can comprehensively process various types of information and make intelligent decisions. The development of artificial intelligence technology, especially deep learning and reinforcement learning, has provided a possibility for solving this complex problem. By building an artificial intelligence-based community safety environment supervision system, real-time monitoring, intelligent analysis and rapid response to the community safety status can be achieved, greatly improving the level of community safety management.

[0004] Existing community safety supervision technology has the following major shortcomings: (1) Single data collection: mainly relying on video surveillance, lacking the collection and integration of multi-dimensional data such as environmental parameters and equipment status; (2) Limited analytical capabilities: Traditional algorithms have difficulty in handling abnormal behavior identification in complex scenarios, with high false alarm rates and poor adaptability; (3) Rigid response mechanism: The early warning and response strategies are fixed and cannot be adjusted dynamically according to the actual situation, resulting in waste of resources or insufficient response; isolated system operation: each security subsystem works independently, lacks a coordinated linkage mechanism, and is difficult to form an overall prevention and control capability; (4) Low level of intelligence: The decision-making process relies on manual experience, lacks data-based intelligent decision-making support, and has low response efficiency. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a community safety environment monitoring system based on artificial intelligence, which solves the problems raised in the above background technology.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a community safety environment monitoring system based on artificial intelligence, including a data acquisition module, a data processing and analysis module, an intelligent decision-making module, an early warning response module and a system management module; The data acquisition module serves as the system's "perception layer." The video surveillance unit continuously collects video footage of public areas through smart cameras within the community, capturing scenes of human activity and vehicle entry and exit. The environmental sensor unit monitors environmental parameters such as temperature, humidity, and smoke concentration in real time. The audio acquisition unit collects ambient sound. The mobile terminal data interface unit receives location information and emergency assistance signals from residents and managers. The collected heterogeneous data is initially cleaned and formatted by the edge computing node before being transmitted to the data processing and analysis module. The data processing and analysis module specifically includes a feature extraction unit and a behavior recognition unit. The feature extraction unit uses a deep learning model to mine features from pre-processed data. The behavior recognition unit combines information such as time and location to build a behavior analysis model to identify abnormal behaviors such as wandering and gathering. The intelligent decision-making module is used to receive the risk assessment results output by the data processing and analysis module; it specifically includes a risk assessment unit, a decision generation unit, and a strategy optimization unit. The risk assessment unit determines the event risk level in the risk matrix based on the event type and impact range factors. The decision generation unit uses the Markov decision process and combines the community resource allocation to generate the optimal response plan including response measures and personnel scheduling. The strategy optimization unit uses a reinforcement learning algorithm to iteratively optimize the decision model parameters based on historical decision results and current environmental feedback to improve decision accuracy. The early warning response module, based on the plan generated by the intelligent decision-making module, specifically includes a hierarchical early warning unit, an emergency response unit, and a linkage control unit. The hierarchical early warning unit triggers different levels of early warning according to the risk level; the emergency response unit generates a task work order, clearly defining the handling process and responsible personnel; the linkage control unit is linked with community access control, fire protection and other equipment;

[0007] The hierarchical warning units are as follows: (1) Early warning trigger mechanism: receive the risk assessment results output by the intelligent decision-making module in real time, compare them with the preset risk threshold, make a comprehensive judgment based on multi-dimensional information such as event type, occurrence time, and geographical location, and automatically match the corresponding early warning level according to the risk level, which are blue / yellow / orange / red; (2) Differentiated release strategy, based on event impact range and user permissions, accurately target affected groups and responsibility persons, and generate formatted warning information according to different warning levels, including event type, location, level and response suggestions, as follows: Blue warning: SMS, APP push; Yellow warning: SMS, APP push, community notice screen; Orange warning: SMS, APP push, broadcast system, sound and light alarm; Red warning: SMS, APP push, broadcast system, sound and light alarm, phone call; (3) Warning state tracking, record warning receiving time and confirmation status, automatically trigger secondary reminders for unconfirmed users, dynamically adjust warning levels according to event development, update release content and channels, archive warning information after disposal, and generate warning logs for subsequent analysis; The emergency response unit is as follows: (1) Task decomposition and scheduling, according to the warning information, retrieve the corresponding emergency response plan from the plan library, decompose the plan into specific task items, clarify the task target, responsible person, time node, calculate the manpower, material and technical resources required to complete each task, based on personnel skill matrix and equipment status, automatically assign tasks to the corresponding responsible person; (2) Execution and monitoring, responsible persons receive task instructions through mobile terminals, upload execution process and results, monitor the execution status of each link in real time through the task board, automatically identify task overtime risks, and support temporary adjustment and record deviations when the execution process is found to be inconsistent with the actual situation, providing decision support such as historical cases and expert knowledge base for on-site disposal personnel; (3) Effect evaluation and closed loop, set response time, disposal success rate, resource utilization rate and other evaluation indicators, automatically collect time, location, operation record data during task execution, compare preset targets with actual results, generate disposal effect evaluation report, based on evaluation results, propose process optimization suggestions, feedback to the strategy optimization unit; The linkage control unit is as follows: (1) Device linkage configuration, maintain the basic information of all linkable devices in the community, predefine device linkage trigger conditions and execution actions, support visual configuration, set device control permissions corresponding to different levels of warning, ensure safe operation; (2) Real-time linkage execution, match the corresponding linkage rules according to the warning information, convert the linkage actions into device recognizable control instructions, support multi-device parallel linkage, ensure response efficiency, real-time access to device execution status, record execution results; (3) Verification of linkage effect: After linkage is executed, check the status of related equipment to confirm whether the expected effect is achieved. If the equipment does not respond or the execution is abnormal, automatically trigger the backup linkage plan, and record the linkage process and results in detail to provide a basis for subsequent analysis; The system management module specifically includes a user authority management unit and a log recording unit. The user authority management unit controls the access and operation permissions of different personnel to the system; the log recording unit fully records the system operation data and event handling process; The user rights management unit is as follows: (1) User identity authentication: When logging in, users need to verify their password, SMS verification code, and biometrics. After passing the verification, an encrypted session token is generated for subsequent operation authentication. An automatic logout mechanism is set up upon timeout, supporting unified identity authentication across platforms and modules. (2) Dynamic allocation of permissions: Based on the RBAC model, roles such as system administrator, security supervisor, duty officer, and ordinary user are predefined, and operation permissions and resource access scopes are assigned to each role. Administrators are supported to modify role permissions or create custom permission groups in real time according to changes in user responsibilities; (3) Operation audit: capture user login time, IP address, and operation content information, regularly screen for abnormal operation patterns, generate risk reports, and provide accurate retrieval of historical operation records to provide evidence support for security incident investigations; The logging units are as follows: (1) Multi-source log collection, recording technical logs of server operation status, program errors, and resource usage, capturing user login, permission changes, and data modification operations, and collecting device operation information such as sensor data anomalies, device start and stop, and linkage execution results; (2) Log processing and storage: divide storage priorities according to log type and urgency, regularly compress historical logs, delete expired data, release storage resources, and use Elasticsearch technology to achieve efficient log storage and horizontal expansion.

[0008] Optionally, the data acquisition module is composed of a video monitoring unit, an environmental sensor unit, an audio acquisition unit and a mobile terminal data interface unit; The video surveillance unit, specifically a 360-degree panoramic camera deployed at a high point in the community, conducts large-scale scanning. Upon detecting a suspicious target, it triggers the regional camera to track it. Based on the target detection algorithm, it automatically adjusts the focus to clearly capture key information such as facial features and vehicle license plates. In low-light environments, it automatically switches to infrared imaging mode and interacts with the lighting system to increase local brightness. Environmental sensor units, specifically grid-based sensor arrays deployed in community public areas, enable real-time collection of temperature, humidity, PM2.5, and hazardous gas concentrations. When monitoring data exceeds a preset threshold, they automatically trigger nearby cameras for image verification and periodically send heartbeat packets to the master control node to detect device online status and data transmission quality. The audio acquisition unit uses a microphone array to spatially locate abnormal sounds, coordinates the nearest camera to steer toward the sound source, and performs real-time recognition of specific audio patterns, such as cries for help and breaking glass, as auxiliary judgment basis for video analysis. It also uses an adaptive filtering algorithm to eliminate background noise interference and improve voice clarity. The mobile terminal data interface unit obtains the real-time location of residents through community WiFi positioning and Bluetooth beacons, and combines it with electronic fences to determine whether they have crossed the boundary. When residents send a distress signal through the APP, their GPS coordinates are obtained synchronously and the nearest camera is triggered for image confirmation. The daily activity paths of residents are continuously recorded to provide basic data for abnormal behavior identification.

[0009] Optionally, the feature extraction unit further includes a data preprocessing unit, specifically including multimodal data cleaning and data standardization processing; Multimodal data cleaning specifically includes video data cleaning, audio data cleaning, and sensor data cleaning; Video data cleaning: extract dynamic foreground targets through frame difference method, remove static background interference, use median filtering to eliminate salt and pepper noise, improve image clarity, and locate moving targets based on edge detection and morphological operations; Audio data cleaning: Spectral subtraction is used to estimate the noise spectrum and subtract it from the original spectrum. A combination of short-time energy and zero-crossing rate is used to determine speech boundaries. Wiener filtering is used to improve the signal-to-noise ratio of the speech signal. Sensor data cleaning, eliminating high-frequency noise, smoothing data curves, identifying and eliminating abnormal data points that are significantly beyond the normal range, and using linear interpolation or spline interpolation to fill in missing data points; Data standardization processing, Min-Max scaling of features of different dimensions, unifying the numerical range to [0, 1], converting different sensor data formats into the system-unified JSON format, establishing security event classification standards, and mapping raw data to standard event categories.

[0010] Optionally, the feature extraction unit is used for behavioral feature extraction, audio feature extraction, and environmental feature extraction; Behavioral feature extraction: Calculate pixel motion between adjacent frames using the Lucas-Kanade method, predict and update the target motion trajectory using a Kalman filter, and extract bone structure features using human key point detection technology; Audio feature extraction enhances high-frequency signals and compensates for high-frequency attenuation in speech signals. The speech signal is segmented into short frames and a Hamming window is applied to reduce spectral leakage. The time-domain signal is converted to the frequency domain through a fast Fourier transform. A Mel filter bank is applied to convert the linear spectrum into a Mel spectrum. The Mel spectrum is logarithmized and subjected to a discrete cosine transform to obtain MFCC features. This also includes abnormal sound recognition. By calculating short-time energy, zero-crossing rate, and spectral entropy features, the extracted features are compared with predefined abnormal sound templates for similarity, and the final judgment is made based on the matching results of multiple feature dimensions. Environmental feature extraction: After normalizing sensor data such as temperature, humidity, and smoke concentration, a multidimensional feature vector is constructed. The sliding window technology is used to extract the temporal change characteristics of the data, such as the rate of change and trend. Based on the spatial distribution of sensors, the correlation and gradient changes of adjacent sensor data are calculated.

[0011] Optionally, the behavior recognition unit specifically includes abnormal behavior detection and event correlation analysis; Abnormal behavior detection is used to analyze trajectories and group behavior. It segments continuous trajectories according to speed and direction change points, calculating the length, duration, average speed, and other characteristics of each segment. The extracted trajectory features are matched with a predefined abnormal behavior pattern library, and abnormal behavior is determined based on the matching degree and preset thresholds. The crowd density distribution is estimated using a Gaussian mixture model, and the main direction and dispersion of group movement are calculated. Abnormal aggregation events are identified based on density thresholds and spatial clustering algorithms. Event association analysis establishes a timestamp and spatial location index for each event, uses the Apriori algorithm to mine frequently co-occurring event combinations, calculates indicators such as support, confidence, and lift to evaluate the effectiveness of rules, and establishes an evidence theory framework. The detection results of video, audio, and environmental sensors are used as independent evidence, and the basic probability distribution function of each evidence is calculated. The Dempster synthesis rule is applied to fuse multi-source evidence to obtain the final event judgment result.

[0012] Optionally, the risk assessment unit is specifically as follows: (1) Multi-dimensional data input, receiving the abnormal event feature vector output by the data processing and analysis module, including event type, occurrence time, geographical location, severity, and obtaining community basic information, environmental parameters of equipment status, calling the historical security event database, and obtaining the disposal results and impact assessment of similar events; (2) Decomposition of risk dimensions, assessment of the threat level to the life and safety of community residents, calculation of possible direct and indirect economic losses, and analysis of the impact of the incident on community order, public safety, and residents’ psychology; (3) Quantification of risk levels: For each risk dimension, events are divided into five levels based on preset thresholds, namely low, lower, medium, higher, and high. The final risk level is calculated using a weighted summation method, and the weights are dynamically adjusted based on the community safety priority. The risk level is also revised based on the real-time situation.

[0013] Optionally, the decision generation unit is specifically as follows: (1) Plan database matching: extract key features from the risk assessment results to calculate the feature similarity between the current event and each plan in the plan database, and select the plan with a similarity exceeding the threshold as the candidate plan; (2) Resource constraint assessment: obtain real-time information on available resources in the community, including the number of security personnel, equipment status, and material reserves, evaluate the feasibility of each candidate plan under the current resource constraints, and calculate the expected effect and resource consumption ratio of each plan; (3) Decision-making plan generation: parameter adjustment and combination optimization of candidate plans are performed to generate multiple alternative plans, and trade-offs are made between multiple goals such as response speed, disposal effect, and resource consumption. The best plan is selected from the alternative plans based on the decision maker's preferences and preset rules.

[0014] Optionally, the strategy optimization unit is specifically as follows: (1) Evaluation of the treatment effect: establish an evaluation index system including response time, treatment success rate, and resource utilization rate, collect real-time data and final results during the incident treatment process, calculate the actual value of each evaluation index, and compare and analyze it with the expected value; (2) Accumulation of experience and knowledge: storing successful disposal cases in the historical case database, supplementing event characteristics, disposal plans and effect evaluations, mining successful disposal rules and patterns from historical cases, converting the mined rules into reusable knowledge, and updating the decision-making knowledge base; (3) Iterate the decision-making strategy, dynamically adjust the weight parameters of the risk assessment model according to the treatment effect, optimize and update the plans in the plan library, eliminate invalid plans, modify the decision-making rules, and improve the accuracy and adaptability of decisions.

[0015] The present invention provides a community safety environment supervision system based on artificial intelligence, which has the following beneficial effects: This AI-based community safety and environmental monitoring system integrates multi-source data such as video, audio, and environmental sensors to comprehensively perceive the community's safety status and eliminate blind spots in monitoring. The various sensors in the data acquisition module work together to form complementary advantages. For example, when video surveillance detects suspicious behavior, the audio acquisition unit can be linked to perform sound analysis to improve judgment accuracy. Deep learning algorithms are used to model and analyze human behavior, and can identify abnormal behavior patterns such as wandering, gathering, and intrusion. The behavior recognition unit in the data processing and analysis module combines multi-dimensional features such as time, location, and number of people to significantly reduce the false alarm rate. The environmental assessment unit can monitor environmental risk factors such as fire hazards and equipment failures in real time. The intelligent decision-making module, based on a deep reinforcement learning algorithm, can automatically optimize response strategies based on historical data and real-time conditions. The risk assessment unit conducts hierarchical assessments of various events, the decision generation unit generates the optimal response plan, and the strategy optimization unit continuously improves decision-making quality through continuous learning. At the same time, it implements precise hierarchical warnings and initiates different response levels based on the severity of the event. The linkage control unit can be linked with community equipment such as access control, lighting, and broadcasting to achieve automated emergency response, significantly shortening response time and improving disposal efficiency. The system management module provides comprehensive user authority management and system configuration functions. The data visualization unit displays complex security data in the form of intuitive charts, allowing managers to quickly grasp the community security status and make decisions. The system has the ability to continuously learn and optimize, and can adapt to changes in different community environments and security needs. By continuously accumulating data and experience, it improves supervision efficiency and forms a virtuous cycle of intelligent security ecosystem. DETAILED DESCRIPTION

[0016] The technical solutions in the embodiments of the present invention will be described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0017] The community safety environment supervision system based on artificial intelligence includes data collection module, data processing and analysis module, intelligent decision-making module, early warning response module and system management module; The data acquisition module serves as the system's "perception layer." The video surveillance unit continuously collects video footage of public areas through smart cameras within the community, capturing scenes of human activity and vehicle entry and exit. The environmental sensor unit monitors environmental parameters such as temperature, humidity, and smoke concentration in real time. The audio acquisition unit collects ambient sound. The mobile terminal data interface unit receives location information and emergency assistance signals from residents and managers. The collected heterogeneous data is initially cleaned and formatted by the edge computing node before being transmitted to the data processing and analysis module. The data acquisition module consists of a video monitoring unit, an environmental sensor unit, an audio acquisition unit, and a mobile terminal data interface unit; The video surveillance unit, specifically a 360-degree panoramic camera deployed at a high point in the community, conducts large-scale scanning. Upon detecting a suspicious target, it triggers the regional camera to track it. Based on the target detection algorithm, it automatically adjusts the focus to clearly capture key information such as facial features and vehicle license plates. In low-light environments, it automatically switches to infrared imaging mode and interacts with the lighting system to increase local brightness. Environmental sensor units, specifically grid-based sensor arrays deployed in community public areas, enable real-time collection of temperature, humidity, PM2.5, and hazardous gas concentrations. When monitoring data exceeds a preset threshold, they automatically trigger nearby cameras for image verification and periodically send heartbeat packets to the master control node to detect device online status and data transmission quality. The audio acquisition unit uses a microphone array to spatially locate abnormal sounds, coordinates the nearest camera to steer toward the sound source, and performs real-time recognition of specific audio patterns, such as cries for help and breaking glass, as auxiliary judgment basis for video analysis. It also uses an adaptive filtering algorithm to eliminate background noise interference and improve voice clarity. The mobile terminal data interface unit obtains the real-time location of residents through community WiFi positioning and Bluetooth beacons, and combines it with electronic fences to determine whether they have crossed the boundary. When residents send a distress signal through the app, their GPS coordinates are obtained synchronously and the nearest camera is triggered for image confirmation. The unit continuously records the residents' daily activity paths and provides basic data for abnormal behavior identification. The data processing and analysis module specifically includes a feature extraction unit and a behavior recognition unit. The feature extraction unit uses a deep learning model to mine features from pre-processed data. The behavior recognition unit combines information such as time and location to build a behavior analysis model to identify abnormal behaviors such as wandering and gathering. The feature extraction unit also includes a data preprocessing unit, which specifically includes multimodal data cleaning and data standardization processing; Multimodal data cleaning specifically includes video data cleaning, audio data cleaning, and sensor data cleaning; Video data cleaning: extract dynamic foreground targets through frame difference method, remove static background interference, use median filtering to eliminate salt and pepper noise, improve image clarity, and locate moving targets based on edge detection and morphological operations; Audio data cleaning: Spectral subtraction is used to estimate the noise spectrum and subtract it from the original spectrum. A combination of short-time energy and zero-crossing rate is used to determine speech boundaries. Wiener filtering is used to improve the signal-to-noise ratio of the speech signal. Sensor data cleaning, eliminating high-frequency noise, smoothing data curves, identifying and eliminating abnormal data points that are significantly beyond the normal range, and using linear interpolation or spline interpolation to fill in missing data points; Data standardization processing, Min-Max scaling of features of different dimensions, unifying the numerical range to [0, 1], converting different sensor data formats into the system-unified JSON format, establishing security event classification standards, and mapping raw data to standard event categories.

[0018] The feature extraction unit is used for behavioral feature extraction, audio feature extraction and environmental feature extraction; Behavioral feature extraction: Calculate pixel motion between adjacent frames using the Lucas-Kanade method, predict and update the target motion trajectory using a Kalman filter, and extract bone structure features using human key point detection technology; Audio feature extraction enhances high-frequency signals and compensates for high-frequency attenuation in speech signals. The speech signal is segmented into short frames and a Hamming window is applied to reduce spectral leakage. The time-domain signal is converted to the frequency domain through a fast Fourier transform. A Mel filter bank is applied to convert the linear spectrum into a Mel spectrum. The Mel spectrum is logarithmized and subjected to a discrete cosine transform to obtain MFCC features. This also includes abnormal sound recognition. By calculating short-time energy, zero-crossing rate, and spectral entropy features, the extracted features are compared with predefined abnormal sound templates for similarity, and the final judgment is made based on the matching results of multiple feature dimensions. Environmental feature extraction: After normalizing sensor data such as temperature, humidity, and smoke concentration, a multidimensional feature vector is constructed. A sliding window technique is used to extract temporal variation characteristics of the data, such as rate of change and trend. Based on the spatial distribution of sensors, the correlation and gradient change of adjacent sensor data are calculated. The behavior recognition unit specifically includes abnormal behavior detection and event correlation analysis; Abnormal behavior detection is used to analyze trajectories and group behavior. It segments continuous trajectories according to speed and direction change points, calculating the length, duration, average speed, and other characteristics of each segment. The extracted trajectory features are matched with a predefined abnormal behavior pattern library, and abnormal behavior is determined based on the matching degree and preset thresholds. The crowd density distribution is estimated using a Gaussian mixture model, and the main direction and dispersion of group movement are calculated. Abnormal aggregation events are identified based on density thresholds and spatial clustering algorithms. Event correlation analysis: Establish a timestamp and spatial location index for each event, use the Apriori algorithm to mine frequently co-occurring event combinations, calculate indicators such as support, confidence, and lift to evaluate the effectiveness of rules, and establish an evidence theory framework. Detection results from video, audio, and environmental sensors are used as independent evidence. The basic probability distribution function of each piece of evidence is calculated, and the Dempster synthesis rule is applied to fuse multi-source evidence to obtain the final event judgment result. The intelligent decision-making module is used to receive the risk assessment results output by the data processing and analysis module; it specifically includes a risk assessment unit, a decision generation unit, and a strategy optimization unit. The risk assessment unit determines the event risk level in the risk matrix based on the event type and impact range factors. The decision generation unit uses the Markov decision process and combines the community resource allocation to generate the optimal response plan including response measures and personnel scheduling. The strategy optimization unit uses a reinforcement learning algorithm to iteratively optimize the decision model parameters based on historical decision results and current environmental feedback to improve decision accuracy. The risk assessment units are as follows: (1) Multi-dimensional data input, receiving the abnormal event feature vector output by the data processing and analysis module, including event type, occurrence time, geographical location, severity, and obtaining community basic information (building layout, population density), equipment status (firefighting facilities, monitoring points) and environmental parameters, calling the historical security event database to obtain the disposal results and impact assessment of similar events; (2) Decomposition of risk dimensions, assessment of the threat level of the incident to the life safety of community residents, calculation of possible direct and indirect economic losses, and analysis of the impact of the incident on community order, public safety, and residents’ psychology; (3) Quantification of risk levels: For each risk dimension, events are divided into five levels based on preset thresholds, namely low, lower, medium, higher, and high. The final risk level is calculated using a weighted summation method, and the weights are dynamically adjusted based on the community safety priority. At the same time, the risk level is revised based on real-time situations (such as weather conditions and holiday factors).

[0019] The decision generation unit is as follows: (1) Plan database matching: extract key features (event type, risk level, geographical location) from the risk assessment results, calculate the feature similarity between the current event and each plan in the plan database, and select the plan with a similarity exceeding the threshold as the candidate plan; (2) Resource constraint assessment: obtain real-time information on available resources in the community, including the number of security personnel, equipment status, and material reserves, evaluate the feasibility of each candidate plan under the current resource constraints, and calculate the expected effect and resource consumption ratio of each plan; (3) Decision-making plan generation: parameter adjustment and combination optimization of candidate plans are performed to generate multiple alternative plans, and trade-offs are made between multiple goals such as response speed, disposal effect, and resource consumption. The best plan is selected from the alternative plans based on the decision maker's preferences and preset rules.

[0020] The strategy optimization unit is as follows: (1) Evaluation of the treatment effect: establish an evaluation index system including response time, treatment success rate, and resource utilization rate, collect real-time data and final results during the incident treatment process, calculate the actual value of each evaluation index, and compare and analyze it with the expected value; (2) Accumulation of experience and knowledge: storing successful disposal cases in the historical case database, supplementing event characteristics, disposal plans and effect evaluations, mining successful disposal rules and patterns from historical cases, converting the mined rules into reusable knowledge, and updating the decision-making knowledge base; (3) Iterate the decision-making strategy, dynamically adjust the weight parameters of the risk assessment model according to the treatment effect, optimize and update the plans in the plan library, eliminate invalid plans, modify the decision-making rules, and improve the accuracy and adaptability of decisions; The early warning response module, based on the plan generated by the intelligent decision-making module, specifically includes a hierarchical early warning unit, an emergency response unit, and a linkage control unit. The hierarchical early warning unit triggers different levels of early warning according to the risk level; the emergency response unit generates a task work order, clearly defining the handling process and responsible personnel; the linkage control unit is linked with community access control, fire protection and other equipment; The hierarchical warning units are as follows: (1) Early warning trigger mechanism: receive the risk assessment results output by the intelligent decision-making module in real time, compare them with the preset risk threshold, make a comprehensive judgment based on multi-dimensional information such as event type, occurrence time, and geographical location, and automatically match the corresponding early warning level according to the risk level, which are blue / yellow / orange / red; (2) Differentiated release strategy: Based on the scope of the event and user permissions, accurately locate the affected people and the responsible persons, and generate formatted warning information according to different warning levels, including event type, location, level and response suggestions, as follows: Blue alert: SMS, APP push; Yellow alert: SMS, APP push, community notice screen; Orange alert: SMS, APP push, broadcast system, sound and light alarm; Red alert: SMS, APP push, broadcast system, sound and light alarm, phone notification; (3) Tracking the warning status, recording the time of warning receipt and confirmation status, automatically triggering a second reminder for unconfirmed users, dynamically adjusting the warning level according to the development of the event, synchronously updating the release content and channels, archiving the warning information after the disposal is completed, and generating a warning log for subsequent analysis;

[0021] The emergency response units are as follows: (1) Task decomposition and scheduling: retrieve the corresponding emergency response plan from the plan library based on the early warning information, decompose the plan into specific task items, clarify the task objectives, responsible persons, and time nodes, calculate the human, material, and technical resources required to complete each task, and automatically assign tasks to the corresponding responsible persons based on the personnel skill matrix and equipment status; (2) Execution and monitoring: the responsible person receives task instructions through the mobile terminal, uploads the execution process and results, monitors the execution status of each link in real time through the task dashboard, automatically identifies the risk of task timeout, and supports temporary adjustments and records deviations when it is found that the plan does not match the actual situation during the execution process, providing decision-making support such as historical cases and expert knowledge bases for on-site disposal personnel; (3) Effect evaluation and closed loop: set evaluation indicators such as response time, disposal success rate, and resource utilization rate, automatically collect time, location, and operation record data during task execution, compare preset goals with actual results, generate disposal effect evaluation reports, and propose process optimization suggestions based on the evaluation results, which are fed back to the strategy optimization unit; The linkage control unit is as follows: (1) Device linkage configuration: maintain the basic information (type, location, interface protocol) of all linked devices in the community, predefine device linkage trigger conditions and execution actions, support visual configuration, set device control permissions corresponding to different levels of warnings, and ensure safe operation; (2) Real-time linkage execution: matching the corresponding linkage rules according to the early warning information, converting the linkage actions into control instructions that can be recognized by the equipment, supporting parallel linkage of multiple equipment, ensuring response efficiency, obtaining the equipment execution status in real time, and recording the execution results; (3) Verification of linkage effect: After linkage is executed, check the status of related equipment to confirm whether the expected effect is achieved. If the equipment does not respond or the execution is abnormal, automatically trigger the backup linkage plan, and record the linkage process and results in detail to provide a basis for subsequent analysis; The system management module specifically includes a user authority management unit and a log recording unit. The user authority management unit controls the access and operation permissions of different personnel to the system; the log recording unit fully records the system operation data and event handling process; The user rights management unit is as follows: (1) User identity authentication: When logging in, users need to verify their password, SMS verification code (or dynamic token) and biometrics (such as fingerprint, face recognition). After verification, an encrypted session token is generated for subsequent operation authentication. An automatic logout mechanism is set up upon timeout, supporting unified identity authentication across platforms and modules. (2) Dynamic allocation of permissions: Based on the RBAC (role-authority control) model, roles such as system administrator, security supervisor, duty officer, and ordinary user are predefined. Each role is assigned operation permissions (such as data viewing, parameter configuration, emergency decision-making, etc.) and resource access scope (specific area data, equipment control). Administrators are supported to modify role permissions or create custom permission groups in real time according to changes in user responsibilities; (3) Operation audit: capture user login time, IP address, operation content (such as configuration modification, triggering warning) information, regularly screen abnormal operation patterns (such as high-frequency sensitive operations, unauthorized access attempts), generate risk reports, and provide accurate retrieval functions for historical operation records to provide evidence support for security incident investigations; The logging units are as follows: (1) Multi-source log collection, recording technical logs of server operation status, program errors, and resource usage, capturing user login, permission changes, and data modification operations, and collecting device operation information such as sensor data anomalies, device start and stop, and linkage execution results; (2) Log processing and storage: divide storage priorities according to log type (error, warning, normal) and urgency, regularly compress historical logs, delete expired data, release storage resources, and use Elasticsearch technology to achieve efficient log storage and horizontal expansion.

[0022] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. The community safety environment supervision system based on artificial intelligence is characterized by: It includes data acquisition module, data processing and analysis module, intelligent decision-making module, early warning response module and system management module; The data acquisition module serves as the system's "perception layer." The video surveillance unit continuously collects video footage of public areas through smart cameras within the community, capturing scenes of human activity and vehicle entry and exit. The environmental sensor unit monitors temperature, humidity, and smoke concentration in real time. The audio acquisition unit collects ambient sounds; the mobile terminal data interface unit receives the location information of residents and managers, as well as emergency help signals. The collected heterogeneous data is initially cleaned and format-converted by the edge computing node before being transmitted to the data processing and analysis module. The data processing and analysis module specifically includes a feature extraction unit and a behavior recognition unit. The feature extraction unit uses a deep learning model to mine features from pre-processed data. The behavior recognition unit combines time and location with a constructed behavior analysis model to identify abnormal wandering and gathering behaviors. The intelligent decision-making module is used to receive the risk assessment results output by the data processing and analysis module; it specifically includes a risk assessment unit, a decision generation unit, and a strategy optimization unit; The risk assessment unit determines the event risk level in the risk matrix based on the event type and impact scope factors; The decision-making unit generates an optimal response plan including response measures and personnel scheduling based on the Markov decision process and the allocation of community resources. The strategy optimization unit uses reinforcement learning algorithms to iteratively optimize decision model parameters based on historical decision results and current environmental feedback to improve decision accuracy. The early warning response module, based on the plan generated by the intelligent decision-making module, specifically includes a hierarchical early warning unit, an emergency response unit, and a linkage control unit. The hierarchical early warning unit triggers different levels of early warning according to the risk level; the emergency response unit generates a task work order, clearly defining the handling process and responsible personnel; the linkage control unit is linked with the community access control and fire protection equipment; The hierarchical warning units are as follows: (1) Early warning trigger mechanism: receive the risk assessment results output by the intelligent decision-making module in real time, compare them with the preset risk threshold, make a comprehensive judgment based on the event type, occurrence time, and geographical location information, and automatically match the corresponding early warning level according to the risk level, which are blue / yellow / orange / red; (2) Differentiated release strategy: Based on the scope of the event and user permissions, accurately locate the affected people and the responsible persons, and generate formatted warning information according to different warning levels, including event type, location, level and response suggestions, as follows: Blue alert: SMS, APP push; Yellow alert: SMS, APP push, community notice screen; Orange alert: SMS, APP push, broadcast system, sound and light alarm; Red alert: SMS, APP push, broadcast system, sound and light alarm, phone notification; (3) Tracking the warning status, recording the time of warning receipt and confirmation status, automatically triggering a second reminder for unconfirmed users, dynamically adjusting the warning level according to the development of the event, synchronously updating the release content and channels, archiving the warning information after the disposal is completed, and generating a warning log for subsequent analysis; The emergency response units are as follows: (1) Task decomposition and scheduling: retrieve the corresponding emergency response plan from the plan library based on the early warning information, decompose the plan into specific task items, clarify the task objectives, responsible persons, and time nodes, calculate the human, material, and technical resources required to complete each task, and automatically assign tasks to the corresponding responsible persons based on the personnel skill matrix and equipment status; (2) Execution and monitoring: the responsible person receives task instructions through the mobile terminal, uploads the execution process and results, monitors the execution status of each link in real time through the task dashboard, automatically identifies the risk of task timeout, and supports temporary adjustments and records deviations when it is found that the plan does not match the actual situation during the execution process, providing decision-making support for on-site disposal personnel based on historical cases and expert knowledge base; (3) Effect evaluation and closed loop: set evaluation indicators such as response time, disposal success rate, and resource utilization rate, automatically collect time, location, and operation record data during task execution, compare preset targets with actual results, generate disposal effect evaluation reports, and propose process optimization suggestions based on the evaluation results, which are fed back to the strategy optimization unit; The linkage control unit is as follows: (1) Device linkage configuration: maintain the basic information of all linked devices in the community, predefine device linkage trigger conditions and execution actions, support visual configuration, set device control permissions corresponding to different levels of warnings, and ensure safe operation; (2) Real-time linkage execution: matching the corresponding linkage rules according to the early warning information, converting the linkage actions into control instructions that can be recognized by the equipment, supporting parallel linkage of multiple equipment, ensuring response efficiency, obtaining the equipment execution status in real time, and recording the execution results; (3) Verification of linkage effect: After linkage is executed, check the status of related equipment to confirm whether the expected effect is achieved. If the equipment does not respond or the execution is abnormal, automatically trigger the backup linkage plan, and record the linkage process and results in detail to provide a basis for subsequent analysis; The system management module specifically includes a user authority management unit and a log recording unit. The user authority management unit controls the access and operation permissions of different personnel to the system; the log recording unit fully records the system operation data and event handling process; The user rights management unit is as follows: (1) User identity authentication: When logging in, users need to verify their password, SMS verification code, and biometrics. After passing the verification, an encrypted session token is generated for subsequent operation authentication. An automatic logout mechanism is set up upon timeout, supporting unified identity authentication across platforms and modules. (2) Dynamic allocation of permissions: Based on the RBAC model, system administrators, security supervisors, on-duty personnel, and ordinary users are predefined, and operation permissions and resource access scopes are assigned to each role. Administrators are supported to modify role permissions or create custom permission groups in real time based on changes in user responsibilities. (3) Operation audit: capture user login time, IP address, and operation content information, regularly screen for abnormal operation patterns, generate risk reports, and provide accurate retrieval of historical operation records to provide evidence support for security incident investigations; The logging units are as follows: (1) Multi-source log collection, recording technical logs of server operation status, program errors, and resource usage, capturing user login, permission changes, and data modification operations, and collecting device operation information such as sensor data anomalies, device start and stop, and linkage execution results; (2) Log processing and storage: divide storage priorities according to log type and urgency, regularly compress historical logs, delete expired data, release storage resources, and use Elasticsearch technology to achieve efficient log storage and horizontal expansion.

2. The artificial intelligence-based community safety environment monitoring system according to claim 1 is characterized by: The data acquisition module consists of a video monitoring unit, an environmental sensor unit, an audio acquisition unit and a mobile terminal data interface unit; The video surveillance unit, specifically a 360-degree panoramic camera deployed at a high point in the community, conducts large-scale scanning. Upon detecting a suspicious target, it triggers the regional camera to track it. Based on the target detection algorithm, it automatically adjusts the focus to clearly capture key information such as facial features and vehicle license plates. In low-light environments, it automatically switches to infrared imaging mode and interacts with the lighting system to increase local brightness. Environmental sensor units, specifically grid-based sensor arrays deployed in community public areas, enable real-time collection of temperature, humidity, PM2.5, and hazardous gas concentrations. When monitoring data exceeds a preset threshold, they automatically trigger nearby cameras for image verification and periodically send heartbeat packets to the master control node to detect device online status and data transmission quality. The audio acquisition unit uses a microphone array to spatially locate abnormal sounds, coordinates the nearest camera to steer toward the sound source, and performs real-time recognition of specific audio patterns, such as cries for help and breaking glass, as auxiliary judgment basis for video analysis. It also uses an adaptive filtering algorithm to eliminate background noise interference and improve voice clarity. The mobile terminal data interface unit obtains the real-time location of residents through community WiFi positioning and Bluetooth beacons, and combines it with electronic fences to determine whether they have crossed the boundary. When residents send a distress signal through the APP, their GPS coordinates are obtained synchronously and the nearest camera is triggered for image confirmation. The daily activity paths of residents are continuously recorded to provide basic data for abnormal behavior identification.

3. The artificial intelligence-based community safety environment monitoring system according to claim 1 is characterized by: The feature extraction unit also includes a data preprocessing unit, which specifically includes multimodal data cleaning and data standardization processing; Multimodal data cleaning specifically includes video data cleaning, audio data cleaning, and sensor data cleaning; Video data cleaning: extract dynamic foreground targets through frame difference method, remove static background interference, use median filtering to eliminate salt and pepper noise, improve image clarity, and locate moving targets based on edge detection and morphological operations; Audio data cleaning: Spectral subtraction is used to estimate the noise spectrum and subtract it from the original spectrum. A combination of short-time energy and zero-crossing rate is used to determine speech boundaries. Wiener filtering is used to improve the signal-to-noise ratio of the speech signal. Sensor data cleaning, eliminating high-frequency noise, smoothing data curves, identifying and eliminating abnormal data points that are significantly beyond the normal range, and using linear interpolation or spline interpolation to fill in missing data points; Data standardization processing, Min-Max scaling of features of different dimensions, unifying the numerical range to [0, 1], converting different sensor data formats into the system-unified JSON format, establishing security event classification standards, and mapping raw data to standard event categories.

4. The artificial intelligence-based community safety environment monitoring system according to claim 1 is characterized by: The feature extraction unit is used for extracting behavioral features, audio features and environmental features; Behavioral feature extraction: Calculate pixel motion between adjacent frames using the Lucas-Kanade method, predict and update the target motion trajectory using a Kalman filter, and extract bone structure features using human key point detection technology; Audio feature extraction: enhance high-frequency signals and compensate for high-frequency attenuation of speech signals. Speech signals are segmented into short frames and Hamming windows are applied to reduce spectral leakage. Time-domain signals are converted to frequency-domain representations using fast Fourier transforms. Mel filter banks are applied to convert linear spectra to Mel spectra. The logarithm of the Mel spectra is taken and discrete cosine transforms are performed to obtain MFCC features. It also includes abnormal sound recognition. By calculating short-time energy, zero-crossing rate, and spectral entropy features, the extracted features are compared with predefined abnormal sound templates for similarity, and the final judgment is made by integrating the matching results of multiple feature dimensions. Environmental feature extraction: After normalizing the temperature, humidity, and smoke concentration data, a multidimensional feature vector is constructed. The sliding window technology is used to extract the temporal change characteristics of the data, such as the rate of change and trend. Based on the spatial distribution of sensors, the correlation and gradient changes of adjacent sensor data are calculated.

5. The artificial intelligence-based community safety environment monitoring system according to claim 1 is characterized by: The behavior recognition unit specifically includes abnormal behavior detection and event correlation analysis; Abnormal behavior detection is used to analyze trajectories and group behaviors. Continuous trajectories are segmented according to speed and direction change points, and the length, duration, and average speed of each segment are calculated. The extracted trajectory features are matched with a predefined abnormal behavior pattern library, and abnormal behavior is determined based on the matching degree and preset thresholds. The crowd density distribution is estimated through the Gaussian mixture model, the main direction and dispersion of group movement are calculated, and abnormal gathering events are identified based on density thresholds and spatial clustering algorithms; Event association analysis establishes a timestamp and spatial location index for each event, uses the Apriori algorithm to mine frequently co-occurring event combinations, calculates support, confidence, and lift indicators to evaluate the effectiveness of rules, and establishes an evidence theory framework. The detection results of video, audio, and environmental sensors are used as independent evidence, and the basic probability distribution function of each evidence is calculated. The Dempster synthesis rule is applied to fuse multi-source evidence to obtain the final event judgment result.

6. The artificial intelligence-based community safety environment monitoring system according to claim 1 is characterized by: The risk assessment unit is as follows: (1) Multi-dimensional data input, receiving the abnormal event feature vector output by the data processing and analysis module, including event type, occurrence time, geographical location, severity, and obtaining community basic information, environmental parameters of equipment status, calling the historical security event database, and obtaining the disposal results and impact assessment of similar events; (2) Decomposition of risk dimensions, assessment of the threat level of the incident to the life safety of community residents, calculation of direct and indirect economic losses, and analysis of the impact of the incident on community order, public safety, and residents’ psychology; (3) Quantification of risk levels: For each risk dimension, events are divided into five levels based on preset thresholds, namely low, lower, medium, higher, and high. The final risk level is calculated using a weighted summation method, and the weights are dynamically adjusted based on the community safety priority. The risk level is also revised based on the real-time situation.

7. The artificial intelligence-based community safety environment monitoring system according to claim 1 is characterized by: The decision generation unit is specifically as follows: (1) Plan database matching: extract key features from the risk assessment results to calculate the feature similarity between the current event and each plan in the plan database, and select the plan with a similarity exceeding the threshold as the candidate plan; (2) Resource constraint assessment: obtain real-time information on available resources in the community, including the number of security personnel, equipment status, and material reserves, evaluate the feasibility of each candidate plan under the current resource constraints, and calculate the expected effect and resource consumption ratio of each plan; (3) Decision-making plan generation: parameter adjustment and combination optimization of candidate plans are performed to generate multiple alternative plans, and trade-offs are made between multiple goals such as response speed, disposal effect, and resource consumption. The best plan is selected from the alternative plans based on the decision maker's preferences and preset rules.

8. The artificial intelligence-based community safety environment monitoring system according to claim 1 is characterized by: The strategy optimization unit is specifically as follows: (1) Evaluation of the treatment effect: establish an evaluation index system including response time, treatment success rate, and resource utilization rate, collect real-time data and final results during the incident treatment process, calculate the actual value of each evaluation index, and compare and analyze it with the expected value; (2) Accumulation of experience and knowledge: storing successful disposal cases in the historical case database, supplementing event characteristics, disposal plans and effect evaluations, mining successful disposal rules and patterns from historical cases, converting the mined rules into reusable knowledge, and updating the decision-making knowledge base; (3) Iterate the decision-making strategy, dynamically adjust the weight parameters of the risk assessment model according to the treatment effect, optimize and update the plans in the plan library, eliminate invalid plans, modify the decision-making rules, and improve the accuracy and adaptability of decisions.