Basic-level community public security event perception linkage system based on edge computing

Abstract

The invention relates to the technical field of situation awareness of edge computing, and discloses a grassroots community public security event awareness linkage system based on edge computing, which comprises a plurality of edge devices, state interaction is carried out on a public channel according to the frequency change of heartbeat pulse signals, and when any device makes low-confidence judgment on a potential high-risk event, the potential high-risk event is subjected to state interaction. The method comprises the following steps: when a plurality of devices respond to each other to form channel resonance, instantaneously increasing heartbeat frequency to wake up adjacent devices to carry out cross validation, and when the plurality of devices respond to form channel resonance, completing distributed confirmation of events and generating a linkage instruction. According to the invention, complex event content collaboration among the plurality of edge devices is converted into common perception of a single physical channel state; therefore, the response mode of the system is converted from a passive lag type of waiting for single strong evidence to an active prediction type of gathering multi-point weak evidence, and high-reliability early warning can be completed in the extremely early stage of a disaster.

Description

Technical Field

[0001] The present invention relates to a grassroots community public security incident perception linkage system based on edge computing, belonging to the technical field of situational awareness of edge computing. Background Art

[0002] Currently, in the field of community security monitoring, edge computing-based systems have become a commonly adopted technical architecture. They usually use the edge pre-processing-central decision-making operation mode, that is, edge devices distributed in various locations perform preliminary event recognition and aggregate structured data and video streams to the cloud or central server for in-depth analysis and final decision-making. This method demonstrates its value in improving data processing efficiency.

[0003] However, when trying to apply this technical architecture on a large scale to grassroots communities with limited network bandwidth, sensitive transformation costs and scarce operation and maintenance resources, its design compromises and economic constraints become real obstacles. In order to ensure the accuracy of central decision-making, the system needs to rely on continuously uploaded high-definition video data streams, which imposes a heavy burden on the already limited networks in many old communities or remote areas. In order to circumvent this burden, the compromise solutions of reducing video quality or reducing device density directly affect the basic perception capabilities of the system, making this technical architecture fall into the dilemma of high cost investment and reduced effectiveness. To get out of this dilemma, one improvement path is to enhance the processing capabilities of a single edge device so that it can independently and accurately judge early weak risk signals. However, this will not only lead to a significant increase in the hardware cost of edge devices, which is contrary to the goal of universality, but more importantly, it still cannot break away from the limitation of relying on a single perspective for independent judgment. It fails to effectively utilize the weak but synchronous anomaly information perceived by multiple adjacent devices at the same time for events with spatial continuity such as the spread of smoke in the early stage of a fire. However, due to the information barriers of the existing architecture, it is impossible to establish such low-cost distributed instantaneous verification logic.

[0004] Specifically, the prior art mainly has the following deficiencies: 1. The dependence of the system on high-bandwidth networks and central computing power leads to a contradiction between the cost and feasibility of large-scale deployment in basic communities; 2. Each edge device is an independent information island, lacking a lightweight horizontal coordination mechanism, and cannot aggregate and verify multiple weak abnormal signals associated in space; 3. The final confirmation of the event requires a decision-making process through the central server, which causes communication and processing delays, making it difficult to meet the requirements for instantaneous response to emergencies. Therefore, how to build a new system operation mode that can realize instantaneous and reliable distributed consensus on weak and uncertain information perceived by multiple edge devices without relying on high-bandwidth networks and central servers through a communication and coordination mechanism that does not need to transmit complex data content, thereby solving the problem of early warning of potential high-risk events under the condition of limited cost and power consumption, has become a technical problem to be solved by the present application. SUMMARY

[0005] The present application provides a kind of based on edge computing's basic community public security event perception linkage system, its main purpose is to solve how to build a new type of system, without relying on high-bandwidth networks and central servers, through a lightweight communication coordination mechanism that does not need to transmit complex data content, realize the instantaneous and reliable distributed consensus on weak and uncertain information perceived by multiple edge devices, thereby solving the problem of early warning of potential high-risk events under the condition of limited cost and power consumption.

[0006] To achieve the above object, the present application provides a kind of based on edge computing's basic community public security event perception linkage system, system includes multiple edge devices, each edge device includes:

[0007] perception module, it is configured to obtain real-time environmental data, and based on event identification model to real-time environmental data processing to generate a confidence value;

[0008] communication module, it is configured to send heartbeat pulse signal on a public channel with first frequency;

[0009] processing module is connected with perception module and communication module, processing module is configured as: when confidence value is lower than first threshold value but higher than second threshold value, control communication module sends the frequency of heartbeat pulse signal from first frequency instantaneous promotion to second frequency, to form wake-up signal;When communication module listens to the existence of wake-up signal from another edge device on public channel, control perception module cross-verification to the real-time environmental data currently acquired;And when in the stored confirmation time window, listen to the frequency of heartbeat pulse signal on public channel exceeds first resonance threshold value, confirm high-risk event and generate linkage instruction.

[0010] Preferably, the processing module is further configured to: the first threshold value is a confidence determination value of the event recognition model for independently triggering a high-level alarm; and the second threshold value is a confidence determination value of the event recognition model for starting preliminary event detection in a state without external excitation.

[0011] Preferably, the processing module and the communication module are further configured to: when the communication module listens to a pulse signal with energy exceeding the interference threshold value, trigger the processing modules of multiple edge devices in the group to record respective time stamps of the arrival of the pulse signal; based on the time delay difference between the multiple time stamps, determine the relative spatial direction of the source of the pulse signal; and instruct the communication modules of all edge devices in the group to synchronously switch to a communication channel in the candidate channel that is far away from the relative spatial direction as an updated common channel.

[0012] Preferably, the sensing module and the processing module are further configured to: after a restart of any edge device, the restarted device and other edge devices in the group that receive the restart beacon of the device all collect transient environmental background readings through respective sensing modules; the processing module of the restarted device compares its own transient environmental background reading with the median of the transient environmental background readings of the other edge devices in the group; and when the difference between the transient environmental background reading of the restarted device and the median exceeds a calibration threshold value, the processing module of the restarted device calibrates the internal recognition reference parameters of the event recognition model based on the median.

[0013] Preferably, the processing module is further configured to: continuously receive heartbeat pulse signals periodically transmitted by other edge devices in the device group and count the arrival time intervals of the heartbeat pulse signals, and then calculate the standard deviation σ of the arrival time intervals; and determine the duration T of the confirmation time window according to the following rule w , T w = T b -kσ, wherein T b is the base duration of the confirmation time window, and k is a calibration coefficient stored in the processing module.

[0014] Preferably, the processing module is further configured to: the setting of the first resonance threshold value corresponds to the frequency of the identifiable heartbeat pulse signals on the common channel within the confirmation time window, which is equivalent to the communication modules of at least two edge devices being in a state of transmitting heartbeat pulse signals at the second frequency at the same time.

[0015] Preferably, the linkage instruction includes at least one of the following: generating alarm data confirming the high-level risk and sending it to the central management platform with the highest network priority; activating an audible and visual alarm device located near the event occurrence location; and sending a start instruction to a nearby fire-fighting facility.

[0016] Preferably, the communication module is further configured as follows: the public channel is a designated frequency of a low-power wide area network; and the heartbeat pulse signal is a physical layer signal that does not carry business data, and its information is transmitted by changing the sending frequency.

[0017] Preferably, the processing module is further configured to: have a built-in wake-up state timer; when the duration of the communication module sending the heartbeat pulse signal at the second frequency exceeds the fatigue time threshold, the processing module forces the communication module to return to sending the heartbeat pulse signal at the first frequency, and generates a device self-test maintenance signal.

[0018] Preferably, the perception module is further configured as follows: cross-validation includes, within a limited time after receiving the wake-up signal, the perception module temporarily adjusts the confidence threshold used for judgment within the event recognition model or calls a preset lightweight verification model to review the real-time environmental data to make a collaborative judgment.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] 1. By establishing a collaborative verification mechanism based on channel resonance, the system can generate early warnings after aggregating weak evidence from multiple devices. Specifically, when the potential risk in a certain area is still in its infancy and can only be detected by individual devices with low confidence, the device can broadcast its alert status to neighboring devices by changing the frequency of its own heartbeat pulse signal rather than transmitting complex event content. This asymmetric wake-up signal, mediated by physical layer state changes, can incentivize neighboring devices to perform forced cross-verification. Once multiple devices detect the same anomaly and successively enter an accelerated heartbeat state, the pulse frequency on the channel will resonate. As a result, without the central platform's knowledge, the confirmation and linkage of high-risk events are completed in a decentralized manner at the initial stage of the event. Its response logic is no longer subject to the inherent delays of data upload and central decision-making.

[0021] 2. The present invention internalizes the adaptive calibration of communication channels and sensors themselves into an autonomous capability of the system, significantly enhancing its operational stability and reliability throughout its life cycle in real complex environments. For example, the system does not blindly avoid sudden strong electromagnetic pulse interference, but creatively transforms it into an opportunity to passively locate the interference source. The location of the interference source is determined by analyzing the arrival delay difference of the pulse signal between different devices, and then collaboratively switching to a communication channel away from the interference. This method enables the core collaborative mechanism of the system to be maintained in harsh electrical environments. At the same time, the system uses device restart events to achieve self-calibration by comparing the environmental background readings of the restarted device with those of other devices in the group. By comparing the median of the environmental background readings of the restarted device and the adjacent device group at the same time, the system can diagnose and compensate for the aging drift of a single sensor caused by long-term operation, avoiding the huge cost of large-scale offline manual calibration and ensuring the long-term consistency of the perception benchmark.

[0022] 3. By deeply mining and reusing implicit information from inter-device communication, the system is able to dynamically assess and adapt to its own network status. Rather than relying on a fixed time window to wait for collaborative confirmation, the system continuously counts the jitter in the arrival timing of periodic heartbeat signals between neighboring devices, using this as a yardstick to characterize the real-time health of the network channel. Based on this yardstick, the confirmation time window is dynamically and reversely adjusted. That is, the more unstable the network, the shorter the confirmation window. This design may seem counterintuitive, but it actually uses logical certainty to deal with physical uncertainty. It mandates that in unstable networks, collaborative confirmation must be completed in a shorter time, otherwise it will quickly switch to a redundant and secure path, thus making the collaborative behavior of the entire system highly predictable under various network conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a schematic diagram of the architecture and process of the grassroots community public security incident perception linkage system of the present invention;

[0024] Figure 2 A radar chart comparing the performance of the collaborative perception system of the present invention and the traditional architecture;

[0025] Figure 3 This is a flow chart of the core operating logic and adaptive adjustment mechanism of the present invention.

[0026] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0027] In order to make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be described in detail below in conjunction with specific embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0028] The present invention provides a grassroots community public safety event perception linkage system based on edge computing. Its overall architecture consists of multiple edge devices deployed in the target monitoring area. Each independent edge device consists of three core functional units: a perception module, a communication module, and a processing module. Among them, the processing module serves as a control center to interact with the perception module and the communication module for data and instructions, while multiple edge devices use a preset public channel to conduct decentralized information collaboration in a manner based on physical layer state changes, together forming a distributed situational awareness network; in a specific application scenario, for example, deploying this system in the corridors of old communities with limited network infrastructure and sensitive to renovation costs to perform situational awareness of early fire risks, the universal technical challenge it faces is that the weak smoke and other signals in the early stages of a fire are highly uncertain. Even if a single camera device captures the image, its built-in event recognition model may give a confidence value far below the independent alarm threshold due to weak signals or environmental interference. Such weak information is usually directly filtered in traditional architectures, thereby delaying the early warning. To meet this challenge, each edge device in the present invention is configured to be in normal standby mode. Under this condition, its communication module will continuously send a physical layer heartbeat pulse signal that does not carry any business data at a preset low reference frequency, i.e., the first frequency, at a specified frequency point, such as a low-power wide area network. The heartbeat pulse signal is only used to maintain the online status of the device in the network. At the same time, the perception module of the device will continuously obtain real-time environmental data, such as video image streams, and call its built-in event recognition model to analyze the data to generate a confidence value for a specific high-risk event, such as smoke. The processing module will determine the confidence value. When the value does not reach a level sufficient to trigger an independent high-level alarm, the device will generate a confidence value. When the heartbeat pulse signal is higher than the first threshold reported by the user, but higher than a preset low wake-up trigger threshold representing a potential abnormality, that is, the second threshold, the processing module will not report or discard this uncertain information, but interpret it as a weak signal that requires collaborative verification by adjacent nodes, and then control the communication module to instantly increase the sending frequency of the heartbeat pulse signal from the first frequency to a preset high wake-up frequency, that is, the second frequency. In this way, the system converts a low-confidence judgment at the application level into a clear frequency state change at the physical communication level, forming a wake-up signal that does not require complex data encoding and transmission.

[0029] In an environment where nodes are densely deployed in grassroots communities, the state changes emitted by a device will inevitably be perceived by multiple devices in its physical vicinity. However, if a lightweight and instantaneous coordination mechanism is to be built between edge devices, it is necessary to avoid the delay and network overhead brought by the traditional complex communication protocol based on the request-response mode. In view of this, this system utilizes the characteristics of shared channel broadcasting. When the communication module of any edge device monitors the pulse frequency on the public channel and jumps from the first frequency to the second frequency, that is, when it receives the wake-up signal, its processing module will be triggered to perform a forced cross-validation process. The cross-validation can be configured as the perception module temporarily adjusting the confidence threshold used for judgment in its event recognition model within a limited time or calling a preset lightweight verification model with lower computational overhead to review the currently acquired real-time environmental data for similar high-risk events. If the neighboring device also obtains a value between the first threshold and the second threshold after cross-validation, If the confidence value is between the two thresholds, the processing module will also control its own communication module to enter the heartbeat acceleration state of sending heartbeat pulse signals at the second frequency. In this way, in a very short period of time, the frequency of the heartbeat pulse signal on the public channel will be superimposed due to the successive responses of multiple devices, forming a channel resonance. The processing module monitors this comprehensive frequency in real time through a built-in pulse frequency counting logic. Once it is found that the comprehensive frequency exceeds a preset first resonance threshold within an internally stored confirmation time window, for example, the setting of the first resonance threshold corresponds numerically to at least two edge devices simultaneously sending heartbeat pulse signals at the second frequency, then the system determines that the occurrence of a high-risk event has been confirmed by distributed consensus; by converging multiple scattered and uncertain weak evidence into a strong signal at the physical layer, the system constructs an early warning response mode that can actively predict, thereby completing highly reliable distributed confirmation in the embryonic stage of high-risk events.

[0030] After confirming the occurrence of a high-risk event, in order to transform this distributed consensus into an effective disposal action, the processing module will immediately generate and execute a linkage instruction. The specific content of the linkage instruction can be configured according to the preset security strategy, which may include at least one or more of the following combinations: generating an alarm data of a confirmed high-level risk and sending it to the remote central management platform with the highest priority in the network, directly activating the sound and light alarm device deployed near the location of the incident for on-site warning or sending a start-up instruction to the nearby fire-fighting facilities with a network interface. In this way, the entire response process forms a closed loop on the edge side, and its response time is no longer subject to the communication and decision-making delay of the central server; however, in real deployment environments such as semi-industrial areas or old communities, public channels will inevitably be subject to instantaneous strong pulse electromagnetic interference generated by the start-up of large motors or the sparking of old lines. This interference may drown out normal heartbeat signals or be misjudged as false resonance, thereby posing a threat to the core coordination mechanism of the system. To ensure the operational stability of the system in harsh electrical environments, this system further integrates an adaptive channel avoidance mechanism, the core of which is to convert interference signals into electromagnetic interference signals. The signal itself serves as an opportunity for channel environment detection. Specifically, when the communication module of any edge device detects a pulse signal on the public channel it monitors, with energy far exceeding that of a normal heartbeat pulse and exceeding a preset interference threshold, its processing module will no longer attempt to decode the signal. Instead, it will immediately trigger an interference localization process within the group. During this process, all devices in the group that receive the strong pulse signal will use their processing modules to accurately record the signal's arrival timestamp and share their respective timestamps through a minimalist information exchange. After collecting multiple timestamps, the processing module of any device can use geometric calculations to estimate the relative spatial location of the pulse interference source within the device group based on the delay difference between them. Once the location of the interference source is determined, the processing module will instruct the communication modules of all devices in the group to synchronously and collectively switch to a pre-selected candidate communication channel that is farther away from the relative spatial location of the located interference source. This updated common channel will continue the event-aware linkage process. By transforming interference sources into temporary location beacons, the system gains the ability to maintain core functional availability in extreme electromagnetic environments.

[0031] Any system based on physical sensors faces the fundamental problem of inaccurate perception benchmarks due to sensor aging drift during long-term operation. This slow performance degradation will cause the confidence of the AI ​​algorithm in processing the same event to decrease year by year, and may eventually lead to the failure of triggering the coordination mechanism. In order to maintain the consistency of its perception capabilities throughout the life cycle of the system, this system also embeds a sensor aging self-calibration mechanism based on group intelligence. This mechanism uses the inevitable restart events of the device due to firmware updates or power outage recovery as a calibration opportunity. Specifically, when any edge device completes the restart, it broadcasts a restart beacon to the group. The neighboring devices that receive the beacon in a healthy state and the restarted device itself will collect a frame of transient environmental background readings in the current environment through their respective perception modules. Subsequently, The processing module of the restarting device will compare its own background reading with the median of the background readings collected from all its neighbors. The median is used to effectively eliminate the impact of abnormal readings that may exist in individual neighbors when calculating the group consensus. Since all devices are in the same physical environment, the background readings of healthy sensors should theoretically be highly consistent. Therefore, if the difference between the restarting device's own reading and the group median exceeds a preset calibration threshold, it can be determined that the device's sensor has drifted significantly. At this time, its processing module will automatically use the group median as a benchmark to reversely compensate and calibrate the recognition benchmark parameters within its event recognition model. This zero-cost, fully automatic internal self-healing mode fundamentally avoids the need for periodic manual on-site calibration of large-scale distributed devices.

[0032] Considering that the network status of wireless communication channels such as LPWAN that the system relies on has the characteristics of non-steady-state fluctuations, the dynamic changes in delay jitter make it difficult for a fixed confirmation time window to simultaneously meet the response speed and reliability under extreme network conditions. In order to enable the system's collaborative logic to dynamically adapt to the real-time physical characteristics of its communication link, the processing module is further configured to execute an adaptive confirmation window calibration procedure. In this procedure, the processing module will continuously receive and count the arrival time intervals of heartbeat pulse signals periodically sent by other edge devices in its device group, and then calculate the standard deviation σ of these time intervals within a sliding time window. The size of this standard deviation σ directly reflects the degree of delay jitter of the current network channel and becomes a quantitative indicator of network health; accordingly, the duration T of the confirmation time window in the system w It is no longer a static constant, but is dynamically adjusted according to the following rules: w =T b ―kσ, where T bis a basic duration of the confirmation time window, and k is a dimensionless calibration coefficient stored in the processing module and calibrated through offline experiments. The logic of this procedure is that when the network is more unstable and the jitter is greater, the system will actively shorten the confirmation window T. w , thereby forcing collaborative confirmation to be completed in a shorter and more certain time. If it cannot be completed, it will switch to the preset redundant safety path more quickly. By using logical certainty to deal with physical uncertainty, the collaborative behavior of the entire system maintains a high degree of predictability and logical robustness under various network conditions. Accordingly, in order to prevent individual devices from being in an accelerated heartbeat state for a long time due to continuous detection of low-confidence events or reception of wake-up signals, thereby causing unnecessary power consumption and possibly affecting the normal order of the channel, an awakening state timer is also set inside the processing module. When the duration of the communication module sending heartbeat pulse signals at the second frequency exceeds a preset fatigue time threshold, the processing module will force the communication module to return to the normal state of sending heartbeat pulse signals at the first frequency, and at the same time generate a device self-test maintenance signal for the background system to check.

[0033] The collaborative judgment and system operation parameters of the present invention are set and adaptively adjusted through a set of built-in deterministic procedures; cross-validation specifically calls an inter-frame optical flow motion detection model, which is used to quantify the pixel displacement rate between consecutive video frames. Its built-in optical flow activation threshold T flow During the initial calibration phase of the device, a background video stream without any events is collected and the mean value μ of the optical flow rate between frames is calculated. bg and standard deviation σ bg , and finally by T flow =μ bg +3σ bg The fatigue time threshold T of the wake-up timer is determined by the procedure. fatigue , then the maximum communication hop number H between any two nodes in the network topology is max The maximum single-hop transmission delay t specified by the protocol hopmax , through T fatigue =2·H max ·t hopmax The core thresholds of the adaptive channel avoidance function and the sensor aging self-calibration function also follow the deterministic procedure; the interference threshold used to judge the external strong pulse interference is measured by the received signal strength indicator RSSI, and its value RSSI interference By making the maximum number of devices in the group send heartbeat signals at the second frequency at the same time under the reference channel environment, the maximum resonance signal intensity RSSI generated is measured. maxresonance , and by RSSI interference =RSSI maxresonanceThe 6dB rule is generated, where 6dB is a fixed signal isolation margin. The topological information of the relative geometric positions of the devices in the group required for performing the time difference positioning of the interference source comes from the topology self-discovery rule automatically executed when the system is first deployed. In this rule, each device exchanges RSSI readings and the gateway device executes a multi-dimensional scaling analysis algorithm to solve and solidify a relative coordinate map without manual intervention; correspondingly, the calibration threshold T of the sensor aging self-calibration calib By performing accelerated aging tests on the same batch of products, the statistical distribution of the background reading difference Δ relative to a golden sample used as a reference device is quantified, and the mean μ is taken. Δ and standard deviation σ Δ , and finally by T calib =μ Δ +2σ Δ The output of this procedure is a measurable physical aging effect as input and a statistically significant decision boundary as output.

[0034] Example 1: In an old community corridor environment with aging lines and cluttered with debris, the perception module of edge device A deployed at one end of the corridor obtains real-time environmental data of faint smoke in the corner of its monitoring range. Because the smoke is thin and affected by dim light, its built-in event recognition model processes the data and generates a confidence value indicating a smoke event of 0.3, which is lower than the first threshold of 0.7 used to independently trigger a high-level alarm, but higher than the set second threshold of 0.2. In this case, the processing module of edge device A immediately controls its communication module to instantly increase the sending frequency of the heartbeat pulse signal from the first frequency as the low base frequency to the second frequency as the high wake-up frequency, thereby forming a wake-up signal on the public channel.

[0035] Edge device B, deployed at the other end of the corridor, has a communication module that monitors the wake-up signal from edge device A on the public channel. Its processing module then controls the perception module to perform forced cross-validation on the currently acquired real-time environmental data. After review, its event recognition model also obtains a low confidence value of 0.4, which is also between the first threshold and the second threshold. The processing module of edge device B also correspondingly controls its communication module to increase the sending frequency of the heartbeat pulse signal to the second frequency; within the system's preset 500-millisecond confirmation time window, the comprehensive frequency of the heartbeat pulse signal on the public channel exceeds the first resonance threshold due to the joint action of edge device A and edge device B. The setting of the first resonance threshold numerically corresponds to the state where at least two edge devices are simultaneously sending heartbeat pulse signals at the second frequency. Accordingly, , the occurrence of high-risk events has been confirmed; this collaborative method transforms the technical requirements for high-precision judgment of a single signal into consensus confirmation of multiple spatiotemporally correlated weak signals, thereby achieving high sensitivity and high reliability within a single architecture; any edge device that detects that the first resonance threshold is triggered, its processing module generates and executes linkage instructions, reports the confirmed fire alarm information to the central management platform with the highest network priority, and simultaneously activates the sound and light alarm device located in the corridor; thus, the system no longer passively waits for a single device to produce a highly certain event judgment, but instead perceives continuous weak anomalies in space at the same time by converging multiple devices, completing early warning in the very early stages of the disaster, and the system's response mode changes from a lagging judgment of a single strong evidence to a predictive perception of multiple weak evidence.

[0036] Example 2: To verify the response efficiency of the collaborative sensing mechanism under weak signal conditions, a test platform was established. The platform was located in a 10-meter-long closed test chamber. An edge device was deployed every 2 meters along the central axis, totaling 5 devices, forming a test group. The system was configured to instantly increase the heart rate when any device made a judgment on a potential event with a confidence level higher than 0.2. A control group with the same structure was also set up. Its 5 edge devices were configured with a conventional edge preprocessing-central decision-making architecture, that is, events were reported to the central server only when the single-machine confidence level exceeded a fixed high threshold of 0.7. The test environment used a smoke-generating device with controllable concentration to simulate the smoke spread process, and the smoke concentration was calibrated with the light attenuation rate.

[0037] In the experiment, the key parameter settings of the built-in event recognition model of each edge device follow the following procedures. The setting of the second threshold for triggering the wake-up signal needs to balance the detection sensitivity and idle power consumption. The setting process is to run the device continuously for 24 hours in a clean environment without smoke, record the confidence value fluctuation caused by the sensor background noise, take the maximum value and add a 15% safety margin to obtain the second threshold of 0.2; the confirmation time window for determining the final confirmation of the event has a basic duration T b The setting of needs to balance the immediacy of response and network fault tolerance. Its value is related to the theoretical maximum single-packet transmission delay of the adopted LPWAN communication protocol and is set to 1.5 times of this delay, that is, 500 milliseconds. After the test is started, the smoke device increases the smoke concentration at a rate of 0.1% OPM per second from zero. At the same time, the system response time of the two groups of equipment from the appearance of smoke to the first confirmation alarm is recorded. The test is repeated 10 times and the average value is taken. The key process data are shown in Table 1.

[0038] Table 1: Comparison of response time between the experimental group and the control group at different smoke concentrations.

[0039]

[0040] When the smoke concentration reached 1.5% OPM, the average confidence level of a single device was 0.34, which did not reach the alarm threshold of the control group. However, in the experimental group, the confidence levels of at least two devices exceeded the second threshold of 0.2, triggering channel resonance through increased heartbeat frequency and completing distributed confirmation within 3.5 seconds. In the control group, after the smoke concentration rose to 3.5% OPM, causing the confidence level of a certain device to exceed the high threshold of 0.7, the first alarm was generated at 12.8 seconds. This difference stems from the experimental group's operating mechanism, which changes the basis for event confirmation from the absolute value judgment of a single data source to the judgment of the superposition of energy states formed by multiple data sources at the physical channel layer.

[0041] Example 3: This example combines Figures 1 to 3 , to illustrate the implementation of a grassroots community public security incident perception linkage system based on edge computing, such as Figure 1As shown, the system includes an edge device group consisting of multiple edge devices, wherein edge device 1, edge device 2 to edge device N are illustrated. Each device has a built-in perception module, a processing module and a communication module, and multiple devices interact with each other through a common channel using a heartbeat pulse signal; the event perception process on the right reveals the operating logic of the system, that is, when a low-confidence event with a confidence level between the second threshold and the first threshold is detected, the system increases the first frequency to the second frequency to wake up the adjacent devices for cross-verification. When the channel frequency exceeds the first resonance threshold to form a channel resonance, the high-risk event is confirmed and a linkage instruction is generated. The final linkage output points to the central management platform for a high-priority alarm, drives the sound and light alarm device for on-site warning, and issues a start-up instruction to the fire-fighting facilities.

[0042] like Figure 2 As shown, Figure 2 In the form of a radar chart, the collaborative perception system of the present invention is intuitively compared with the traditional architecture in six key performance dimensions. These six dimensions are response speed, detection sensitivity, power consumption efficiency, network dependence, deployment cost and maintenance convenience. The collaborative perception system outlined by the solid line in the figure is significantly superior to the traditional architecture outlined by the dotted line in terms of response speed, detection sensitivity, power consumption efficiency and maintenance convenience, and at the same time shows lower objective requirements in network dependence and deployment cost.

[0043] like Figure 3 As shown in the figure, after edge device A and edge device B each obtain environmental data and make a low-confidence judgment, they interact with each other based on the heartbeat frequency through the public channel. If the channel resonance condition is met, that is, the frequency exceeds the first resonance threshold, the system determines that the distributed consensus is completed and generates a linkage instruction, thereby activating the sound and light alarm device for on-site prompts and reporting to the central management platform; in addition, the figure also illustrates two adaptive functions triggered by specific events. One is when the trigger: when the device is restarted, the system will start the sensor aging self-calibration mechanism based on collective wisdom; the other is when the trigger: when strong pulse interference occurs, the system will start the adaptive channel avoidance function based on the time difference positioning of the pulse interference source. After locating the interference source, this function will output instructions to update the public channel.

[0044] Example 4: When the system is deployed in an electrical environment with strong pulse electromagnetic interference, its common channel is at risk of being suppressed by the interference signal. The processing module of the system has an adaptive channel avoidance function based on time difference positioning of the pulse interference source; in this functional procedure, when the communication module of the edge device continuously monitors the common channel, if the energy of its received signal exceeds the preset interference threshold within a sampling period, and the value of the interference threshold is set higher than the upper limit of the channel resonance energy that can be formed by multiple edge devices simultaneously sending heartbeat pulse signals at the second frequency, then the processing module will determine this signal as external pulse interference and immediately record the local arrival timestamp of the leading edge of the pulse signal; the device, together with other devices in the group that detect the same interference pulse, broadcasts a positioning signaling containing its own device ID and the recorded arrival timestamp to the common channel within a preset time window. The positioning signaling is sent using a high-redundancy modulation method with forward error correction coding to combat channel fading in an interference environment.

[0045] After any edge device in the group receives at least two positioning signals from other devices, its processing module calculates the relative spatial orientation of the pulse interference source through the hyperbolic positioning algorithm based on the delay difference between the collected multiple arrival timestamps and combined with the pre-stored topological information representing the relative geometric positions of the devices in the group; after determining the orientation of the interference source, the processing module selects a communication channel with the largest difference from the calculated interference source azimuth from a preset candidate channel list as the updated common channel; then, the processing module broadcasts an instruction containing the new channel frequency parameters and the synchronous switching time to all devices in the group, instructing all devices to switch to the new common channel at the specified time and restore the event perception linkage function. This procedure enables the system to use the interference signal to passively detect the channel environment and actively avoid it when encountering strong external pulse interference, thereby maintaining the operation continuity of the core coordination mechanism.

[0046] Example 5: In order to cope with the perception baseline drift caused by physical aging of a single edge device sensor, the system's processing module has a self-calibration function based on group consensus. When any edge device is restarted, the device broadcasts a restart beacon containing its own ID to the device group to which it belongs. Within a preset time window after receiving the beacon, other healthy devices in the group collect and send back their respective transient environmental background readings. The reading is quantified as a value calculated by the mean square deviation of the brightness of the central area of ​​the current original image frame; after collecting at least two valid readings from neighbors, the restarting device compares its own background reading with the median of all received neighbor readings. If the difference between the two exceeds the preset calibration threshold, the restarting device reversely calibrates the baseline parameters of its internal event recognition model based on the median.

[0047] Adaptive confirmation time window Tw =T b The calibration coefficient k in kσ is determined through an offline calibration procedure conducted on a test platform with controllable network jitter. By injecting different levels of network delay and packet loss, a series of discrete network jitter states are generated. The actual value of the standard deviation σ of the heartbeat signal arrival time interval is measured in each state. Correspondingly, in each jitter state, repeated experiments are conducted to find the longest confirmation time window T in which the cooperative wake-up signal can be received with a probability of more than 99%. w ; After obtaining multiple groups (σ,T w ) data points, a linear regression analysis is performed on these data points to obtain the value of the calibration coefficient k, which is then solidified into the processing module of the edge device.

[0048] Example 6: Before deployment, the system uses a set of offline standardized calibration procedures to determine its core logical parameters. The procedure sets thresholds for the event recognition model built into the edge device. It uses a benchmark data set containing a large number of labeled positive and negative samples, and runs the model on the data set to determine the first threshold for independently triggering a high-level alarm. The value is set to the product of the highest confidence value generated by the model on all negative samples and a preset safety factor; at the same time, the implementation method of cross-validation is defined as follows: when a device receives a wake-up signal, its processing module temporarily enables a lightweight verification model, which analyzes the optical flow change rate of the target area in the current image frame. If the rate exceeds the preset optical flow threshold, a collaborative judgment is made.

[0049] The calibration procedure also sets the parameters of the cooperative mechanism, wherein the determination of the first resonance threshold is realized in the processing module by a sliding window pulse counter, which counts the pulses in the confirmation time window T. w The heartbeat pulse signal received in the time is counted. When the count value exceeds a preset pulse count value, it is determined that the first resonance threshold is reached. The setting of the pulse count value is based on the values ​​of the first frequency and the second frequency and the confirmation time window T w The fatigue time threshold is set based on the calculation of the maximum delay of event propagation within the network, and its value is set to twice the theoretical maximum time required to complete a wake-up and response between any two devices with the longest distance in the group.

[0050] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0051] Finally, it should be noted that the above examples are merely intended to illustrate the technical solutions of the present application and not to limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. A grassroots community public security incident perception linkage system based on edge computing, characterized by: The system includes multiple edge devices, each of which includes: a perception module configured to acquire real-time environmental data and process the real-time environmental data based on the event recognition model to generate a confidence value; a communication module configured to transmit a heartbeat pulse signal at a first frequency on a common channel; A processing module is connected to the perception module and the communication module, and the processing module is configured to: when the confidence value is lower than the first threshold but higher than the second threshold, control the communication module to instantaneously increase the sending frequency of the heartbeat pulse signal from the first frequency to the second frequency to form a wake-up signal; when the communication module monitors the presence of a wake-up signal from another edge device on the public channel, control the perception module to cross-verify the currently acquired real-time environmental data; and when the frequency of the heartbeat pulse signal monitored on the public channel exceeds the first resonance threshold within the stored confirmation time window, confirm that a high-risk event has occurred and generate a linkage instruction.

2. The grassroots community public security incident perception linkage system based on edge computing according to claim 1 is characterized in that: The processing module is further configured as follows: the first threshold is the confidence judgment value of the event recognition model for independently triggering a high-level alarm; the second threshold is the confidence judgment value of the event recognition model for starting preliminary event detection in the absence of external stimulation.

3. The grassroots community public security incident perception linkage system based on edge computing according to claim 1 is characterized in that: The processing module and the communication module are further configured as follows: when the communication module monitors a pulse signal whose energy exceeds the interference threshold, the processing modules of multiple edge devices in the trigger group record the arrival timestamps of the pulse signals; the processing module determines the relative spatial orientation of the pulse signal source based on the delay difference between the multiple arrival timestamps; and the processing module instructs the communication modules of all edge devices in the group to synchronously switch to a communication channel in the candidate channel that is far away from the relative spatial orientation as the updated common channel.

4. The grassroots community public security incident perception linkage system based on edge computing according to claim 1 is characterized in that: The perception module and the processing module are further configured as follows: after any edge device is restarted, the restarted device and other edge devices in the group that receive the restart beacon of the device collect transient environmental background readings through their respective perception modules; the processing module of the restarted device compares its own transient environmental background reading with the median of the transient environmental background readings of other edge devices in the group; and when the difference between its own transient environmental background reading and the median exceeds the calibration threshold, the processing module of the restarted device calibrates the internal recognition benchmark parameters of the event recognition model based on the median.

5. The grassroots community public security incident perception linkage system based on edge computing according to claim 1 is characterized in that: The processing module is further configured to: continuously receive heartbeat pulse signals periodically sent by other edge devices in its device group and count the arrival time intervals of the heartbeat pulse signals, and then calculate the standard deviation σ of the arrival time intervals; The duration of the confirmation time window T is determined according to the following rules w , T w =T b ―kσ, where T b To determine the basic duration of the time window, k is a calibration coefficient stored in the processing module.

6. The grassroots community public security incident perception linkage system based on edge computing according to claim 1 is characterized in that: The processing module is further configured to: set the first resonance threshold, corresponding to the frequency of the heartbeat pulse signal identifiable on the common channel within the confirmation time window, which is equivalent to the communication modules of at least two edge devices being in a state of sending heartbeat pulse signals at the second frequency at the same time.

7. The grassroots community public security incident perception linkage system based on edge computing according to claim 1 is characterized in that: The linkage instruction includes at least one of the following: generating alarm data confirming a high-level risk and sending it to the central management platform with the highest network priority; activating the sound and light alarm device located near the location of the incident; and sending a start instruction to nearby fire-fighting facilities.

8. The grassroots community public security incident perception linkage system based on edge computing according to claim 1 is characterized in that: The communication module is further configured as follows: the public channel is a designated frequency point of a low-power wide area network; and the heartbeat pulse signal is a physical layer signal that does not carry business data, and its information is transmitted by changing the sending frequency.

9. The grassroots community public security incident perception linkage system based on edge computing according to claim 1 is characterized in that: The processing module is further configured as: a built-in wake-up state timer; when the duration of the communication module sending the heartbeat pulse signal at the second frequency exceeds the fatigue time threshold, the processing module forces the communication module to resume sending the heartbeat pulse signal at the first frequency and generates a device self-test maintenance signal.

10. The grassroots community public security incident perception linkage system based on edge computing according to claim 1 is characterized in that: The perception module is further configured as follows: cross-validation includes, within a limited time after receiving the wake-up signal, the perception module temporarily adjusts the confidence threshold used for judgment within the event recognition model or calls a preset lightweight verification model to review the real-time environmental data to make a collaborative judgment.

Images