Intelligent pet positioning fence edge high-precision positioning method based on multi-modal fusion

Abstract

The application relates to the technical field of wireless positioning monitoring, and particularly discloses an intelligent pet positioning fence edge high-precision positioning method based on multi-modal fusion, which comprises the following steps: acquiring a multi-modal positioning data set, wherein the positioning data set comprises satellite positioning data reported by an intelligent pet edge terminal through a cellular network and near-field positioning data collected by a local area entity fence node from a near-field signal of the intelligent pet edge terminal and returned through a wireless link. The application dynamically adjusts the reporting refresh frequency of the positioning data by calculating a behavior comprehensive deviation degree in the local area through the edge terminal, reduces the reporting frequency when the pet is in a daily activity state to save device power, and automatically increases the refresh frequency when a behavior deviation occurs to guarantee the timeliness of position tracking.

Description

Technical Field

[0001] This invention relates to the field of wireless positioning and monitoring technology, specifically to a high-precision positioning method for the edge of an intelligent pet positioning fence based on multimodal fusion. Background Technology

[0002] Currently, conventional pet monitoring systems primarily use a single satellite positioning technology to obtain location data and report the coordinates to a backend server at fixed time intervals via terminal devices. After receiving the coordinate data, the server compares it with pre-defined static electronic fence boundaries. If the coordinates exceed the set boundaries, an out-of-bounds warning message is sent to the user's terminal.

[0003] In practical applications, the aforementioned static and singular location monitoring mechanism has limitations:

[0004] On the one hand, the fixed data reporting frequency of the terminal cannot adapt to the dynamic changes in pet behavior. Continuous high-frequency reporting will cause the device to consume power quickly, while low-frequency reporting will easily cause delays in out-of-bounds alarms.

[0005] On the other hand, single satellite positioning is prone to signal drift at the edge of buildings, and static electronic fences ignore the differences in individual pet behavior patterns, movement states, and physiological indicators. This leads to the system being prone to false alarms or missed alarms when the positioning signal is unstable or the pet exhibits abnormal behavior. Furthermore, single alarm notifications lack a hierarchical processing mechanism, and users cannot directly issue interactive commands through the system for rapid response after receiving the notification. Summary of the Invention

[0006] This invention aims to at least partially solve one of the technical problems in related technologies. Therefore, the objective of this invention is to propose a high-precision edge positioning method for intelligent pet location fences based on multimodal fusion, in order to improve the accuracy of boundary crossing detection and danger alarms.

[0007] To achieve the above objectives, a first aspect of the present invention proposes a high-precision positioning method for the edge of a smart pet positioning fence based on multimodal fusion, applied to a wireless positioning and monitoring system consisting of a smart pet edge terminal, local physical fence nodes, a cloud service platform, and a mobile user terminal. The method includes:

[0008] Acquire a multimodal positioning data set, which includes satellite positioning data reported by the smart pet edge terminal via a cellular network, and near-field positioning data collected by the local physical fence node from the smart pet edge terminal and transmitted back via a wireless link;

[0009] Locally on the smart pet edge terminal, the comprehensive deviation of the current behavior is calculated, and the location data reporting refresh frequency of the smart pet edge terminal is dynamically adjusted according to the level of the comprehensive deviation, so as to perform wireless resource and location scheduling on the edge side.

[0010] On the cloud service platform, density clustering is performed based on historical trajectory data to construct individual behavior baselines, and the boundary parameters of the four-level partitioned polygonal electronic fence are dynamically calibrated based on the individual behavior baselines.

[0011] A comprehensive risk score is calculated by integrating features from location, motion, health, and behavioral deviation dimensions, and cross-validation of boundary crossing events is performed using the multimodal positioning dataset.

[0012] Based on the cross-validation results of the boundary crossing event and the comprehensive risk score, the boundary crossing risk level is determined. The corresponding level of alarm notification and quick action command are sent to the mobile user terminal through the wireless push channel, and an interactive closed loop is formed after receiving the control signaling feedback from the user based on the quick action command.

[0013] To achieve the above objectives, a second aspect of the present invention proposes a high-precision positioning system for the edge of an intelligent pet positioning fence based on multimodal fusion, comprising an interconnected intelligent pet edge terminal, a local physical fence node, a cloud service platform, and a mobile user terminal;

[0014] The local physical fence node is used to collect near-field signals from the smart pet edge terminal and transmit near-field positioning data back to the cloud service platform via a wireless link.

[0015] The intelligent pet edge terminal is used to report satellite positioning data to the cloud service platform via a cellular network; and to calculate the comprehensive deviation of the current behavior locally, and dynamically adjust the location data reporting refresh frequency of the intelligent pet edge terminal according to the level of the comprehensive deviation, so as to perform wireless resource and positioning scheduling on the edge side.

[0016] The cloud service platform is used to acquire a multimodal positioning data set composed of satellite positioning data and near-field positioning data; perform density clustering based on historical trajectory data to construct an individual behavior baseline, and dynamically calibrate the boundary parameters of the four-level partitioned polygonal electronic fence according to the individual behavior baseline; calculate a comprehensive risk score by fusing features of location dimension, motion dimension, health dimension, and behavior deviation dimension, and perform cross-verification of boundary crossing events in virtual and real dual-layer positioning in combination with the multimodal positioning data set; determine the boundary crossing risk level according to the result of the cross-verification of boundary crossing events and the comprehensive risk score, and send alarm notifications and quick action commands of the corresponding level to the mobile user terminal through a wireless push channel;

[0017] The mobile user terminal is used to receive alarm notifications and quick action commands of the corresponding level, and send the control signaling fed back by the user based on the quick action commands to the cloud service platform to form an interactive closed loop.

[0018] To achieve the above objectives, a third aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory. When the computer program is executed by the processor, it implements the above-described high-precision positioning method for intelligent pet positioning fence edges based on multimodal fusion.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0020] In practical use, the technical solution of this invention dynamically adjusts the reporting refresh frequency of positioning data by calculating the comprehensive deviation of behavior locally on the edge terminal. When the pet is in a daily activity state, the reporting frequency is reduced to save device power, and when behavior deviation occurs, the refresh frequency is automatically increased to ensure the timeliness of location tracking.

[0021] Meanwhile, the cloud service platform automatically constructs individual behavioral baselines using historical trajectories to dynamically calibrate polygonal electronic fence parameters. It also integrates multi-dimensional features such as location, movement, health, and behavioral deviations, and performs cross-validation using multimodal positioning data formed by cellular networks and near-field signals. This enables the system to issue alarm notifications and quick action commands corresponding to the risk level to user terminals based on the cross-validation results and comprehensive risk scores. Users can directly respond with control signals after receiving alarms, forming an interactive closed loop. This improves the accuracy of boundary crossing detection and danger alarms, and enhances the overall response efficiency of the monitoring system to emergencies. Attached Figure Description

[0022] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein:

[0023] Figure 1 This is a flowchart illustrating the high-precision edge positioning method for intelligent pet positioning fences based on multimodal fusion provided by the present invention.

[0024] Figure 2 This is a graph showing the conversion data between near-field signal strength and estimated distance based on Kalman filtering in the high-precision positioning method for intelligent pet positioning fences based on multimodal fusion provided by this invention.

[0025] Figure 3This is a schematic diagram of the behavior comprehensive deviation threshold triggering and terminal positioning refresh frequency adaptive state ladder in the intelligent pet positioning fence edge high-precision positioning method based on multimodal fusion provided by the present invention.

[0026] Figure 4 This invention provides a high-precision positioning method for the edge of an intelligent pet positioning fence based on multimodal fusion, which includes trajectory cleaning based on DBSCAN clustering and generation of a topology bitmap for a four-level partitioned dynamic electronic fence.

[0027] Figure 5 This is a three-dimensional surface mapping of the remaining battery power, uplink signal-to-noise ratio and forwarding priority weight of adjacent heterogeneous edge nodes in the high-precision positioning method of intelligent pet positioning fence based on multimodal fusion provided by the present invention.

[0028] Figure 6 This is a graph showing the exponential decay curve of the backoff time of the ad hoc temporary data link based on forwarding priority weight in the high-precision positioning method for intelligent pet positioning fences based on multimodal fusion provided by this invention.

[0029] Figure 7 This is a diagram showing the nonlinear mapping relationship between the comprehensive deviation and the upper limit exponent of the competition window in the high-precision positioning method for intelligent pet positioning fences based on multimodal fusion provided by this invention;

[0030] Figure 8 This is a schematic diagram illustrating the implementation of the intelligent pet positioning fence edge high-precision positioning system based on multimodal fusion provided by the present invention;

[0031] Figure 9 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0032] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0033] The following description, with reference to the accompanying drawings, outlines a high-precision edge positioning method, system, and electronic device for intelligent pet positioning fences based on multimodal fusion, according to embodiments of the present invention.

[0034] Example 1:

[0035] This embodiment discloses a high-precision positioning method for the edge of a smart pet positioning fence based on multimodal fusion. This method is applied to a wireless positioning and monitoring system composed of a smart pet edge terminal, a local physical fence node, a cloud service platform, and a mobile user terminal.

[0036] At the hardware physical architecture level of the system, the smart pet edge terminal is a miniaturized wearable device worn on the pet's neck or back. It integrates a GPS / BDS dual-mode radio frequency module for receiving satellite positioning signals, an NB-IoT (narrowband Internet of Things) or LTE-M baseband chip for accessing cellular networks, a low-power Bluetooth (BLE) or Wi-Fi module for near-field communication, a microelectromechanical system (MEMS) six-axis inertial sensor, and optical heart rate and other vital signs sensors.

[0037] The local physical fence nodes are deployed in fixed locations such as user homes, yards, and pet parks, specifically manifested as Bluetooth gateways, Wi-Fi probes, or LoRa base stations with fixed physical coordinates. The cloud service platform is deployed on a distributed cloud server, responsible for handling massive concurrent data access, geographic information system (GIS) calculations of spatial data, and model training for machine learning algorithms. The mobile user terminal is typically a smartphone or tablet computer owned by the user, with a mobile application installed to complement the wireless positioning and monitoring system.

[0038] like Figure 1 As shown, the method in this embodiment includes the following steps:

[0039] Phase 1: Collection and identity association of multimodal positioning datasets.

[0040] Specifically, during system initialization and daily operation, the primary task is to establish a reliable wireless communication link and identity authentication system, and to acquire a multimodal positioning data set.

[0041] The multimodal positioning data set includes satellite positioning data reported by the smart pet edge terminal via a cellular network, and near-field positioning data collected by the local physical fence node from the smart pet edge terminal and transmitted back via a wireless link.

[0042] To ensure the accuracy of multimodal data tracing in complex network environments, the cloud service platform establishes a multimodal wireless identifier mapping table that includes pet chip identifiers, terminal device identifiers, and network media access control addresses, thereby completing the wireless identity binding between the smart pet edge terminal and the mobile user terminal.

[0043] For example, in the underlying database structure of this mapping table, the unique code of the RFID microchip implanted in the pet is used as the primary key to map and associate the IMEI (International Mobile Equipment Identity), Bluetooth MAC address, Wi-Fi MAC address of the smart pet edge terminal, as well as the user account ID and operating system device push token of the mobile user terminal.

[0044] All identification fields maintain a one-to-one mapping relationship in the database, ensuring global uniqueness. Whether the smart pet edge terminal accesses via an NB-IoT base station in an open outdoor area or via a Bluetooth channel of a local physical fence node in an indoor blind spot, the cloud service platform can retrieve the multimodal wireless identification mapping table to uniformly resolve and aggregate the physical layer network media access control addresses or terminal device identifiers from different sources to the corresponding individual pet, thereby eliminating data silos in heterogeneous networks.

[0045] Optionally, regarding the near-field positioning data generation mechanism, when the smart pet edge terminal enters the signal coverage area of ​​the local physical fence node, the terminal's near-field communication module will send a wireless broadcast beacon packet with a MAC address according to a preset broadcast interval.

[0046] The local physical fence node continuously collects the near-field signal strength value of the smart pet edge terminal, converts the near-field signal strength value into distance features through a logarithmic path loss model, and generates the near-field positioning data by combining the fixed geographic coordinates of the local physical fence node.

[0047] To enable those skilled in the art to more clearly understand the conversion process from near-field signal intensity to distance characteristics, a specific engineering implementation of the logarithmic path loss model is introduced here.

[0048] Due to multipath effects and signal fading in indoor or complex obstacle environments, the main control MCU of the local physical fence node needs to perform Kalman filtering or moving average filtering on multiple continuously acquired near-field signal strength values ​​to remove transient noise and obtain smoothed near-field signal strength values, which are then substituted into the following formula to calculate the distance:

[0049] ;

[0050] In the formula, This represents the estimated physical distance between the local physical fence node and the smart pet edge terminal, in meters. This represents the reference near-field signal strength test value when the transmitter and receiver are one meter apart, in units of... ; This represents the near-field signal strength value actually received by the local physical fence node and after filtering and smoothing, in units of... ; The path loss index represents the degree of signal attenuation in the current environment. In open indoor spaces, this value is usually 2.0, while in complex indoor spaces with multiple partition walls, this value can be 3.5 to 4.5.

[0051] Through the above calculations, the local physical fence node can not only confirm the existence of the terminal, but also obtain relatively accurate radial distance data. Combined with the node's own known latitude and longitude coordinates or indoor two-dimensional plane coordinates, it generates near-field positioning data containing node coordinates and distance radius.

[0052] like Figure 2 The test simulation results of the local physical fence node continuously tracking the edge terminal of the smart pet in a complex multipath interference environment are shown. The horizontal axis of the figure is the sampling time in seconds, the left vertical axis is the received signal strength in decibels and milliwatts, and the right vertical axis is the estimated physical distance in meters.

[0053] The light blue curve in the figure represents the original acquired signal, which is subject to a sudden and irregular drop from -60 dB / mW to -95 dB / mW due to the multipath effect caused by indoor partitions and object reflections. Directly using this data can easily lead to misjudgment of location out of bounds.

[0054] The red curve generated after algorithm smoothing represents the Kalman filter signal. This curve effectively suppresses most of the transient noise and presents a stable waveform characteristic. However, when the pet terminal moves away rapidly from the 10th to the 20th second, the filtered signal curve objectively shows a lag delay of about 1.5 seconds during the follow-down process. This response delay reflects the engineering trade-off made by the filtering algorithm in balancing data smoothness and state sensitivity.

[0055] The green curve obtained by transforming the filtered signal through the logarithmic path loss model represents the estimated physical distance. Its value steadily increases from the initial 2 meters to around 12 meters as the target moves. This estimated physical distance curve can accurately reflect the core behavioral trend of the pet moving away from the node, providing reliable spatial data support for the cross-verification of virtual and real positioning. However, during the static activity maintenance phase from the 20th to the 60th second, the curve still exhibits a slight oscillation of about 0.5 meters up and down. This indicates that the current transformation mapping mechanism still has room for optimization in the face of continuous microwave interference in a static environment, and there is still room to further introduce an environmentally adaptive dynamic calibration factor to improve the static positioning accuracy.

[0056] Phase 2: Terminal behavior deviation analysis and resource scheduling based on edge computing.

[0057] In real-world business scenarios, smart pet edge terminals are limited by their size, resulting in extremely limited lithium polymer battery capacity, typically between 300mAh and 500mAh. If the terminal continuously wakes up the GPS / BDS module and cellular radio frequency module at a fixed high frequency for location reporting, the terminal will run out of power within hours, completely losing its monitoring function.

[0058] Therefore, in this embodiment, edge computing logic is deployed locally on the smart pet edge terminal to calculate the comprehensive deviation of the current behavior. The location data reporting refresh frequency of the smart pet edge terminal is dynamically adjusted according to the level of the comprehensive deviation in order to perform wireless resource and location scheduling on the edge side.

[0059] Specifically, the step of calculating the comprehensive deviation of the current behavior locally on the smart pet edge terminal includes the following multi-dimensional feature analysis steps:

[0060] First, the terminal needs to obtain the individual behavior baseline issued by the cloud service platform. The individual behavior baseline includes historical high-frequency activity areas, baseline speed ranges, and inactive time periods. To accommodate the limited RAM and Flash storage space of the edge terminal, the cloud service platform will reduce the dimensionality of the complex geographic polygon vector data into a GeoHash encoding set or a minimum bounding rectangle (MBR) coordinate set before issuing the data, so that the terminal can quickly retrieve it.

[0061] During feature extraction, the terminal's main control MCU determines whether the current location of the smart pet edge terminal has deviated from the historical high-frequency activity area to determine the area deviation. If the latest satellite positioning coordinates fail to match any GeoHash encoding block of the historical high-frequency activity area, the marked area has deviated.

[0062] Subsequently, the MCU parses the ground speed field or integral IMU acceleration data in the satellite positioning message to calculate the proportion by which the current moving speed exceeds the baseline speed range to determine the speed deviation. Simultaneously, the terminal retrieves the timestamp from its built-in real-time clock (RTC) to determine whether the current time belongs to the inactive period (such as the early morning sleep period), and combines this with the IMU's motion status flag to determine whether the current state is abnormal, thereby determining the time period deviation.

[0063] After discretizing and normalizing the three single-dimensional features mentioned above, the terminal assigns corresponding preset weight coefficients to the region deviation, the speed deviation, and the time period deviation, and uses the weighted sum as the comprehensive deviation of the current behavior. The specific calculation model in this field is as follows:

[0064] ;

[0065] In the formula, This represents the overall deviation calculated so far, and its value is constrained to a closed interval between 0 and 1. The value is 1 when the region deviates from the quantified value and leaves the historical high-frequency activity area, and 0 when it does not leave the area. This indicates the speed deviation from the quantified value. Its value is equal to the difference between the current actual speed and the upper limit of the baseline speed range, and then divided by the upper limit of the baseline speed range. If it does not exceed the upper limit, the value is 0. This indicates that the time period deviates from the quantified value. The value is 1 when continuous motion occurs during an inactive period, and 0 otherwise. Represents the preset weighting coefficients for the region dimension; Represents the preset weighting coefficients for the velocity dimension; This represents the preset weighting coefficients for the time period dimension, where the sum of all preset weighting coefficients is 1. .

[0066] The practical function of this formula is to converge multi-dimensional discrete behavioral anomaly signals into a single continuous evaluation index through linear weighting, providing a clear mathematical criterion for subsequent baseband state machine switching.

[0067] Based on the aforementioned comprehensive deviation, the terminal executes the step of dynamically adjusting the location data reporting refresh frequency of the smart pet edge terminal according to the level of the comprehensive deviation. Specifically:

[0068] When the overall deviation is lower than the first deviation threshold, it is determined to be a low deviation state. The terminal sets the location data reporting refresh frequency to a basic low-frequency mode to reduce terminal communication power consumption. For example, at this time, the pet may be lying quietly in a familiar yard, and the baseband chip enters a deep sleep mode (such as the PSM mode of NB-IoT), only maintaining a heartbeat report once every 2 hours.

[0069] When the overall deviation is greater than or equal to the first deviation threshold and less than the second deviation threshold, it is determined to be a medium deviation state, and the terminal increases the location data reporting refresh frequency to the normal monitoring mode. At this time, the pet may move beyond expectations within the normal area, and the baseband chip switches to extended discontinuous reception mode (eDRX), set to wake up and report its location once every 10 minutes.

[0070] When the overall deviation is greater than or equal to the second deviation threshold, it is determined to be a high deviation state. The terminal switches the location data reporting refresh frequency to real-time tracking mode to improve the reporting accuracy of the location data. At this time, the pet may have escaped the familiar area and is moving quickly. The baseband chip maintains a full-duplex connection and forcibly increases the refresh frequency to once every 10 seconds.

[0071] Wherein, the second deviation threshold is greater than the first deviation threshold. Through the aforementioned closed-loop scheduling at the edge, the system autonomously achieves an adaptive energy consumption balance mechanism for high-risk, high-frequency and safe, low-frequency operations without relying on cloud computing power intervention.

[0072] like Figure 3 The simulation results of adaptive state switching of the intelligent pet edge terminal during the computing resource scheduling process are shown. The horizontal axis of the figure is the running time in minutes, the left vertical axis is the comprehensive deviation degree (dimensionless), and the right vertical axis is the refresh frequency state level (level).

[0073] The blue solid line in the figure represents the comprehensive deviation curve of the terminal's local continuous calculation output. The two dark gray dashed lines mark the first deviation threshold of 0.3 and the second deviation threshold of 0.7, respectively. The red stepped curve represents the refresh rate status level that is dynamically adjusted by the deviation.

[0074] As can be seen from the above, in the initial stage of 0 to 20 minutes, the overall deviation fluctuates around 0.15 and does not reach the first deviation threshold. At this time, the red state curve remains at the first level, indicating that the terminal is in the basic low frequency mode to reduce power consumption.

[0075] After 20 minutes, as the overall deviation increases and exceeds the first deviation threshold, the status curve jumps to the second level, and the terminal enters the normal monitoring mode.

[0076] After 40 minutes, the overall deviation exceeded the second deviation threshold, and the state curve climbed to the third level, triggering a high-frequency real-time tracking mode.

[0077] Through the above waveform transformation, it is objectively demonstrated that edge computing power can realize the tiered on-demand allocation of network resources based on the abnormal behavior of living entities.

[0078] Meanwhile, the local waveforms from the 29th to the 31st minute show that when the overall deviation briefly drops below 0.3 due to minor environmental disturbances, the red state curve rapidly falls back to the first level and then rebounds instantly. This frequent ping-pong jump effect near the critical threshold indicates that the current hard-decision triggering logic has limitations in dealing with boundary data fluctuations, demonstrating that the system still has room for improvement in subsequent iterations by introducing state transition hysteresis intervals or time delay confirmation mechanisms to enhance network stability and smoothness.

[0079] Phase 3: Cloud baseline construction and dynamic calibration of four-level zoned electronic fences.

[0080] On the cloud service platform side, the accumulated massive historical trajectory data contains in-depth characteristics of individual pet behavior patterns. In this embodiment, on the cloud service platform, density clustering is performed based on the historical trajectory data to construct an individual behavior baseline, and the boundary parameters of the four-level partitioned polygonal electronic fence are dynamically calibrated according to the individual behavior baseline.

[0081] Specifically, the process of constructing an individual behavior baseline by performing density clustering based on historical trajectory data includes rigorous spatial data cleaning and mining steps.

[0082] Because satellite positioning experiences multipath reflections in urban canyon environments, it can cause trajectory points to jump. Therefore, the cloud-based system first filters out positioning drift noise points in the historical trajectory data using a velocity threshold method to generate a set of valid trajectory points. The engineering filtering principle is as follows: retrieve the latitude and longitude coordinates and timestamps of two adjacent historical trajectory points, and use a spherical distance calculation model to calculate the actual instantaneous velocity.

[0083] For example, the instantaneous velocity filtering calculation formula is as follows:

[0084] ;

[0085] In the formula, This represents the actual instantaneous velocity between adjacent trajectory points, expressed in meters per second. Represents the Earth's mean radius constant; and These represent the latitude and longitude in radians of the preceding trajectory points, respectively; and These represent the latitude and longitude values ​​in radians for subsequent trajectory points, respectively. and These represent the absolute timestamps of the preceding and subsequent trajectory point records, respectively, in seconds.

[0086] When calculated If the speed exceeds the physiological movement limit threshold of the pet breed (e.g., 20 meters per second for dogs), the subsequent trajectory point is determined to be a positioning drift noise point and is removed, thereby ensuring that the subsequent clustering algorithm is not affected by extreme outliers.

[0087] Based on this, the cloud uses a density-based spatial clustering algorithm (specifically the DBSCAN algorithm) to perform high-frequency region clustering on the set of effective trajectory points, extracts multiple high-frequency activity regions, and generates the minimum hull convex polygon for each high-frequency activity region.

[0088] The DBSCAN algorithm does not require pre-specifying the number of clusters. Instead, it automatically identifies highly dense connected regions by configuring the neighborhood radius (Eps) and the minimum number of neighborhood samples (MinPts) of the core point. After identification, the cloud calls computational geometry algorithms such as Graham Scan or divide-and-conquer to extract the boundary contour of each high-frequency active region point set, generating a minimum hull convex polygon as an abstract regional feature.

[0089] It should also be noted that behavioral characteristics have not only spatial attributes but also temporal attributes.

[0090] The cloud platform divides a single natural day into multiple equal-length time periods, such as one hour per period, resulting in 24 time periods in total. It then calculates the average movement speed and activity duration within each time period to generate behavioral baselines for each time period. These high-frequency activity areas are combined with the time-specific behavioral baselines to generate the individual behavioral baseline. To adapt to changes in pets' behavioral habits with seasons or physiological cycles, the cloud platform uses a sliding window mechanism to update the individual behavioral baseline daily.

[0091] For example, the length of the sliding window can be set to 14 calendar days. At midnight every day, the cloud will automatically discard old trajectory data from 14 days ago and incorporate the latest trajectory data from the previous day, re-triggering the above clustering and feature extraction operations to achieve dynamic iteration and self-evolution of the baseline model.

[0092] After generating individual behavior baselines, the cloud platform adaptively plans the physical monitoring boundaries accordingly.

[0093] Traditional electronic fences are mostly user-drawn single circles or static polygons, lacking flexibility. In contrast, the four-level zoned polygonal electronic fence described in this embodiment is divided from the inside out into a core security zone, a regular activity zone, a warning buffer zone, and a danger zone.

[0094] The core security zone and the danger zone are typically defined by users using mobile devices with absolute static coordinates based on actual geographical barriers (such as house walls or nearby roads). The intermediate flexible boundary layer is dynamically generated using an algorithm. The regular activity zone is generated by taking the union of the polygons corresponding to the multiple high-frequency activity areas and then extending it outward by a preset margin (e.g., a fixed extension of 5 meters). The dynamic boundary width of the warning buffer zone is obtained by multiplying the basic buffer zone width by a dynamic scaling factor, which is the ratio of the average movement speed during the daytime active period in the time-segmented behavioral characteristic baseline to the standard reference speed. The dynamic calculation formula for its width is as follows:

[0095] ;

[0096] In the formula, This represents the calculated dynamic boundary width of the early warning buffer zone, in meters. This represents the system's preset base buffer width constant, for example, set to 15 meters; This represents the pet's historical average movement speed during its active daytime hours, extracted from the behavioral characteristic baseline of the specified time periods, in meters per second. This represents the standard reference speed constant for pets of this size class.

[0097] In practical applications, this formula works as follows: If a pet usually moves very fast and has strong sprinting power, its historical average moving speed is relatively high, and the calculated dynamic ratio coefficient is greater than 1. The system will automatically widen the warning buffer zone, for example, to 25 meters, so that sufficient response and warning time can be obtained in advance when the pet initiates a high-speed sprint to cross the boundary. Conversely, if the pet is old and weak and moves slowly, the warning buffer zone will be automatically narrowed to ensure the advance warning while releasing more space for normal activities and reducing false alarms caused by the small range.

[0098] like Figure 4 The diagram shows a topological bitmap of a cloud service platform that processes historical trajectory data and automatically generates zone fences. The horizontal and vertical axes of the diagram are both in meters.

[0099] In the graph, the scattered light gray crosshairs represent positioning drift noise data caused by multipath interference, while the dense dark blue solid circles represent valid high-frequency activity trajectory points retained after cleaning by the density clustering algorithm. The spatial distribution of the graph shows that the density clustering algorithm eliminated isolated outlier data far from the activity center.

[0100] In the construction of the polygonal fence, the rectangle formed by the solid green line and its light-colored fill area in the center-left area represents the core security zone statically defined by the user.

[0101] The polygon formed by the dark yellow solid line and its filled area represents the regular activity area. This area is generated by extracting the minimum hull convex polygon from the effective trajectory points and expanding it outward by a preset margin.

[0102] The orange solid-line polygon that extends further outward represents the early warning buffer zone, and its boundary is calculated by combining a dynamic scaling factor.

[0103] The vertical strip marked with a solid red line on the far right is the user-defined danger zone.

[0104] The nested distribution of the aforementioned fences demonstrates the technical effectiveness of the adaptive dynamic spatial calibration mechanism, which can provide a flexible and safe isolation buffer based on the actual movement trajectory of live pets.

[0105] Simultaneous observation of the graph reveals that within the range of 40 to 80 meters on the horizontal axis and 20 to 60 meters on the vertical axis, although there are no valid trajectory points in dark blue, the large blank area is still completely included by the dark yellow regular activity area due to the system's current use of a basic convex polygon envelope algorithm. This phenomenon of excessive spatial envelopment indicates that the current boundary fitting logic has limitations when dealing with complex terrains with significant concavity features, demonstrating the potential for further upgrades and improvements by introducing concave polygon contour edge fitting technology to achieve higher-precision boundary fitting.

[0106] Phase 4: Cross-verification of comprehensive risk scoring and virtual-real dual-layer positioning.

[0107] As the smart pet edge terminal continuously reports data, the cloud service platform needs to perform core risk assessment logic. Specifically, the cloud integrates location, movement, health, and behavioral deviation features to calculate a comprehensive risk score, and combines this with the multimodal positioning data set to perform cross-validation of boundary crossing events in both virtual and real-world positioning.

[0108] First, multi-dimensional features are extracted from the cloud. The location dimension feature is extracted from the area attributes of the current position of the smart pet edge terminal relative to the four-level polygonal electronic fence. The parity of the intersection points between the coordinate points and the boundaries of each polygon level is calculated using the ray casting algorithm to determine the region in which the pet is located. Instantaneous acceleration and rate of change of direction are extracted as motion dimension features. By analyzing the reported IMU data, it is determined whether there are panicked escape actions such as running or sudden turns. Physiological indicators collected by the vital signs sensors built into the smart pet edge terminal (such as abnormal heart rate spikes collected by the PPG sensor) are retrieved as health dimension features.

[0109] Next, the cloud maps the location dimension features, movement dimension features, health dimension features, and the comprehensive deviation degree into normalized single-dimensional risk scores, and performs a weighted sum based on preset weight coefficients for location dimension, movement dimension, health dimension, and behavior deviation dimension to obtain a comprehensive risk score used for boundary judgment.

[0110] For example, the formula for calculating risk aggregation is as follows:

[0111] ;

[0112] In the formula, This represents the overall risk score used to determine the level of transgression, with a value between 0 and 100. This represents the risk score based on location, with discrete values ​​determined by the region (0 for core area, 20 for regular area, 60 for buffer zone, and 100 for restricted area). This represents the risk score for the motion dimension mapped from the instantaneous acceleration variance. This represents a health dimension risk score mapped based on the proportion of physiological indicators such as heart rate that deviate from the normal heart rate range. This indicates the deviation risk score of behaviors reported by edge terminals or reviewed by the cloud. Each corresponds to a preset weight coefficient for each dimension, and the sum of the weight coefficients for each dimension is 1.

[0113] The application of this multi-dimensional perception model breaks the traditional logic of judging boundary crossing based on a single spatial location. For example, even if the pet's coordinates are temporarily within the warning buffer zone (moderate location risk), if its heart rate spikes abnormally (extremely high health risk) and it is moving outwards at high acceleration (extremely high movement risk), the overall risk score will still be pushed up to above the danger threshold, thereby achieving very early detection of potential escape events.

[0114] Meanwhile, in response to the practical problem of errors in the positioning medium, this embodiment also performs cross-verification of boundary crossing events by combining the multimodal positioning data set for virtual and physical dual-layer positioning, including the execution of positioning priority determination rules: when the local physical fence node detects the near-field signal of the smart pet edge terminal, and the strength of the near-field signal is higher than the near-field connection threshold, it is directly determined that the current location is within a safe area. The priority of this physical near-field determination is higher than the spatial calculation determination result of the four-level partition polygon electronic fence.

[0115] In practical applications, satellite positioning signals are prone to multipath drift of hundreds of meters when penetrating balconies or near building glass curtain walls, causing cloud-based spatial calculations to misjudge that a pet has instantly moved outside the danger zone. In such cases, cross-verification using near-field signals from physical nodes, provided that near-field high-frequency shortwave signals such as Bluetooth or Wi-Fi maintain a strong connection (typically within an indoor coverage range of 10 to 20 meters), can provide highly reliable proof of presence at the physical topology level, directly overturning the erroneous satellite virtual positioning results.

[0116] Optionally, when the physical near-field determination is within a safe area, the cloud service platform sends a control signal to the smart pet edge terminal, overriding the refresh frequency strategy based on the comprehensive deviation and forcing the smart pet edge terminal into a sleep positioning mode. This top-down signaling intervention not only suppresses false deviation calculations caused by satellite signal drift but also greatly blocks unnecessary network wake-ups, significantly extending the device's battery life.

[0117] Phase 5: Physiological state linkage and interactive hierarchical alarm closed loop.

[0118] This embodiment fully considers the physiological uncertainties of live animals and also includes a linkage mechanism between health data and fence control.

[0119] Specifically, when the cloud service platform detects that the physiological indicators in the health dimension features exceed the normal physiological threshold, such as a dog's resting heart rate suddenly exceeding 140 beats / minute or an abnormally high body temperature, it determines that there is an abnormal health condition. The system will automatically increase the weight coefficient of the health dimension, making it easier to trigger a high-risk alarm for the total score if any slight spatial deviation exceeds the limit.

[0120] Simultaneously, the cloud sends wireless scheduling commands to the smart pet edge terminal to force an increase in the location data reporting refresh frequency, and shrinks the physical boundary of the four-level zone polygonal electronic fence according to a preset ratio to tighten the alarm threshold. In the cloud-based geographic information engine, the shrinkage of the fence boundary physical area follows the following area shrinkage conversion formula:

[0121] ;

[0122] In the formula, This represents the total physical area of ​​the polygonal fence after recalibration under abnormal health conditions, expressed in square meters. The initial total physical area of ​​the polygonal fence generated from the original calculation of the individual behavior baseline; This represents the system's preset health abnormality penalty coefficient, used to adjust contraction sensitivity; This indicates the percentage by which the current physiological indicator deviates from the normal physiological threshold baseline.

[0123] The significance of this linkage mechanism lies in the fact that when pets are in a physiologically vulnerable period, such as when they are sick, stressed, in heat, or injured, the risk of them getting lost increases exponentially. After sensing this inherent physiological abnormality, the system proactively compresses the safe activity range (physical boundary) inward. This means that as soon as the pet steps into what is normally considered a safe edge, the system will issue an early warning, thus buying caregivers the golden intervention time during the most dangerous window.

[0124] After determining the final boundary crossing risk level, the following steps are executed: determining the boundary crossing risk level based on the cross-validation results of the boundary crossing event and the comprehensive risk score, and sending a corresponding level of alarm notification and quick action command to the mobile user terminal via the wireless push channel. Specifically:

[0125] When the risk level of exceeding the boundary is determined to be low, the status update is only performed within the application pre-installed on the mobile user terminal, without triggering system-level notification pushes, thus avoiding unnecessary frequency disturbances to the user's daily work and life.

[0126] When the risk level of crossing the boundary is determined to be medium risk, a normal notification signal and an in-application location viewing command are issued, and the user can see the normal prompt by pulling down the notification bar.

[0127] When the risk level of exceeding the boundary is determined to be high-risk, the device is in a high-risk state. The cloud service platform sends a strong alert notification signal by calling the highest priority service of system-level push channels such as Apple APNs or Android FCM, causing the device to vibrate or emit a specific alarm sound. More importantly, the push payload data packet carries quick action instructions for one-click navigation and forced activation of high-frequency positioning based on the Notification Service Extension framework.

[0128] On the system notification display interface of the mobile user terminal, in response to the user's click operation on the shortcut command, the operating system will bypass the cold start and UI rendering time of the application's main process and directly call the background wireless network communication interface to send control signals to the cloud service platform, without needing to wake up and enter the application's main interface. This forms a low-latency interactive closed loop: after hearing the high-risk alarm, the user does not need to unlock the phone to find the APP, skip the startup advertisement and find the corresponding control button. They only need to press or click the "Powerful Tracking" button on the lock screen notification interface, and the control signal will immediately penetrate the cloud and be sent to the communication baseband of the pet terminal with extremely low interaction latency, causing the device's buzzer to start or the LED to flash, greatly reducing the latency of user operation.

[0129] In summary, the solution presented in this embodiment effectively overcomes the technical problems faced by existing technologies in specific application scenarios. Existing technologies employ fixed positioning frequencies and static virtual coordinate fences, which cannot solve the serious satellite drift false alarm problem caused by multipath effects in urban scenarios, nor can they handle the battery life collapse problem caused by high-frequency positioning, and they lack adaptive adjustment mechanisms for the uncontrollable physiological behaviors of live animals.

[0130] The embodiments of the present invention have produced positive technical effects by upgrading the traditional passive, simple spatial determination to a multimodal adaptive sensing system based on biological behavioral baselines:

[0131] First, the edge-side refresh frequency scheduling based on comprehensive deviation and the cloud-based virtual-physical dual-layer cross-verification are combined to shield satellite drift noise at the physical layer and implement hibernation for non-essential activities, which significantly improves the overall battery life of the micro terminal.

[0132] Secondly, the cloud-based four-level dynamic fence, based on DBSCAN trajectory clustering and sliding window extraction, can self-correct as pets' habits change, and couples health signs with the fence contraction coefficient (dynamically contracting according to health status), achieving high-precision physical control that varies from pet to pet and changes with the time.

[0133] Finally, by bypassing system-level shortcut commands in the app interface, an air interface response link with extremely low latency was constructed. This solution closely meets the stringent requirements of high accuracy, long endurance, and real-time intervention capabilities in actual monitoring operations, and has significant practical application value and industrialization potential.

[0134] Example 2:

[0135] In actual monitoring operations, when pets escape abnormally due to fright or other stress reactions, their escape routes often deviate from the usual human activity areas and are very likely to enter physical spaces without cellular network coverage, such as underground garages, narrow alleys between dense buildings, culverts, or remote forest areas.

[0136] In such extreme scenarios, if relying solely on a single direct uplink, even if the edge terminal calculates an extremely high overall deviation locally, it will be unable to report multimodal positioning data and alarm signals to the cloud service platform, thus causing the communication loop of the entire wireless positioning and monitoring system to break.

[0137] To address the aforementioned technical challenges, this embodiment introduces a self-organizing network relay transmission mechanism in edge network blind spots to maintain the communication closed loop of the wireless positioning and monitoring system.

[0138] According to the technical solution of this embodiment, the self-organizing network relay transmission mechanism includes the following steps:

[0139] Specifically, the smart pet edge terminal continuously monitors its own cellular network uplink signal quality parameters.

[0140] In the physical layer implementation of the baseband chip, the NB-IoT or LTE-M RF module built into the smart pet edge terminal periodically measures the reference signal received power and reference signal received quality of the serving cell and neighboring cells during paging cycles or connected-state data transmissions. The edge terminal's main control microprocessor reads and parses the aforementioned RF channel parameters in real time by directly issuing standard AT commands (such as AT+CSQ or proprietary RF status query commands) to the cellular baseband module.

[0141] The above parameters directly reflect the air interface link loss and signal-to-noise ratio levels between the terminal and the cellular base station, and are the core objective indicators for assessing whether the terminal is in a network coverage blind spot.

[0142] It should be noted that when the uplink signal quality parameter of the cellular network is lower than the preset connectivity threshold, and the overall deviation is in the high deviation state, the near-field communication module built into the smart pet edge terminal switches from near-field broadcast mode to self-organizing network scanning mode. This triggering mechanism employs a strict double-conditional AND judgment.

[0143] In engineering, the preset connectivity threshold is typically defined as the critical level at which the baseband chip cannot maintain the most basic control plane signaling interaction. When the signal quality falls below this connectivity threshold, it means that the terminal has essentially been removed from the effective coverage of the cellular network.

[0144] However, in the conventional logic of existing technologies, after a terminal goes offline, it usually enters deep sleep to save power and initiates a periodic blind search process with gradually increasing intervals. But in this embodiment, combined with the comprehensive deviation calculated locally in Embodiment 1, if the terminal determines that it is not only offline but its current behavior has also triggered a high deviation state, it means that the pet is in a very high-risk escape process. Entering sleep at this time would severely delay the rescue opportunity.

[0145] Therefore, the terminal's main control microprocessor immediately triggers a hardware interrupt, sending a mode switching command to the built-in near-field communication module (such as a Bluetooth Low Energy (BLE) module or a Wi-Fi module) via the internal integrated circuit (I2C) or Serial Peripheral Interface (SPI) bus. Upon receiving the command, the near-field communication module suspends its current passive broadcast mode (Broadcaster / Peripheral Role), which is only acting as a peripheral device passively sending broadcast beacon packets, and switches to an active listening and connection initiation mode (Observer / Central Role), thereby changing the access permission of the communication medium from passive waiting to active addressing.

[0146] For example, after completing the hardware mode switch, the smart pet edge terminal encapsulates the unreported multimodal positioning data and the comprehensive deviation into an emergency relay data frame. The emergency relay data frame is attached with a survival time tag and is broadcast to the surrounding area through the near-field communication module that has switched to the self-organizing network scanning mode.

[0147] During the protocol encapsulation process at the data link layer, in order to adapt to the limitation of the maximum transmission unit of the near-field communication protocol, the main control microprocessor performs binary structure compression and serialization processing on the unreported satellite positioning coordinates, near-field positioning signal strength, timestamp, and quantized comprehensive deviation value to form the application layer payload.

[0148] Subsequently, a specific identifier field is added to the header of the payload. This identifier field is defined as an emergency relay request in the protocol stack, thus forming a complete emergency relay data frame.

[0149] In particular, to prevent the emergency relay data frame from being forwarded infinitely in a complex ad hoc network environment, thereby triggering a "broadcast storm" that exhausts the computing power and bandwidth of all nodes, a time-to-live tag must be attached to the header of the data frame.

[0150] The survival time tag can be a timestamp validity period or a hop count decrementing counter (HopLimit). The near-field communication module broadcasts this emergency relay data frame to the physical space at maximum transmission power and high frequency on a preset emergency broadcast channel (such as channels 37, 38, and 39 of BLE), expecting a response from potential surrounding nodes.

[0151] Optionally, when adjacent heterogeneous edge nodes within the cellular network coverage area detect the emergency relay data frame through their own near-field communication protocol, the adjacent heterogeneous edge nodes calculate and generate forwarding priority weights based on their own remaining power level and the current uplink signal-to-noise ratio parameter of the cellular network.

[0152] The adjacent heterogeneous edge nodes here refer to third-party terminal devices with wireless communication relay capabilities that happen to intersect with the escaped pet in physical space. These nodes may be other pets that are safe and wearing homogeneous collars, smartphones carried by community residents walking by, low-speed unmanned delivery vehicles with vehicle-to-everything (V2X) relay functions, or smart street light gateways deployed on urban infrastructure.

[0153] Because these heterogeneous nodes have different hardware resources and different network conditions, the system cannot require all nodes that hear the broadcast to unconditionally initiate relaying, otherwise it would lead to serious uplink channel conflicts and redundant transmissions. Therefore, after parsing the emergency relay request identifier, the listening node will start a local evaluation algorithm to calculate the forwarding priority weight that represents its own relay capability.

[0154] To accurately quantify this evaluation process, this embodiment introduces the following specific calculation model for forwarding priority weight:

[0155] ;

[0156] In the formula, This represents the forwarding priority weight calculated by adjacent heterogeneous edge nodes; this value is a dimensionless continuous numerical value. This represents the estimated remaining battery energy of the adjacent heterogeneous edge node, in milliampere-hours (mAh). This indicates the design nominal battery energy of the adjacent heterogeneous edge node under full charge, in milliampere-hours (mAh). This represents the actual measured uplink signal-to-noise ratio between the adjacent heterogeneous edge node and the cellular base station, as currently monitored by the node, in decibels. This represents the standard reference signal-to-noise ratio constant preset by the system for cellular networks, used to normalize the actual signal-to-noise ratio. This represents the weighting adjustment coefficient for the power consumption dimension; This represents the weight adjustment coefficient for the signal dimension.

[0157] Through the calculation using the above formula, a node's calculated forwarding priority weight will only be at a high level when it has sufficient remaining power (enough to support additional baseband transmission power consumption) and is within cellular network coverage with signal quality exceeding the baseline threshold (ensuring reliable delivery of relay data to the cloud in a single operation). The direct application of this mathematical model in practical engineering is to achieve the survival of the fittest among relay resources, automatically selecting the highest-performing "stepping stone" nodes with the strongest forwarding capabilities in the current physical space.

[0158] like Figure 5 The simulation results of a three-dimensional evaluation of forwarding priority weights are presented when adjacent heterogeneous edge nodes initiate ad hoc network relay transmissions in edge network blind spots. In the figure, the horizontal axis represents the remaining battery energy percentage (dimensionless), the vertical axis represents the normalized uplink signal-to-noise ratio (dimensionless), and the vertical axis represents the forwarding priority weights (dimensionless).

[0159] The three-dimensional colored surface in the figure, which gradually changes from a cool dark blue tone to a warm yellow tone, represents the priority weight mapping relationship under different combinations of power and communication network status.

[0160] Observing the surface morphology, it can be seen that when the remaining battery energy ratio and the normalized uplink signal-to-noise ratio gradually increase and are in the high range, the surface rises synchronously and presents a yellow peak state. The weight value eventually reaches the saturation upper limit of 0.9. This objectively reflects that the system can accurately select heterogeneous nodes with sufficient power and excellent channel quality as emergency relay jumpers, ensuring the reliability of multimodal positioning data transmission.

[0161] Meanwhile, when the remaining battery energy ratio drops to a critical position of 0.2, the three-dimensional surface exhibits a vertical step-like downward waveform transformation, meaning that the priority weight is directly forced to zero. This vertical drop characteristic based on fixed numerical hard truncation indicates that the current evaluation model lacks a smooth transition when dealing with edge nodes in a critical energy state. This can easily lead to the direct elimination of some nodes that still have short-term relay capabilities, potentially missing the opportunity to establish temporary data links in a physically sparse node space. This demonstrates that the mapping mechanism has room for improvement in the future by introducing a smooth decay function to enhance the tolerance of relay node selection and the resilience of network connectivity.

[0162] Specifically, in response to the forwarding priority weight being higher than a preset relay threshold, the adjacent heterogeneous edge node establishes a temporary data link with the smart pet edge terminal, extracts the emergency relay data frame, and transmits the emergency relay data frame to the cloud service platform through the cellular network communication link of the adjacent heterogeneous edge node.

[0163] In actual Media Access Control (MAC) layer communication sequences, considering that there may be multiple heterogeneous nodes around an escaped pet that meet the above-mentioned preset relay threshold, if these nodes simultaneously attempt to establish a connection with the pet terminal, strong co-channel interference and channel collision will occur.

[0164] To address this concurrent access collision issue, adjacent heterogeneous edge nodes will calculate a temporary data link backoff time window based on the calculated forwarding priority weights before initiating a connection request.

[0165] For example, the control algorithm for this backoff time window is as follows:

[0166]

[0167] In the formula, This represents the backoff delay time, in milliseconds, that adjacent heterogeneous edge nodes need to wait before initiating a temporary data link establishment request. This indicates the default maximum latency constant of the protocol, for example, set to 500 milliseconds; This represents an exponentially decaying adjustment factor used to control the convergence of backoff time with priority weights, and its value is a positive number. This represents the time of tiny random jitter generated by the hardware random number generator, in milliseconds, used to prevent continuous collisions between two nodes with the same priority.

[0168] Based on the mathematical monotonicity of this formula, the forwarding priority weights of adjacent heterogeneous edge nodes can be determined. The higher the value, the smaller the exponent in the formula, and the faster the calculated backoff delay time. The shorter the countdown, the better. This means that the best nodes will end their backoff countdown as quickly as possible, seize the wireless channel first, and send a connection request (Connect_REQ).

[0169] After successfully establishing a temporary data link, such as a Bluetooth GATT connection or a Wi-Fi Direct connection, adjacent heterogeneous edge nodes extract the received emergency relay data frame from the underlying buffer. In the application layer logic, the adjacent heterogeneous edge nodes do not decrypt or tamper with the core service payload of this data frame. Instead, they encapsulate it as an opaque data block in the communication payload of protocols such as MQTT or CoAP, and directly invoke their own cellular network uplink channel to transparently transmit the data block to the designated cloud service platform gateway address.

[0170] like Figure 6 The simulation results of the backoff time generated by adjacent heterogeneous edge nodes based on their own evaluation status after receiving an emergency relay broadcast are shown. The horizontal axis of the figure is the forwarding priority weight with a dimension of 1, and the vertical axis is the backoff delay time with a unit of milliseconds.

[0171] The blue solid line in the figure, which shows a continuous and smooth downward trend, represents the theoretical retreat decay curve. The orange-red solid dots distributed around the curve represent the simulated scatter points of the actual retreat time after superimposing random hardware jitter.

[0172] Observing the waveform transformation in the figure, it can be seen that when the forwarding priority weight is in the low range of 0 to 0.3, the backoff delay time remains in the high range of 300 milliseconds to 500 milliseconds. This means that inferior nodes with insufficient power or congested channels will back off for a long time and actively give up the right to occupy the wireless medium.

[0173] As the weights increase, the theoretical curve exhibits a steep exponential decline. When the weights reach the high priority range of 0.8 to 1.0, the theoretical backoff delay time is rapidly compressed to an extremely low level of 10 to 50 milliseconds. This non-linear hierarchical division in the time domain effectively ensures that high-quality nodes can seize the channel and establish temporary data links as quickly as possible, objectively demonstrating the technical effectiveness of the system in breaking communication conflicts in complex blind zone environments.

[0174] However, observing the scatter distribution in the lower right corner of the graph reveals that in the high-priority interval with a weight greater than 0.8, the orange-red scatter points representing the actual backoff time are highly dense and severely compressed into an extremely narrow bottom time window. Because the system uses a random jitter time with a fixed amplitude, the actual backoff times of multiple high-priority nodes in this interval are highly likely to overlap. This overly crowded time slot distribution in the high-weight region indicates that the current backoff algorithm still faces a small probability of co-frequency collisions when dealing with extremely dense concurrent responses from high-priority heterogeneous nodes. This demonstrates that the proposed solution has room for improvement in future evolution by introducing an adaptive jitter time dynamic widening mechanism that is negatively correlated with priority weight, thereby further widening the time slot gap between high-priority nodes.

[0175] It should also be noted that the cloud service platform parses the emergency relay data frames that arrive through the transmission, and continues to perform the step of calculating a comprehensive risk score based on the parsed data by fusing features of location dimension, motion dimension, health dimension and behavior deviation dimension, while attaching a blind zone relay identifier to the generated alarm notification.

[0176] In the backend microservice architecture of the cloud system, when the message gateway receives a data packet reported by a heterogeneous third-party node IP address, it routes the data stream to a specific relay unpacking service node based on the emergency relay request identifier in the packet header. The cloud uses the asymmetric encryption private key negotiated during the initial identity binding phase to decrypt the ciphertext payload of the emergency relay data frame and extract the original status data, such as the positioning coordinates and overall deviation, which were originally accumulated on the edge terminal due to blind spots.

[0177] Subsequently, the cloud computing node resumes the core assessment process from Example 1, re-inputting the extracted feature data into the risk assessment model to calculate the comprehensive risk score at this moment. Furthermore, since this data is not directly reported by the pet edge terminal, its reporting path itself implies the extremely dangerous spatial context information that the target has left the regular communication coverage area.

[0178] Therefore, in the subsequent high-risk alert notifications sent to mobile user terminals, the cloud will forcibly write a specific blind zone relay identifier status bit. During the rendering of the mobile user terminal's front-end interface, this status bit will trigger special visual and interactive effects, such as changing the regular red boundary crossing warning to a "Last Known Location of Blind Zone" or "Relay Recovery Route" alert with a specific icon, thus providing valuable decision-making reference information for the user's subsequent physical search and rescue operations.

[0179] In summary, based on an objective comparison with existing technologies, conventional pet monitoring devices generally suffer from inherent defects when encountering network blind spots, such as prolonged interruption of location information, unknown status during offline periods, and delayed historical alarms after communication is restored.

[0180] The technical solutions described in this invention, by deeply coupling the network uplink channel quality with the comprehensive deviation of biological behavior calculated locally by the terminal using state machine logic, not only solve the link reconstruction problem in a network-free environment at the hardware level, but also endow the entire IoT system with biological stress response attributes. When a pet is in a safe state and occasionally encounters signal obstruction, the system can remain silent to maintain continuous operation.

[0181] When a real danger occurs accompanied by a communication disruption, the system can automatically organize surrounding heterogeneous computing resources to construct a temporary escape communication channel through specific backoff algorithms and priority weighting mechanisms. This solution, while ensuring data transmission reliability and preventing channel collision storms, significantly extends the monitoring boundaries of traditional single-base station coverage systems. It achieves significant and objective technological progress in ensuring tracking success rates in complex environments and reducing the permanent loss rate of pets, demonstrating strong engineering feasibility and industry application value.

[0182] Example 3:

[0183] In real-world monitoring environments, such as large pet theme parks, animal breeding facilities, veterinary hospitals, or stray animal shelters, there is typically a very high density of devices deployed. In these scenarios, dozens or even hundreds of animals wearing the aforementioned smart pet edge terminal may simultaneously gather within the radio frequency coverage of a single local physical fence node (such as a high-power low-power Bluetooth gateway or an indoor Wi-Fi probe).

[0184] Under normal conditions, each terminal reports data according to its own low-frequency cycle, and the channel can still bear the load; however, when sudden external stimuli occur (such as loud construction noise, thunderstorms, or stray dogs entering), it is very easy to trigger a mass stress response.

[0185] According to the technical logic of the aforementioned embodiments, mass stress causes a large number of terminals to instantly jump their locally calculated comprehensive deviation to a high deviation state, thereby simultaneously triggering an increase in the location data reporting refresh frequency. This massive concurrent upload requests will cause severe co-channel interference and packet collision at the physical channel layer, resulting in continuous blockage of the radio frequency receiving channel of the local physical fence node. At this time, data packets from high-risk terminals that actually physically cross the boundary will be lost due to continuous physical layer backoff or transmission failure, causing severe delays or partial paralysis of the monitoring system.

[0186] To address the technical challenges in this specific scenario, this method, specifically for multi-terminal concurrent coverage scenarios, also includes a local wireless channel resource priority scheduling mechanism based on comprehensive deviation.

[0187] For example, the first step of this scheduling mechanism is to objectively and quantitatively perceive the current air interface environment. The local physical fence node continuously counts the wireless channel load rate within its own wireless coverage area.

[0188] In the interaction between the hardware physical layer and the media access control layer (MAC layer), the baseband RF module of the local physical fence node performs the Clear Channel Assessment (CCA) mechanism. The microcontroller (MCU) samples the channel energy frequently within a preset time observation window (e.g., every 1000 milliseconds as a statistical period).

[0189] The microcontroller counts the number of sampling time slots where the Received Signal Strength Indication (RSSI) exceeds the Energy Detection Threshold. The ratio of this number of time slots, which is considered "busy," to the total number of time slots in the observation window, is defined as the wireless channel load factor. This parameter objectively reflects the density of current space wireless medium occupancy.

[0190] Optionally, the local physical fence node will maintain a status monitoring timer locally and send a channel congestion status signaling to the cloud service platform when the wireless channel load rate exceeds a preset congestion threshold.

[0191] In engineering settings, the preset congestion threshold is typically set as an empirical critical value between 75% and 85%. When the channel load rate exceeds this threshold, it indicates that the traditional Carrier Sense Multiple Access / Collision Avoidance (CSMA / CA) mechanism has approached its theoretical throughput limit, and continuing to rely on blind retransmissions from terminals will lead to a sharp decline in channel utilization.

[0192] At this point, the local physical fence node immediately suspends the transmission queue of downlink regular data packets and constructs a high-priority channel congestion status signaling. The header of this signaling frame not only contains a congestion flag but also includes the current node's unique device identifier (MAC address) and the currently observed precise wireless channel load rate. This signaling is pushed to the cloud service platform with extremely low transmission latency via the local physical fence node's high-speed broadband backhaul link (such as Gigabit Ethernet or a 5G cellular uplink channel), thereby triggering the cloud's global resource reallocation logic.

[0193] It should be noted that, in response to the channel congestion status signaling, the cloud service platform extracts the comprehensive deviation of all smart pet edge terminals currently within the coverage area of ​​the local physical fence node, and uses the comprehensive deviation as the descending sorting criterion to generate a channel access priority sequence.

[0194] Upon receiving the congestion status signal, the load balancing gateway of the cloud service platform will immediately wake up the corresponding resource scheduling microservice. The microservice module first queries the distributed in-memory database to retrieve the device context information of all smart pet edge terminals whose last location node belonged to the congested local physical fence node within a preset time period (e.g., 5 minutes).

[0195] Subsequently, the cloud extracts the latest overall deviation value cached by these terminals in the previous round of communication. Since the overall deviation value directly reflects the pet's current level of abnormal behavior and risk of crossing boundaries, it is incorporated into the underlying network scheduling, enabling cross-layer collaboration between business logic and communication medium access control. The cloud service platform executes a quicksort or mergesort algorithm to strictly sort all extracted terminals in descending order according to the magnitude of the overall deviation value, generating a one-dimensional channel access priority sequence.

[0196] In this sequence, the smaller the index value (the higher the ranking) of the terminal, the higher its overall deviation, and the more urgent the risk of out-of-bounds escape it faces. Therefore, it needs to obtain higher transmission privileges in the subsequent channel allocation.

[0197] Specifically, in order to transform the sorting results in the cloud into executable physical layer rules on the edge side, the cloud service platform generates asymmetric medium access control parameters based on the channel access priority sequence, and encapsulates the asymmetric medium access control parameters into resource allocation signaling and sends them to the corresponding smart pet edge terminals.

[0198] The so-called asymmetric media access control parameters refer to breaking the principle of equal competition for channels among all terminals in the original protocol, and artificially allocating differentiated backoff time ranges and communication time slot resources to devices with different risk levels.

[0199] For example, to accurately map the overall deviation of each terminal to its upper limit exponent of the contention window in the CSMA / CA protocol, this embodiment introduces a first resource control mapping algorithm. The complete formula of this algorithm is as follows:

[0200] ;

[0201] In the formula, This represents the upper limit index of the competition window calculated and generated by the cloud service platform for a specific smart pet edge terminal; This represents the system's preset minimum window exponent constant, which determines the minimum physical layer backoff time allowed by the protocol. This indicates the adjustable range step size of the competition window index; This represents the relative value of the overall deviation of the specific smart pet edge terminal under the current congested local node, after normalization within the cluster. It is different from the absolute value calculated locally by the terminal and is specifically used to characterize the relative high-risk ranking of the terminal in the current concurrent group. The value range is between 0 and 1. This represents the exponential smoothing factor used to adjust the curvature of the mapped curve, and is used to control the sensitivity of the nonlinear mapping.

[0202] Based on the inverse proportionality of the above formula, it is clear that: when the overall deviation of the terminal... The larger the value (i.e., the higher the risk of exceeding the boundary in a high-deviation state), the closer the product term in the second half of the equation approaches zero, thus increasing the upper limit exponent of the generated competition window. The smaller.

[0203] The upper limit of the physical layer backoff time is exponentially compressed, meaning that when the terminal detects a busy channel, the generated random backoff time will be extremely short, enabling channel probing and preemption at a higher frequency. In the specific mapping logic, when... When the value is less than the preset median exponent, the corresponding backoff time interval constitutes the first backoff time window described in this embodiment; when When the index is greater than or equal to the median value, the corresponding longer backoff time interval constitutes the second backoff time window described in this embodiment, thereby ensuring that the underlying hardware parameter settings correspond completely to the backoff time window defined above.

[0204] like Figure 7 The simulation mapping results of the cloud service platform allocating asymmetric media access control parameters in a multi-terminal concurrent scenario are shown. In the figure, the horizontal axis is the relative value of the normalized comprehensive deviation, with a dimension of 1, and the vertical axis is the competition window upper limit exponent, with a dimension of 1.

[0205] In the figure, the blue dashed line represents the theoretical continuous mapping curve, and the orange-red stepped solid line represents the actual discrete stepping curve after hardware rounding.

[0206] Observing the waveform trend, it can be seen that as the relative value of the normalized overall deviation gradually increases from 0 to 1, the theoretical curve exhibits a non-linear, inversely proportional, accelerating downward trend, causing the value of the actual deviation curve to decrease from a maximum of 10 to a basic lower limit of 2. This non-linear downward characteristic intuitively reflects that when the channel is congested, the system can significantly compress the upper limit exponent of the contention window corresponding to terminals in a high deviation state, thereby allocating them a shorter first backoff time window to prioritize channel preemption and effectively reducing the data transmission delay of high-risk terminals.

[0207] Meanwhile, observing the orange-red solid line in the high-risk deviation range of 0.6 to 1.0 on the horizontal axis reveals that, due to the hardware's physical constraint that the contention window index must be an integer, the actual discrete stepping curve forms a wide, flat, stepped platform within this range. This coarse-grained discrete quantization effect results in multiple high-risk terminals with subtle risk differences within this range being assigned the exact same backoff window index. Consequently, even in extremely dense concurrent contention, the risk of collisions at the same frequency still exists. This objectively indicates that the current resource mapping mechanism can be further improved by introducing a decimal-level dynamic time compensation factor or an adaptive subdivision step size strategy to smooth out quantization errors. This demonstrates that there is still room for further improvement in enhancing the anti-collision isolation of high-priority nodes.

[0208] Alternatively, relying solely on the CSMA / CA mechanism, which shortens the backoff window, may still result in a small probability of collisions under extremely dense concurrency. To provide deterministic transmission guarantees for terminals at the highest risk level, the cloud system also allocates dedicated time slots based on the Time Division Multiple Access (TDMA) system to the top-ranking devices.

[0209] Here, a second time slot resource allocation algorithm is introduced to calculate the allocation ratio of dedicated, non-contact time slots:

[0210] ;

[0211] In the formula, This represents the percentage of a dedicated, non-contending time slot allocated to a specific smart pet edge terminal within its respective scheduling cycle. This represents the scheduling sensitivity tuning parameter, which takes the value of a real number greater than 0. The larger the value, the more drastically the allocation result is tilted towards high-priority terminals. This represents the normalized ranking score of the current terminal in the channel access priority sequence (the higher the ranking, the higher the score). This represents the total number of active smart pet edge terminals currently participating in resource reallocation; This represents the normalized ranking score of the nth smart pet edge terminal in the channel access priority sequence.

[0212] This type of Softmax exponential allocation model ensures that channel time slot resources are exponentially concentrated towards high-risk devices, providing accurate data payload support for generating resource allocation signaling. The cloud service platform packages and serializes the generated contention window index and dedicated time slot configuration parameters into resource allocation signaling, and quickly distributes it to various terminals through the downlink broadcast channel or unicast control channel of the local physical fence node.

[0213] It is also important to note that after receiving the resource allocation signaling, each of the aforementioned smart pet edge terminals adjusts its local data transmission mechanism according to the asymmetric media access control parameters. After parsing and verifying the validity of the resource allocation signaling, the main control microprocessor of the smart pet edge terminal performs a hot update of the MAC layer state machine by modifying the hardware registers of the baseband chip or overwriting the memory variables of the underlying protocol stack.

[0214] Specifically, the adjustment rules for the underlying mechanism in this embodiment are as follows:

[0215] On the one hand, for smart pet edge terminals with a comprehensive deviation in the high deviation state, a dedicated non-contact time slot and a first backoff time window are allocated to ensure priority transmission of out-of-bounds data.

[0216] When the terminal determines that its configuration falls into the high deviation state range, the microprocessor will configure a local high-precision hardware timer based on the received time slot start time and duration parameters.

[0217] When the microsecond-level hardware timer is triggered, the baseband RF module bypasses any carrier sensing action and forcibly activates the RF transmitter within its allocated, contention-free time slot, directly pushing data packets containing critical multimodal positioning data such as satellite positioning coordinates and instantaneous acceleration into the air interface. This approach, similar to reserving a channel, fundamentally eliminates the possibility of data collisions at the air interface.

[0218] Meanwhile, even if the data in this dedicated time slot is not successfully acknowledged by the node due to sudden channel fading (no ACK feedback), the terminal uses an extremely short first backoff time window (determined by the upper limit exponent of the small contention window in the aforementioned formula) when it falls back to the CSMA / CA contention phase. After waiting for an extremely short random number of microseconds, the terminal will attempt to transmit again, forming a dual transmission privilege of absolute guarantee and priority contention.

[0219] On the other hand, for smart pet edge terminals with a comprehensive deviation in the low deviation state, a shared competition time slot and a second backoff time window are allocated to reduce the probability of collision of co-frequency signals through physical layer backoff.

[0220] If the terminal confirms that it is in a low-deviation state after parsing the parameters, such as pets sleeping safely in a crowded cage, although the number is large but the risk is extremely low, its baseband chip will be prohibited from sending any signal during the period of the dedicated time slot.

[0221] Conversely, these low-risk terminals must compete for the channel using standard or degraded carrier sensing mechanisms within the shared contention time slots specifically allocated by the local entity fence nodes.

[0222] Even more stringent is the fact that these terminals are allocated a significantly amplified second backoff time window. During actual communication, as soon as the channel is detected to be busy, the backoff counter inside these low-risk terminals will generate an extremely long random waiting period, actively relinquishing the right to occupy the wireless medium; wherein, the duration of the second backoff time window is longer than the duration of the first backoff time window.

[0223] This asymmetric weighting process effectively dilutes and disperses the concurrent conflicts of a large number of low-value periodic heartbeat packets or redundant location data on the timeline.

[0224] Optionally, after the above-mentioned control signaling and terminal parameter adjustment are completed, the local physical fence node, as the receiving end of the network center node, receives multimodal positioning data reported by each smart pet edge terminal in time slots under the communication mechanism of the above-mentioned asymmetric medium access control parameters, so as to alleviate channel congestion and maintain the positioning concurrency capability in multi-terminal scenarios.

[0225] The radio frequency receivers of the local physical fence nodes operate according to a scheduling schedule synchronized with the cloud. During the contention-free TDMA time slot phase, the nodes sequentially and deterministically receive alarm data packets transmitted by high-risk pets; while during the shared contention time slot phase, the nodes process heartbeat data packets from low-risk terminals that arrive randomly after random backoff.

[0226] Once the load rate falls below the healthy level within a certain time window, the node will report the status signaling again to notify the cloud to restore the normal fair competition mode, thus forming a closed loop of elastic and self-healing underlying communication resource scheduling.

[0227] In general, existing conventional IoT devices typically rely solely on the standard Binary Exponential Backoff (BEB) algorithm to alleviate congestion when faced with a large number of concurrent collisions. While this mechanism can reduce the probability of collisions to some extent, it is essentially indiscriminate and equal. It cannot recognize the urgency of the business logic carried behind data packets, resulting in a situation where a pet terminal in extreme danger of getting lost receives a completely equal probability of channel preemption at the physical layer compared to a pet terminal in absolute safety that is sound asleep. Under severe congestion, this "fairness" often leads to unacceptable transmission delays for high-risk data, or even the data being discarded by the system.

[0228] The solution in this embodiment uses the concept of cross-layer design to transmit the comprehensive deviation (i.e., individual behavioral risk assessment index) located in the application layer to the physical layer and the media access control layer, which serves as the decision-making basis for directly controlling the allocation of wireless channel resources and timing scheduling.

[0229] In real-world scenarios with numerous terminals, this mechanism not only prevents overall network paralysis during signaling storms but also ensures that limited wireless frequency resources are always allocated to the highest-risk nodes through an asymmetric resource allocation strategy that uses dedicated, contention-free time slots and dynamically adjusted backoff time windows. This significantly reduces the absolute physical delay from the occurrence of high-risk boundary crossing events to successful data reporting, solidifying the reliability of the entire multimodal intelligent positioning and monitoring system from its underlying communication foundation. It possesses strong industrial practicality and significant technological advancement attributes.

[0230] Example 4:

[0231] In response to the objective technical problems mentioned in the background art, such as the conventional pet monitoring system using a single satellite positioning and fixed time interval reporting leading to rapid power consumption or delayed boundary alarms, and the static fixed electronic fence ignoring individual differences of pets easily causing false alarms and missed alarms, this embodiment provides a high-precision positioning system for the edge of an intelligent pet positioning fence based on multimodal fusion.

[0232] This system, through the coordinated linkage of distributed hardware modules and the reasonable allocation of edge computing power, concretizes the processing steps in the method embodiment into physical devices for actual operation, thereby realizing a closed loop of dynamic frequency control and high-precision interactive alarm.

[0233] like Figure 8 As shown in the figure, the intelligent pet positioning fence edge high-precision positioning system based on multimodal fusion disclosed in this embodiment includes an interconnected intelligent pet edge terminal, a local physical fence node, a cloud service platform, and a mobile user terminal.

[0234] In the actual network topology, the above four main physical entities interact with each other through wireless wide area network (WWAN), wireless local area network (WLAN) and short-range wireless personal area network (WPAN), forming a complete Internet of Things (IoT) entity system that includes data perception, edge decision-making, cloud evaluation and user terminal presentation.

[0235] On the near-field layer side of data perception, the local entity fence node is used to collect near-field signals from the smart pet edge terminal and transmit near-field positioning data back to the cloud service platform via a wireless link.

[0236] In practical applications, this local physical fence node is specifically a fixed wireless gateway device deployed at the boundaries of a user's indoor or outdoor yard, or in areas where pets congregate. In terms of hardware selection, this node typically uses a Bluetooth (BLE) gateway supporting the Bluetooth Low Energy protocol stack, a Wi-Fi probe device, or a LoRa IoT base station. Internally, the local physical fence node is equipped with a microprocessor (such as an ARM Cortex-A series or M series main control chip) and an omnidirectional radio frequency antenna array.

[0237] During operation, the node continuously scans the surrounding radio frequency channels, captures physical layer broadcast packets emitted by smart pet edge terminals within its coverage area, and extracts the Media Access Control (MAC) address and Received Signal Strength Indicator (RSSI) from them.

[0238] After the microprocessor performs Kalman filtering or moving average filtering to reduce noise in the original RSSI signal, it converts it into a spatial distance quantity through a logarithmic path loss model, and combines it with the fixed latitude and longitude coordinates pre-calibrated by the gateway to package it into a near-field positioning data packet containing a valid timestamp.

[0239] Subsequently, the node transmits the aforementioned data back to the cloud service platform via a high-bandwidth wireless link such as an internally integrated gigabit Ethernet interface, fiber broadband modem, or 5G / 4G cellular network interface, thereby providing physical-level high-precision on-site positioning proof in complex indoor environments where satellite signals are obstructed.

[0240] On the mobile layer side of data perception, the smart pet edge terminal is used to report satellite positioning data to the cloud service platform through the cellular network; and to calculate the comprehensive deviation of the current behavior locally, and dynamically adjust the location data reporting refresh frequency of the smart pet edge terminal according to the level of the comprehensive deviation, so as to perform wireless resource and positioning scheduling on the edge side.

[0241] The terminal is designed as a smart collar, vest tracker, or pendant that can be worn directly by live pets. Internally, it integrates an ultra-low power microcontroller (MCU), a GPS / BDS dual-mode Global Navigation Satellite System (GNSS) receiver chip, a narrowband Internet of Things (NB-IoT) or LTE-M cellular baseband chip, a microelectromechanical system (MEMS) six-axis inertial sensor, and an optical vital signs detection module (such as a PPG heart rate sensor).

[0242] The terminal's embedded microcontroller runs a lightweight real-time operating system (RTOS) and executes edge computing logic locally. To resolve the contradiction in the background technology where fixed-frequency reporting leads to battery depletion or positioning delays, the MCU reads sensor data in real time and calculates the overall deviation of behavior with the support of local computing resources.

[0243] When a pet is in a normal sleeping or calm activity state, the overall deviation level is low, and the MCU controls the cellular baseband chip to enter power-saving mode (PSM), reporting data at an extremely low frequency to reduce power consumption. When the inertial sensor detects abnormal acceleration, or coordinate system comparison reveals a deviation from the normal area, causing the overall deviation level to increase, the MCU directly triggers an interrupt signal at the hardware level, waking up the RF module and switching to high-frequency real-time tracking mode. This edge scheduling mechanism at the hardware level effectively ensures the timeliness of location data in high-risk situations.

[0244] At the core decision-making level of data processing and control, the cloud service platform is used to acquire a multimodal positioning data set composed of satellite positioning data and near-field positioning data; perform density clustering based on historical trajectory data to construct an individual behavior baseline, and dynamically calibrate the boundary parameters of the four-level partitioned polygonal electronic fence according to the individual behavior baseline; calculate a comprehensive risk score by fusing features of location dimension, motion dimension, health dimension, and behavior deviation dimension, and perform cross-verification of boundary crossing events in combination with the multimodal positioning data set; determine the boundary crossing risk level based on the results of the cross-verification of boundary crossing events and the comprehensive risk score, and send alarm notifications and quick action commands of the corresponding level to the mobile user terminal through a wireless push channel.

[0245] Physically, this cloud service platform is a cloud computing infrastructure comprised of a load balancing gateway, a cluster of multiple high-performance computing servers, a distributed relational database, and a non-relational in-memory cache database. Its internally deployed geospatial data engine (GIS Engine) periodically extracts massive amounts of historical trajectory data from the database, calls the DBSCAN density clustering algorithm module for batch processing calculations, thereby extracting the high-frequency activity areas of each independent terminal and generating baseline boundary vector data for the electronic fence.

[0246] Meanwhile, the risk assessment microservice continuously monitors and performs multi-dimensional weighted calculations on streaming data such as spatial location, physiological health parameters, and motion acceleration reported by the terminal. This cloud-based computing mechanism directly overcomes the mechanical false alarms caused by a single static fence in the background technology, enabling the control boundary of the fence and risk assessment to be adaptively adjusted according to the specific behavioral characteristics and real-time vital signs of each pet. On this basis, the cloud uses the physical layer near-field signal returned by the local area gateway as a high-priority verification reference to logically correct spatial drift caused by satellite positioning, significantly improving the accuracy of boundary crossing detection.

[0247] On the user interaction layer of the system, the mobile user terminal is used to receive alarm notifications and quick action commands of corresponding levels, and send the control signaling fed back by the user based on the quick action commands to the cloud service platform to form an interactive closed loop. The mobile user terminal here is generally a smartphone or tablet computer carried by the user. The device runs a dedicated client application (APP) and is connected to the system-level message push service interface of the mobile operating system (such as Apple's APNs or Android's FCM framework).

[0248] In practice, to address the lack of hierarchical processing and rapid interaction channels in the background technology, when the cloud determines that a high-risk overstepping is occurring and issues an instruction, the underlying network module of the mobile user terminal's operating system intercepts the specific payload and directly renders a high-priority, strong reminder pop-up window with physical touch buttons for "one-click navigation" or "force high-frequency positioning" on the device's lock screen or system notification bar.

[0249] After the user touches the command button, the background process of the mobile device's operating system can directly call the wireless communication baseband without waiting for the main application to start cold. The control signal is then transmitted back to the cloud service platform via the cellular network, and the cloud then transmits the command to the smart pet edge terminal.

[0250] Based on the above system architecture, in actual use, the system in this embodiment significantly reduces hardware power consumption in daily operation and extends the device's battery life by configuring local computing power on the smart pet edge terminal to dynamically adjust the network frequency. At the same time, by leveraging the cluster computing power of the cloud service platform, it performs virtual and real dual-layer cross-verification of multimodal positioning data and couples biometrics to calculate multidimensional risk scores, effectively filtering out erroneous positioning drift noise in complex radio frequency environments.

[0251] The system not only improves the objective accuracy of electronic fence boundary judgment and danger alarm, but also builds a quick interaction channel through mobile user terminals that does not require waking up the APP, which greatly reduces the system response time from the discovery of danger to the issuance of intervention instructions. It provides a set of engineering hardware system solutions with high robustness and low latency in the field of wireless positioning and monitoring technology.

[0252] Example 5:

[0253] Corresponding to the above embodiments, the present invention also proposes an electronic device.

[0254] like Figure 9 The diagram shows a structural schematic of an electronic device according to the present invention. The electronic device 100 includes a processor 101 and a memory 103. The processor 101 and the memory 103 are connected, for example, via a bus 102. Optionally, the electronic device 100 may further include a transceiver 104. It should be noted that in practical applications, the transceiver 104 is not limited to one unit, and the structure of this electronic device 100 does not constitute a limitation on the embodiments of the present invention.

[0255] Processor 101 may be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA, or other programmable logic device, transistor logic device, hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in connection with this disclosure. Processor 101 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0256] Bus 102 may include a pathway for transmitting information between the aforementioned components. Bus 102 may be a PCI bus or an EISA bus, etc. Bus 102 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 9 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0257] The memory 103 stores a computer program corresponding to the high-precision edge positioning method for intelligent pet positioning fences based on multimodal fusion in the above embodiments of the present invention. This computer program is controlled and executed by the processor 101. The processor 101 executes the computer program stored in the memory 103 to implement the content shown in the aforementioned method embodiments.

[0258] Among them, electronic devices 100 include, but are not limited to: mobile terminals such as laptops and PADs (tablet computers) and fixed terminals such as desktop computers. Figure 9 The electronic device 100 shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of the present invention.

[0259] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A high-precision positioning method for the edge of a smart pet positioning fence based on multimodal fusion, applied to a wireless positioning and monitoring system consisting of a smart pet edge terminal, local physical fence nodes, a cloud service platform, and a mobile user terminal, characterized in that... The method includes: Acquire a multimodal positioning data set, which includes satellite positioning data reported by the smart pet edge terminal via a cellular network, and near-field positioning data collected by the local physical fence node from the smart pet edge terminal and transmitted back via a wireless link; Locally on the smart pet edge terminal, the comprehensive deviation of the current behavior is calculated. The comprehensive deviation is obtained by weighted summation of the regional deviation of the current location, the speed deviation of the current movement speed, and the time period deviation of the current time. The location data reporting refresh frequency of the smart pet edge terminal is dynamically adjusted according to the level of the comprehensive deviation in order to perform wireless resource and location scheduling on the edge side. On the cloud service platform, density clustering is performed based on historical trajectory data to construct individual behavior baselines that include historical high-frequency activity areas, baseline speed ranges, and inactive time periods. The boundary parameters of the four-level partitioned polygonal electronic fence are dynamically calibrated based on the individual behavior baselines. A comprehensive risk score is calculated by integrating features from location, motion, health, and behavioral deviation dimensions, and cross-validation of boundary crossing events is performed using the multimodal positioning dataset. Based on the cross-validation results of the boundary crossing event and the comprehensive risk score, the boundary crossing risk level is determined. The corresponding level of alarm notification and quick action command are sent to the mobile user terminal through the wireless push channel, and an interactive closed loop is formed after receiving the control signaling feedback from the user based on the quick action command.

2. The method according to claim 1, characterized in that, The acquisition of the multimodal positioning data set includes: The cloud service platform establishes a multimodal wireless identifier mapping table that includes pet chip identifier, terminal device identifier, and network media access control address, and completes the wireless identity binding between the smart pet edge terminal and the mobile user terminal. The local physical fence node continuously collects the near-field signal strength value of the smart pet edge terminal, converts the near-field signal strength value into distance features through a logarithmic path loss model, and generates the near-field positioning data by combining the fixed geographic coordinates of the local physical fence node.

3. The method according to claim 1, characterized in that, The step of calculating the overall deviation of the current behavior locally on the smart pet edge terminal includes: Obtain the individual behavior baseline issued by the cloud service platform; Determine whether the current position of the smart pet edge terminal has deviated from the historical high-frequency activity area to determine the degree of area deviation; Calculate the percentage by which the current moving speed exceeds the baseline speed range to determine the speed deviation. Determine whether the current time belongs to the inactive period and is in an active state to determine the period deviation; The region deviation, the speed deviation, and the time period deviation are each assigned a corresponding preset weight coefficient, and the weighted sum is used as the comprehensive deviation of the current behavior, wherein the sum of each preset weight coefficient is one.

4. The method according to claim 1, characterized in that, The step of dynamically adjusting the location data reporting refresh frequency of the smart pet edge terminal based on the level of the comprehensive deviation includes: When the overall deviation is lower than the first deviation threshold, it is determined to be a low deviation state, and the location data reporting refresh frequency is set to a basic low frequency mode to reduce terminal communication power consumption. When the overall deviation is greater than or equal to the first deviation threshold and less than the second deviation threshold, it is determined to be a medium deviation state, and the location data reporting refresh frequency is increased to the normal monitoring mode. When the overall deviation is greater than or equal to the second deviation threshold, it is determined to be a high deviation state, and the location data reporting refresh frequency is switched to real-time tracking mode to improve the reporting accuracy of location data. Wherein, the second deviation threshold is greater than the first deviation threshold.

5. The method according to claim 1, characterized in that, The method of constructing an individual behavioral baseline based on density clustering of historical trajectory data, which includes information on historical high-frequency activity areas, baseline speed ranges, and inactive periods, includes: The location drift noise points in the historical trajectory data are filtered out by the velocity threshold method to generate a set of valid trajectory points; A density-based spatial clustering algorithm is used to perform high-frequency region clustering on the set of effective trajectory points, extracting multiple high-frequency active regions and generating the minimum hull convex polygon of each high-frequency active region. A single natural day is divided into multiple equal-length time periods. The average movement speed and activity duration ratio in each time period are calculated to generate a time-segmented behavioral characteristic baseline. The high-frequency activity area is then combined with the time-segmented behavioral characteristic baseline to generate the individual behavioral baseline. The individual behavioral baseline is updated daily using a sliding window mechanism.

6. The method according to claim 5, characterized in that, The four-level zoned polygonal electronic fence is divided into, from the inside out, a core security zone, a regular activity zone, an early warning buffer zone, and a dangerous restricted zone. The regular activity area is generated by taking the union of the polygons corresponding to the multiple high-frequency activity areas and then expanding them outward by a preset margin. The dynamic boundary width of the warning buffer is obtained by multiplying the basic buffer width by a dynamic scaling factor, which is the ratio of the average moving speed during the daytime active period in the time-segmented behavioral characteristic baseline to the standard reference speed.

7. The method according to claim 1, characterized in that, The integrated risk score is calculated by fusing features from location, movement, health, and behavioral deviation dimensions, including: Extract the regional attributes of the current position of the smart pet edge terminal relative to the four-level partitioned polygonal electronic fence as the position dimension feature; Instantaneous acceleration and rate of change of direction of motion are extracted as motion dimension features; Physiological indicators collected by the vital signs sensor built into the smart pet edge terminal are used as health dimension features; The location dimension features, movement dimension features, health dimension features, and the comprehensive deviation are each mapped to a normalized single-dimensional risk score. The comprehensive risk score used for boundary judgment is obtained by weighting and summing the preset weight coefficients for the location dimension, movement dimension, health dimension, and behavior deviation dimension, where the sum of the weight coefficients of each dimension is one.

8. The method according to claim 1, characterized in that, The cross-validation of boundary-crossing events in the virtual-real dual-layer positioning based on the multimodal positioning dataset includes executing positioning priority determination rules: When the local physical fence node detects the near-field signal of the smart pet edge terminal, and the strength of the near-field signal is higher than the near-field connection threshold, it is directly determined that it is currently in a safe area. The priority of this physical near-field determination is higher than the spatial calculation determination result of the four-level partition polygon electronic fence. When the physical near-field determination is triggered and the location is within a safe area, the cloud service platform sends a control signal to the smart pet edge terminal, covering the refresh frequency strategy based on the comprehensive deviation and forcing the smart pet edge terminal to enter a sleep positioning mode.

9. The method according to claim 7, characterized in that, It also includes a mechanism linking health data with fence control: When the cloud service platform detects that the physiological indicators in the health dimension features exceed the normal physiological threshold, it determines that there is an abnormal health condition and automatically increases the weight coefficient of the health dimension. Simultaneously, wireless scheduling commands are sent to the smart pet edge terminal to increase the frequency of location data reporting and refresh, and the physical boundaries of the four-level zone polygonal electronic fence are reduced by a preset ratio to tighten the alarm threshold.

10. The method according to claim 1, characterized in that, The step of determining the boundary risk level based on the cross-validation results of the boundary violation event and the comprehensive risk score, and sending a corresponding level of alarm notification and quick action command to the mobile user terminal via a wireless push channel includes: When the risk level of crossing the boundary is determined to be low risk, the status update is only performed in the application pre-installed on the mobile user terminal, and no notification push is triggered. When the risk level of the boundary crossing is determined to be medium risk, a normal notification signaling and an in-application location viewing instruction are issued. When the risk level of crossing the boundary is determined to be high risk, a strong reminder notification signal is issued, along with quick action instructions for one-click navigation and forced activation of high-frequency positioning; In the system notification display interface of the mobile user terminal, in response to the user's click operation on the shortcut action command, the wireless network interface is directly invoked to send control signaling to the cloud service platform without waking up and entering the main interface of the application.

Images