Smart park safety management method based on cloud edge cooperative computing

Abstract

The invention relates to the technical field of smart park management and cloud computing, in particular to a smart park safety management method based on cloud edge cooperative computing. According to the method, the multi-source data is acquired through the terminal device which is calibrated periodically, and the lightweight AI model of the edge node performs preliminary analysis according to the screening threshold and uploads the suspected hidden danger data. And the cloud platform performs deep analysis by using the prediction model, evaluates the security situation level and triggers early warning and emergency response. Meanwhile, the cloud model recognition accuracy, the edge model false alarm rate and the actual hidden danger proportion in the uploaded data are comprehensively calculated, the safety management characterization value is generated, the system state is judged based on the safety management characterization value, then the edge screening threshold value and the data uploading strategy are adaptively and dynamically adjusted, and a continuously-optimized management closed loop is formed. The safety management efficiency of the smart park is improved.

Description

Technical Field

[0001] This invention relates to the technical field of smart park management and cloud computing, and in particular to a smart park security management method based on cloud-edge collaborative computing. Background Technology

[0002] With the acceleration of urbanization and the popularization of Internet of Things (IoT) technology, smart parks, as integrated complexes combining office, residential, production, and leisure functions, are facing increasingly diversified and sophisticated security management needs. Currently, smart park security management mainly relies on traditional security systems and a single computing model.

[0003] Cloud-edge collaborative computing technology combines the powerful computing power and massive storage resources of the cloud with the real-time data processing capabilities of edge nodes, effectively compensating for the shortcomings of single computing models. However, the application of existing cloud-edge collaborative technologies in smart park security management is still in its early stages. It has not yet formed a complete collaborative mechanism to meet the personalized needs of park security management, such as multi-scenario adaptation, multi-system linkage, and hierarchical early warning, thus failing to fully leverage the advantages of cloud-edge collaboration.

[0004] Chinese Patent Publication No. CN120163694A discloses a smart park security management system, including a data sensing module, an edge computing module, an intelligent analysis module, and a command execution module. This invention uses the data sensing module to collect relevant security data from multiple sources in real time, and the edge computing module to preprocess the collected data locally, enabling the perception of the overall security situation within the park. The intelligent analysis module then analyzes and processes the collected data, generating tiered early warning signals. Finally, the command execution module executes the corresponding early warning and emergency response operations.

[0005] It is evident that existing technologies have the following problems: due to the slow response and poor data collaboration in smart park management solutions, the efficiency of commanding park security management is relatively low. Summary of the Invention

[0006] To address this, the present invention provides a smart park security management method based on cloud-edge collaborative computing, which overcomes the problem of low efficiency in commanding park security management caused by the slow response and poor data collaboration in existing smart park management schemes.

[0007] To achieve the above objectives, this invention provides a smart park security management method based on cloud-edge collaborative computing, comprising:

[0008] Multiple periodically calibrated terminal sensing devices are used to collect multi-source data on park security, including video image data, environmental sensor data, personnel access data, and equipment operation data;

[0009] The edge node analyzes by using a local lightweight AI model, and screens out suspected hidden danger data based on a set screening threshold;

[0010] The edge node uploads the suspected hidden danger data and the key data to the cloud management platform based on an uploading strategy, the cloud management platform uses a prediction model to deeply analyze the data uploaded by the edge node, to evaluate the security situation level of the managed park;

[0011] Based on the security situation level, the managed park is warned, and the corresponding terminal equipment and the park security subsystem are started for emergency response;

[0012] Based on the recognition accuracy of the prediction model of the cloud management platform for security hidden dangers, the false positive rate of the lightweight AI model, and the proportion of actual hidden dangers in the suspected hidden danger data uploaded by the edge node, a security management representation value is calculated;

[0013] Based on the security management representation value, the security management state of the smart park is determined, the screening threshold is adjusted based on the security management state, and the uploading strategy is adjusted based on the security management state after the screening threshold is adjusted.

[0014] Further, the method further comprises: the cloud management platform periodically optimizes and updates the prediction model by using the suspected hidden danger data uploaded by the edge node and historical multi-source data; the prediction model after optimization and update is lightweight processed to generate a lightweight model adapted to the computing power of the edge node, and the lightweight model is distributed to the edge node through a secure channel.

[0015] Further, the method further comprises: the cloud management platform periodically adjusts the data uploading strategy and the edge node computing task allocation according to the edge node computing power load and the network bandwidth state.

[0016] Further, the process of determining the security management state of the smart park based on the security management representation value comprises: if the security management representation value is less than a preset representation value, it is determined that the security management state is unqualified; if the security management representation value is greater than or equal to the preset representation value, it is determined that the security management state is qualified.

[0017] Further, the process of adjusting the screening threshold based on the security management state comprises: if the security management state is unqualified, calculating the difference between the false positive rate of the lightweight AI model and a preset false positive rate; if the difference is greater than a preset difference, adjusting the screening threshold based on the ratio of the difference to the preset difference.

[0018] Further, the process of adjusting the screening threshold based on the ratio of the difference to the preset difference comprises: increasing the screening threshold based on the ratio of the difference to the preset difference, and the increase amplitude of the screening threshold is proportional to the ratio.

[0019] Further, the process of adjusting the upload strategy based on the adjusted security management state after adjusting the screening threshold comprises: if the security management state after adjusting the screening threshold is unqualified, repeatedly adjusting the screening threshold at least once until the number of adjustments is less than a preset number of times and the security management state is qualified or the number of adjustments is equal to the preset number of times; if the security management state after stopping adjustment is unqualified, obtaining data upload delay times at multiple historical moments; calculating the average value of the multiple data upload delay times, and if the average value is greater than a preset average value, adjusting the key frame interval in the upload strategy based on the difference between the average value and the preset average value.

[0020] Further, the process of adjusting the key frame interval in the upload strategy based on the difference between the average value and the preset average value comprises: increasing the key frame interval in the upload strategy based on the difference between the average value and the preset average value, and the increase amplitude of the key frame interval is proportional to the difference.

[0021] Further, the method further comprises: increasing the compression rate of the background area in the video image data based on the ratio of the key frame interval to a preset interval, and the increase amplitude of the compression rate is proportional to the ratio.

[0022] Further, the method further comprises: if the security management state after adjusting the compression rate of the background area in the video image data is unqualified, obtaining security management characteristic values at multiple historical moments; calculating the average value and variance of the multiple security management characteristic values, and if the average value is less than a preset average value and the variance is less than a preset variance, reducing the calibration period of the terminal perception device based on the ratio of the average value to the preset average value, and the reduction amplitude of the calibration period is inversely proportional to the ratio.

[0023] Compared with the prior art, the beneficial effects of the present application are that the present application collects multi-source data by the terminal device with periodic calibration, and the lightweight AI model of the edge node performs preliminary analysis according to the screening threshold and uploads suspected hidden danger data. The cloud platform uses a prediction model for in-depth analysis, evaluates the security situation level and triggers early warning and emergency response. At the same time, the present application generates a security management characteristic value by comprehensively calculating the cloud model recognition accuracy, the edge model false positive rate and the proportion of actual hidden dangers in the uploaded data, and determines the system state based on the security management characteristic value, and then dynamically adjusts the edge screening threshold and the data upload strategy to form a continuously optimized management closed loop. The present application improves the safety management efficiency of the smart park.

[0024] Further, the present application periodically optimizes and updates the prediction model by using the suspected hidden danger data uploaded by the edge node and the historical multi-source data, which can make the prediction model more effectively analyze and make the analysis result more accurate, thereby further improving the early warning efficiency of the management park and further improving the safety management efficiency of the smart park.

[0025] Further, the application adjusts the data uploading strategy and the edge node computing task allocation based on the computing power load and network bandwidth of the edge node, can more effectively perform data uploading and task allocation, makes the work efficiency of the cloud management platform and the edge node more effective, and thus further improves the safety management efficiency of the smart park.

[0026] Further, the application determines whether the safety management state is qualified based on the comparison result of the safety management characteristic value and the preset characteristic value, can more accurately determine the safety management state, thereby making subsequent adjustment more effective, and thus further improving the safety management efficiency of the smart park.

[0027] Further, the application determines the reason why the safety management state is unqualified based on the difference between the false positive rate of the lightweight AI model and the preset false positive rate, can determine the reason why the safety management state is unqualified based on whether the AI model of the edge node produces a large number of false positives in preliminary screening, thereby making subsequent adjustment more effective, and thus further improving the safety management efficiency of the smart park.

[0028] Further, the application adjusts the screening threshold based on the ratio of the difference value and the preset difference value, can make non-hidden danger data not be uploaded, thereby making the analysis of the cloud management platform more accurate, and thus further improving the safety management efficiency of the smart park.

[0029] Further, the application determines the reason why the safety management state is unqualified based on the average value of the data uploading delay time at multiple historical moments, can determine whether the safety management state is unqualified due to insufficient bandwidth for data uploading, thereby making subsequent adjustment more effective, and thus further improving the safety management efficiency of the smart park.

[0030] Further, the application adjusts the key frame interval in the uploading strategy based on the difference between the average value and the preset average value, can make key data uploading more effective, thereby making the cloud management platform more accurate and more timely in analysis, and thus further improving the safety management efficiency of the smart park.

[0031] Further, the application increases the compression rate of the background area in the video image data based on the ratio of the key frame interval and the preset interval, can more effectively upload hidden danger key information, thereby making the cloud management platform more effective in analysis, and thus further improving the safety management efficiency of the smart park.

[0032] Further, the application adjusts the calibration cycle of the terminal perception device based on the average value of the safety management characteristic values at multiple historical moments, can more accurately obtain data of a specific area or type, thereby making the analysis of the edge node and the cloud management platform more accurate, and thus further improving the safety management efficiency of the smart park. BRIEF DESCRIPTION OF DRAWINGS

[0033] Figure 1 A structure schematic diagram of a smart park safety management system based on cloud edge collaborative computing according to an embodiment of the present application;

[0034] Figure 2 A step flow chart of a smart park safety management method based on cloud edge collaborative computing according to an embodiment of the present application;

[0035] Figure 3 A step flow chart of a step of determining based on a comparison result of a safety management characteristic value and a preset characteristic value according to an embodiment of the present application;

[0036] Figure 4 A step flow chart of a step of determining based on a safety management state after adjusting a screening threshold according to an embodiment of the present application. DETAILED DESCRIPTION

[0037] In order to make the objects and advantages of the present application clearer, the present application will be further described below in conjunction with embodiments. It should be understood that the specific embodiments described herein merely serve to explain the present application and do not limit the present application.

[0038] The preferred embodiments of the present application will be described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments merely serve to explain the technical principles of the present application and are not intended to limit the protection scope of the present application.

[0039] It should be noted that in the description of the present application, unless explicitly defined and limited, the terms “mounting”, “connection” and “linking” should be understood in a broad sense, for example, it can be fixed connection, or detachable connection, or integral connection; it can be mechanical connection, or electrical connection; it can be direct connection, or indirect connection through an intermediate medium, or internal connection of two elements. Those skilled in the art can understand the specific meaning of the above terms in the present application according to the specific circumstances.

[0040] Please refer to Figure 1 As shown in FIG. 1, which is a structure schematic diagram of a smart park safety management system based on cloud edge collaborative computing according to an embodiment of the present application.

[0041] The system comprises a collection unit, a screening unit, an evaluation unit, a processing unit, a calculation unit and an analysis unit.

[0042] The collection unit collects multi-source data of park safety by using a plurality of periodically calibrated terminal sensing devices, including video image data, environmental sensing data, personnel access data and equipment operation data;

[0043] The screening unit is connected with the collection unit, which is used for the edge node to analyze by using the local lightweight AI model, and screen out suspected hidden danger data based on the set screening threshold;

[0044] The evaluation unit is connected with the screening unit, which is used for the edge node to upload the suspected hidden danger data and key data to the cloud management platform based on the uploading strategy, and the cloud management platform uses the prediction model to deeply analyze the data uploaded by the edge node, so as to evaluate the security situation level of the managed park;

[0045] The processing unit is connected with the evaluation unit, which is used for prewarning the managed park based on the security situation level, and starting the emergency response of the corresponding terminal equipment and the park security subsystem in linkage;

[0046] The calculation unit is connected with the screening unit, the evaluation unit and the processing unit respectively, which is used for calculating the security management representation value based on the recognition accuracy of the prediction model of the cloud management platform for security hidden danger, the false positive rate of the lightweight AI model and the proportion of actual hidden danger in the suspected hidden danger data uploaded by the edge node;

[0047] The analysis unit is connected with the calculation unit, which is used for determining the security management state of the smart park based on the security management representation value, adjusting the screening threshold based on the security management state, and adjusting the uploading strategy based on the security management state after adjusting the screening threshold.

[0048] Specifically, first, the cloud management platform is built, including deploying cloud server cluster, distributed database, big data analysis engine and AI model training module, the cloud management platform is responsible for global security situation analysis, AI model training and updating, edge node resource scheduling, security strategy formulation and issuing; at the same time, the cloud security protection module is built, realizing data encryption storage, access permission control and network attack protection.

[0049] Secondly, the edge node cluster is deployed, including: according to the geographical zoning of the park such as office area, residential area, production area and public leisure area, deploying edge node, each edge node is configured with edge gateway, edge server and local cache module; the edge node and the cloud management platform establish communication connection through 5G or optical fiber network, and at the same time, the edge node and the terminal sensing equipment in the jurisdiction area establish local communication through protocol; the edge nodes realize interconnection through local area network, forming edge collaborative network.

[0050] The terminal perception device deployment includes: deploying terminal perception devices in key positions in the park, including video monitoring cameras, infrared sensors, smoke sensors, temperature and humidity sensors, access control card readers, electronic fence detectors, etc.; and numbering and parameter configuring the terminal perception devices, specifying the attribution edge node and data collection frequency of each device, to ensure that the devices can stably collect and upload data.

[0051] Specifically, the multi-source data collected by each terminal perception device for park safety includes video image data such as personnel flow, vehicle driving, facility status, environmental sensing data such as temperature, humidity, smoke concentration, harmful gas concentration, personnel access data such as access card record, face recognition information, device operation data such as security device voltage, current, and running time.

[0052] Specifically, the edge node uses a local lightweight AI model such as a target detection YOLOv5 model and an abnormal behavior recognition model such as a lightweight convolutional neural network to preliminarily analyze the preprocessed data, obtain the confidence of the hidden danger data, and filter out suspected hidden danger data based on a filtering threshold, such as unauthorized personnel entering, smoke concentration exceeding the standard, and device operation anomaly.

[0053] Specifically, the edge node uploads the suspected hidden danger data after preliminary analysis and key core data such as important area video images and personnel dense area sensing data to the cloud management platform in real time; for regular non-key data such as normal environmental sensing data and regular personnel access records, they are stored in the local cache module of the edge node, and the cloud management platform can retrieve them as needed according to the demand. The cloud management platform dynamically adjusts the data upload strategy and edge node computing task allocation according to the computing power load and network bandwidth state of the edge node, to realize the optimal scheduling of cloud and edge resources.

[0054] Specifically, the specific process of the evaluation unit operation is: first, the cloud platform receives data packets from different edge nodes, and annotates all the time and data to a specific position and time axis of a unified park digital map. Each data packet contains: suspected hidden danger metadata, specifically event type, location (edge node ID or region coordinates), timestamp, confidence, and associated original data index such as which frames of which video; key data slices, including key data segments associated with the hidden danger uploaded for verification, such as 10 seconds of video clips before and after triggering the alarm, and 5 minutes of readings of related sensors; context state data, including the load and network status of the edge node, etc.

[0055] Secondly, the video slices, sensor data streams, device states and other information at the same space-time point are input into a multi-modal fusion model such as a multi-modal Transformer, and the model verification outputs a'verified hidden danger event' with a type label and risk attributes; according to the geographical location and time proximity, the related events are clustered, and a time series prediction model such as a long short-term memory network is used to analyze the historical time series and sensor indicators of each region, and output the real-time risk score and short-term risk prediction risk of each region.

[0056] Finally, a dynamic space-time risk map is constructed, in which the key areas, important assets and entrances and exits in the park are taken as nodes, the dynamic space-time risk map is input into a space-time graph neural network, a global situation tensor is output, and a park safety situation index is calculated based on the global situation tensor, and the park safety situation index is mapped to a safety situation level.

[0057] Specifically, the prediction model of the cloud management platform relies on the cloud's strong computing power and a large amount of historical data to optimize the parameters, enhance the features or optimize the structure of the high-precision model to complete the training and optimization, ensuring that the model can accurately identify cross-regional hidden dangers and predict global security trends; then, the cloud performs pruning, quantization, knowledge distillation and other lightweight processing on the optimized high-precision model to generate a lightweight model that adapts to the computing power of the edge node, ensuring that the edge node can run efficiently, and the cloud replaces the old version of the model by issuing the lightweight model to the edge node through a secure channel, thereby upgrading the edge node model.

[0058] Please refer to Figure 2 As shown in the figure, it is a step flow chart of the intelligent park safety management method based on cloud edge collaborative computing according to the embodiment of the application.

[0059] The step flow of the intelligent park safety management based on cloud edge collaborative computing specifically includes:

[0060] S1, a plurality of periodically calibrated terminal sensing devices are used to collect multi-source data of park safety by the acquisition unit, including video image data, environmental sensing data, personnel access data and device operation data;

[0061] S2, the screening unit connected with the acquisition unit analyzes based on the edge node using a local lightweight AI model, and screens out suspected hidden danger data based on the set screening threshold;

[0062] S3, the evaluation unit connected with the screening unit uploads the suspected hidden danger data and key data to the cloud management platform based on the edge node based on the uploading strategy, and the cloud management platform uses a prediction model to deeply analyze the data uploaded by the edge node to evaluate the safety situation level of the managed park;

[0063] S4, based on the security situation level, the processing unit connected with the evaluation unit gives an early warning to the management park, and the corresponding terminal device and the park security subsystem start emergency response;

[0064] S5, through the calculation unit connected with the screening unit, the evaluation unit and the processing unit respectively, based on the prediction model of the obtained cloud management platform, the identification accuracy of the security hidden danger, the false positive rate of the lightweight AI model and the proportion of actual hidden dangers in the suspected hidden danger data uploaded by the edge node, the security management characteristic value is calculated; wherein, the security management characteristic value , wherein a, b, c are the identification accuracy of the prediction model for the security hidden danger, the false positive rate of the lightweight AI model and the proportion of actual hidden dangers in the suspected hidden danger data uploaded by the edge node, respectively, the weight coefficient of the corresponding parameter, which is set based on the actual demand and ensures that the sum of the three weight parameters is 1.

[0065] S6, through the analysis unit connected with the calculation unit, based on the security management characteristic value, the security management state of the smart park is determined, the screening threshold is adjusted based on the security management state, and the uploading strategy is adjusted based on the security management state after adjusting the screening threshold.

[0066] Please refer to Figure 3 , which is the step flow chart of the determination of the embodiment of the application based on the comparison result of the security management characteristic value and the preset characteristic value.

[0067] Specifically, based on the hardware performance limit of the image sensor of the edge camera, such as low-illumination sensitivity, dynamic range, thermal imaging accuracy and actual complex environment fault tolerance requirements, such as distinguishing between intruding personnel and animals, vegetation shaking, adverse weather interference, and combining with a number of historical data obtained by statistical analysis in the historical early warning disposal process, such as false positive rate under different time period illumination conditions, multi-dimensional feature mode of various real intrusion events, time sequence distribution law of environmental interference factors, subsequent corresponding preset or critical parameter value setting is carried out.

[0068] Specifically, the preset characteristic value L0=85%, then the comparison process based on the security management characteristic value L and the preset characteristic value L0 is as follows:

[0069] If the security management characteristic value L is less than the preset characteristic value L0, it is determined that the security management state is unqualified;

[0070] If the security management characteristic value L is greater than or equal to the preset characteristic value L0, it is determined that the security management state is qualified.

[0071] Specifically, in the case of determining that the safety management state is unqualified, the difference between the false positive rate of the lightweight AI model and the preset false positive rate is calculated; if the difference is greater than the preset difference, it means that the edge model produces a large number of false positives in the preliminary screening, resulting in a large amount of non-hidden danger data being uploaded, and the screening threshold is adjusted based on the ratio of the difference to the preset difference, wherein the preset difference is set based on the comprehensive analysis of the false positive rate statistical characteristics of the lightweight AI model in the historical normal operation period and the upper limit of system resource tolerance: first, by collecting the false positive rate data of the model in a long-term operation under various typical scenarios such as different time periods, light conditions and crowd density, the fluctuation range is calculated to determine the natural fluctuation benchmark of the model performance; second, combined with the resource carrying capacity of the cloud management platform such as network bandwidth utilization, computing load threshold and business fault tolerance demand, a maximum allowed false positive rate deviation limit is set; finally, the preset difference is taken as the smaller value between the natural fluctuation benchmark and the resource tolerance limit, that is, the preset difference is set to 0.2%, and the preset ratio P0 of the difference to the preset difference is set to 1.2, then the comparison process of the ratio P of the difference to the preset difference and the preset ratio P0 is as follows:

[0072] If the ratio P of the difference to the preset difference is less than or equal to the preset ratio P0, the screening threshold is adjusted to 1.37 times the original screening threshold.

[0073] If the ratio P of the difference to the preset difference is greater than the preset ratio P0, the screening threshold is adjusted to 2.19 times the original screening threshold.

[0074] Please refer to Figure 4 The figure is a step flow chart for determining the safety management state based on the adjusted screening threshold according to the embodiment of the application.

[0075] Specifically, in the case that the safety management state is unqualified after adjusting the screening threshold, the screening threshold is adjusted at least once until the adjustment times is less than the preset times and the safety management state is qualified or the adjustment times is equal to the preset times; if the safety management state is unqualified after stopping adjustment, the data upload delay time of multiple historical moments is obtained; the average value of the multiple data upload delay times is calculated, and if the average value is greater than the preset average value, it means that the upload bandwidth is insufficient, which may cause critical data upload delay, loss or compression distortion, affecting the timeliness and accuracy of cloud analysis; at the same time, it may hinder the model update, so the key frame interval in the upload strategy is adjusted based on the difference between the average value and the preset average value, wherein the setting process of the preset average value is as follows: first, according to the data freshness window required for the cloud prediction model to complete a deep analysis and the overall system response time delay target, the maximum average delay that can be tolerated in the data upload stage is deduced reversely; second, referring to the theoretical transmission time of the baseline bandwidth guaranteed in the network service level agreement (SLA) when transmitting a typical data packet size, and adding the redundancy coefficient of conventional network fluctuation and protocol overhead, the preset average value is set to 1.5s, and the preset difference Q0 between the average value and the preset average value is set to 0.2s, then the comparison process based on the difference Q between the average value and the preset average value and the preset difference Q0 is as follows:

[0076] If the difference Q between the average value and the preset average value is less than or equal to the preset difference Q0, the key frame interval in the upload strategy is adjusted to 1.5 times of the original key frame interval, wherein the adjusted key frame interval retains one decimal place.

[0077] If the difference Q between the average value and the preset average value is greater than the preset difference Q0, the key frame interval in the upload strategy is adjusted to 2.1 times of the original key frame interval, wherein the adjusted key frame interval retains one decimal place.

[0078] Specifically, in order to ensure that the hidden danger key information is not lost, further expand the compression rate of the background area in the upload strategy, then adjust the compression rate of the background area in the video image data based on the ratio of the key frame interval to the preset interval, wherein the setting of the preset key frame interval is based on the comprehensive balance of the maximum event missing time window tolerated by the business and the sustainable processing capacity of the edge node, the value is first determined by the maximum data delay allowed by the cloud to complete an effective deep analysis, thereby deducing the time consumption of edge processing and uploading; at the same time, the highest frame processing frequency of the edge AI model when running stably on the target hardware needs to be determined through stress testing, therefore, the preset interval of the key frame is set to 2s, and the preset ratio R0 of the key frame interval to the preset interval is set to 0.83, then the comparison process based on the ratio R of the key frame interval to the preset interval and the preset ratio R0 is as follows:

[0079] If the ratio R of the key frame interval to the preset interval is less than or equal to a preset ratio R0, the compression rate of the background region in the video image data is adjusted to 1.39 times of the original compression rate, wherein the adjusted compression rate is rounded up.

[0080] If the ratio R of the key frame interval to the preset interval is greater than the preset ratio R0, the compression rate of the background region in the video image data is adjusted to 2.16 times of the original compression rate, wherein the adjusted compression rate is rounded up.

[0081] Specifically, in the case that the safety management state after adjusting the compression rate of the background region in the video image data is unqualified, the safety management characteristic values of multiple historical moments are obtained; the average value and the variance of the multiple safety management characteristic values are calculated, and if the average value is less than a preset average value and the variance is less than a preset variance, it indicates that the terminal perception device has conditions such as camera contamination, sensor drift, device failure, etc., causing the data quality of a specific region or type to continuously decrease, and then the calibration period of the terminal perception device is adjusted based on the ratio of the average value to the preset average value, wherein a preset ratio T0 of the average value to the preset average value is 0.89, and the comparison process based on the ratio T of the average value to the preset average value and the preset ratio T0 is as follows:

[0082] If the ratio T of the average value to the preset average value is less than or equal to the preset ratio T0, the calibration period of the terminal perception device is adjusted to 0.77 times of the original calibration period, wherein the adjusted calibration period is rounded up.

[0083] If the ratio T of the average value to the preset average value is greater than the preset ratio T0, the calibration period of the terminal perception device is adjusted to 0.89 times of the original calibration period, wherein the adjusted calibration period is rounded up.

[0084] In order to make the purpose and advantages of the present application more clear and obvious, the present application will be further described below in conjunction with examples; it should be understood that the specific examples described herein are only used to explain the present application, and do not limit the present application.

[0085] Example 1

[0086] This embodiment takes a large smart park with an area of 500,000 square meters as an example, which covers four main areas of office area, residential area, production area and public leisure area, and builds a smart park safety management architecture based on cloud-edge collaborative computing:

[0087] 8 high-performance servers are used to configure distributed database to store park security management data, deploy Spark big data analysis engine and TensorFlow AI model training framework; cloud security protection module uses SSL encryption protocol to realize data transmission encryption, implements permission management based on RBAC (role-based access control) model, and deploys firewall and intrusion detection system to prevent network attacks.

[0088] According to 4 main areas, each of which is deployed with an edge node, each edge node is configured with an industrial edge gateway, an edge server is configured with an Intel Core i7 processor, 128GB memory, 2TB SSD storage and a local cache module 1TB mechanical hard disk; the four edge nodes are interconnected through a gigabit local area network, and each edge node is connected to the cloud management platform through a 5G network.

[0089] 20 high-definition video surveillance cameras and 10 access control devices are deployed at the entrance of the office area, elevator entrance and important office; 15 video surveillance cameras and 8 access control devices are deployed at the unit door and parking lot of the residential area; 12 video surveillance cameras, 15 smoke sensors, 10 temperature and humidity sensors and 8 electronic fence detectors are deployed in the production area workshop and warehouse; 18 video surveillance cameras and 6 environmental sensors are deployed in the public leisure area square and road; all terminal sensing devices are configured with a unique device number, and communicate with the corresponding regional edge node through WiFi / BLE protocol, and the data collection frequency is set as: video image data 15 frames / second, environmental sensing data 1 time / minute, and personnel access data real-time collection.

[0090] Example 2

[0091] The terminal sensing device collects data according to the preset configuration, i.e. the smoke sensor in the warehouse of the production area collects smoke concentration of 0.8 mg / m³, the threshold value of which is 0.5 mg / m³, and the temperature data collected simultaneously is 38°C; the video surveillance camera in the warehouse collects a picture of a person moving without wearing a safety helmet; the access control device records that 5 people have entered or exited the area in the past 1 hour.

[0092] After receiving the above data, the edge node in the production area eliminates one temperature abnormal fluctuation data, converts the smoke concentration and temperature data into JSON format, converts the video image data into H.265 encoding format, and compresses the video image data to a ratio of 4:1; the local lightweight YOLOv5 target detection model is used to analyze the video image data, and the suspected safety hazard of "person without wearing safety helmet" is identified, and the smoke concentration exceeding standard data, video clip of person without wearing safety helmet and personnel access record are marked as key data.

[0093] The production area edge node uploads the above-mentioned key data to the cloud management platform in real time through the 5G network, and the normal temperature and humidity data of other areas in the production area are stored in the local cache; after receiving the data, the cloud management platform uses the Spark big data analysis engine to analyze the correlation between the smoke concentration exceeding the standard and the personnel without safety helmets, and evaluates the fire risk level of the area as "greater hidden danger" through the high-precision fire risk prediction model; at the same time, the cloud management platform dispatches the office area and the residential area edge node to share the recent park fire hazard history data, and further verifies the hidden danger level.

[0094] The technical scheme of the present application has been described in combination with the preferred embodiments shown in the drawings, but it is easy for those skilled in the art to understand that the protection scope of the present application is obviously not limited to these specific embodiments. Those skilled in the art can make equivalent changes or replacements to the related technical features without departing from the principles of the present application, and the technical scheme after the changes or replacements will fall within the protection scope of the present application.

[0095] The above description is only the preferred embodiments of the present application and is not intended to limit the present application; for those skilled in the art, the present application can have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A smart park security management method based on cloud-edge collaborative computing, characterized in that, The method comprises the following steps: Collecting multi-source data of park safety by using a plurality of periodically calibrated terminal sensing devices, including video image data, environmental sensing data, personnel access data and equipment operation data; An edge node analyzes the data using a local lightweight AI model and screens out suspected hidden danger data based on a set screening threshold; The edge node uploads the suspected hidden danger data and key data to a cloud management platform based on an upload strategy, and the cloud management platform uses a prediction model to perform in-depth analysis on the data uploaded by the edge node to evaluate the safety situation level of the managed park; Based on the safety situation level, the park is warned, and the corresponding terminal equipment and park security subsystem are started for emergency response; Based on the recognition accuracy of the prediction model of the cloud management platform, the false positive rate of the lightweight AI model, and the proportion of actual hidden dangers in the suspected hidden danger data uploaded by the edge node, a safety management representation value is calculated; Based on the safety management representation value, the safety management state of the smart park is determined, the screening threshold is adjusted based on the safety management state, and the upload strategy is adjusted based on the safety management state after adjusting the screening threshold. 2.The smart park security management method based on cloud edge collaboration computing according to claim 1, wherein, The method further comprises the following steps: The cloud management platform periodically optimizes and updates the prediction model using the suspected hidden danger data uploaded by the edge node and historical multi-source data; The optimized and updated prediction model is processed to generate a lightweight model that adapts to the computing power of the edge node, and the lightweight model is distributed to the edge node through a secure channel. 3.The smart park security management method based on cloud-edge collaborative computing according to claim 2, characterized in that, The method further comprises the following steps: The cloud management platform periodically adjusts the data upload strategy and the edge node computing task allocation according to the edge node computing power load and network bandwidth state.

4. The cloud-edge collaborative computing-based smart park security management method according to claim 3, characterized in that, The process of determining the safety management state of the smart park based on the safety management representation value comprises: If the safety management representation value is less than a preset representation value, the safety management state is determined to be unqualified; If the safety management representation value is greater than or equal to the preset representation value, the safety management state is determined to be qualified.

5. The cloud-edge collaborative computing-based smart park security management method according to claim 4, characterized in that, The process of adjusting the screening threshold based on the safety management state comprises: If the safety management state is unqualified, calculate the difference between the false positive rate of the lightweight AI model and a preset false positive rate; If the difference is greater than a preset difference, adjust the screening threshold based on the ratio of the difference to the preset difference.

6. The cloud-edge collaborative computing-based smart park security management method according to claim 5, characterized in that, The process of adjusting the screening threshold based on the ratio of the difference to the preset difference comprises: Increase the screening threshold based on the ratio of the difference to the preset difference, and the increase amplitude of the screening threshold is proportional to the ratio.

7. The cloud-edge collaborative computing-based smart park security management method according to claim 6, characterized in that, The process of adjusting the upload strategy based on the safety management state after adjusting the screening threshold comprises: If the safety management state after adjusting the screening threshold is unqualified, adjust the screening threshold at least once, until the adjustment times are less than a preset number of times and the safety management state is qualified or the adjustment times are equal to the preset number of times, and stop adjusting; If the safety management state after stopping adjusting is unqualified, obtain the data upload delay time at a plurality of historical moments; Calculate the average value of the plurality of data upload delay times, and if the average value is greater than a preset average value, adjust the key frame interval in the upload strategy based on the difference between the average value and the preset average value. 8.The cloud-edge collaborative computing based smart park security management method of claim 7, wherein, The process of adjusting the key frame interval in the upload strategy based on the difference between the average value and the preset average value comprises: The key frame interval in the uploading strategy is increased based on a difference between the average value and a preset average value, and the increase amplitude of the key frame interval is proportional to the difference. 9.The cloud-edge collaborative computing based smart park security management method of claim 8, wherein, The method further comprises: The compression rate of the background region in the video image data is increased based on a ratio of the key frame interval to a preset interval, and the increase amplitude of the compression rate is proportional to the ratio.

10. The cloud-edge collaborative computing-based smart park security management method of claim 9, wherein, The method further comprises: If the security management state after adjusting the compression rate of the background region in the video image data is unqualified, the security management characteristic values of a plurality of historical moments are obtained; The average value and the variance of the plurality of security management characteristic values are calculated, if the average value is less than a preset average value and the variance is less than a preset variance, the calibration period of the terminal perception device is reduced based on a ratio of the average value to the preset average value, and the reduction amplitude of the calibration period is inversely proportional to the ratio.

Images