Behavior monitoring and early warning system for tourists in zoo

Abstract

The invention relates to the technical field of intelligent monitoring, in particular to a behavior monitoring and early warning system for tourists in a zoo, which comprises a data acquisition layer, an edge calculation layer, a central processing layer and an application service layer. Through the multi-source data fusion and deep learning technology, the purposes of park dead-corner-free monitoring, multi-dimensional perception, accurate abnormal behavior recognition and accurate behavior context understanding can be achieved, and then the coverage and accuracy of safety management are greatly improved; a multi-level early warning mechanism from prompt to emergency can be provided, early warning information can be accurately distributed, differentiated intervention suggestions can be provided according to specific conditions, the intervention effect can be continuously evaluated, and accurate and effective safety management can be realized; the method can predict security requirements, dynamically adjust resource configuration, optimize patrol paths and recommend proper emergency plans, and can enable limited security resources to play the maximum utility, thereby improving the safety management efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent monitoring, and particularly relates to a behavior monitoring and early warning system for tourists in a zoo. BACKGROUND

[0002] The technical field of intelligent monitoring aims to use information technology, communication technology and intelligent algorithms to monitor, analyze and manage various environments and scenes in real time, detect, identify and prevent security problems by using video monitoring systems, sensor networks, data analysis and artificial intelligence algorithms, improve monitoring efficiency and accuracy by intelligently analyzing collected data, and apply to public security, traffic management, home security and other fields to predict and warn abnormal events and provide support for timely response and management.

[0003] Traditional scenic spot monitoring technology has not achieved the effect of combining monitoring and timely alarm. At present, in the face of high-density crowd, tourists' high-risk abnormal behavior and complex environment, it lacks intelligent analysis and risk assessment capability, cannot respond to differentiated early warning solutions in time, causes delay in problem handling and unreasonable allocation of resources, and thus reduces the safety management efficiency.

[0004] Therefore, a behavior monitoring and early warning system for tourists in a zoo is provided. SUMMARY

[0005] Based on this, it is necessary to provide a behavior monitoring and early warning system for tourists in a zoo in view of the above technical problems.

[0006] The application provides a behavior monitoring and early warning system for visitors in a zoo, comprising: a data acquisition layer for acquiring various channel acquisition data; wherein the various channel acquisition data comprises video data, passenger flow data, environmental data, ticketing data and mobile application data; an edge computing layer for preprocessing, image processing and target detection processing and data fusion processing of the acquired various channel acquisition data to obtain preliminary processing data; a central processing layer for using deep learning technology to perform visitor behavior identification and analysis processing on the preliminary processing data based on predefined visitor behavior categories to obtain visitor behavior identification results, analyze the categories and quantities of abnormal behaviors and generate visitor abnormal behavior analysis results; based on a visitor density calculation algorithm, the current visitor density data of each functional area in the scenic area is calculated; based on the visitor abnormal behavior analysis results and the current visitor density data of each functional area in the scenic area, combined with environmental data and current time data, a current safety risk assessment score is generated; and based on the current safety risk assessment score, combined with the various channel acquisition data, safety situation awareness data is generated; an application service layer for displaying passenger flow data, current visitor density data of each functional area in the scenic area and current safety risk assessment score, responding to visitor abnormal behavior analysis results or current safety risk assessment score, triggering corresponding early warning information and publishing differentiated intervention recommendation measure information, and providing historical data analysis, trend prediction and decision support.

[0007] Optionally, the data acquisition layer comprises: a video data acquisition module for installing video monitoring devices covering each functional area in the park and acquiring video data through the video monitoring devices; a passenger flow data acquisition module for installing passenger flow counting devices at park entrances, exits and passages connecting different functional areas, and installing Wi-Fi probes and Bluetooth beacons at passages connecting different functional areas to obtain passenger flow data including the number and time distribution information of visitors entering the park, the flow and residence time information of visitors in each functional area, the visitor distribution information and visitor movement path information of the park; an environmental data acquisition module for acquiring environmental data in a predefined time window, wherein the environmental data comprises meteorological data, noise decibel data, illumination intensity data and air quality index data; a ticketing system interface module for interfacing with the park ticketing system to acquire ticketing data; and a mobile application data acquisition module for collecting mobile application data including visitor location data and visitor behavior data through the official APP of the park under the condition of user authorization, and the mobile application data is obtained after data privacy protection processing.

[0008] Optionally, the edge computing layer comprises: a data preprocessing module, configured to clean and standardize the acquired passenger flow data, environment data, ticket data and mobile application data to obtain preprocessed data; an edge computing device, installed beside the video monitoring device, configured to use target detection and tracking technology to perform image processing and target detection processing on the acquired video data to obtain tourist detection and tracking data, the tourist detection and tracking data comprising tourist position data and tourist quantity data in the video data, tourist moving track data in the video sequence, group tourist data and family unit data in the video data, and target focus crowd data in the video data; and a local analysis unit, configured to perform data fusion processing on the preprocessed data and the tourist detection and tracking data to obtain preliminary processed data.

[0009] Optionally, the central processing layer comprises: a central server, configured to use deep learning technology to perform tourist behavior recognition and analysis processing on the preliminary processed data based on a pre-defined tourist behavior category to obtain a tourist behavior recognition result, wherein the tourist behavior category comprises normal behavior, abnormal behavior, emergency behavior and special behavior; a tourist density calculation algorithm, configured to calculate current tourist density data of each functional area in the scenic spot; based on the tourist abnormal behavior analysis result and the current tourist density data of each functional area in the scenic spot, combined with environment data and current time data, a current safety risk assessment score is generated; an analysis engine, configured to generate security situation awareness data according to the current safety risk assessment score combined with multi-channel collected data; and a data storage module, configured to store historical multi-channel collected data, historical tourist behavior recognition results, historical safety risk assessment scores and historical early warning information.

[0010] Optionally, the tourist density calculation algorithm comprises a regional density calculation algorithm and a weighted density calculation algorithm, wherein if the uniform distribution of tourists in each functional area is considered, the regional density calculation algorithm is used to calculate the average density of tourists in the functional area, the formula of the regional density calculation algorithm is: ; in the formula, D represents the average density of tourists, N represents the number of tourists in the functional area, and A represents the area of the functional area; if the uneven distribution of tourists in each functional area is considered, the weighted density calculation algorithm is used to calculate the weighted density of tourists in all sub-regions of the functional area, the formula of the weighted density calculation algorithm is: ; in the formula, , wherein D represents the weighted density of tourists, , wherein w d represents the weight of the d th sub-region, , wherein N d represents the number of tourists in the d th sub-region, and m represents the number of sub-regions in the functional area; the calculated average density of tourists or weighted density of tourists is used as the current tourist density data.

[0011] Optionally, the current security risk assessment score is generated based on the tourist abnormal behavior analysis result and the current tourist density data of each functional area in the scenic area, combined with the environmental data and the current time data, including: based on the tourist abnormal behavior analysis result and the current tourist density data of each functional area in the scenic area, combined with the environmental data and the current time data, using a risk assessment algorithm to calculate the current security risk assessment score; the formula of the risk assessment algorithm is: ; wherein, represents the current security risk assessment score, represents the risk score based on the density factor, represents the risk score based on the abnormal behavior, represents the risk score based on the environmental factor, represents the risk score based on the time factor, respectively represent the weight coefficients based on the density factor, the abnormal behavior, the environmental factor and the time factor; wherein, the formula of the risk score based on the density factor is: ; wherein, represents the current tourist density data; represents the security density threshold; vj represents the risk growth parameter, and the value range is 1.5-2.5; the formula of the risk score based on the abnormal behavior is: ; wherein, represents the risk score based on the abnormal behavior, represents the severity score of the i-th abnormal behavior, represents the weight of the i-th abnormal behavior, and n represents the number of detected abnormal behaviors; the risk score based on the environmental factor and the risk score based on the time factor are obtained by mapping the pre-defined environmental factor impact on risk value and the time factor impact on risk value respectively.

[0012] Optionally, the application service layer includes: a real-time monitoring interface module for displaying passenger flow data, current tourist density data of each functional area in the scenic area and current security risk assessment score; an early warning management module for triggering corresponding early warning information in response to the tourist abnormal behavior analysis result or the current security risk assessment score; an intervention control interface module for releasing differentiated intervention suggestion measure information by the administrator for the triggered early warning information; a data analysis platform for statistical and trend prediction analysis on the stored historical multi-channel collected data, historical tourist behavior recognition result, historical security risk assessment score and historical early warning information to obtain an analysis result; based on the analysis result, dynamically adjusting resource allocation, optimizing the patrol path, and generating a visual emergency plan report.

[0013] Optionally, the triggering corresponding early warning information in response to the tourist abnormal behavior analysis result or the current security risk assessment score comprises: in response to the tourist abnormal behavior analysis result, when it is detected that the tourist abnormal behavior analysis result exceeds a pre-defined corresponding abnormal behavior type threshold value, triggering corresponding level early warning information based on a multi-level early warning mechanism based on anomaly detection; in response to the current security risk assessment score, when it is detected that the current security risk assessment score exceeds a pre-defined early warning threshold value, triggering corresponding level early warning information based on a multi-level early warning mechanism based on a threshold value; wherein the multi-level early warning mechanism based on anomaly detection and the multi-level early warning mechanism based on a threshold value both comprise a four-level early warning mechanism, respectively a prompt level, a warning level, an alarm level and an emergency level, wherein the prompt level adopts a blue indicator light for early warning, indicating a slight abnormality, needing attention but not intervening for the time being, the warning level adopts a yellow indicator light for early warning, indicating an obvious abnormality, suggesting taking preventive measures, the alarm level adopts an orange indicator light for early warning, indicating a serious abnormality, needing immediate intervention, and the emergency level adopts a red indicator light for early warning, indicating a critical situation, needing to start an emergency plan.

[0014] Optionally, the triggering condition of the multi-level early warning mechanism based on a threshold value adopts a early warning triggering algorithm to realize the triggering process, and a formula of the early warning triggering algorithm is: ; in the formula, W represents the four-level early warning mechanism, and takes a value range of 0, 1, 2 and 3, respectively corresponding to the prompt level, the warning level, the alarm level and the emergency level, represents the current security risk assessment score, respectively represent three pre-defined early warning threshold values from low to high, for distinguishing and triggering corresponding early warning mechanisms; the early warning threshold values are obtained by dynamic adjustment using a dynamic adjustment algorithm, and a formula of the dynamic adjustment algorithm is: ; in the formula, represents the i-th level early warning threshold value at t time, represents a basic threshold value, and the basic threshold value is a pre-defined early warning threshold value, represents the current security risk assessment score, represents a security risk assessment score of a reference time period, represents an adjustment coefficient, and takes a value range of 0.3-0.7.

[0015] Optionally, the dynamic adjustment of the resource configuration is realized by optimizing the number of security personnel in the functional area, and the optimized patrol path is realized by calculating an optimal patrol path using a weighted graph algorithm.

[0016] The application provides a behavior monitoring and early warning system for visitors in a zoo, which has the following beneficial effects: 1. Comprehensive monitoring and accurate identification: through multi-source data fusion and deep learning technology, the purpose of realizing park no-dead-angle monitoring, multi-dimensional perception, accurate identification of abnormal behavior and accurate understanding of behavior context can be achieved, thereby greatly improving the coverage and accuracy of safety management; 2. Intelligent risk assessment and prediction: real-time park safety risk calculation is performed by comprehensively considering various factors, future safety situation is predicted based on historical data, and the evaluation model is continuously optimized through adaptive learning, which can change safety management from passive response to active prevention; 3. Differentiated early warning and intervention: a multi-level early warning mechanism from prompt to emergency is provided, early warning information is accurately distributed, differentiated intervention suggestions are provided according to specific circumstances, and intervention effect is continuously evaluated, realizing accurate and effective safety management; 4. Specific high-risk behavior identification: special identification algorithms are designed for high-risk behaviors such as feeding and invading safety areas, high-priority processing is given, a rapid response mechanism is started, and targeted education and guidance are provided, which can effectively reduce the incidence of safety incidents; 5. Resource optimization configuration: security demand is predicted, resource configuration is dynamically adjusted, patrol path is optimized, and suitable emergency plan is recommended, which can make the limited security resources play the maximum effect, thereby improving the safety management efficiency; 6. Data-driven decision making: safety trend analysis, risk correlation discovery, situation visualization display and data-based decision making suggestions are provided, so that safety management is changed from experience-driven to data-driven, further improving the scientificity of management decision; 7. Balance of visitor experience: non-intrusive monitoring technology is adopted, targeted and accurate intervention is implemented, education and guidance are mainly implemented, information transparency is maintained, dynamic monitoring and efficient management of scenic spot risks are realized, and visitor experience and safety guarantee are improved. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 It is a system principle diagram of the application.

[0018] Figure 2 It is a system principle diagram of the data acquisition layer of the application.

[0019] Figure 3 It is a system principle diagram of the edge computing layer of the application.

[0020] Figure 4 It is a system principle diagram of the central processing layer of the application.

[0021] Figure 5 It is a system principle diagram of the application service layer of the application. DETAILED DESCRIPTION

[0022] In order to make the objects, technical solutions and advantages of the present application clearer, further detailed description will be made to the present application with reference to the specific embodiments and the accompanying drawings. It should be understood that the specific embodiments described herein are only used to explain the present application, and should not be used to limit the present application.

[0023] Reference is made to the accompanying drawings Figures 1-5 A behavior monitoring and early warning system 100 for visitors in a zoo, the system comprising a data acquisition layer 110, an edge computing layer 120, a central processing layer 130 and an application service layer 140.

[0024] In some embodiments, the data acquisition layer 110 is configured to acquire a plurality of channel data; wherein the plurality of channel data comprises video data, passenger flow data, environmental data, ticketing data and mobile application data.

[0025] Further, reference is made to the accompanying drawings Figure 2 The data acquisition layer 110 comprises a video data acquisition module 1110 configured to install video monitoring devices covering each functional area in the park, and acquire video data through the video monitoring devices; a passenger flow data acquisition module 1120 configured to install passenger flow counting devices at the entrances and exits of the park and the passages connecting different functional areas, and install Wi-Fi probes and Bluetooth beacons at the passages connecting different functional areas, to obtain passenger flow data including the number and time distribution information of visitors entering the park, the flow and residence time information of visitors in each functional area, the distribution information and movement path information of visitors in the park; an environmental data acquisition module 1130 configured to acquire environmental data in a predefined time window, wherein the environmental data comprises meteorological data (including but not limited to temperature data, humidity data, wind speed data, precipitation data, etc.), noise decibel data, illumination intensity data and air quality index data; a ticketing system interface module 1140 configured to interface with the ticketing system of the park to acquire ticketing data; and a mobile application data acquisition module 1150 configured to collect mobile application data including visitor location data and visitor behavior data through the official APP of the park under the condition that the user authorizes, and the mobile application data is obtained after data privacy protection processing.

[0026] Further, the video monitoring devices include, but are not limited to, high-definition cameras, thermal imaging cameras, panoramic cameras, etc.; wherein the high-definition cameras are used to collect video data of the activities of tourists, can support 1080p or higher resolution, and can be flexibly selected according to actual needs, which are not limited here; the thermal imaging cameras are used to provide thermal imaging monitoring in areas with insufficient light, so as to ensure all-weather monitoring capability; the panoramic cameras can be deployed in key areas to provide 360-degree dead-angle-free monitoring, and the key areas include, but are not limited to, functional areas such as the sightseeing area, the animal exhibition area, and the rest area. In addition, the collection strategies for the video data include: 1. Area coverage: ensuring that there is no monitoring blind area in the key areas; 2. Resolution adjustment: adjusting the video resolution according to the importance of the functional areas; 3. Frame rate optimization: adjusting the frame rate according to the scene characteristics to balance the monitoring effect and storage demand; 4. View angle optimization: optimizing the installation angle of the video monitoring device to maximize the monitoring effect.

[0027] Further, the data sources of the visitor flow data include: installing visitor flow counting devices (such as infrared counters) at the entrances and exits of the park and the passages connecting different functional areas to obtain the number and time distribution information of the visitors entering the park and the visitor flow and residence time information in each functional area, installing Wi-Fi probes at the passages connecting different functional areas to obtain the visitor distribution information of the park, and installing Bluetooth beacons at the passages connecting different functional areas to obtain the visitor moving path information.

[0028] Further, the defined time window in the environmental data collection module 1130 can be set for regular monitoring and abnormal situations respectively, for regular monitoring, the collection can be performed once every 5-15 minutes, and for abnormal situation monitoring, the collection frequency can be increased when abnormal data is detected, such as once every 1-2 minutes.

[0029] Further, the ticket data obtained through the ticket system interface module 1140 includes, but is not limited to, ticket type distribution, group visitor information, and target special population (elderly, children) proportion data.

[0030] Further, the mobile application data obtained through the mobile application data collection module 1150 includes, but is not limited to, user location, query content, and residence time information, which needs to be obtained under the authorization of the user and after data privacy protection processing, and the data privacy protection processing includes: 1. Data desensitization: desensitizing personal identity information; 2. Authorized collection: clearly informing the user of the data collection purpose and obtaining authorization; 3. Data encryption: using encryption technology to protect data transmission and storage; 4. Access control: strictly controlling data access permissions.

[0031] In some embodiments, the edge computing layer 120 is configured to preprocess, image process and target detect the collected multi-channel data, and perform data fusion processing to obtain preliminary processed data.

[0032] Further, with reference to the accompanying drawings Figure 3 The edge computing layer 120 includes a data preprocessing module 1210 configured to clean and standardize the collected passenger flow data, environmental data, ticketing data and mobile application data to obtain preprocessed data; an edge computing device 1220 installed beside the video monitoring device and configured to perform image processing and target detection processing on the collected video data by using target detection and tracking technology to obtain tourist detection and tracking data, the tourist detection and tracking data including tourist position data and tourist quantity data in the video data, tourist moving track data in the video sequence, group tourist data and family unit data in the video data, and target focus crowd data in the video data; and a local analysis unit 1230 configured to perform data fusion processing on the preprocessed data and the tourist detection and tracking data to obtain preliminary processed data.

[0033] Further, by setting the data preprocessing module 1210, the collected passenger flow data, environmental data, ticketing data and mobile application data can be cleaned and standardized to obtain preprocessed data.

[0034] Further, by setting the edge computing device 1220 and installing it beside the video monitoring device, the collected video data can be image processed and target detected, and the image processing and target detection processing can employ technologies including but not limited to target detection technology (such as improved YOLOv5 or Faster R-CNN model), target tracking technology (such as DeepSORT or ByteTrack algorithm), feature extraction technology (extracting appearance features for cross-video monitoring device feature tracking) and ID management technology (allocating unique IDs for tracked targets to handle ID switching and recovery) to achieve the purpose of tourist detection and tracking. The specific detection and tracking content includes personnel detection (for identifying tourist positions and quantities in the video), multi-target tracking (for tracking tourist moving tracks in the video sequence), group analysis (for identifying group tourists and family units) and special crowd identification (for identifying children, the elderly and other people who need special attention), and thus the tourist detection and tracking data including tourist position data and tourist quantity data in the video data, tourist moving track data in the video sequence, group tourist data and family unit data in the video data, and target focus crowd data in the video data.

[0035] Further, by setting the local analysis unit 1230, data fusion processing can be realized on the pre-processed data and the tourist detection and tracking data, so as to reduce the data running burden of the central server 1310, wherein the data fusion processing includes but is not limited to steps such as time-space alignment, multi-modal fusion, data complement, conflict resolution, etc.; wherein the time-space alignment is used to align the data from different sources according to time and space; the multi-modal fusion is used to fuse multi-modal data such as video, audio, environment, etc.; the data complement is used to utilize the complementary advantages of different data sources to improve the analysis accuracy; the conflict resolution is used to solve the conflict of different data sources by using a weighted fusion strategy, and the specific data fusion algorithm includes but is not limited to Kalman filter (used for fusion and smoothing of time series data), Dempster-Shafer evidence theory (used for processing uncertain data fusion), Bayesian network (used for constructing probability dependency relationship between data), deep learning method (using deep neural network for feature-level fusion).

[0036] In some embodiments, the central processing layer 130 is configured to: adopt a deep learning technology, based on a pre-defined tourist behavior category, to perform tourist behavior recognition and analysis processing on the preliminary processed data, to obtain a tourist behavior recognition result, and to analyze the category and quantity of abnormal behaviors, to generate a tourist abnormal behavior analysis result; based on a tourist density calculation algorithm, to calculate the current tourist density data of each functional area in the scenic area; based on the tourist abnormal behavior analysis result and the current tourist density data of each functional area in the scenic area, in combination with environmental data and current time data, to generate a current safety risk assessment score; and based on the current safety risk assessment score, in combination with multi-channel collected data, to generate safety situation awareness data.

[0037] Further, referring to the accompanying drawings Figure 4 The central processing layer 130 includes: a central server 1310 configured to adopt a deep learning technology, based on a pre-defined tourist behavior category, to perform tourist behavior recognition and analysis processing on the preliminary processed data, to obtain a tourist behavior recognition result, wherein the tourist behavior category includes normal behavior, abnormal behavior, emergency behavior and special behavior; based on a tourist density calculation algorithm, to calculate the current tourist density data of each functional area in the scenic area; based on the tourist abnormal behavior analysis result and the current tourist density data of each functional area in the scenic area, in combination with environmental data and current time data, to generate a current safety risk assessment score; an analysis engine 1320 configured to, according to the current safety risk assessment score, in combination with multi-channel collected data, to generate safety situation awareness data; and a data storage module 1330 configured to store historical multi-channel collected data, historical tourist behavior recognition results, historical safety risk assessment scores and historical early warning information.

[0038] Further, the central server 1310 adopts a deep learning technology to perform tourist behavior recognition and risk assessment processes; in the process of performing tourist behavior recognition, it is necessary to first define tourist behavior categories, including normal behavior, abnormal behavior, emergency behavior, and special behavior. The normal behavior category includes but is not limited to walking, watching, taking pictures, and resting, the abnormal behavior category includes but is not limited to climbing fences, entering safety areas, feeding animals, chasing and playing, and damaging facilities, the emergency behavior category includes but is not limited to falling, arguing, and gathering and shouting, and the special behavior category includes but is not limited to long-time retention, repeated back-and-forth, and getting lost. For these four types of tourist behavior, the behavior recognition methods that can be used include: 1) pose estimation, used to identify the body posture and action of the tourist; 2) spatiotemporal feature extraction, used to analyze the time and space features of the action; 3) context analysis, used to understand the behavior in combination with environmental and location information; and 3) behavior classification, used to determine the behavior type using a classification model. Secondly, based on the predefined tourist behavior categories, the tourist behavior recognition and analysis processing are performed on the preliminary processed data to obtain tourist behavior recognition results, including tourist normal behavior analysis results, tourist abnormal behavior analysis results, tourist emergency behavior analysis results, and tourist special behavior results.

[0039] Further, the detection process for the abnormal behavior category includes rule violation detection (for detecting behaviors that violate park rules), safety risk behavior (for detecting behaviors that may cause safety problems), abnormal pattern recognition (for identifying behaviors that deviate from normal patterns), and group anomaly detection (for detecting group abnormal behaviors). Meanwhile, the abnormal detection methods that can be used include: rule-based detection method (using predefined rules to identify explicit rule violations), statistical-based detection method (using statistical methods to identify behaviors that deviate from normal distribution), learning-based detection method (using unsupervised learning methods to identify abnormal patterns), and context-aware detection method (considering environmental and time factors to make abnormal judgments).

[0040] Further, an abnormal behavior detection algorithm can be used to identify whether the tourist has abnormal behavior, wherein the feeding behavior recognition algorithm includes: (1) calculating the pose abnormality score (i.e., the deviation of the current pose from the normal pose pattern), the specific formula is: , wherein represents the pose abnormality score, represents the bth key point feature of the current pose, represents the normal mean of the bth key point feature, represents the standard deviation of the bth key point feature, and K represents the number of key point features; and (2) calculating the spatiotemporal abnormality score (i.e., detecting the abnormal pattern of the tourist in time and space), the specific formula is: , wherein a spatio-temporal anomaly score, a spatial anomaly score, a temporal anomaly score, a weight of a spatial factor, with a value range of 0-1.

[0041] For example, when the system monitors the behavior of a visitor near the elephant exhibition area, the posture analysis obtains a posture anomaly score of 1.8 (assuming that the normal threshold of the posture anomaly score is 1.0, and the posture anomaly score is obviously higher than the normal threshold 1.0); the spatial anomaly score of the visitor is 0.7 (located in the semi-prohibited area), the temporal anomaly score is 0.9 (staying in the area for too long), and the spatial weight is set to 0.6; then the spatio-temporal anomaly score = 0.6 x 0.7 + (1-0.6) x 0.9 = 0.78, combined with the posture anomaly and the spatio-temporal anomaly, the system judges that the visitor may be trying to feed the behavior, if a four-level warning mechanism is set for the feeding behavior type, respectively, the prompt level, the warning level, the alarm level and the emergency level, wherein the threshold range of the prompt level is <0.25, the threshold range of the warning level is >0.25 and <0.5, the threshold range of the alarm level is >0.5 and <0.75, and the threshold range of the emergency level is >0.75 and <1, based on the calculated spatio-temporal anomaly score 0.78 in the threshold range of the emergency level, the emergency level of the warning is triggered.

[0042] Further, the present application provides corresponding identification and processing mechanisms for two high-risk behaviors in abnormal behavior types: feeding behavior and invading safety area behavior.

[0043] The identification target for feeding behavior includes: posture features (identify typical throwing posture and action sequence), object detection (detect food or other throwing objects), position association (analyze the relative position relationship between the visitor and the animal exhibition area), behavior sequence (analyze the preparation action before feeding and the reaction after feeding), and the feeding behavior identification process that can be used includes: pre-screening (preliminary screening based on position and basic behavior), fine-grained analysis (fine-grained posture analysis on suspicious behavior), object tracking (track the trajectory of possible throwing objects), and comprehensive judgment (comprehensive judgment combined with multiple features).

[0044] Further, a feeding behavior identification algorithm can be used to identify whether a visitor has a feeding behavior, wherein the feeding behavior identification algorithm includes: (1) extracting key features of the feeding behavior, and the specific formula is: , wherein, represents an arm posture feature, represents an arm movement speed feature, represents a distance feature from the exhibition area, represents a food object detection feature, representing behavior duration feature; (2) calculating the feeding behavior score, the specific formula is: , wherein, representing the feeding behavior score, representing the normalized value of the cth feeding behavior feature; representing the weight of the cth feeding behavior feature.

[0045] For example, when the system detects the behavior of a visitor, the extracted key features, the corresponding normalized values, and the corresponding weights are as follows: (1) arm posture feature: normalized value 0.85 (highly consistent with the throwing posture), weight 0.3; (2) arm movement speed feature: normalized value 0.7 (speed and direction consistent with the throwing action), weight 0.25; (3) distance from the exhibition area feature: normalized value 0.9 (very close to the exhibition area fence), weight 0.2; (4) food object detection feature: normalized value 0.6 (detecting an object that may be food), weight 0.15; (5) behavior duration feature: normalized value 0.5 (moderate duration), weight 0.1; then the feeding behavior score = 0.3x0.85+0.25x0.7+0.2x0.9+0.15x0.6+0.1x0.5=0.75, which indicates that the behavior is likely to be a feeding behavior, and assuming that the feeding behavior score 0.75 is within the threshold range of the preset emergency level, the system should trigger the corresponding warning.

[0046] For the recognition target of the intrusion into the safety area behavior, the target includes: boundary detection (defining and monitoring the boundary of the safety area), boundary crossing judgment (detecting whether the visitor crosses the safety boundary), intention analysis (analyzing the behavior pattern and intention of the visitor approaching the boundary), risk assessment (evaluating the risk level of the intrusion behavior); at the same time, the intrusion behavior recognition method that can be used includes: virtual fence (defining a virtual safety boundary in the system), trajectory analysis (analyzing the relationship between the visitor's moving trajectory and the boundary), behavior prediction (predicting the possible moving direction and intention of the visitor), multi-person association (analyzing group intrusion behavior).

[0047] Strategies for handling high-risk behaviors include: priority escalation (giving high-risk behaviors the highest priority), rapid warning (activating a rapid warning mechanism to notify relevant personnel in the shortest possible time), automatic intervention (activating automatic intervention when possible, such as voice warnings), and evidence preservation (automatically saving relevant video evidence for subsequent processing). At the same time, rapid response mechanisms that can be adopted include: dedicated channels (setting up dedicated communication channels for high-risk warnings), proximity dispatch (dispatching the nearest security personnel for rapid response), plan activation (automatically activating the corresponding emergency plan), and coordinated response (coordinating a joint response among multiple departments). In addition, behavioral education and guidance can also be used for warnings, such as real-time education (immediate education for illegal tourists), information push (pushing safety reminders to tourists in relevant areas), sign reinforcement (dynamically adjusting the display method of safety reminder signs), and positive guidance (providing demonstration and guidance of correct behavior). Specific education and guidance implementation methods include: targeted broadcasting (playing safety reminders to specific areas), mobile push (pushing safety information to park APP users), electronic display screens (displaying reminder information on electronic display screens in relevant areas), and interactive education (providing safety education through interactive methods).

[0048] Furthermore, the tourist density calculation algorithm can be used to calculate the current tourist density data of each functional area in the scenic area. The tourist density calculation algorithm includes the regional density calculation algorithm and the weighted density calculation algorithm. Among them, if the uniform distribution of tourists in each functional area is considered, the regional density calculation algorithm is used to calculate the average density of tourists in the functional area. The formula of the regional density calculation algorithm is: Where D represents the average density of tourists, N represents the number of tourists in the functional area, and A represents the area of ​​the functional area. If the uneven distribution of tourists in each functional area is considered, a weighted density calculation algorithm is used to calculate the weighted density of tourists in all sub-areas within the functional area. The formula of the weighted density calculation algorithm is: Where, represents the weighted density of tourists, represents the weight of the d-th sub-region, represents the number of tourists in the d-th sub-area, and m represents the number of sub-areas in the functional area; the calculated average tourist density or tourist weighted density is used as the current tourist density data.

[0049] Further, by calculating the current tourist density data of each functional area in the scenic spot, traffic trend analysis, congestion prediction and capacity assessment can be performed, wherein the traffic trend analysis is specifically analyzing the time variation trend of passenger flow, the congestion prediction is specifically predicting the area and time where congestion may occur, and the capacity assessment is specifically assessing the actual carrying capacity and saturation of each area. Meanwhile, the techniques for realizing density analysis include but are not limited to density estimation techniques (using density map estimation method to calculate crowd density), flow statistics techniques (statistically counting the number of people passing through a specific area within a unit time), trend analysis techniques (using time series analysis method to predict flow variation), and threshold setting techniques (setting density warning threshold according to the characteristics of the area).

[0050] Further, by combining the tourist abnormal behavior analysis result and the current tourist density data of each functional area in the scenic spot with the environmental data and the current time data, the current safety risk assessment score can be obtained, and the implementation process specifically includes: based on the tourist abnormal behavior analysis result and the current tourist density data of each functional area in the scenic spot, combining the environmental data and the current time data, using a risk assessment algorithm to calculate the current safety risk assessment score; the formula of the risk assessment algorithm is: ; in the formula, the current safety risk assessment score is represented by S, the risk score based on the density factor is represented by Sd, the risk score based on the abnormal behavior is represented by Sa, the risk score based on the environmental factor is represented by Se, the risk score based on the time factor is represented by St, respectively represent the weight coefficients based on the density factor, the abnormal behavior, the environmental factor and the time factor; wherein the formula of the risk score based on the density factor is: ; in the formula, the current tourist density data is represented by D, the safety density threshold is represented by Ds, and vj represents a risk growth parameter, and the value range is 1.5-2.5; the formula of the risk score based on the abnormal behavior is: ; in the formula, the risk score based on the abnormal behavior is represented by Sa, the severity score of the i-th abnormal behavior is represented by Si, the weight of the i-th abnormal behavior is represented by wi, and n represents the number of detected abnormal behaviors; the risk score based on the environmental factor and the risk score based on the time factor are obtained by mapping the pre-defined environmental factor influence degree value on risk and the time factor influence degree value on risk.

[0051] For example, when a panda viewing area has the following risk assessment parameters set: density weight 0.4, behavior weight 0.3, environment weight 0.2, time weight 0.1; the current visitor density is 0.8 person / m2, the safety density threshold is 1.2 person / m2, the risk growth parameter is 2; two abnormal behaviors are detected: fence jumping attempt (severity 0.9, weight 0.6) and feeding behavior (severity 0.7, weight 0.4); the environment risk score is 0.3 (good weather), the time risk score is 0.5 (close to closing time); then the risk score based on the density factor = = 0.445, the risk score based on abnormal behavior = = 0.6688, the current safety risk assessment score = = 0.48864, which indicates that the current risk is at a medium level, close to the warning threshold (assuming 0.5).

[0052] Further, by setting the analysis engine 1320, the analysis and prediction of security situation awareness data can be realized, specifically including: predicting the security situation in the near future, simulating the security situation under different scenarios, predicting the overall security situation awareness of the park; for predicting the security situation in the near future, it can include: passenger flow prediction (predicting passenger flow changes in the next 1-4 hours), hotspot prediction (predicting possible tourist gathering hotspots), behavior prediction (predicting possible abnormal behavior types and locations), resource demand prediction (predicting the demand distribution of security personnel and resources), at the same time, the technology for predicting the security situation in the near future can include: time series analysis (using ARIMA or LSTM model to analyze time series data), machine learning model (using random forest or gradient boosting tree for prediction), deep learning method (using deep neural network to capture complex patterns), integrated prediction (fusing the prediction results of multiple models to improve accuracy); for simulating the security situation under different scenarios, it can include: peak period simulation (simulating the passenger flow distribution during holiday peak period), emergency simulation (simulating the impact of emergencies such as severe weather), evacuation simulation (simulating evacuation path and time in emergency situations), intervention effect simulation (simulating the effect of different intervention measures), at the same time, the technology for simulating the security situation under different scenarios can include: rule-based simulation (using pre-defined rules for simple scenario simulation), agent-based simulation (using multi-agent system to simulate complex scenarios), historical data-driven (scenario extrapolation based on historical data), expert knowledge fusion (fusing expert experience to construct scenarios); for predicting the overall security situation awareness of the park, it can include: global situation (overall security condition assessment of the park), regional situation (security condition comparison of each functional area), time evolution (time change trend of security situation), correlation analysis (correlation between different security factors), at the same time, the technology for predicting the overall security situation awareness of the park can include: data visualization (using intuitive visualization to display security situation), multi-dimensional analysis (analyzing security situation from time, space, behavior, etc.), anomaly highlighting (highlighting abnormal and risk points), decision suggestion (providing decision suggestions based on situation analysis).

[0053] Further, a time series model can be used to predict future short-term passenger flow, wherein the formula of the time series model is: ; in the formula, F(t+h) represents the predicted passenger flow at time t+h, F(t-i+1) represents the historical passenger flow, ε(t-j+1) represents the historical prediction error, and respectively represent the model parameters corresponding to the historical passenger flow and the model parameters corresponding to the historical prediction error, and respectively represent the model order corresponding to the historical passenger flow and the model order corresponding to the historical prediction error, and ε(t+h) represents the prediction error; if the influence of external factors is considered, the external factors need to be fused for passenger flow prediction, and the specific formula is: ; in the formula, represents the basic prediction value, represents the kth external factor (such as weather, holiday, etc.), represents the influence coefficient of the kth external factor, and m represents the number of considered external factors.

[0054] For example, when the system uses an ARIMA(2, 1, 1) model to predict the passenger flow of a certain area, the model parameters are φ1=0.7, φ2=0.2, and θ1=0.3; the passenger flow at the current time t is 200 people, the passenger flow at t-1 is 180 people, and the prediction error at t-1 is 10 people; considering external factors: weather (sunny, influence coefficient 0.1) and time (weekend, influence coefficient 0.2), the predicted passenger flow at t+1 is: the basic predicted passenger flow=0.7×200+0.2×180+0.3×10-(0.7+0.2-1)×180=197 people, the predicted passenger flow considering external factors=197+0.1×200+0.2×200=257 people, which indicates that considering the sunny and weekend factors, the passenger flow of the area is expected to increase significantly.

[0055] In some embodiments, the application service layer 140 is configured to display passenger flow data, current visitor density data of each functional area in the scenic area, and a current safety risk assessment score, trigger corresponding warning information and publish differentiated intervention suggestion measure information in response to visitor abnormal behavior analysis results or the current safety risk assessment score, and provide historical data analysis, trend prediction, and decision support.

[0056] Further, with reference to the accompanying Figure 5 , the application service layer 140 includes: a real-time monitoring interface module 1410 configured to display passenger flow data, current visitor density data of each functional area in the scenic area, and a current safety risk assessment score; a warning management module 1420 configured to trigger corresponding warning information in response to visitor abnormal behavior analysis results or the current safety risk assessment score; an intervention control interface module 1430 configured to publish differentiated intervention suggestion measure information by an administrator for the triggered warning information; and a data analysis platform 1440 configured to perform statistical and trend prediction analysis on stored historical multi-channel collected data, historical visitor behavior recognition results, historical safety risk assessment scores, and historical warning information to obtain analysis results; based on the analysis results, dynamically adjust resource allocation, optimize patrol paths, and generate a visual emergency plan report.

[0057] Further, by setting the real-time monitoring interface module 1410, the information including but not limited to passenger flow data, current visitor density data of each functional area in the scenic area, current security risk assessment score, park passenger flow distribution, hot spot area, etc. can be visually displayed.

[0058] Further, by setting the early warning management module 1420, in response to the visitor abnormal behavior analysis result or the current security risk assessment score, a multi-level early warning mechanism from prompt to emergency can be triggered, and the early warning information can be accurately distributed to provide differentiated intervention suggestions according to specific conditions.

[0059] Further, in response to the visitor abnormal behavior analysis result or the current security risk assessment score, the corresponding early warning information is triggered, including: in response to the visitor abnormal behavior analysis result, when it is detected that the visitor abnormal behavior analysis result exceeds a pre-defined corresponding abnormal behavior type threshold, a multi-level early warning mechanism based on anomaly detection is triggered to trigger early warning information of the corresponding level; in response to the current security risk assessment score, when it is detected that the current security risk assessment score exceeds a pre-defined early warning threshold, a multi-level early warning mechanism based on threshold is triggered to trigger early warning information of the corresponding level. Wherein, the multi-level early warning mechanism based on anomaly detection and the multi-level early warning mechanism based on threshold both include: a four-level early warning mechanism, respectively prompt level, warning level, alarm level and emergency level, wherein the prompt level adopts a blue indicator light for early warning, indicating a slight abnormality, attention is needed but no intervention is needed at present, the warning level adopts a yellow indicator light for early warning, indicating a significant abnormality, suggesting to take preventive measures, the alarm level adopts an orange indicator light for early warning, indicating a serious abnormality, needing immediate intervention, and the emergency level adopts a red indicator light for early warning, indicating a critical situation, needing to start an emergency plan.

[0060] Further, the trigger condition of the multi-level early warning mechanism based on threshold adopts a early warning trigger algorithm to realize the trigger process, and the formula of the early warning trigger algorithm is: ; in the formula, W represents the four-level early warning mechanism, and the value range is 0, 1, 2, 3, respectively corresponding to the prompt level, the warning level, the alarm level and the emergency level, represents the current security risk assessment score, respectively represent three pre-defined early warning thresholds from low to high, which are used to distinguish and trigger the corresponding early warning mechanism; the early warning threshold is obtained by dynamic adjustment using a dynamic adjustment algorithm, and the formula of the dynamic adjustment algorithm is: ; in the formula, represents the i-th level early warning threshold at time t, represents the basic threshold, and the basic threshold is a pre-defined early warning threshold, represents the current security risk assessment score, represents the security risk assessment score of the reference period, represents an adjustment coefficient, and the value range is 0.3-0.7.

[0061] For example, when the basic early warning threshold of the system is set as: , , ; it is currently a weekend afternoon, and the historical data shows that the average risk score of this period is 0.45, while the average risk score of the benchmark period (working day) is 0.35; the adjustment coefficient γ = 0.5, and the dynamically adjusted early warning threshold is: , , , which means that during the weekend peak period, the system will appropriately increase the early warning threshold to avoid excessive low-level early warnings.

[0062] Further, in addition to the aforementioned multi-level early warning mechanism based on anomaly detection and the multi-level early warning mechanism based on threshold, a rule-based early warning mechanism and a manual-triggered early warning mechanism can also be included, wherein the rule-based early warning mechanism is triggered when specific rule conditions are met, and the manual-triggered early warning mechanism is manually triggered by security personnel.

[0063] Further, by setting the intervention control interface module 1430, the administrator can be allowed to issue targeted reminders and control measures. Differentiated intervention suggestions can be provided according to the triggered early warning information, and the intervention effect can be continuously evaluated to achieve precise and effective safety management.

[0064] Further, the system distributes early warning information to relevant personnel, including: (1) security personnel: receiving early warning information through mobile terminals, (2) management personnel: receiving early warning through management platform and mobile application; (3) tourists: receiving safety reminders through broadcast, information screen or APP push; (4) external agencies: sending early warning to police, medical institutions and other external agencies when necessary; the distribution strategy adopted when the early warning information is distributed to relevant personnel includes: (1) role orientation: sending relevant early warning according to personnel role orientation; (2) area orientation: sending early warning to staff in specific areas; (3) level filtering: determining the scope of recipients according to the level of early warning; (4) information customization: customizing early warning content and form according to the recipient. In addition, the system provides intervention suggestions for different situations, such as (1) information prompt: providing safety reminders through broadcast, information screen; (2) personnel scheduling: suggesting adjusting security personnel deployment; (3) flow control: suggesting limiting access to specific areas or guiding diversion; (3) emergency response: suggesting starting emergency evacuation or rescue measures; at the same time, the generation process of intervention measures includes: (1) rule-based: using predefined rules to generate intervention suggestions; (2) case-based: referring to the handling method of similar historical cases; (3) simulation-based: evaluating the effect of different intervention measures through scenario simulation; (4) expert system: integrating expert knowledge to generate intervention suggestions; and in order to evaluate the actual effect of intervention measures, the following content can be evaluated, including: (1) immediate effect: evaluating the immediate change of risk indicators after intervention; (2) short-term effect: evaluating the risk change 15-30 minutes after intervention; (3) long-term effect: analyzing the long-term impact of intervention measures on similar situations; (4) side effect evaluation: evaluating the negative effects that may be brought by intervention measures; at the same time, the effect evaluation methods that can be used include: (1) before and after comparison: comparing the changes of risk indicators before and after intervention; (2) A / B testing: testing the effect of different intervention measures in similar situations; (3) tourist feedback: collecting tourist feedback on intervention measures; (4) long-term trend: analyzing the impact of intervention measures on long-term safety trend.

[0065] Further, by setting up a data analysis platform 1440, historical data analysis and decision support functions can be provided; for historical data analysis functions, the following content can be analyzed: (1) behavior pattern analysis: analyzing long-term patterns and trends of tourist behavior; (2) hot spot area analysis: identifying long-term tourist gathering hot spots; (3) risk factor analysis: identifying key factors affecting safety risk; (4) intervention effect analysis: evaluating the long-term effect of different intervention measures, and the analysis methods that can be used include: (1) statistical analysis: using statistical methods to analyze data distribution and correlation; (2) pattern mining: using data mining techniques to discover hidden patterns; (3) correlation analysis: analyzing the correlation between different factors; (4) trend prediction: predicting long-term safety trend changes.

[0066] Further, for the decision support function, it can include three aspects of security management optimization, report and visualization management, and knowledge base construction management; among them, for security management optimization, it can be used to optimize the following contents: (1) personnel configuration optimization: optimize the number and distribution of security personnel; (2) facility layout suggestion: provide security facility layout optimization suggestions; (3) rule adjustment suggestion: propose rule adjustment suggestions based on data analysis; (4) training demand analysis: identify the training needs of security personnel, and the optimization methods used can include: (1) simulation optimization: test the effects of different configurations through simulation; (2) sensitivity analysis: analyze the influence degree of different factors on the security situation; (3) cost-benefit analysis: evaluate the cost and benefit of different optimization schemes; (4) expert review: combine expert experience to review optimization suggestions; for report and visualization management, the content includes: (1) real-time dashboard: display key indicators of current security situation; (2) trend report: generate regular reports on security trends; (3) event analysis report: in-depth analysis of important security events; (4) decision support view: provide intuitive data view for management decisions, while the visualization implementation methods include: (1) interactive chart: provide interactive data charts; (2) heat map: use heat map to display risk and passenger flow distribution; (3) timeline view: show the time change of security situation; (4) relationship network: visualize the relationship between different factors; for the knowledge base construction aspect in knowledge base construction management, the following types of knowledge bases are constructed, including: (1) case base: records typical security events and handling methods; (2) rule base: maintains security rules and intervention strategies; (3) model base: saves and updates various analysis and prediction models; (4) expert knowledge: captures and encodes the experience and knowledge of security management experts, and for the knowledge base management aspect in knowledge base construction management, the following contents are managed: (1) knowledge acquisition: acquire knowledge from actual cases and expert experience; (2) knowledge organization: organize knowledge by category and relevance; (3) knowledge update: update knowledge according to new data and analysis results; (4) knowledge application: apply knowledge to decision-making and early warning generation.

[0067] Further, the dynamic adjustment of resource configuration is achieved by optimizing the number of security personnel in the functional area, and the optimized patrol path is calculated by a weighted graph algorithm.

[0068] Further, for the security personnel optimization configuration aspect, the number of security personnel required by each functional area can be calculated to optimize the security personnel configuration, and the specific formula is: ; In the formula, represents the number of security personnel required by the dth functional area, represents the basic configuration number, represents the security risk assessment score, denotes the normalized visitor density, denotes the historical event frequency, denote the weight coefficients of the security risk assessment score, the visitor density and the historical event frequency, respectively.

[0069] Further, for the patrol path optimization aspect, the optimization process of the patrol path can be achieved by a weighted graph algorithm, and the formula of the weighted graph algorithm is: ; in the formula, denotes the optimal patrol path, denotes the set of all possible paths, denotes the distance between the node x and the node y; denotes the path importance weight, denotes the current security risk assessment score on the path.

[0070] For example, if the basic security configuration of the fierce beast area of the zoo is 3 people, the current security risk assessment score of the area is 0.6, the normalized visitor density is 0.8, the historical event frequency is 0.4, the weight coefficients are set as e=0.5, g=0.3, =0.2, then the required number of security personnel is: required personnel=3×(1+0.5×0.6+0.3×0.8+0.2×0.4)=4.86≈5 people, which indicates that considering the current risk and the passenger flow, 5 security personnel are required in the fierce beast area, which is 2 more than the basic configuration.

[0071] Obviously, those skilled in the art should understand that the above embodiments of the steps of the present application can be executed in a manner different from the present application, and the simulation methods and experimental equipment include but are not limited to the above description. The above steps of the present application can be executed in a different order in some cases, and the steps shown or described above can be executed separately. Therefore, the present application is not limited to any specific combination of hardware and software.

[0072] The above is a further detailed description of the present application in combination with specific embodiments, and the specific implementation of the present application should not be limited to these descriptions. For those skilled in the art to which the present application belongs, without departing from the concept of the present application, a number of simple deductions or substitutions can be made, which should be regarded as falling within the protection scope of the present application.

Claims

1. A behavior monitoring and early warning system for visitors in a zoo, characterized in that: include: The data collection layer is used to obtain data collected from multiple channels; wherein the data collected from multiple channels includes video data, passenger flow data, environmental data, ticketing data and mobile application data; The edge computing layer is used to perform pre-processing, image processing, target detection processing, and data fusion processing on the data collected from multiple channels to obtain preliminary processed data; The central processing layer is used to use deep learning technology to identify and analyze tourist behavior based on predefined types of tourist behavior on the preliminary processed data, obtain tourist behavior identification results, analyze the types and number of abnormal behaviors, and generate abnormal tourist behavior analysis results; calculate the current tourist density data of each functional area in the scenic area based on the tourist density calculation algorithm; generate the current safety risk assessment score based on the abnormal tourist behavior analysis results and the current tourist density data of each functional area in the scenic area, combined with environmental data and current time data; and generate security situation awareness data based on the current safety risk assessment score and combined with data collected from multiple channels; The application service layer is used to display passenger flow data, current tourist density data of each functional area in the scenic area, and current safety risk assessment scores. In response to the analysis results of abnormal tourist behavior or the current safety risk assessment scores, it triggers corresponding early warning information and publishes differentiated intervention recommendation measures, as well as provides historical data analysis, trend forecasting, and decision support.

2. A behavior monitoring and early warning system for visitors in a zoo according to claim 1, characterized in that: The data acquisition layer includes: Video data acquisition module, used to install video surveillance equipment covering various functional areas in the park and obtain video data through the video surveillance equipment; The passenger flow data collection module is used to install passenger flow counting devices at the park entrance, exit, and passages connecting different functional areas, as well as Wi-Fi probes and Bluetooth beacons at the passages connecting different functional areas. This module obtains passenger flow data including the number of visitors entering the park and their time distribution, the flow rate and length of stay of visitors in each functional area, the distribution of visitors in the park, and the movement paths of visitors. An environmental data acquisition module is used to obtain environmental data within a predefined time window, wherein the environmental data includes meteorological data, noise decibel data, light intensity data, and air quality index data; Ticketing system interface module, used to connect with the park ticketing system to obtain ticketing data; The mobile application data collection module is used to collect mobile application data including visitor location data and visitor behavior data through the park's official APP with user authorization, and the mobile application data is obtained after data privacy protection processing.

3. A behavior monitoring and early warning system for visitors in a zoo according to claim 1, characterized in that: The edge computing layer includes: The data preprocessing module is used to clean and standardize the acquired passenger flow data, environmental data, ticketing data, and mobile application data to obtain preprocessed data; An edge computing device, installed next to the video surveillance device, is used to perform image processing and target detection processing on the acquired video data using target detection and tracking technology to obtain visitor detection and tracking data. The visitor detection and tracking data includes visitor location data and visitor quantity data in the video data, visitor movement trajectory data in the video sequence, group visitor data and family unit data in the video data, and target audience data in the video data; The local analysis unit is used to perform data fusion processing on the pre-processed data and the visitor detection and tracking data to obtain preliminary processed data.

4. A behavior monitoring and early warning system for visitors in a zoo according to claim 1, characterized in that: The central processing layer includes: The central server is configured to employ deep learning technology to perform tourist behavior identification and analysis on the initially processed data based on predefined tourist behavior categories, thereby obtaining tourist behavior identification results, wherein the tourist behavior categories include normal behavior, abnormal behavior, emergency behavior, and special behavior; calculate the current tourist density data for each functional area within the scenic area based on a tourist density calculation algorithm; and generate a current safety risk assessment score based on the abnormal tourist behavior analysis results and the current tourist density data for each functional area within the scenic area, combined with environmental data and current time data; The analysis engine is used to generate security situation awareness data based on the current security risk assessment score and data collected from multiple channels; The data storage module is used to store historical data collected from multiple channels, historical visitor behavior recognition results, historical safety risk assessment scores and historical warning information.

5. A behavior monitoring and early warning system for visitors in a zoo according to claim 4, characterized in that: The tourist density calculation algorithm includes a regional density calculation algorithm and a weighted density calculation algorithm. If the uniform distribution of tourists in each functional area is considered, the regional density calculation algorithm is used to calculate the average density of tourists in the functional area. The formula of the regional density calculation algorithm is: Where D represents the average density of tourists, N represents the number of tourists in the functional area, and A represents the area of ​​the functional area; If the uneven distribution of tourists in each functional area is taken into account, a weighted density calculation algorithm is used to calculate the weighted density of tourists in all sub-areas within the functional area. The formula of the weighted density calculation algorithm is: Where, represents the weighted density of tourists, represents the weight of the d-th sub-region, represents the number of tourists in the dth sub-area, and m represents the number of sub-areas in the functional area; The calculated average tourist density or tourist weighted density is used as the current tourist density data.

6. A behavior monitoring and early warning system for visitors in a zoo according to claim 4, characterized in that: The current safety risk assessment score is generated based on the analysis results of abnormal tourist behavior and the current tourist density data of each functional area in the scenic area, combined with environmental data and current time data, including: Based on the analysis results of abnormal tourist behavior and the current tourist density data of each functional area in the scenic area, combined with environmental data and current time data, a risk assessment algorithm is used to calculate the current safety risk assessment score; The formula of the risk assessment algorithm is: Where, Indicates the current security risk assessment score. represents the risk score based on the density factor, represents the risk score based on abnormal behavior, represents the risk score based on environmental factors, represents the risk score based on the time factor, Respectively represent the weight coefficients based on density factors, abnormal behavior, environmental factors, and time factors; The formula for the risk score based on density factors is: Where, Indicates the current tourist density data; represents the safety density threshold; vj represents the risk growth parameter, ranging from 1.5 to 2.5; The formula for the risk score based on abnormal behavior is: Where, represents the risk score based on abnormal behavior, represents the severity score of the i-th abnormal behavior, represents the weight of the i-th abnormal behavior, and n represents the number of abnormal behaviors detected; The risk score based on environmental factors and the risk score based on time factors are obtained by mapping with the predefined environmental factor impact value and the time factor impact value on the risk, respectively.

7. A behavior monitoring and early warning system for visitors in a zoo according to claim 1, characterized in that: The application service layer includes: Real-time monitoring interface module, used to display passenger flow data, current visitor density data of each functional area in the scenic area and current safety risk assessment score; An early warning management module is used to trigger corresponding early warning information in response to the analysis results of abnormal tourist behavior or the current safety risk assessment score; The intervention control interface module is used for administrators to issue differentiated intervention suggestions based on triggered warning information; The data analysis platform is used to conduct statistical and trend forecasting analysis on the stored historical data collected from multiple channels, historical tourist behavior recognition results, historical safety risk assessment scores and historical warning information to obtain analysis results; based on the analysis results, it dynamically adjusts resource allocation, optimizes patrol routes, and generates visual emergency plan reports.

8. A behavior monitoring and early warning system for visitors in a zoo according to claim 7, characterized in that: The triggering of corresponding warning information in response to the analysis result of abnormal behavior of tourists or the current security risk assessment score includes: in response to the analysis result of abnormal behavior of tourists, when it is detected that the analysis result of abnormal behavior of tourists exceeds the predefined threshold value of the corresponding abnormal behavior type, triggering the warning information of the corresponding level based on the multi-level warning mechanism of abnormal detection; in response to the current security risk assessment score, when it is detected that the current security risk assessment score exceeds the predefined warning threshold value, triggering the warning information of the corresponding level based on the multi-level warning mechanism of the threshold value; Among them, the multi-level early warning mechanism based on anomaly detection and the multi-level early warning mechanism based on thresholds both include: a four-level early warning mechanism, namely prompt level, warning level, alarm level and emergency level. Among them, the prompt level uses a blue indicator light for early warning, indicating a minor abnormality that requires attention but no intervention for the time being; the warning level uses a yellow indicator light for early warning, indicating an obvious abnormality and it is recommended to take preventive measures; the alarm level uses an orange indicator light for early warning, indicating a serious abnormality that requires immediate intervention; the emergency level uses a red indicator light for early warning, indicating a critical situation that requires the activation of an emergency plan.

9. A behavior monitoring and early warning system for visitors in a zoo according to claim 8, characterized in that: The triggering condition of the threshold-based multi-level warning mechanism adopts a warning triggering algorithm to implement the triggering process. The formula of the warning triggering algorithm is: Where W represents the four-level early warning mechanism, with a value range of 0, 1, 2, and 3, corresponding to the prompt level, warning level, alarm level, and emergency level, respectively. Indicates the current security risk assessment score. Represents three predefined warning thresholds from low to high, which are used to distinguish and trigger corresponding warning mechanisms; The warning threshold is obtained by dynamic adjustment using a dynamic adjustment algorithm. The formula of the dynamic adjustment algorithm is: Where, represents the i-th level warning threshold at time t, Indicates the basic threshold, which is a predefined warning threshold. Indicates the current security risk assessment score. represents the security risk assessment score for the baseline period, Indicates the adjustment coefficient, ranging from 0.3 to 0.

7.

10. The behavior monitoring and early warning system for visitors in a zoo according to claim 7, characterized in that: The dynamic adjustment of resource configuration is achieved by optimizing the number of security personnel in the functional area, and the optimization of patrol paths is achieved by calculating the optimal patrol path through a weighted graph algorithm.

Images