Campsite intelligent safety management method, device and equipment and storage medium

Abstract

This invention discloses a method, device, equipment, and storage medium for intelligent safety management of campsites, belonging to the field of Internet of Things and artificial intelligence safety management technology. It includes: collecting multimodal sensor data from a multifunctional integrated terminal deployed at the campsite, the multimodal sensor data including video image data, physiological health data, and location data; performing multi-task parallel inference on the video image data to obtain behavioral anomaly detection results, fire and smoke identification results, and crowd density distribution; fusing the physiological health data, location data, and behavioral anomaly detection results from multiple sources to generate a comprehensive risk score; when the comprehensive risk score exceeds a preset threshold, determining the trigger level and scheduling a linkage response instruction set, and sending the linkage response instruction set to the corresponding integrated terminal. This invention achieves proactive safety early warning driven by multimodal data fusion, effectively reducing emergency response latency and improving the overall safety management efficiency of the campsite.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things and artificial intelligence safety management technology, and in particular to a method, device, equipment and storage medium for intelligent safety management of campsites. Background Technology

[0002] In recent years, the outdoor camping tourism industry has developed rapidly, and the number of campsites has continued to grow. However, campsite safety management faces unique challenges such as complex environments, dense populations, and continuous day and night operations, and existing technologies have the following main shortcomings.

[0003] First, the existing campsite safety management system is mainly based on single-function modules. Subsystems such as video surveillance, personnel management, and health monitoring are deployed independently, and there is a lack of data sharing mechanisms between the systems. This results in data silos between the subsystems, making it impossible for managers to make comprehensive judgments on safety risks based on multi-dimensional data and making it difficult to achieve cross-system collaborative early warning.

[0004] Second, existing video surveillance systems primarily rely on video recording and manual review, allowing for post-incident review only and lacking real-time intelligent analysis capabilities. While some systems incorporate image recognition modules, tasks such as behavior recognition and smoke / fire detection operate independently, preventing the parallel output of detection results for multiple types of abnormal events within a single inference iteration. Therefore, real-time performance and computational resource utilization efficiency need improvement.

[0005] Third, existing systems typically trigger only a fixed alarm when detecting a single abnormal signal, lacking a comprehensive risk assessment mechanism based on multi-source data fusion. Single events such as abnormal physiological health, boundary crossing, abnormal behavior, and environmental hazards may generate false alarms due to insufficient confidence, while combinations of multiple low-confidence signals often indicate higher security risks, and existing systems cannot effectively handle such situations.

[0006] Fourth, when an emergency occurs, the existing system lacks a graded response mechanism based on the severity of the event, as well as the ability to coordinate resources based on the event location. Rescue personnel cannot obtain accurate event location and on-site images in the first instance, resulting in a long delay in emergency response and potentially missing the best opportunity for handling the situation.

[0007] Therefore, there is an urgent need for a technical solution for campsite safety management that can solve the above problems. Summary of the Invention

[0008] In order to solve one or more problems existing in the prior art, this application provides a method, device, equipment and storage medium for intelligent safety management of campsites to overcome the shortcomings of the prior art. The technical problem to be solved by the present invention is achieved through the following technical solution.

[0009] According to a first aspect of this application, a smart safety management method for campsites is provided, the method comprising:

[0010] Multimodal sensor data is collected from multifunctional integrated terminals deployed at the campsite. The multimodal sensor data includes video image data, physiological health data and location data reported by wearable devices worn by tourists.

[0011] Perform multi-task parallel inference on the video image data to obtain abnormal behavior detection results, fire and smoke identification results, and crowd density distribution map;

[0012] The physiological health data, location data, abnormal behavior detection results, fire and smoke identification results, and crowd density distribution map are fused from multiple sources to generate a comprehensive regional risk score.

[0013] The alarm trigger level is determined based on the comparison between the comprehensive risk score of the region and the preset risk threshold.

[0014] Based on the alarm trigger level, a linkage response instruction set is generated and sent to the multi-functional integrated terminal. The linkage response instruction set includes at least one of camera scheduling instructions, lighting control instructions, broadcast instructions, and security notification instructions.

[0015] Preferably, the step of performing multi-task parallel inference on the video image data to obtain abnormal behavior detection results, fire and smoke identification results, and crowd density distribution map includes:

[0016] The video image data is input into a multi-task convolutional neural network, which includes a shared backbone network and a behavior detection head, a fire and smoke detection head, and a density estimation head connected to the shared backbone network.

[0017] The behavior detection head outputs abnormal behavior detection results, and the abnormal behavior includes at least one of falling, fighting, crossing boundaries, and gathering of people.

[0018] The fire and smoke detection head outputs fire and smoke identification results, wherein the fire and smoke identification results include flame confidence and smoke confidence.

[0019] The density estimation head outputs a distribution map of pedestrian density in each monitored area.

[0020] Preferably, the step of performing multi-source fusion processing on the physiological health data, location data, abnormal behavior detection results, fire and smoke identification results, and crowd density distribution map to generate a comprehensive regional risk score includes:

[0021] Anomaly scoring is performed on the physiological health data reported by each wearable device, including heart rate, blood oxygen saturation, and blood pressure.

[0022] The location data of the wearable device is mapped to the corresponding monitoring area to obtain the tourist distribution vector within the area;

[0023] Based on the abnormal behavior detection results, flame confidence, smoke confidence, and population density distribution map within the area, the environmental risk component of each monitoring area is calculated;

[0024] The physiological health abnormality score and the environmental risk component are weighted and summed according to preset weights to obtain the regional comprehensive risk score.

[0025] Preferably, determining the alarm trigger level based on the comparison result between the regional comprehensive risk score and the preset risk threshold includes:

[0026] When the comprehensive risk score of the area is greater than or equal to the first threshold and less than the second threshold, the alarm triggering level is determined to be a level one alarm, triggering the notification of nearby security personnel and the retrieval of camera footage.

[0027] When the comprehensive risk score of the area is greater than or equal to the second threshold, the alarm trigger level is determined to be a level 2 alarm, triggering full-area broadcast, lighting linkage, and retrieval of multiple camera images; wherein, the second threshold is greater than the first threshold.

[0028] Preferably, before acquiring multimodal sensing data from the multifunctional integrated terminal deployed at the campsite, the method further includes:

[0029] Receive an entry registration request sent by a wearable device, the entry registration request carrying the wearable device identifier;

[0030] Based on the wearable device identifier, a tourist profile is established, which includes the mapping relationship between the device identifier and the tourist's identity information, as well as the tourist's health baseline data.

[0031] The tourist health baseline data will be sent to the cloud-based tourist profile database for personalized baseline comparison during subsequent physiological health abnormality scoring.

[0032] Preferably, the multi-functional integrated terminal is equipped with a unique physical identifier. After generating and issuing a linkage response instruction set to the multi-functional integrated terminal, the method further includes:

[0033] Receive a call event message reported by the multi-functional integrated terminal in response to the triggering of the emergency call button, the call event message containing the physical number of the triggering terminal;

[0034] The location of the call is determined by querying a pre-stored terminal location coordinate mapping table based on the physical number.

[0035] Based on the location of the call, a notification containing the location coordinates is pushed to the nearest security personnel's terminal, and a two-way intercom establishment command is sent to the triggering terminal.

[0036] According to a second aspect of this application, a campsite intelligent safety management device employing the above-described campsite intelligent safety management method is provided, the device comprising:

[0037] The data acquisition module is used to collect multimodal sensor data from a multifunctional integrated terminal deployed at the campsite. The multimodal sensor data includes video image data, physiological health data and location data reported by wearable devices worn by tourists.

[0038] The reasoning and analysis module is used to perform multi-task parallel reasoning on the video image data to obtain abnormal behavior detection results, fire and smoke recognition results, and crowd density distribution map;

[0039] The fusion scoring module is used to perform multi-source fusion processing on the physiological health data, location data, abnormal behavior detection results, fire and smoke identification results, and crowd density distribution map to generate a comprehensive regional risk score.

[0040] The level determination module is used to determine the alarm trigger level based on the comparison result between the comprehensive risk score of the area and the preset risk threshold.

[0041] The linkage response module is used to generate and send a linkage response instruction set to the multi-functional integrated terminal based on the alarm trigger level.

[0042] Preferably, the multi-functional integrated terminal is equipped with a unique physical number identifier, which is used to query a pre-stored terminal location coordinate mapping table to obtain the current location coordinates of the multi-functional integrated terminal.

[0043] According to a third aspect of this application, an electronic device is provided, including a memory and a processor, characterized in that the memory stores a computer program, and the processor executes the computer program to implement the above-described intelligent safety management method for campsites.

[0044] According to a fourth aspect of this application, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the above-described intelligent safety management method for campsites.

[0045] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0046] This invention proposes a parallel inference scheme for multi-task convolutional neural networks, which simultaneously completes three types of tasks—behavior detection, fire and smoke recognition, and crowd density estimation—by sharing a backbone network. Compared to the serial execution of three independent detection models, this significantly reduces single-frame inference latency and improves the efficiency of computing resource utilization.

[0047] This invention constructs a regional comprehensive risk scoring mechanism that integrates multi-source heterogeneous data, mapping video analysis results, physiological health data, and location data to a comprehensive risk dimension. This realizes the transformation from hard threshold alarms triggered by a single event to continuous risk assessment driven by multi-dimensional fusion, effectively reducing the false alarm rate caused by noise from a single sensor.

[0048] This invention designs a linkage response scheduling mechanism based on risk scoring and grading. It automatically matches response resources such as camera scheduling, lighting control, broadcasting, and security personnel notification according to the alarm level, achieving precise matching of response resources with the severity of the event. Compared with the fixed alarm method of existing systems, it effectively reduces emergency response latency.

[0049] This invention designs a location resolution mechanism based on the physical number of a multi-functional integrated terminal, which achieves second-level accurate location of emergency call events, ensures the traceability of event location in the event of partial network interruption, and improves the positioning accuracy of emergency rescue. Attached Figure Description

[0050] To more clearly illustrate the embodiments of this application or the existing technical solutions, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 This is a flowchart of an intelligent safety management method for campsites according to Embodiment 1 of this application;

[0052] Figure 2 This is a structural block diagram of a campsite intelligent safety management system according to Embodiment 2 of this application;

[0053] Figure 3 This is a structural block diagram of a computer-readable storage medium according to Embodiment 3 of this application;

[0054] Figure 4 This is a structural block diagram of an electronic device according to Embodiment 4 of this application. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0056] Example 1: A smart safety management method for campsites, such as Figure 1 As shown, the method execution flow at time T0 is as follows:

[0057] Step 1: Collect multimodal sensor data from the multifunctional integrated terminal deployed at the campsite. The multimodal sensor data includes video image data, physiological health data and location data reported by wearable devices worn by tourists.

[0058] In this step, real-time operating data of the vehicle charger is collected by sensors, and the load mutation frequency and power fluctuation amplitude are extracted from the real-time operating data based on the sliding window algorithm.

[0059] Among them, the Multi-Functional Integrated Terminal (MFIT) can refer to a composite edge sensing device that uses a smart light pole as a physical carrier and integrates a high-definition camera, a voice broadcasting module, an LED display, a one-button call button, a WiFi AP module, a smart lighting controller, a weather sensor, a smart water and electricity pile, and an IoT communication module. Its deployment density is dynamically configured according to the size of the campsite. The coverage radius of a single pole is 50-80 meters for small campsites, 60-100 meters for medium campsites, and large campsites are deployed in a hierarchical manner according to the core area and the edge area.

[0060] Video image data is captured in real time by the high-definition camera on the terminal, with a resolution of 2560×1440@25fps, and supports smart265 encoding and ROI enhancement; physiological health data, including heart rate, blood oxygen saturation and blood pressure, are continuously collected and reported periodically by the IP68 waterproof smart bracelet worn by tourists.

[0061] Location data is obtained by the wristband through GPS + Beidou + LBS multi-mode positioning technology. A single positioning base station supports concurrent detection of 500 electronic tags, with a detection range of up to 100 meters in open areas.

[0062] In this step, when tourists enter the campsite, management personnel distribute wearable devices (smart bracelets) to them and bind their identity information. After the wearable device is powered on, it sends an entry registration request to the server. This request carries the wearable device identifier (device MAC address) and the tourist's identity information. Upon receiving the entry registration request, the server creates a tourist profile, which contains a mapping relationship between the device identifier and the tourist's identity information.

[0063] After visitors enter, wearable devices continuously collect physiological data for 10 minutes while at rest. The server calculates the visitor's mean resting heart rate, normal range for blood oxygen saturation, and normal range for blood pressure, storing this data as the visitor's personalized health baseline in a cloud-based visitor profile database. Subsequent physiological health abnormality scoring is based on this personalized baseline rather than a uniform fixed threshold, improving the accuracy of abnormality detection and effectively reducing false alarms caused by individual differences.

[0064] Step 2: Perform multi-task parallel inference on the video image data to obtain abnormal behavior detection results, fire and smoke recognition results, and crowd density distribution map;

[0065] In this step, performing multi-task parallel inference on the video image data to obtain abnormal behavior detection results, fire and smoke identification results, and crowd density distribution map specifically includes:

[0066] Video image frames are input into a multi-task convolutional neural network. This network includes a shared backbone and three detection heads connected to it: a behavior detection head, a fire-smoke detection head, and a density estimation head. The shared backbone uses a deep residual network architecture to extract multi-scale feature maps from the images. These extracted feature maps are then fed into the three independent detection heads for task-specific inference. By using the feature extraction from the shared backbone, the processing latency per frame is significantly reduced compared to using three independent models to infer from the same image frame separately.

[0067] The behavior detection head outputs abnormal behavior detection results. The behavior detection head uses an anchor-based target detection sub-network to detect and estimate the pose of human targets in video frames, and identifies abnormal behaviors such as falling (abnormal changes in the aspect ratio of the human body), fighting (high-frequency overlapping movement of multiple targets), crossing boundaries (targets crossing the preset electronic fence boundary line), and crowd gathering (the number of targets in the area exceeds a set density threshold), and outputs the bounding box coordinates, category label, and confidence score of the detected targets.

[0068] The fire and smoke detection head outputs fire and smoke recognition results, which include flame confidence and smoke confidence. The fire and smoke detection head performs color and texture feature analysis on the extracted feature map, captures the dynamic characteristics of the flame through temporal inter-frame difference, and outputs the confidence scores for detected flame and smoke in the current frame, respectively. The confidence scores range from 0 to 1.

[0069] The density estimation head outputs a population density distribution map for each monitored area. Based on a convolutional density estimation algorithm, the density estimation head maps video frames to a density map scaled proportionally to the original resolution. The value of each pixel in the density map reflects the population density of the corresponding area. By integrating the density map by region, the estimated number of people and the density level for each preset monitored area are obtained.

[0070] Among them, multi-task parallel inference can refer to running multiple dedicated detection heads in parallel on the same neural network backbone structure, sharing the underlying visual feature extraction capability, and avoiding redundant calculations; abnormal behavior detection results can refer to the structured judgment results output after analyzing the target human posture, motion trajectory and interaction relationship in video frames, including at least one of falling, fighting, crossing boundaries and gathering of people; fire and smoke recognition results can refer to the flame confidence and smoke confidence output after jointly modeling the flame pixel area and smoke diffusion pattern in video frames; and the crowd density distribution map can refer to the two-dimensional spatial density heat map output on a monitoring area basis, where the value of each pixel represents the weighted statistics of the number of people passing through the corresponding geographic grid per unit time.

[0071] This application extracts common visual features from video images through a shared backbone network, and uses behavior detection heads, fire and smoke detection heads, and density estimation heads to complete abnormal behavior recognition, fire and smoke probability assessment, and crowd density modeling, respectively, achieving one-time analysis of multi-dimensional safety elements in video data. On this basis, the structured abnormal behavior information output by the behavior detection head, the two-dimensional confidence index output by the fire and smoke detection head, and the spatial density distribution map output by the density estimation head together constitute a standardized data interface that can be directly consumed by the subsequent fusion and scoring module. This enables the system to simultaneously acquire three core indicators—personal behavior risk, fire hazard risk, and spatial carrying capacity risk—within millisecond response time, significantly improving the completeness, timeliness, and interpretability of campsite safety situation awareness.

[0072] Step 3: Perform multi-source fusion processing on the physiological health data, location data, abnormal behavior detection results, fire and smoke identification results, and crowd density distribution map to generate a comprehensive regional risk score;

[0073] In this step, the physiological health data, location data, abnormal behavior detection results, fire and smoke identification results, and crowd density distribution map are fused from multiple sources to generate a comprehensive regional risk score. Specifically, this includes:

[0074] First, a personalized anomaly score is assigned to the physiological health data reported by each wearable device. Based on the visitor's baseline health data (including resting heart rate, normal blood oxygen saturation range, and normal blood pressure range) stored during device registration, the server calculates the deviation of the currently reported heart rate, blood oxygen saturation, and blood pressure values ​​from the baseline. These deviations are then normalized and weighted to obtain a single device's physiological health anomaly score, H_score, with a weighting of 0.4 for heart rate anomalies, 0.4 for blood oxygen anomalies, and 0.2 for blood pressure anomalies.

[0075] Secondly, the location data of wearable devices is mapped to the corresponding monitoring areas to obtain the tourist distribution vector within each area. Each monitoring area corresponds to the coverage area of ​​one or more multi-functional integrated terminals. Based on the location coordinates reported by the wearable devices, the server assigns tourists to their respective monitoring areas, counts the number of tourists in each area and the number of tourists with physiological health abnormalities, and forms the regional tourist distribution vector.

[0076] Next, based on the reasoning results of step S102, the environmental risk component E_score for each monitoring area is calculated. The environmental risk component is obtained by weighted summation of the mean confidence score of abnormal behavior detection (weight 0.3), flame confidence score (weight 0.4), smoke confidence score (weight 0.1), and the normalized value of the regional population density level (weight 0.2).

[0077] Finally, the mean physiological health abnormality score H_score of all tourists in the region and the environmental risk component E_score are weighted and summed according to preset weights (H_score weight 0.45, E_score weight 0.55) to obtain the regional comprehensive risk score R_score, which ranges from 0 to 1.

[0078] This application constructs a regional comprehensive risk scoring mechanism with spatiotemporal consistency and multi-source representativeness by spatially aligning, semantically normalizing, and weighting individual physiological health data collected by wearable devices with group environmental situation data obtained from video analysis. By leveraging the accurate characterization of individual tourist status through abnormal physiological health scores, combined with the joint modeling of the triple threats of behavior, fire, and crowd flow in the monitored area by environmental risk components, and relying on preset weights to flexibly adjust the contribution of risk elements, the system can significantly improve the sensitivity and discriminativeness of the regional risk score when a tourist's sudden physical discomfort coincides with abnormal surrounding environment (such as crowd gathering or initial fire). This effectively overcomes the false alarm and missed alarm problems caused by information fragmentation in traditional single-source alarm systems, providing a reliable data foundation for hierarchical alarms and accurate responses.

[0079] Step 4: Determine the alarm trigger level based on the comparison result between the comprehensive risk score of the area and the preset risk threshold;

[0080] In this step, determining the alarm trigger level based on the comparison result between the comprehensive risk score of the area and the preset risk threshold specifically includes: when the comprehensive risk score of the area is greater than or equal to the first threshold and less than the second threshold, the alarm trigger level is determined to be a level one alarm, triggering notification to nearby security personnel and retrieval of camera footage; when the comprehensive risk score of the area is greater than or equal to the second threshold, the alarm trigger level is determined to be a level two alarm, triggering full-area broadcasting, lighting linkage, and retrieval of multiple camera footage; wherein, the second threshold is greater than the first threshold.

[0081] The preset risk thresholds include a first threshold and a second threshold, both of which are static configuration parameters, with the second threshold being greater than the first threshold; the alarm triggering levels include two discrete states: a first-level alarm and a second-level alarm; the comparison result can refer to the classification conclusion obtained by successively determining the numerical magnitude of the regional comprehensive risk score and the two thresholds; this step does not involve the generation of new data, but only executes the condition judgment logic.

[0082] This application can determine the alarm trigger level by, for example, sequentially comparing the regional comprehensive risk score with a first threshold and a second threshold using an if-else branch structure; it can also determine the alarm trigger level by mapping the score to a predefined level code using a lookup table method; further, it can determine the alarm trigger level by using a finite state machine to transition between three risk ranges (low, medium, and high). This application obtains a response strategy identifier that matches the severity of the risk based on any of the above methods.

[0083] This application achieves precise mapping of alarm trigger levels through a two-tiered risk assessment structure consisting of a first and a second threshold, combined with real-time input of regional comprehensive risk scores. By leveraging first-level alarms to target nearby security resources and key viewpoints, it improves the efficiency of verification and closed-loop handling capabilities for medium-risk events. Furthermore, by utilizing second-level alarms to trigger broadcasts, lighting, and multiple video feeds across the entire area, it enhances the breadth of situational awareness, on-site visibility, and the effectiveness of personnel guidance for high-risk events. On this basis, the relative magnitudes of the two thresholds ensure the hierarchical clarity and execution certainty of the response strategy, thus systematically solving technical problems in traditional campsite security management such as the mismatch between response intensity and risk level, inefficient resource allocation, and poor visibility of nighttime operations. This significantly improves the rationality, timeliness, and effectiveness of emergency response.

[0084] Step 5: Based on the alarm trigger level, generate and send a linkage response instruction set to the multi-functional integrated terminal. The linkage response instruction set includes at least one of camera scheduling instructions, lighting control instructions, broadcast instructions, and security notification instructions.

[0085] Before collecting multimodal sensor data from the multifunctional integrated terminal deployed at the campsite, this step further includes: receiving an entry registration request sent by a wearable device, the entry registration request carrying a wearable device identifier; establishing a visitor profile based on the wearable device identifier, the visitor profile containing a mapping relationship between the device identifier and visitor identity information and visitor health baseline data; sending the visitor health baseline data to a cloud-based visitor profile database for personalized baseline comparison during subsequent physiological health abnormality scoring; the multifunctional integrated terminal is equipped with a unique physical identification number. After generating and sending a linkage response instruction set to the multifunctional integrated terminal, the method further includes: receiving a call event message reported by the multifunctional integrated terminal in response to the emergency call button triggering, the call event message containing the physical number of the triggering terminal; querying a pre-stored terminal location coordinate mapping table based on the physical number to determine the call location; and based on the call location, pushing a handling notification containing location coordinates to the nearest security personnel terminal and sending a two-way intercom establishment instruction to the triggering terminal.

[0086] Among them, the linkage response instruction set can refer to a set of control command sets with clear execution subjects, objects of action, operation types, and timing constraints; the camera scheduling instruction is used to control the pan / tilt rotation angle, focal length zoom factor, and video stream push target address of the camera of the specified terminal; the lighting control instruction is used to adjust the brightness percentage, switch state, and time segment mode of the LED light group of the specified terminal; the broadcast instruction is used to send a preset voice text or real-time voice stream to the voice broadcast module of the specified terminal; the security notice instruction is used to push a structured event notice containing location coordinates to the mobile terminal of the specified security personnel; at least one can mean that the instruction set must include one or more items, and each instruction can be dynamically combined according to the alarm level.

[0087] For a level-1 alarm, the server generates a linkage response instruction set containing the following instructions: push an alarm notice (including the alarm area number and comprehensive risk score) to the terminal of the security personnel closest to the alarm area; send a camera image retrieval instruction to the multi-functional integrated terminal covering the alarm area, and push the monitoring image of this area to the management platform.

[0088] For a level-2 alarm, the server generates a linkage response instruction set containing the following instructions: push an emergency alarm notice to all online security personnel terminals; send a lighting control instruction (illuminance increased to the maximum value) to the multi-functional integrated terminal covering the alarm area and its adjacent areas; send a broadcast instruction to the multi-functional integrated terminal within the alarm area, and play a pre-recorded safety evacuation prompt through the voice broadcast module; at the same time, push multiple live monitoring images to the management platform.

[0089] In this application, by taking the video image data, physiological health data, and location data collected by the multi-functional integrated terminal as multi-source inputs, extracting structured semantic information through multi-task parallel inference, generating a regional comprehensive risk score through spatio-temporal alignment and weighted fusion, then grading and determining the alarm level according to a preset threshold, and finally driving multi-category terminal devices such as cameras, lighting, broadcasting, and notifications to execute differential linkage responses, it realizes a full-link automation closed-loop of perception - analysis - decision - response; with the help of a multi-modal data cross-verification mechanism, it significantly reduces the false alarm rate of single sensors; relying on a multi-task shared backbone structure, it ensures low-latency inference performance in high-throughput scenarios; based on a hierarchical response strategy based on risk scores, it avoids resource waste caused by indiscriminate alarms; and the cross-device collaborative control driven by the instruction set shortens the emergency response from the minute level to the second level, effectively solving the core technical problems of low intelligence, response lag, and system fragmentation in campsite safety management.

[0090] Embodiment 2: A campsite intelligent safety management device

[0091] As Figure 2As shown, the intelligent security management device includes: a data acquisition module, a reasoning and analysis module, a fusion scoring module, a level determination module, and a linkage response module, wherein:

[0092] The data acquisition module is used to collect multimodal sensor data from a multifunctional integrated terminal deployed at the campsite. The multimodal sensor data includes video image data, physiological health data and location data reported by wearable devices worn by tourists. The data output by this module is in the format of structured time series frames, which includes timestamps, raw values ​​of multi-channel video / physiological / location / environment, and preliminary extracted abnormal behavior tags, fire and smoke confidence scores, and average crowd density fields. The frame header contains a CRC16 checksum to ensure transmission integrity.

[0093] The inference analysis module is used to perform multi-task parallel inference on the video image data to obtain abnormal behavior detection results, fire and smoke recognition results, and crowd density distribution map. This module can be integrated into a GPU-accelerated environment and supports parallel feature extraction. The abnormal behavior detection result is defined as a JSON structure containing bounding box coordinates, behavior category labels, and confidence scores. The fire and smoke recognition result is defined as a pair of double floating-point values ​​consisting of flame confidence and smoke confidence. The crowd density distribution map is defined as a two-dimensional heatmap matrix with the same spatial resolution as the input image, where each pixel value represents the relative intensity of crowd density within the corresponding geographic grid.

[0094] The fusion scoring module is used to perform multi-source fusion processing on the physiological health data, location data, abnormal behavior detection results, fire and smoke identification results, and crowd density distribution map to generate a regional comprehensive risk score. This module runs on a high-performance CPU cluster and supports multi-threaded spatiotemporal alignment. The multi-source fusion processing adopts a spatiotemporal alignment rule to unify to a 5-second sliding time window and a 10m × 10m geographic grid. The physiological health abnormality score is the weighted average of the physiological abnormality scores of all tourists in the region. The environmental risk component is a weighted synthesis of the behavioral risk sub-component, the fire risk sub-component, and the crowd risk sub-component. The regional comprehensive risk score ranges from [0, 100] with a precision to one decimal place.

[0095] The level determination module is used to determine the alarm trigger level based on the comparison result of the comprehensive risk score of the area and the preset risk threshold. This module adopts a memory-resident lookup table mechanism to achieve millisecond-level response. The first threshold and the second threshold are non-negative real numbers, and their sum does not constitute a constraint, but strictly satisfies that the second threshold > the first threshold. The alarm trigger level output is an enumerated variable, which takes the value of a first-level alarm or a second-level alarm, and is accompanied by a corresponding handling strategy identifier.

[0096] The linkage response module is used to generate and send a linkage response instruction set to the multi-functional integrated terminal based on the alarm trigger level. This module adopts a heterogeneous computing architecture, with the instruction template matching part deployed in Redis cache and the instruction serialization and encryption part deployed in a dedicated security coprocessor. The linkage response instruction set includes at least one of camera scheduling instructions, lighting control instructions, broadcast instructions, and security notification instructions. The instruction set is encrypted and sent via HTTPS POST, and the terminal returns the execution status including the success_flag and execution_latency_ms fields.

[0097] The working process and principle of this device are as follows: the data acquisition module continuously acquires multi-source sensor data and completes primary feature extraction, outputting structured time-series frames to the inference analysis module; the latter completes multi-task parallel inference, outputting structured abnormal behavior, two-dimensional fire and smoke confidence scores, and spatial density heatmaps, and transmits the results to the fusion scoring module; this module integrates physiological health abnormality scores and environmental risk components to generate a comprehensive regional risk score, which is then sent to the level determination module; the latter determines the alarm level based on preset thresholds and sends the results to the linkage response module; this module integrates alarm levels and regional spatial attributes to generate an instruction set containing device ID, operation type, parameter values, and execution time limits, which is then sent to the corresponding multi-functional integrated terminal to drive cameras, lighting, broadcasting, and notification devices to perform precise actions, forming an autonomous closed-loop control system covering the entire chain of perception, analysis, decision-making, execution, and feedback.

[0098] Example 3

[0099] like Figure 3 As shown, this application also provides a computer-readable storage medium, which may be included in the electronic device described in the above embodiments, or it may exist independently and not assembled into the electronic device. The computer-readable storage medium stores one or more programs that, when used by one or more processors, execute the deep learning-based medical image defect detection method described in this application.

[0100] Example 4:

[0101] like Figure 4 As shown, this application provides an electronic device that can be a server, terminal, or a security processor in a Trusted Execution Environment (TEE). The electronic device includes a processor, memory, input / output interfaces (I / O), and a communication interface.

[0102] The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the input / output interface. The processor provides computational and control capabilities, supporting instruction-level parallelism, out-of-order execution, and hardware-assisted cryptographic acceleration instruction set extensions. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a constant data table; the constant data table includes the original S-box mapping data for symmetric cryptographic algorithms, in address-value pairs, where the address corresponds to input 'a' and the value corresponds to output 'b', satisfying Sa=b. The internal memory provides the environment for the operating system and computer programs in the non-volatile storage media to run. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals or security modules via a network or bus, supporting PCIe, USB, SPI, or TrustZone secure channel protocols. When executed by the processor, the computer program implements a data processing method for cryptographic operations.

[0103] It should be noted that the above detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0104] In the detailed description above, reference has been made to the accompanying drawings, which form part of this document. In the drawings, similar symbols typically identify similar parts unless the context otherwise indicates otherwise. The illustrated embodiments described in the detailed specification, drawings, and claims are not intended to be limiting. Other embodiments may be used and other changes may be made without departing from the spirit or scope of the subject matter presented herein.

[0105] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A camping site intelligent safety management method, characterized in that, The method includes: Multimodal sensor data is collected from multifunctional integrated terminals deployed at the campsite. The multimodal sensor data includes video image data, physiological health data and location data reported by wearable devices worn by tourists. Perform multi-task parallel inference on the video image data to obtain abnormal behavior detection results, fire and smoke identification results, and crowd density distribution map; The physiological health data, location data, abnormal behavior detection results, fire and smoke identification results, and crowd density distribution map are fused from multiple sources to generate a comprehensive regional risk score. The alarm trigger level is determined based on the comparison between the comprehensive risk score of the region and the preset risk threshold. Based on the alarm trigger level, a linkage response instruction set is generated and sent to the multi-functional integrated terminal. The linkage response instruction set includes at least one of camera scheduling instructions, lighting control instructions, broadcast instructions, and security notification instructions.

2. The method of claim 1, wherein, The process of performing multi-task parallel inference on the video image data to obtain abnormal behavior detection results, fire and smoke identification results, and crowd density distribution maps includes: The video image data is input into a multi-task convolutional neural network, which includes a shared backbone network and a behavior detection head, a fire and smoke detection head, and a density estimation head connected to the shared backbone network. The behavior detection head outputs abnormal behavior detection results, and the abnormal behavior includes at least one of falling, fighting, crossing boundaries, and gathering of people. The fire and smoke detection head outputs fire and smoke identification results, wherein the fire and smoke identification results include flame confidence and smoke confidence. The density estimation head outputs a distribution map of pedestrian density in each monitored area.

3. The method of claim 2, wherein, The process of fusing the physiological health data, location data, abnormal behavior detection results, fire and smoke identification results, and crowd density distribution map into a multi-source data to generate a comprehensive regional risk score includes: Anomaly scoring is performed on the physiological health data reported by each wearable device, including heart rate, blood oxygen saturation, and blood pressure. The location data of the wearable device is mapped to the corresponding monitoring area to obtain the tourist distribution vector within the area; Based on the abnormal behavior detection results, flame confidence, smoke confidence, and population density distribution map within the area, the environmental risk component of each monitoring area is calculated; The physiological health abnormality score and the environmental risk component are weighted and summed according to preset weights to obtain the regional comprehensive risk score.

4. The method of claim 1, wherein, The step of determining the alarm trigger level based on the comparison result of the regional comprehensive risk score and the preset risk threshold includes: When the comprehensive risk score of the area is greater than or equal to the first threshold and less than the second threshold, the alarm triggering level is determined to be a level one alarm, triggering the notification of nearby security personnel and the retrieval of camera footage. When the comprehensive risk score of the area is greater than or equal to the second threshold, the alarm trigger level is determined to be a level 2 alarm, triggering full-area broadcast, lighting linkage, and retrieval of multiple camera images; wherein, the second threshold is greater than the first threshold.

5. The method of claim 1, wherein, Before acquiring multimodal sensing data from the multifunctional integrated terminal deployed at the campsite, the process also includes: Receive an entry registration request sent by a wearable device, the entry registration request carrying the wearable device identifier; Based on the wearable device identifier, a tourist profile is established, which includes the mapping relationship between the device identifier and the tourist's identity information, as well as the tourist's health baseline data. The tourist health baseline data will be sent to the cloud-based tourist profile database for personalized baseline comparison during subsequent physiological health abnormality scoring.

6. The method of claim 1, wherein, The multi-functional integrated terminal is equipped with a unique physical identifier. After generating and issuing a linkage response instruction set to the multi-functional integrated terminal, the method further includes: Receive a call event message reported by the multi-functional integrated terminal in response to the triggering of the emergency call button, the call event message containing the physical number of the triggering terminal; The location of the call is determined by querying a pre-stored terminal location coordinate mapping table based on the physical number. Based on the location of the call, a notification containing the location coordinates is pushed to the nearest security personnel's terminal, and a two-way intercom establishment command is sent to the triggering terminal.

7. A camping site intelligent safety management device, characterized in that, The device includes: The data acquisition module is used to collect multimodal sensor data from a multifunctional integrated terminal deployed at the campsite. The multimodal sensor data includes video image data, physiological health data and location data reported by wearable devices worn by tourists. The reasoning and analysis module is used to perform multi-task parallel reasoning on the video image data to obtain abnormal behavior detection results, fire and smoke recognition results, and crowd density distribution map; The fusion scoring module is used to perform multi-source fusion processing on the physiological health data, location data, abnormal behavior detection results, fire and smoke identification results, and crowd density distribution map to generate a comprehensive regional risk score. The level determination module is used to determine the alarm trigger level based on the comparison result between the comprehensive risk score of the area and the preset risk threshold. The linkage response module is used to generate and send a linkage response instruction set to the multi-functional integrated terminal based on the alarm trigger level.

8. The camping ground intelligent safety management device according to claim 7, characterized in that, The multi-functional integrated terminal is equipped with a unique physical number identifier, which is used to query a pre-stored terminal location coordinate mapping table to obtain the current location coordinates of the multi-functional integrated terminal.

9. An electronic device comprising a memory and a processor, characterized in that The memory stores a computer program, and when the processor executes the computer program, it implements the campsite intelligent safety management method as described in any one of claims 1 to 6.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed by the processor, it implements the campsite intelligent safety management method as described in any one of claims 1 to 6.

Images