Alarm method and device for outdoor safety

Abstract

The invention provides an outdoor safety-oriented alarm method and device, and the method comprises the steps: collecting current environment data and behavior data of a user in two successive periods, the behavior data in the two successive periods being first behavior data and second behavior data; inputting the current environment data and the first behavior data into a behavior baseline model to obtain conventional behavior data, and calculating a difference value between the second behavior data and the conventional behavior data to obtain a state deviation degree; calling a preset API to obtain environment early warning data, obtaining geographic information data based on the user positioning data and the local map data, and fusing the environment early warning data and the geographic information data to calculate and obtain an environment risk coefficient and a risk weight; inputting the state deviation degree, the environmental risk coefficient and the risk weight into an adaptive threshold function to obtain an alarm score; and if the alarm score is higher than the alarm threshold value, spreading alarm information. According to the invention, automation and intellectualization of outdoor safety alarm are realized.

Description

Technical Field

[0001] This application relates to the field of safety alarm technology, specifically to an alarm method and device for outdoor safety. Background Technology

[0002] With the rapid development of society and economy and the improvement of people's living standards, outdoor activities such as hiking, mountaineering, and trail running are becoming increasingly popular. However, the outdoor environment is usually complex, changeable, and unpredictable. Participants may not only face environmental risks such as sudden severe weather and geological disasters, but may also fall into dangerous situations due to personal reasons such as getting lost, hypothermia, physical exhaustion, or sudden illness.

[0003] Currently, common outdoor safety technologies mainly include location monitoring based on electronic fences, information services for pushing weather alerts, and wearable devices with functions such as fall detection. These technologies can trigger alarms under specific conditions. However, these existing technologies have significant drawbacks. They typically treat environmental risks separately from the user's own behavioral state, failing to achieve a comprehensive assessment. Summary of the Invention

[0004] To overcome the shortcomings of the prior art, this application provides an alarm method and device for outdoor safety.

[0005] The specific technical solution is as follows: An alarm method for outdoor safety includes: Collect current environmental data and user behavior data in two consecutive periods, wherein the behavior data in the two consecutive periods are referred to as first behavior data and second behavior data, respectively. The current environmental data and the first behavioral data are input into the behavioral baseline model to obtain regular behavioral data, and the difference between the second behavioral data and the regular behavioral data is calculated to obtain the state deviation. The system calls a preset API to obtain environmental early warning data, and simultaneously obtains geographic information data based on user location data and local map data. The environmental early warning data and the geographic information data are then fused together to calculate the environmental risk coefficient and risk weight. The state deviation, the environmental risk coefficient, and the risk weight are input into an adaptive threshold function to obtain an alarm score; if the alarm score is higher than the alarm threshold, an alarm message is propagated.

[0006] In one embodiment, calculating risk weights by fusing the environmental early warning data and the geographic information data includes: Based on the aforementioned environmental early warning data and geographic information data, the dominant environmental risk categories are identified; Risk weights are determined from predefined configuration rules based on the dominant environmental risk category.

[0007] In one embodiment, the predefined configuration rules include: If the dominant risk is a type I risk, then the weight of motion state data is increased and the weight of physiological state data is decreased, wherein the type I risk is a local-specific risk related to a specific geographical location; If the dominant risk is a second type of risk, then the weight of the physiological state data is increased by an amplification function, wherein the second type of risk is a global universal risk that is related to a wide area.

[0008] In one embodiment, calculating an environmental risk coefficient by fusing the environmental early warning data and the geographic information data includes: A first risk level is determined based on the environmental early warning data, and a second risk level is determined based on the geographic information data; According to a predefined fusion strategy, the first risk level and the second risk level are combined to calculate the comprehensive risk level; The comprehensive risk level is mapped to a numerical environmental risk coefficient.

[0009] In one embodiment, the behavioral baseline model is established in the following manner: Collect historical behavioral data and historical environmental data of users within a preset security period; The historical behavior data is preprocessed and features are extracted to obtain a user behavior feature sequence; Using the historical environmental data as a condition variable and the user behavior feature sequence as a training target, a deep temporal prediction network is trained to obtain the behavior baseline model.

[0010] In one embodiment, the adaptive threshold function is constructed as follows: Using the state deviation as the basic input, the state deviation is linearly modulated by the environmental risk coefficient, and then the modulated result is weighted and fused using the risk weight to obtain the final alarm score.

[0011] In one embodiment, the propagation of alarm information includes: If a cellular network signal is available, an alarm message will be sent through that cellular network; if no cellular network signal is available, the alarm message will be automatically switched to the satellite communication module. Simultaneously, sound alarms, screen flashing, and vibration alerts are activated locally.

[0012] In one embodiment, the alarm method further includes: Receive user feedback data; The user feedback data is associated and stored with the corresponding first behavior data, second behavior data, environmental early warning data, and geographic information data to form a calibration training set; the behavior baseline model is incrementally trained based on the calibration training set. Based on the statistical results of the accuracy feedback instruction, the parameters of the adaptive threshold function are optimized.

[0013] In one embodiment, the alarm method further includes: Based on user feedback data collected during historical periods, the historical false alarm rate and historical false alarm rate are calculated. The alarm threshold value is dynamically adjusted based on the historical false alarm rate and the historical false alarm rate. When the historical false alarm rate exceeds the first threshold, the alarm threshold value is increased, and when the historical false alarm rate exceeds the second threshold, the alarm threshold value is decreased.

[0014] An alarm device for outdoor safety includes: The data acquisition module is used to collect current environmental data and user behavior data in two consecutive periods, wherein the behavior data in the two consecutive periods are the first behavior data and the second behavior data, respectively. The first calculation module is used to input the current environmental data and the first behavioral data into the behavioral baseline model to obtain regular behavioral data, and to calculate the difference between the second behavioral data and the regular behavioral data to obtain the state deviation. The second calculation module is used to call a preset API to obtain environmental early warning data, and at the same time obtain geographic information data based on user location data and local map data. The environmental early warning data and the geographic information data are then fused to calculate the environmental risk coefficient and risk weight. The risk assessment module is used to input the state deviation, the environmental risk coefficient, and the risk weight into an adaptive threshold function to obtain an alarm score; The propagation module is used to propagate alarm information when the alarm score is higher than the alarm threshold value.

[0015] This application has at least the following beneficial effects: This application provides an alarm method for outdoor safety, comprising: collecting current environmental data and user behavior data in two consecutive periods, wherein the behavior data in the two consecutive periods are respectively first behavior data and second behavior data; inputting the current environmental data and the first behavior data into a behavior baseline model to obtain regular behavior data, and calculating the difference between the second behavior data and the regular behavior data to obtain a state deviation degree; calling a preset API to obtain environmental early warning data, and simultaneously obtaining geographic information data based on user location data and local map data, and fusing the environmental early warning data and geographic information data to calculate an environmental risk coefficient and a risk weight; inputting the state deviation degree, environmental risk coefficient, and risk weight into an adaptive threshold function to obtain an alarm score; and if the alarm score is higher than an alarm threshold value, propagating an alarm message.

[0016] This application, by introducing a behavioral baseline model and state deviation calculation, can dynamically identify abnormal changes in user behavior, enabling the system to quantitatively assess whether user behavior deviates from the normal pattern. This allows for timely detection of abnormal behavior that may be caused by sudden illness, accidents, or poor environmental adaptation, enhancing the initiative and accuracy of safety monitoring and effectively reducing outdoor risks caused by abnormal behavior.

[0017] Secondly, this application utilizes a fusion mechanism of environmental early warning data (such as weather alerts and natural disaster information) and geographic information data (such as terrain complexity and distribution of hazardous areas) to comprehensively assess environmental risks. It fully leverages real-time environmental information and static geographic features, ensuring that risk assessment considers not only macro-level environmental threats but also the geographic hazard of the user's specific location, thus improving the comprehensiveness and accuracy of risk identification. This reduces false alarms and missed alarms, providing users with more reliable risk warnings.

[0018] Moreover, this application adopts an adaptive threshold function, which overcomes the limitations of the fixed threshold mechanism, enabling the alarm system to adapt to changes in different user behavior patterns and environmental conditions. It realizes dynamic calculation and flexible adjustment of alarm scores, improves the sensitivity and adaptability of alarms, ensures timely dissemination of alarm information in truly high-risk situations, and avoids unnecessary interference.

[0019] Overall, this application achieves automation and intelligence in outdoor safety alarms by integrating behavioral analysis, environmental risk assessment, and intelligent decision-making mechanisms. This application forms a closed loop from data collection to alarm propagation, enabling rapid response to complex and changing outdoor environments and timely dissemination of alarm information to users or rescue personnel. This effectively reduces safety risks during outdoor activities and enhances user safety. It not only improves alarm efficiency but also provides a scalable solution for outdoor safety monitoring, suitable for various application scenarios. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A flowchart illustrating the alarm method for outdoor safety provided in this embodiment. Figure 1 ; Figure 2 A flowchart illustrating the alarm method for outdoor safety provided in this embodiment. Figure 2 ; Figure 3 A flowchart illustrating the alarm method for outdoor safety provided in this embodiment. Figure 3 ; Figure 4 This is a schematic diagram of the module of the alarm device for outdoor safety provided in this embodiment.

[0022] Figure label: 1-Data Acquisition Module; 2-First Calculation Module; 3-Second Calculation Module; 4-Risk Assessment Module; 5-Propagation Module. Detailed Implementation

[0023] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0024] In the description of this application, it should be noted that the terms "vertical", "up", "down", "horizontal", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application.

[0025] In the description of this application, it should also be noted that, unless otherwise expressly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0026] like Figures 1 to 3 As shown, this embodiment provides an alarm method for outdoor safety, including: Collect current environmental data and user behavior data in two consecutive periods. The behavior data in the two consecutive periods are the first behavior data (e.g., data within time window T-1) and the second behavior data (e.g., data within the immediately adjacent current time window T). Input the current environmental data and the first behavior data into the behavior baseline model to obtain the regular behavior data, and calculate the difference between the second behavior data and the regular behavior data to obtain the state deviation. The system calls a preset API to obtain environmental early warning data, and at the same time obtains geographic information data based on user location data and local map data. The environmental risk coefficient and risk weight are calculated by integrating the environmental early warning data and geographic information data. The state deviation, environmental risk coefficient, and risk weight are input into the adaptive threshold function to obtain the alarm score; if the alarm score is higher than the alarm threshold, the alarm information is propagated.

[0027] This embodiment, by introducing a behavioral baseline model and state deviation calculation, can dynamically identify abnormal changes in user behavior. This enables the system to quantitatively assess whether user behavior deviates from the normal pattern, thereby timely detecting abnormal behavior that may be caused by sudden illness, accidents, or poor environmental adaptation. This enhances the initiative and accuracy of safety monitoring and effectively reduces outdoor risks caused by abnormal behavior.

[0028] Secondly, this embodiment utilizes a fusion mechanism of environmental early warning data (such as weather alerts and natural disaster information) and geographic information data (such as terrain complexity and distribution of hazardous areas) to comprehensively assess environmental risks. It fully leverages real-time environmental information and static geographic features, ensuring that risk assessment considers not only macro-level environmental threats but also the geographic hazard of the user's specific location, thus improving the comprehensiveness and accuracy of risk identification. This reduces false alarms and missed alarms, providing users with more reliable risk warnings.

[0029] Moreover, this embodiment adopts an adaptive threshold function, which overcomes the limitations of the fixed threshold mechanism, enabling the alarm system to adapt to changes in different user behavior patterns and environmental conditions. It realizes dynamic calculation and flexible adjustment of alarm scores, improves the sensitivity and adaptability of alarms, ensures timely dissemination of alarm information in truly high-risk situations, and avoids unnecessary interference.

[0030] Overall, this embodiment achieves automation and intelligence in outdoor safety alarms by integrating behavioral analysis, environmental risk assessment, and intelligent decision-making mechanisms. This embodiment forms a closed loop from data collection to alarm propagation, enabling rapid response to complex and ever-changing outdoor environments and timely dissemination of alarm information to users or rescue personnel. This effectively reduces safety risks during outdoor activities and enhances user safety. It not only improves alarm efficiency but also provides a scalable solution for outdoor safety monitoring, suitable for various application scenarios.

[0031] Behavioral data includes, but is not limited to, the user's three-axis acceleration, angular velocity, heading, cadence, heart rate, and blood oxygen saturation.

[0032] Environmental data includes, but is not limited to, ambient temperature, humidity, and air pressure.

[0033] Among them, the behavioral baseline model is a time-series prediction model trained on a long short-term memory network (LSTM), which is configured to predict the user's normal behavioral data in the next time window given historical environmental conditions and behavioral patterns.

[0034] Among them, the state deviation is a quantified error value, which can be obtained by calculating the Euclidean distance or Mahalanobis distance between the vectors of each dimension between the second behavior data and the regular behavior data. The larger the value, the greater the difference between the user's current state and the normal state predicted by the model.

[0035] The local map data consists of vector maps pre-installed on the device or downloaded online, containing information such as terrain, elevation, water areas, and cliff boundaries.

[0036] This includes calling preset APIs to obtain environmental early warning data, such as calling the API of the National Meteorological Administration Early Warning Center and the API of the local geological disaster early warning center to obtain early warning types and levels for rainstorms, lightning, strong winds, and landslides.

[0037] Among them, geographic information data is obtained based on user location data and local map data, including: obtaining geographic information data such as the distance between the user and the nearest cliff edge, the elevation and slope of the location, and whether there is water in the surrounding area based on user location data provided by GPS / BeiDou module and vector maps pre-installed in the device or downloaded online (containing information such as terrain, elevation, water area, cliff boundary, etc.).

[0038] Among them, the environmental risk coefficient is a normalized value, for example, between 0 and 1, used to characterize the overall degree of danger of the external environment; the risk weight is a vector, whose elements correspond to the importance coefficients of different behavioral data dimensions (such as heart rate weight, acceleration weight, and altitude weight).

[0039] In one embodiment, the behavioral baseline model is established in the following way: Collect historical behavioral data and historical environmental data of users within a preset security period; Historical behavior data is preprocessed and features are extracted to obtain user behavior feature sequences; Using historical environmental data as conditional variables and user behavior feature sequences as training targets, a deep temporal prediction network is trained to obtain a behavioral baseline model.

[0040] Specifically, the preset safety period is greater than or equal to two weeks.

[0041] Specifically, historical behavior data is preprocessed, including but not limited to noise reduction, filtering, and timestamp alignment.

[0042] Specifically, feature extraction is performed on historical behavioral data, including but not limited to extracting mean heart rate and acceleration variance.

[0043] The behavioral baseline model construction process provided in this embodiment enables the model to learn and memorize users' personalized behavioral patterns in safe states and different environments. This provides a precise and personalized benchmark for calculating state deviation, allowing the system to identify truly abnormal states for a specific user, rather than merely states deviating from the group average, thus greatly improving the accuracy of individual risk identification.

[0044] In one embodiment, the adaptive threshold function is constructed as follows: Using the state deviation as the basic input, the state deviation is linearly modulated by the environmental risk coefficient, and then the modulated result is weighted and fused using risk weights to obtain the final alarm score.

[0045] The specific implementation of the adaptive threshold function is as follows: Alarm score = State deviation_i × (1 + Environmental risk coefficient) × Risk weight_i, where i represents the i-th dimension of the behavioral data.

[0046] This embodiment clarifies the mathematical structure of the adaptive threshold function, the core of which lies in using the environmental risk coefficient as a modulator to dynamically scale the severity of abnormal behavior. This allows the alarm threshold to adjust according to changes in environmental risk levels, avoiding disturbing alarms caused by normal physiological fluctuations in safe environments, while maintaining high alertness for minor anomalies in dangerous environments, thus achieving an optimal balance between alarm sensitivity and specificity.

[0047] like Figure 2 As shown, in one embodiment, risk weights are calculated by fusing environmental early warning data and geographic information data, including: Identify dominant environmental risk categories based on environmental early warning data and geographic information data; Risk weights are determined from predefined configuration rules based on the dominant environmental risk category.

[0048] Specifically, the identification principles for environmental risk categories based on environmental early warning data and geographic information data include: Principle one: Compare the first risk level determined based on environmental early warning data with the second risk level determined based on geographic information data. The risk category corresponding to the higher risk level will be preliminarily determined as the dominant risk category.

[0049] For example, if the system receives a red alert for heavy rain (the first risk level is mapped to level 4), and simultaneously determines through geographic information data that the user is located on a gentle slope grassland (the second risk level is mapped to level 1), then according to this principle, level 4 is higher than level 1. Therefore, the dominant risk category should be determined as the second type of risk related to weather (global and universal risk).

[0050] Principle 2: When two risk levels are the same or similar (e.g., the difference is within 1 level), the risk category with higher immediacy and urgency should be prioritized as the dominant risk.

[0051] For example, if the system receives a yellow gale warning (first risk level mapped to level 2), and simultaneously detects that a user is standing on the edge of a fast-flowing river (second risk level mapped to level 2), although the risk levels are the same, the risk of falling into the water is immediate and fatal, while the impact of a gale is relatively continuous and gradual. Therefore, according to this principle, the dominant risk category should be determined as type I risk (local specific risk).

[0052] In one embodiment, the predefined configuration rules include: If the dominant risk is Type I risk, then the weight of motion state data is increased and the weight of physiological state data is decreased. Type I risk refers to local-specific risks related to a specific geographical location (e.g., near a cliff or the edge of water). In this scenario, the user's posture, movement speed, and direction are more critical risk indicators. If the dominant risk is a type II risk, then the weight of physiological state data is increased by an amplification function. Type II risks are global and universal risks that are related to a wide area (e.g., heat waves or extreme cold waves). In this scenario, physiological signs such as the user's heart rate and blood oxygen are more important for risk assessment.

[0053] This embodiment dynamically adjusts risk weights based on the dominant risk type, enabling the alarm system to focus on the most critical monitoring indicators in the current situation. For example, on a cliff edge, the system focuses more on the user's movement to prevent falls; in a high-temperature environment, it focuses more on the user's physiological reactions to prevent heatstroke. The dynamic focusing mechanism provided by this embodiment greatly enhances the contextual relevance and scientific rigor of alarm decisions, further reducing the possibility of misjudgments.

[0054] Specifically, a baseline weight vector W_base=[W_motion_base,W_physio_base,...] is preset for each dimension of data, where W_motion_base is the baseline weight for motion state data (such as acceleration, posture, and velocity), and W_physio_base is the baseline weight for physiological state data (such as heart rate and blood oxygen).

[0055] When the dominant risk is identified as Type I, the system increases the weight of the motion state data according to the severity of the risk (e.g., the distance d between the user and the cliff edge) using a boosting function, as shown below: The new motion state weights W_motion_new = W_motion_base × (1 + α / d); Where α is an adjustment coefficient greater than 0.

[0056] The smaller the distance d between the user and the cliff edge, the larger the calculated weight W_motion_new, making the user's actions more noticeable as they get closer to danger.

[0057] The weights of the new physiological state data are calculated based on the ratio of the weights of the original physiological state data to the remaining weights excluding the weights of the original movement state data, and the remaining weights excluding the weights of the new movement state data.

[0058] Specifically, the amplification function is an exponential amplification function to strongly correlate widespread environmental risks (such as high temperatures) with core physiological indicators (such as heart rate and blood oxygen). When a heatwave strikes, the system pays particular attention to whether users exhibit signs of heatstroke or exhaustion, rather than their gait. This ensures that the system can sensitively detect early abnormal changes in the user's internal state when facing diffuse and cumulative environmental threats, thereby providing early warning and buying valuable time for intervention and self-rescue.

[0059] The exponential scaling function is shown below: The new physiological state weight W_physio_new = W_physio_base × γ^S; Where γ is a base greater than 1 (e.g., 1.5), and S is the first risk level determined based on environmental early warning data (e.g., S=2 for orange high temperature warning, S=3 for red high temperature warning).

[0060] like Figure 3 As shown, in one embodiment, the environmental risk coefficient is calculated by integrating environmental early warning data and geographic information data, including: The first risk level is determined based on environmental early warning data, and the second risk level is determined based on geographic information data; Based on a predefined fusion strategy, the first risk level and the second risk level are combined to calculate the comprehensive risk level. The overall risk level is mapped to a numerical environmental risk coefficient.

[0061] Among them, determining the first risk level based on environmental early warning data includes mapping blue, yellow, orange, and red rainstorm warnings to risk levels 1 to 4, respectively.

[0062] Among them, the second risk level is determined based on geographic information data, including: according to the distance between the user and the dangerous terrain: greater than 100 meters is level 0, 50-100 meters is level 1, and less than 50 meters is level 2.

[0063] The predefined fusion strategy is a weighted average strategy. The first risk level is R1, with a corresponding weight coefficient of W1; the second risk level is R2, with a corresponding weight coefficient of W2; the overall risk level R = W1*R1 + W2*R2, where W1 + W2 = 1. Weight coefficients are assigned values ​​based on prior knowledge or application scenario configuration. For example, in mountainous areas with changeable weather, W1 = 0.7 and W2 = 0.3 can be set; in fixed areas with complex terrain, W1 = 0.4 and W2 = 0.6 can be set.

[0064] Among them, the comprehensive risk levels 0 to 4 are mapped to environmental risk coefficients of 0, 0.25, 0.5, 0.75, and 1.0, respectively.

[0065] This embodiment comprehensively assesses risk levels through a weighted approach, resulting in a more comprehensive and balanced overall risk assessment. This avoids the bias that may result from a single risk source dominating the judgment, making the system's risk perception more nuanced and in line with reality. It is particularly suitable for scenarios with high requirements for false alarm control and continuous and stable monitoring.

[0066] In one embodiment, the alarm message is propagated, including: If a cellular network signal is available, an alarm message will be sent through that cellular network; if no cellular network signal is available, the alarm message will be automatically switched to the satellite communication module. At the same time, sound alarms, screen flashing, and vibration alerts are activated locally to remind users who may be unaware or to attract the attention of people nearby.

[0067] This embodiment provides a multi-layered, highly reliable alarm propagation scheme. Through a combination of primary and backup communication links (cellular network and satellite communication), it ensures that alarm information can be reliably transmitted even in outdoor environments with vastly different signal coverage. Local alarms serve as the last line of defense, directly providing strong sensory stimulation to users and requesting on-site assistance, thus forming a comprehensive, closed-loop alarm protection system.

[0068] Specifically, if a cellular network signal is available, alarm information containing detailed location, alarm type, and snapshots of key sensor readings will be sent through the cellular network to the cloud monitoring center and preset emergency contacts.

[0069] Specifically, if there is no cellular network signal, it will automatically switch to the satellite communication module (such as the Beidou short message module) to send a short message containing core information (such as latitude and longitude, SOS code).

[0070] In one embodiment, the alarm method further includes: Receive user feedback data; specifically, after receiving an alarm, the user can click the "false alarm" or "confirm" button. User feedback data is associated and stored with corresponding first behavior data, second behavior data, environmental early warning data, and geographic information data to form a calibration training set; the behavior baseline model is incrementally trained based on the calibration training set to optimize its prediction accuracy. Based on the statistical results of the accuracy feedback instructions, the parameters (such as weight allocation rules) of the adaptive threshold function are optimized.

[0071] This embodiment introduces an important feedback and self-optimization mechanism. It can learn from real user feedback, continuously revise its behavioral baseline model and decision-making logic, and thus become more accurate and personalized over time. This effectively solves the prediction bias problem that may exist in the initial stage of machine learning models, enabling the system to have the ability to continuously evolve.

[0072] In one embodiment, the alarm method further includes: Based on user feedback data collected during historical periods, the historical false alarm rate and historical false alarm rate are calculated. The alarm threshold is dynamically adjusted based on the historical false alarm rate and the historical false alarm rate. When the historical false alarm rate exceeds the first threshold, the alarm threshold is increased; when the historical false alarm rate exceeds the second threshold, the alarm threshold is decreased.

[0073] In one embodiment, a fixed evaluation period (e.g., weekly or monthly) is set, and the historical false alarm rate (FPR) is calculated based on user feedback data collected within that period. hist ) and historical underreporting rate (FNR) hist Threshold definition: Preset first threshold (FPR) target ), representing the maximum tolerable false alarm rate; a preset second threshold (FNR) target ), representing the maximum tolerable false negative rate.

[0074] At the end of each evaluation period, the calculated FPR will be... hist With FPR target Comparison, FNR hist With FNR target Compare the values ​​to determine whether to adjust the alarm threshold value T.

[0075] Specifically, when FPR hist Greater than FPR target At that time, the alarm threshold value is adjusted and increased based on the following formula: T new =T old +μ×(FPR hist FPR target ) Where μ is a preset scaling factor.

[0076] This design allows the threshold value to be raised more significantly as the false alarm rate exceeds the target value, making adjustments more targeted and efficient.

[0077] Specifically, when FNR hist Greater than FNR target At that time, the alarm threshold value is adjusted and lowered based on the following formula: T new =T old -ν×(FNR hist FNR target ) Where ν is a preset proportional coefficient.

[0078] This design ensures that the threshold value decreases more as the false negative rate exceeds the target value, allowing the system to recover to a sensitive state more quickly.

[0079] Specifically, the alarm threshold value T is located within a preset range [T]. min ,T max Within [the specified range], any adjusted new alarm threshold value T new All are located within this range.

[0080] This embodiment monitors the two key performance indicators, false alarm rate and false negative rate, and the system can automatically adjust its alarm tendency to ensure that it maintains the best overall performance in long-term operation, adapt to different users' tolerance for false alarms and false negatives, and achieve stability and optimization of the system's overall performance.

[0081] like Figure 4 As shown, this embodiment also provides an alarm device for outdoor safety, including: Data acquisition module 1 is used to collect current environmental data and user behavior data in two consecutive periods, namely the first behavior data and the second behavior data. The first calculation module 2 is used to input the current environmental data and the first behavior data into the behavior baseline model to obtain the regular behavior data, and to calculate the difference between the second behavior data and the regular behavior data to obtain the state deviation. The second calculation module 3 is used to call a preset API to obtain environmental early warning data, and at the same time obtain geographic information data based on user location data and local map data. It integrates environmental early warning data and geographic information data to calculate environmental risk coefficient and risk weight. Risk assessment module 4 is used to input the state deviation, environmental risk coefficient and risk weight into the adaptive threshold function to obtain the alarm score; The propagation module 5 is used to propagate alarm information when the alarm score is higher than the alarm threshold.

[0082] Those skilled in the art will understand that the modules or steps described above in this application can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computing device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, this application is not limited to any particular combination of hardware and software.

[0083] Note that the above description is merely a preferred embodiment and the technical principles employed in this application. Those skilled in the art will understand that this application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments. Many other equivalent embodiments may be included without departing from the concept of this application, and the scope of this application is determined by the scope of the appended claims.

[0084] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. An outdoor safety-oriented alarming method, characterized by, The method comprises: collecting current environment data and user behavior data in two consecutive periods, wherein the behavior data in the two consecutive periods are first behavior data and second behavior data respectively; inputting the current environment data and the first behavior data into a behavior baseline model to obtain regular behavior data, and calculating a difference between the second behavior data and the regular behavior data to obtain a state deviation degree; calling a preset API to obtain environment early warning data, and simultaneously obtaining geographic information data based on user positioning data and local map data, and fusing the environment early warning data and the geographic information data to calculate an environment risk coefficient and a risk weight; inputting the state deviation degree, the environment risk coefficient and the risk weight into an adaptive threshold function to obtain an alarm score; if the alarm score is higher than an alarm threshold, propagating alarm information.

2. The outdoor security-oriented alarming method according to claim 1, wherein The method of fusing the environment early warning data and the geographic information data to calculate a risk weight comprises: identifying a dominant environment risk category based on the environment early warning data and the geographic information data; determining a risk weight from a predefined configuration rule according to the dominant environment risk category.

3. The outdoor security-oriented alarming method according to claim 2, wherein The predefined configuration rule comprises: if the dominant risk is a first type of risk, increasing the weight of motion state data and decreasing the weight of physiological state data, wherein the first type of risk is a local specificity risk related to a specific geographic location; if the dominant risk is a second type of risk, increasing the weight of physiological state data through an amplification function, wherein the second type of risk is a global universality risk related to a wide area.

4. The outdoor security-oriented alarming method of claim 1, wherein, The method of fusing the environment early warning data and the geographic information data to calculate an environment risk coefficient comprises: determining a first risk level based on the environment early warning data, and determining a second risk level based on the geographic information data; combining the first risk level and the second risk level according to a predefined fusion strategy to calculate a comprehensive risk level; mapping the comprehensive risk level into a numerical environment risk coefficient.

5. The outdoor security oriented alerting method of claim 1, wherein, The behavior baseline model is established by: collecting historical behavior data and historical environment data of a user in a preset safety period; preprocessing and feature extracting the historical behavior data to obtain a user behavior feature sequence; training a deep time series prediction network with the historical environment data as a conditional variable and the user behavior feature sequence as a training target to obtain the behavior baseline model.

6. The outdoor security-oriented alerting method of claim 1, wherein, The adaptive threshold function is constructed by: inputting the state deviation degree as a basic input, linearly modulating the state deviation degree by the environment risk coefficient, and then weighting and fusing the modulated result by the risk weight to obtain a final alarm score.

7. The outdoor security-oriented alerting method of claim 1, wherein, The method of propagating alarm information comprises: if there is a cellular network signal, sending alarm information through the cellular network; if there is no cellular network signal, automatically switching to a satellite communication module to send alarm information; simultaneously, starting a sound alarm, screen flicker and vibration prompt locally and synchronously.

8. The outdoor security oriented alerting method of claim 1, wherein, The alarm method further comprises: receiving user feedback data; The user feedback data is stored in association with the corresponding first behavior data, second behavior data, environment warning data and geographic information data to form a correction training set; and the behavior baseline model is trained based on the correction training set; Based on the statistical results of the accuracy feedback instructions, the parameters of the adaptive threshold function are optimized.

9. The outdoor security-oriented alerting method of claim 1, wherein, The alarm method further includes: Based on the user feedback data collected in the historical period, the historical false positive rate and the historical false negative rate are calculated; According to the historical false positive rate and the historical false negative rate, the alarm threshold value is dynamically adjusted, the alarm threshold value is increased when the historical false positive rate exceeds a first threshold value, and the alarm threshold value is decreased when the historical false negative rate exceeds a second threshold value.

10. An outdoor security oriented alarm device, characterized by It includes: The acquisition module is used for acquiring current environment data and user behavior data in two previous periods, and the behavior data in the two previous periods are first behavior data and second behavior data respectively; The first calculation module is used for inputting the current environment data and the first behavior data into a behavior baseline model to obtain regular behavior data, and calculating the difference between the second behavior data and the regular behavior data to obtain a state deviation degree; The second calculation module is used for calling a preset API to obtain environment warning data, and simultaneously obtaining geographic information data based on user positioning data and local map data, and calculating environment risk coefficients and risk weights by fusing the environment warning data and the geographic information data; The risk judgment module is used for inputting the state deviation degree, the environment risk coefficient and the risk weight into an adaptive threshold function to obtain an alarm score; The propagation module is used for propagating alarm information when the alarm score is higher than an alarm threshold value.

Images