Safety monitoring and warning system and smart watch

Abstract

The invention provides a safety monitoring and warning system and a smart watch. The system is applied to the portable electronic equipment and comprises a data sensing module used for acquiring real-time data of a user or the surrounding environment of the user; the data processing module is used for carrying out edge calculation on the real-time data obtained by the data sensing module to obtain characteristic parameters, and the edge calculation comprises the steps of filtering motion artifacts, carrying out deviation correction on positioning data and carrying out characteristic extraction; the risk assessment module is used for performing dangerous behavior identification and state judgment by adopting a lightweight machine learning model fused with multiple data sources according to the characteristic parameters obtained by the data processing module, and assessing the risk level based on a preset individual dynamic baseline threshold; and the alarm module is used for triggering response actions of different levels according to the assessment result of the risk assessment module in combination with the danger type of the user and the risk level. The intelligent watch is suitable for children and can realize children safety monitoring with low false alarm, fast response and long endurance.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of safety monitoring, and in particular to a safety monitoring and alarm system and a smart watch. BACKGROUND

[0002] With the development of portable smart wearable devices, safety monitoring and alarm systems are gradually applied to vulnerable groups such as children and the elderly. The prior art mostly uses a single sensor such as an acceleration sensor for anomaly detection and uploads data to the background processing through the network. When the user has an accident, such systems have deficiencies in response speed, judgment accuracy and scene adaptability, which can easily cause false positives, false negatives or response delays, affecting the actual use effect and user trust.

[0003] Therefore, there is an urgent need for a safety monitoring and alarm system suitable for portable electronic devices to improve recognition accuracy, response real-time and use endurance, so as to better protect the user. SUMMARY

[0004] Therefore, the present application discloses a safety monitoring and alarm system and a smart watch, which realizes low false alarm, fast response and long endurance of user intelligent safety monitoring.

[0005] In a first aspect, the present application discloses a safety monitoring and alarm system applied to a portable electronic device, which comprises: a data sensing module for acquiring real-time data of a user himself or his surrounding environment, the real-time data comprising several of the following data: heart rate data, motion data, positioning data, illumination intensity data, temperature data and humidity data; a data processing module for performing edge computing on the real-time data acquired by the data sensing module to obtain feature parameters, the edge computing comprising filtering motion artifacts, correcting the positioning data and feature extraction; a risk assessment module for performing dangerous behavior recognition and state judgment using a lightweight machine learning model that fuses multiple data sources according to the feature parameters obtained by the data processing module, and assessing the risk level based on a preset individual dynamic baseline threshold; and an alarm module for triggering different levels of response actions according to the assessment result of the risk assessment module in combination with the dangerous type and risk level of the user.

[0006] Optionally, the safety monitoring and alarming system further comprises an anomaly identification enhancement module for assisting in determining the authenticity of an abnormal state and reducing the false alarm rate during risk assessment, and the anomaly identification enhancement module performs processing operations based on at least one of the following mechanisms: a context awareness mechanism for dynamically adjusting the risk assessment threshold or alert level in combination with the external environment or behavioral scenario information of the user currently in; a multi-source information cross-validation mechanism for consistency verification in combination with real-time data from multiple sensors after detecting a preliminary anomaly to improve the accuracy of anomaly identification; and an individual dynamic baseline adaptive mechanism for dynamically updating the reference baseline of feature parameters based on the historical health data of the user to enhance the adaptability of the model to individual differences.

[0007] Optionally, the data processing module comprises a submodule for performing motion artifact filtering processing, which uses an adaptive filter to eliminate motion interference; a submodule for performing positioning data correction processing, which corrects positioning errors by Kalman filtering fusion of multi-source position information; and a submodule for performing feature parameter extraction and rapid risk assessment, which processes based on a lightweight random forest model constructed based on fusion of heart rate, motion and positioning data.

[0008] Optionally, the data processing module and the risk assessment module cooperatively realize identification of multiple dangerous situations, including: the data processing module extracts impact force peak parameters and body position mutation parameters through dynamic threshold segmentation detection and quaternion attitude solution respectively, and the risk assessment module compares the impact force peak parameters and the body position mutation parameters with corresponding parameters in the preset threshold to determine whether the user has fallen; the data processing module extracts humidity sudden increase parameters and heart rate sudden drop parameters through first-order difference mutation detection and trend slope calculation respectively, and the risk assessment module compares the humidity sudden increase parameters and the heart rate sudden drop parameters with corresponding parameters in the preset threshold to determine whether the user is in a drowning state; the data processing module extracts continuous rotation parameters and step frequency abnormal parameters through fast Fourier transform spectrum analysis and gait cycle correlation matching respectively, and the risk assessment module compares the continuous rotation parameters and the step frequency abnormal parameters with corresponding parameters in the preset threshold to determine whether the user is in a dragged state; and the data processing module extracts heart rate variability parameters and motion stillness parameters through Poincaré scatter plot analysis and zero speed detection respectively, and the risk assessment module compares the heart rate variability parameters and the motion stillness parameters with corresponding parameters in the preset threshold to determine whether the user is at risk of sudden illness.

[0009] Optionally, the risk assessment module performs weighted fusion based on the heart rate anomaly detection result, the motion pattern recognition result and the positioning trajectory analysis result to output an assessment result.

[0010] Optionally, the risk assessment module dynamically adjusts the weighting parameters of the heart rate anomaly detection, the motion pattern recognition and the positioning trajectory analysis according to the environment scenario in which the user is located, the environment scenario including daily activities, night sleep, outdoor sports and unfamiliar areas.

[0011] Optionally, when the portable electronic device is in a network-connected state, the alarm module is configured to transmit the alarm information to the preset guardian electronic mobile terminal through network communication; when the portable electronic device is in a non-network-connected state, the alarm module is configured to directly establish a voice call connection between the guardian electronic mobile terminal through the local communication module to realize the transmission of the alarm information.

[0012] Optionally, if the voice call connection attempt fails (for example, the target terminal is not answered, there is no available signal channel at present, etc.), the alarm module will automatically start a backup processing mechanism, which includes an offline storage and timed retransmission mechanism and a voice call replacement mechanism; in the offline storage and timed retransmission mechanism, the current alarm information (such as physiological parameters, motion data, positioning information, etc.) is stored in an encrypted and compressed form in the non-volatile storage unit of the device. The alarm module then detects the communication state according to the preset strategy, and automatically retries the transmission of the alarm information after detecting the recovery of the communication, until the transmission is successful or the maximum number of retries is exceeded; in the voice call replacement mechanism, in the case where real-time voice call cannot be established, the device will automatically play a pre-recorded help-seeking audio as a prompt sound for emergency events, so as to facilitate the surrounding personnel to perceive the abnormality and improve the warning effect and response efficiency in emergency scenarios.

[0013] Optionally, the response action includes an emergency handling response; when the alarm module triggers the emergency handling response, the following operations are performed: activating the high-frequency positioning function to improve the positioning accuracy; starting the collection of environmental sound for a continuous duration of several seconds; generating a compressed and encrypted data packet containing the current physiological parameters, motion data and environmental information, the compressed and encrypted data packet being encrypted using the SM4 algorithm and performing encryption key negotiation using a temporary session key, wherein the audio data is desensitized at the device end to filter out the frequency band information irrelevant to the dangerous event; performing corresponding processing according to the current network state and the risk level; when it is detected that the network is available, if the risk level is at a low level, the encrypted data packet is sent to the electronic terminal device of the guardian, and a voice call connection is established after a preset delay; if the risk level is at a high level, the delay is skipped, and the voice call connection is directly established, and the encrypted data packet is sent; when it is detected that the network is unavailable, the encrypted data packet is temporarily stored in the non-volatile storage unit of the device, and a voice call connection with the guardian is immediately established through the local communication module.

[0014] Optionally, the data perception module and the data processing module dynamically adjust the data sampling frequency and the calculation period according to different application scenarios; the risk assessment module and the alarm module are in a low-power standby mode in a non-triggering state, and enter an active state only when a pre-warning signal or a timing wake-up is detected.

[0015] In a second aspect, the application discloses an intelligent watch, comprising the safety monitoring and alarm system as disclosed in the first aspect.

[0016] In summary, the safety monitoring and alarm system and the intelligent watch disclosed by the application have at least the following beneficial effects:

[0017] (1) By fusing heart rate, motion, positioning, light, temperature and humidity, and other sensor data, and combining edge computing and lightweight machine learning models, real-time perception and risk identification of the user's state are achieved, significantly improving the accuracy and response speed of the monitoring system, and effectively reducing the possibility of false positives and false negatives;

[0018] (2) By setting an abnormality identification enhancement module, introducing context awareness, multi-source cross-validation, and individual baseline adaptation mechanisms, the system can dynamically adjust the judgment basis according to environmental and individual differences, improving the identification ability of real danger events to adapt to complex and variable actual application scenarios;

[0019] (3) The system supports feature extraction and discrimination for a variety of specific dangerous situations (such as falling, drowning, dragging, and sudden illness), and through multi-dimensional parameter comparison and multi-modal fusion evaluation, it realizes fine-grained monitoring of high-frequency and high-risk behaviors of special groups such as children;

[0020] (4) The alarm module has dual-channel alarm capability of networking and off-network, and can simultaneously perform high-frequency positioning, environmental recording, and information encryption upload when triggering an emergency response, so that key information can be reliably and timely transmitted to the guardian under different communication conditions;

[0021] (5) Through the power consumption adaptation mechanism, while meeting the real-time monitoring requirements, the overall system energy consumption can be effectively reduced, significantly prolonging the device battery life, and providing a more practical and stable operating environment for portable electronic devices. BRIEF DESCRIPTION OF DRAWINGS

[0022] The drawings used in the description of the embodiments of the application are briefly described below.

[0023] Figure 1 A structural schematic diagram of a safety monitoring and alarm system provided by an embodiment of the application is shown.

[0024] Figure 2 A SOS processing flowchart provided by an embodiment of the application is shown.

[0025] Figure 3 A general monitoring flowchart provided by an embodiment of the present application is shown.

[0026] Figure 4 An emergency response flowchart provided by an embodiment of the present application is shown. DETAILED DESCRIPTION

[0027] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the specific embodiments of the present application will be described below with reference to the drawings. The drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained from these drawings without creative labor, and other embodiments can be obtained without departing from the concept of the present application, and adjustments and improvements made within the scope of the present application are within the protection scope of the present application.

[0028] In order to make the drawing simple, only the parts related to the corresponding embodiments are shown in each drawing, which does not represent the actual structure of the product. In addition, in order to make the drawing simple and easy to understand, in some drawings, only some of the parts with the same structure or function are shown, and there can be more or less parts with the same structure or function.

[0029] In the present application, unless otherwise explicitly specified and limited, ordinal words such as "first", "second", etc. are only used to distinguish the description of the associated objects, and cannot be understood as indicating or implying the relative importance or order between the associated objects; in addition, it also does not represent the number of associated objects. "Multiple" includes two or more, and other quantifiers are similar. " / " is used to describe the relationship between the associated objects, which represents the "or" relationship between the associated objects. "And / or" is used to describe the relationship between the associated objects, which includes any combination relationship between the associated objects, for example, "a and / or b" includes: "a alone", "b alone", or "a and b". "One or more" or "at least one" of multiple objects means any object or any combination of multiple objects, for example, "one or more of a1, a2, a3" or "at least one of a1, a2, a3" includes: "a1 alone", "a2 alone", "a3 alone", "a1 and a2", "a1 and a3", "a2 and a3", or "a1, a2 and a3".

[0030] In the prior art, portable electronic devices (such as children's smart watches) have been widely used in the daily care and safety management scenarios of children. Such devices usually collect the motion and position data of the user through hardware such as acceleration sensors and GPS modules, and combine the periodic analysis of the data by the back-end server to determine whether there is an abnormal event. Once a suspicious state is detected, the device will send an alarm message to the guardian through the network. However, practical applications show that such traditional systems still have obvious shortcomings in many aspects, and it is difficult to meet the comprehensive requirements of "real-time, safety and intelligence" in the context of children.

[0031] Firstly, the high false positive rate and false negative rate are the main problems faced by current children safety monitoring systems. Traditional solutions mostly rely on a single acceleration sensor to determine dangerous behaviors such as falling, lacking understanding of the behavior context and cross-validation between data. For example, children may trigger the falling determination condition during daily play, jumping, and vigorous activity, etc., resulting in false positives; and real dangerous behaviors that do not accompany with violent acceleration changes are easily ignored by the system, causing false negatives. Studies show that the average false positive rate of existing systems in the industry is about 15%, which significantly affects the practicality of the system and the trust of the guardian (i.e. the parents of the children).

[0032] Secondly, there is a significant delay in state recognition. Many existing devices save power by uploading data periodically for server-side analysis, and the upload period is usually more than 30 seconds. If the user has a dangerous event during the upload interval, the system will not respond in time, which will miss the best intervention opportunity and seriously affect the timeliness and effectiveness of rescue. In the context of children, real-time response is particularly important, and the risk of delayed recognition cannot be ignored.

[0033] Thirdly, the device has insufficient endurance and poor user experience. Since multiple sensors need to continuously collect data and frequently upload through wireless modules, the device has high power consumption, resulting in a typical endurance time of less than 1 day for traditional children's wearable devices. Frequent charging not only affects the convenience of children's daily wear, but also easily causes forgetting to wear, reducing the sustainability of safety protection.

[0034] Fourthly, the poor environmental adaptability is also a technical shortcoming of traditional systems. In complex outdoor environments, behaviors such as running, falling, and dragging may have similar motion patterns, and it is difficult to accurately distinguish them relying on a single acceleration feature. Existing systems lack multi-modal perception and context analysis capabilities, making it difficult to adapt to the characteristics of children's highly uncertain behaviors, and the misjudgment rate further increases.

[0035] Fifth, the emergency response mechanism is imperfect. Most current systems can only send simple positioning information to parents when an anomaly is detected, and cannot synchronously transmit multi-dimensional on-site data such as heart rate fluctuations, motion trajectories, environmental parameters, etc., which makes it difficult for guardians to quickly determine the severity of the event and the current state, and is not conducive to developing timely and reasonable response strategies, missing effective rescue opportunities.

[0036] Sixth, the algorithm lacks adaptability to the behavior characteristics of children. Most traditional systems use algorithms designed for adult behavior models, without fully considering the physiological parameter differences and behavior characteristics of children. For example, children have a wider range of heart rate fluctuations and no obvious activity patterns, so adult models cannot accurately identify their behavior state, resulting in frequent false positives or identification failures. The differences between individual children are also not fully utilized by existing systems, and the static threshold judgment mechanism limits the intelligence and generalization ability of the system.

[0037] In summary, the existing child safety monitoring and alarm system has obvious shortcomings in precision, timeliness, power consumption, environmental adaptability, multi-dimensional judgment ability, and personalized modeling, and it is difficult to effectively meet the safety protection needs of children in daily use. Therefore, it is urgent to provide a safety monitoring and alarm system for children's portable devices based on multi-modal fusion analysis, which is safer, more responsive, and more durable, to solve the above technical problems and improve the overall user experience and monitoring efficiency.

[0038] The following will be described with reference to the accompanying drawings.

[0039] Please refer to Figure 1 , which shows a structure schematic diagram of a safety monitoring and alarm system provided by an embodiment of the present application. As Figure 1 shown, a safety monitoring and alarm system 100 includes a data sensing module 110, a data processing module 120, a risk assessment module 130, and an alarm module 140. The safety monitoring and alarm system 100 can be applied to a portable electronic device. In this application, the portable electronic device can be a smart watch, a smart bracelet, or other terminal device that is easy to carry, and the user is a child, and the guardian is the parent.

[0040] The data perception module 110 is configured to acquire real-time data of the user or the surrounding environment of the user, and the real-time data includes one or more of heart rate data, motion data, positioning data, light intensity data, temperature data, and humidity data. The data processing module 120 is configured to perform edge computing on the real-time data acquired by the data perception module 110 to obtain feature parameters, and the edge computing includes filtering motion artifacts, correcting the positioning data, and feature extraction. The risk assessment module 130 is configured to perform dangerous behavior recognition and state judgment by using a lightweight machine learning model that fuses multiple data sources according to the feature parameters obtained by the data processing module 120, and assesses the risk level based on a preset individual dynamic baseline threshold. The alarm module 140 is configured to trigger different levels of response actions according to the assessment result of the risk assessment module 130, in combination with the type of danger and the risk level in which the user is located.

[0041] From the hardware layer, various sensors are included in the data perception module, such as a PPG heart rate sensor, a six-axis IMU, a GPS / BDS dual-mode positioning sensor, a light intensity sensor, a temperature sensor, and a humidity sensor. The PPG heart rate sensor can continuously monitor heart rate and blood oxygen changes and detect sudden arrhythmia. The six-axis IMU can identify abnormal actions such as falls and drags. The dual-mode positioning GPS+Beidou dual-system positioning supports ephemeris prediction and rapid cold start. The environment-related sensors monitor temperature, humidity, and light intensity to assist in determining scenarios such as drowning (sudden increase in humidity) and heatstroke (high temperature). It should be noted that although the PPG heart rate sensor in the present application can be used to continuously monitor heart rate and blood oxygen changes and assist in identifying certain physiological abnormal states, the data collected by the device is for reference only and cannot replace professional medical diagnosis. The determination of high-risk physiological states such as arrhythmia in the system should in principle be used as a preliminary warning prompt and must be confirmed by a guardian or combined with subsequent medical means for judgment. The device does not have the attributes of a medical device and does not assume legal responsibility for any form of disease diagnosis, treatment recommendations, or medical intervention results. The setting is intended to enhance the auxiliary function of home safety monitoring, not to replace the professional health assessment system.

[0042] After data acquisition is completed, the data processing module will perform edge computing. In some embodiments of the present application, the data processing module comprises: a submodule for performing motion artifact filtering processing, which uses an adaptive filter to eliminate motion interference; a submodule for performing positioning data correction processing, which corrects positioning errors by Kalman filtering to fuse multiple source position information; a submodule for performing feature parameter extraction and rapid risk assessment, which processes based on a lightweight random forest model constructed based on fused heart rate, motion and positioning data, the model occupies a storage space of about 4KB, and is suitable for resource-limited edge computing devices. It should be noted that the "30ms" refers to the time required for the model to complete the inference operation on the edge device, excluding the pre-sensor sampling time and intermediate processing time. In actual application, the total response delay of the system includes: sensor sampling time (for example, the typical sampling period of a PPG heart rate sensor is 100ms), data preprocessing time (such as filtering, feature extraction), and model inference time. The above multi-stage processing jointly determines the final risk identification delay. The system can dynamically adjust the parameters of each stage according to the task urgency and computing resource situation, to ensure accuracy while considering real-time performance and power consumption performance.

[0043] That is, the data processing module can quickly complete data cleaning, correction and modeling inference locally on the device, avoiding the problem of uploading all raw data to the cloud server for processing, thereby effectively reducing the delay of data transmission and network burden. At the same time, even in a network-free or weak network environment, continuous evaluation and timely warning of the user's state can be achieved. In this way, the system response speed can be improved, millisecond-level data analysis and risk judgment can be realized; the stability and independence of the system can be enhanced, so that it can also stably run in edge devices (such as wearable terminals); and the privacy security can be improved, because a large amount of raw sensitive data does not need to be uploaded to the cloud, reducing the risk of leakage; and the network bandwidth and cloud service cost are reduced, which can meet the demand for low power consumption.

[0044] The risk assessment module is used for intelligent analysis and judgment of the current state of the user. It uses a lightweight machine learning model that fuses multiple data sources (such as heart rate, exercise intensity, geographic location, etc.) to identify whether the user has dangerous behavior and make judgments on their physiological or environmental state based on the feature parameters provided by the data processing module. The model has high computational efficiency and is friendly to edge devices, and can run in real time without relying on cloud computing. At the same time, the system also introduces individual dynamic baseline threshold as an evaluation standard, that is, the risk judgment standard for each user is not fixed, but is dynamically adjusted according to their daily behavior patterns and historical data, thereby improving the accuracy and individual adaptability of risk identification.

[0045] The alarm module is the core module that executes the corresponding response strategy after the risk assessment module completes the judgment. It not only depends on the risk level (such as low, medium, and high) evaluated, but also triggers different levels of response actions in combination with the specific dangerous type (such as falling, high temperature exposure, long time stillness, etc.) of the user. For example, for a serious event with a high risk level, the system may directly issue a sound alarm, send an emergency notification to the guardian or a medical service platform; and for a low or medium risk level, the system may use a vibration reminder or a graphical prompt. Through this targeted response mechanism, the system not only avoids frequent false positives that disturb the user, but also provides timely and effective safety protection in critical moments.

[0046] Please continue to refer to Figure 1 In some embodiments of the present application, the safety monitoring and alarm system 100 further comprises an abnormality identification enhancement module 150 for assisting in determining the authenticity of the abnormal state and reducing the false positive rate in the risk assessment process. The abnormality identification enhancement module performs processing operations based on at least one of the following mechanisms: a context awareness mechanism for dynamically adjusting the risk assessment threshold or alert level in combination with the external environment or behavior scene information of the user currently in; a multi-source information cross-validation mechanism for consistency verification in combination with real-time data from multiple sensors after detecting a preliminary abnormality to improve the accuracy of abnormality identification; and an individual dynamic baseline adaptive mechanism for dynamically updating the reference baseline of feature parameters based on the user's historical health data to enhance the adaptability of the model to individual differences.

[0047] The abnormality identification enhancement module can be used to implement the function of suppressing false positives. Through the combination of various verification and updating mechanisms, the false positive rate is reduced. For example, the context awareness mechanism can exclude known safe scenarios. When the user is in a school scenario and the risk coefficient is less than 0.85, all information that may trigger an alarm is excluded. This is because in a school environment, the user is often in a relatively safe and controllable area and is also under real-time supervision by teachers or monitoring systems. If the risk assessment result shows a low risk coefficient at this time, even if there are local abnormal parameters (such as short-term acceleration mutation or slight heart rate fluctuations), it is more likely to be a false positive caused by normal activities (such as running, jumping, and playing), so the system will actively shield such information to improve the robustness and fault tolerance of the overall judgment. For example, when the coefficient corresponding to the heart rate sensor is less than 0.3 and the motion change (such as acceleration) is greater than 2g 2 , it is not determined that the user has fallen. This is because although the value of acceleration exceeds 2g 2Generally means that a violent impact has occurred, which in a general scenario can represent a fall or a collision, but if at the same time a very small change in heart rate is detected (below the threshold of a normal stress response), it indicates that the user's physiological state remains stable and no typical fall stress response has occurred. Therefore, the system will determine that the violent movement is more likely to be non-dangerous (such as a quick sitting down, a vigorous jump or a brief shaking of the device), thereby avoiding a false alarm of a fall event. Such a two-factor determination mechanism can significantly improve the accuracy and practicality of the alarm system.

[0048] The multi-source information cross-verification mechanism is to start multi-sensor verification when an abnormality is initially detected and lasts for several seconds (e.g., 5 seconds). When most sensors (e.g., more than 2 / 3 of the sensors) confirm, the alarm mechanism is triggered, otherwise it is marked as a false alarm. In this way, the false alarm rate caused by single sensor abnormalities or short-term noise can be significantly reduced, and the accuracy of the alarm and the reliability of the system can be improved.

[0049] The individual dynamic baseline adaptive mechanism dynamically updates the reference baseline of the feature parameters based on the historical health data of the user to enhance the adaptability of the model to individual differences. For example, the heart rate baseline of the user can be updated, and a sliding window of 7 days is set. The following formula is used: baseline = 0.9 * baseline + 0.1 * current_hr. Wherein, the baseline on the left side of the equation is the updated heart rate baseline, the baseline on the right side of the equation is the heart rate baseline before updating, and current_hr is the current measured heart rate. After updating, if the absolute value of the difference between the current heart rate and the heart rate baseline is greater than a preset value (e.g., 30), a prompt message will be generated and sent to the terminal of the guardian, informing the guardian of the situation. In this way, personalized dynamic management of the user's health status can be achieved, and false positives or false negatives caused by unreasonable fixed threshold settings can be avoided.

[0050] In some embodiments of the present application, when the portable electronic device is in a networked state, the alarm module is configured to transmit the alarm information to the preset guardian electronic mobile terminal through network communication; when the portable electronic device is in a non-networked state, the alarm module is configured to directly establish a voice call connection between the guardian electronic mobile terminal through the local communication module to realize the transmission of the alarm information.

[0051] When the portable electronic device is in a networked state, the alarm module can timely transmit the alarm information to the preset guardian electronic mobile terminal through network communication (such as a cellular network, Wi-Fi, etc.), for example, sending an APP notification, a push message or a text reminder; when the device is in a non-networked state, the system automatically switches to establish a voice call connection with the guardian's electronic mobile terminal through a local communication module (such as Bluetooth, near field communication NFC, LoRa, etc.) to realize the instant communication of the alarm information. In this way, whether the device is currently networked or not, the system has an independent and reliable information transmission path, and the risk information can be delivered without interruption.

[0052] This mechanism can significantly improve the stability and continuity of the alarm system, enabling effective notification of guardians in various network environments, especially in practical application scenarios where network coverage is poor or temporarily lost. In addition, the voice call method is more direct and urgent than text notification, which can attract the immediate attention of guardians in critical moments, thereby enhancing the response efficiency and safety assurance capability of the system in emergency situations.

[0053] Figure 2 An SOS processing flowchart provided by an embodiment of the present application is shown. As shown in Figure 2 In some embodiments of the present application, the response action includes an emergency treatment response; when the alarm module triggers the emergency treatment response, the following operations are performed: the high-frequency positioning function is activated to improve the positioning accuracy; the environmental sound collection is started for a duration of several seconds; the compressed and encrypted data packet containing the current physiological parameters, motion data and environmental information is generated; the current network state and risk level are processed accordingly; when the network is detected to be available, if the risk level is at a low level, the encrypted data packet is sent to the guardian's electronic terminal device, and a voice call connection is established after a preset delay; if the risk level is at a high level, the delay is skipped, and the voice call connection is directly established, and the encrypted data packet is sent; when the network is detected to be unavailable, the encrypted data packet is temporarily stored in the non-volatile storage unit of the device, and a voice call connection with the guardian is immediately established through the local communication module.

[0054] The SOS processing flow can flexibly adjust the response strategy according to the risk level and network status, and realize rapid and intelligent response to emergency events. On the one hand, by activating high-frequency positioning and environmental sound collection, the current risk environment can be comprehensively perceived, and the judgment accuracy of the emergency event can be improved. On the other hand, the compressed and encrypted data packaging mechanism helps to efficiently transmit multi-source information under the premise of protecting user privacy, and improves the communication security and data integrity. In addition, dynamically adjusting the communication mode (such as delayed call or immediate call) according to the risk level not only improves the response efficiency of the system in emergency state, but also avoids excessive disturbance to non-emergency events. Even in a network-free environment, the system can ensure the immediate delivery of alarm information through the local communication module, and has reliable help ability even in the most severe scenarios, significantly enhancing the robustness and practicality of the system.

[0055] Among them, the compressed and encrypted data packet is encrypted by using the national standard SM4 algorithm, and the temporary session key is used for encryption key negotiation, wherein the audio data is desensitized at the device end to filter out the frequency band information irrelevant to the dangerous event. By using the national standard SM4 algorithm to encrypt the compressed data packet and combining the temporary session key for key negotiation, the technical scheme disclosed in the present application significantly improves the data security and anti-interception ability of the alarm information in the wireless transmission process, effectively prevents sensitive information from being eavesdropped, tampered with or forged during transmission, conforms to the national commercial encryption standard, and has stronger compliance and practicality. In addition, in terms of audio data processing, the system performs desensitization processing operation at the device end, filters out the frequency band information irrelevant to the dangerous event, not only effectively protects the privacy and safety of the user and the surrounding personnel, avoids disputes caused by irrelevant voice content, but also reduces the volume of audio data, improves the data compression efficiency and upload speed, and saves valuable time window for subsequent alarm response and data return.

[0056] In some embodiments of the present application, if the voice call connection attempt fails (for example, the target terminal is not answered, there is no available signal channel, etc.), the alarm module will automatically start the backup processing mechanism, which includes the offline storage and timing retransmission mechanism and the voice call replacement mechanism; wherein in the offline storage and timing retransmission mechanism, the current alarm information (such as physiological parameters, motion data, positioning information, etc.) is stored in the form of encrypted compression in the non-volatile storage unit of the device. The alarm module then detects the communication state according to the preset strategy, and automatically retries the transmission of the alarm information after detecting the recovery of the communication, until the transmission is successful or the maximum number of retries is exceeded; in the voice call replacement mechanism, in the case where real-time voice call cannot be established, the device will automatically play a pre-recorded help audio as a prompt sound for emergency events, so as to facilitate the surrounding personnel to perceive the abnormality and improve the warning effect and response efficiency in emergency scenarios.

[0057] By the above manner, even in the case that the device is in a non-networking state and the communication condition is limited, the system can still guarantee the reservation and subsequent delivery of the key alarm information, while realizing the instant release of the basic help signal, thereby enhancing the alarm robustness and practicality of the system under extreme conditions.

[0058] In some embodiments of the present application, the alarm mechanism is divided into three levels, corresponding to the response requirements under different risk levels respectively, which embodies the hierarchical processing capability of the system for abnormal situations. For example, the first-level risk strategy mainly deals with slight abnormalities or potential risks, only reminding through the APP pop-up window, and recording the user's action trajectory for subsequent tracing and behavior analysis; the second-level risk strategy faces higher-level risks, sends a clear warning to the user himself through vibration, and starts the high-frequency positioning function (such as 1Hz sampling) to continuously obtain fine location information; the third-level risk strategy targets serious or urgent risks, automatically triggers the SOS mechanism, and sends the on-site recording, location information and real-time medical related data (such as heart rate, blood oxygen, etc.) to the guardian or service platform, realizing all-round and rapid emergency response.

[0059] In this way, the hierarchical strategy can effectively balance the response sensitivity and interference control of the system, avoiding the disturbance of the user and / or guardian caused by frequent false alarms due to slight fluctuations, while immediately taking comprehensive and high-priority emergency measures when a real emergency occurs. The hierarchical mechanism enables the system to have higher intelligence and self-adaptability, improves the user and / or guardian experience, and accurately responds to different risk situations, thereby significantly enhancing the practicality, safety and reliability of the system.

[0060] Please refer to Figure 3 which shows a conventional monitoring flowchart provided by an embodiment of the present application. As shown in Figure 3 After the sensor completes the collection of physiological, motion or environmental data, the system first pre-processes the original signal to remove noise, calibrate errors and improve data quality; then enters the feature extraction stage to extract key indicators such as heart rate fluctuation amplitude, acceleration change trend, geographic displacement, etc., which are used as inputs for the subsequent risk assessment model; then the system performs risk scoring according to the extracted features to determine whether there is potential danger in the current state. On the other hand, after data collection, the system also performs dynamic data management operations in parallel. For example, the current heart rate data is fused and updated with the historical heart rate baseline to form an individualized dynamic baseline, which improves the adaptability of the model to individual differences of the user. At the same time, those normal data judged as low risk will be archived and stored for subsequent model training, reference or traceability analysis. In addition, this part of low-risk data can also be directly used for feature extraction, further enriching the feature samples of the current period, thereby improving the running stability and recognition accuracy of the system under non-emergency conditions.

[0061] The flow realizes the parallel of "real-time analysis" and "long-term accumulation" of data processing, which can quickly respond to abnormal states, and through continuous learning and archiving, enhance the individualization and long-term adaptability of the model. At the same time, the low-risk data is used for feature extraction, which improves the utilization efficiency of system resources, helps to reduce the overall power consumption and improve the robustness of the algorithm.

[0062] Please refer to Figure 4 , which shows an emergency response flowchart provided by an embodiment of the present application. As shown in Figure 4 , the emergency response flow includes a plurality of key links in series, which is used to quickly, safely and efficiently start a comprehensive emergency handling mechanism when a high-risk score is detected. When the system determines that the current risk score reaches the high-risk threshold, the emergency protocol will be activated immediately, and the data encryption and packaging operation will be performed to ensure the confidentiality and integrity of sensitive information during transmission. After the emergency protocol is activated, the system will simultaneously start multi-source data collection, including physiological parameters, motion state, environmental sound, location information, etc., to comprehensively build risk scene perception. After the data encryption and packaging are completed, the flow is divided into two parallel directions: on the one hand, the dual-channel data transmission mechanism is started, and the information is pushed through the main network channel and the local communication channel respectively, which improves the success rate and timeliness of information delivery, and at the same time, the APP sends an explicit alarm notification; on the other hand, the escape assistance mechanism (such as issuing sound, light signal guidance, or navigation to a safe location) is started, and the external environment is continuously perceived, and the rescue state information is updated in real time, so that the subsequent interveners can understand the progress on site.

[0063] The flow takes into account the response speed, information integrity and multi-channel reliability in design, not only improves the emergency efficiency in emergency, but also strengthens the overall protection of the user's personal safety. The dual-channel transmission and continuous state update mechanism significantly reduces the risk of information loss and rescue delay, and the introduction of the escape assistance function reflects the system's ability to coordinate in the "self-help" and "other help" levels, thereby greatly improving the application value and emergency disposal level of the system in actual high-risk scenarios.

[0064] In some embodiments of the present application, the data processing module and the risk assessment module cooperatively realize the identification of multiple dangerous situations, including: the data processing module extracts the impact force peak parameter and the body position mutation parameter through dynamic threshold segmentation detection and quaternion attitude solution respectively, and the risk assessment module compares the impact force peak parameter and the body position mutation parameter with the corresponding parameters in the preset threshold value to determine whether the user has fallen; the data processing module extracts the humidity sudden increase parameter and the heart rate sudden drop parameter through first-order difference mutation detection and trend slope calculation respectively, and the risk assessment module compares the humidity sudden increase parameter and the heart rate sudden drop parameter with the corresponding parameters in the preset threshold value to determine whether the user is in a drowning state; the data processing module extracts the continuous rotation parameter and the step frequency abnormal parameter through fast Fourier transform spectrum analysis and gait cycle correlation matching respectively, and the risk assessment module compares the continuous rotation parameter and the step frequency abnormal parameter with the corresponding parameters in the preset threshold value to determine whether the user has a dragging behavior; the data processing module extracts the heart rate variability parameter and the motion stillness parameter through Poincaré scatter plot analysis and zero speed detection respectively, and the risk assessment module compares the heart rate variability parameter and the motion stillness parameter with the corresponding parameters in the preset threshold value to determine whether the user has a sudden illness risk.

[0065] Due to the large difference in the weight of children, different detection thresholds are set for different children by introducing a weight compensation factor. The standard child weight is set to 30 kg, and the base threshold Threshold_base is determined to be 3.0 g; this data is based on industry average data. The weight compensation formula is established as follows: wherein, Threshold_adjusted is the compensated threshold, Threshold_base is the base threshold for compensation, w is the current child weight, unit is kg, w_ref is the reference weight, for example, it can be set to 30 kg, k is the adjustment coefficient, which can be determined according to experimental calibration, for example, it can be 0.5. For example, for a child with a lighter weight (20 kg), the compensated threshold is 3.0*(1+0.5*(20-30) / 30) = 2.5 g; for a child with a heavier weight (50 kg), the compensated threshold is 3.0*(1+0.5*(50-30) / 30) = 4.0 g. The segmented detection logic includes dynamic window division, that is, the detection window length is adjusted according to the weight (the heavier the weight, the longer the window). Since the falling action of a child with a heavier weight lasts longer, dynamic window division can better adapt to the specific situation of different children, thereby realizing adaptive adjustment of the weight term.

[0066] The dangerous situation can include falling, drowning, (child being) dragged, and disease outbreak. The preset threshold values in the risk assessment module can be set according to the needs to meet the actual situation of different children. Through sliding window peak detection, an impact force peak value is obtained, and a body position mutation parameter is obtained through quaternion attitude solution; when the impact force peak value is greater than 3g on the z-axis and the pitch angle in the body position mutation parameter is greater than 60°, it is determined that the child falls. In addition, an age compensation formula can also be set, similar to the above weight compensation formula, the threshold value after compensation = basic threshold value x (1+k·(age-8) / 8), k is the adjustment coefficient, and the age is taken as the basic value of 8 years old. Through first-order difference mutation detection, a humidity sudden increase parameter is obtained, and a heart rate sudden drop parameter is obtained through trend slope calculation; when the humidity sudden increase parameter is greater than 90% and the change value Δ of the heart rate sudden drop parameter is greater than 40bpm / 10s, it is determined that the child drowns. Here, in order to improve the recognition accuracy of the system in a specific high-risk scene (such as drowning), and to avoid the key physiological characteristics being ignored due to improper sensor weight configuration, the system introduces a dangerous type priority adjustment mechanism and a multi-sensor conflict detection rule. Specifically, when the humidity sudden increase parameter is detected to be greater than 90%, the system automatically identifies this situation as a potential high-risk state of drowning, and forcibly increases the weight of heart rate data in risk assessment to 0.7, in order to pay attention to the heart rate change characteristics, and avoid the motion signal from covering up the core indicators such as heart rate sudden drop due to too high proportion. At the same time, the system also introduces a conflict detection mechanism: when there is obvious contradiction between the motion feature and the heart rate feature in the fusion evaluation (for example, the motion signal prompts normal while the heart rate mutation prompts abnormal), the system will automatically trigger the multi-sensor review process, retrieve the original data for consistency verification, and determine whether to enter a higher level of alarm judgment process according to the rule priority. Through this strategy, the judgment robustness and fault tolerance of the system to high-risk events such as drowning in complex environments can be significantly enhanced.

[0067] Through Fourier transform spectrum analysis, a continuous rotation parameter is obtained, and a step frequency abnormality parameter is obtained through gait cycle correlation matching; when the angular velocity in the continuous rotation parameter is greater than 1.5 rad / s and the step frequency is abnormal (different from the conventional step frequency), it is determined that the child is being dragged. Through Poincaré scatter plot analysis, a heart rate variability parameter is obtained, and a motion stillness parameter is obtained through zero speed detection; when the SDNN in the heart rate variability parameter is less than 30ms and the acceleration variance in the motion stillness parameter is less than 0.1g 2 , it is determined that the child has a sudden illness.

[0068] The embodiment realizes accurate recognition of common dangerous situations of children (including falling, drowning, being dragged, and sudden illness) through a multi-modal fusion analysis method, and combines impact force, body position, humidity, heart rate, rotation speed, step frequency, and heart rate variability and other multi-dimensional parameters for comprehensive judgment, thereby significantly improving the accuracy and robustness of recognition. Each type of dangerous behavior has a clear preset threshold, and dynamic adjustment according to individual differences is supported, which improves the recognition sensitivity and effectively reduces the false positive rate, and can more accurately adapt to the actual safety monitoring needs of different children in different scenes.

[0069] In some embodiments of the present application, the risk assessment module performs weighted fusion based on the heart rate anomaly detection result, the motion pattern recognition result, and the positioning trajectory analysis result to output an evaluation result. The risk assessment module dynamically adjusts the weighted parameters of heart rate anomaly detection, motion pattern recognition, and positioning trajectory analysis according to the environment scenario in which the user is located, including daily activities, nighttime sleep, outdoor exercise, and unfamiliar areas.

[0070] Through heart rate anomaly detection, a heart rate anomaly detection result (hereinafter referred to as output probability P1) is obtained; through motion pattern recognition, a motion pattern recognition result (hereinafter referred to as output probability P2) is obtained; and through positioning trajectory analysis, a motion trajectory analysis result (hereinafter referred to as output probability P3) is obtained. P1, P2, and P3 are input to the fusion engine, and a weighted decision is made, and the evaluation result is determined according to the final output probability after weighting. For example, the evaluation result includes triggering SOS, warning prompt, and continuous monitoring. When the final output probability is greater than or equal to 0.8, the system triggers SOS; when the final output probability is between 0.6 and 0.8, the system gives a warning prompt; and when the final output probability is less than 0.6, the system continuously monitors without giving a warning.

[0071] In addition, the risk assessment module also dynamically adjusts the weights of P1, P2, and P3 in the weighted processing according to the environment scenario in which the user is located. For example, when the child is in daily activities, the heart rate weight and the motion weight can be set to 0.4, and the positioning weight can be set to 0.2; when the child is in nighttime sleep, the motion weight and the positioning weight can be set to 0.2, and the heart rate weight can be set to 0.6; when the child is in outdoor exercise, the heart rate weight can be set to 0.3, the motion weight can be set to 0.5, and the positioning weight can be set to 0.2; and when the child is in an unfamiliar area, the heart rate weight can be set to 0.2, the motion weight can be set to 0.3, and the positioning weight can be set to 0.5.

[0072] In this embodiment, by introducing a multi-source data fusion mechanism in the risk assessment module, the heart rate anomaly detection, motion pattern recognition and positioning trajectory analysis three types of evaluation results are weighted processed, and the weight of each index is dynamically adjusted according to the environment scene where the user is located, which improves the identification accuracy and adaptability of the system to different dangerous situations. The weighted fusion strategy overcomes the limitations of traditional single parameter evaluation method, while maintaining the sensitivity of judgment, effectively reduces the false positive rate.

[0073] In addition, by modeling the behavior characteristics of children in different scenarios, the system can emphasize physiological indicator monitoring in the night sleep scenario, highlight dynamic behavior recognition during outdoor activities, and enhance the sensitivity of location deviation in unfamiliar areas, thereby realizing intelligent evaluation that is more in line with the actual risk distribution. The above strategies not only enhance the flexibility and generalization ability of the evaluation model, but also significantly improve the response accuracy and resource allocation efficiency of the system, so that children can obtain more reliable safety protection in various life situations.

[0074] In some embodiments of the present application, the data sensing module and the data processing module dynamically adjust the data sampling frequency and the calculation period according to different application scenarios; the risk assessment module and the alarm module are in a low-power standby mode in a non-triggering state, and only enter an active state when a pre-warning signal or a timing wake-up is detected.

[0075] In order to further improve the endurance and energy efficiency ratio of the system, the data sensing module and the data processing module are designed to have scene awareness and dynamic control capability. For example, the system can automatically adjust the data sampling frequency of various sensors according to the current use situation (such as daily activities, night sleep, outdoor sports, etc.). For example, in the night sleep scenario, the device will reduce the sampling frequency of the acceleration and positioning module, while appropriately increasing the detection accuracy of the heart rate sensor to focus on monitoring static abnormal physiological characteristics, thereby realizing accurate allocation of resources and minimization of power consumption. In addition, the data processing module also adopts a similar scene-driven strategy, which adjusts the feature extraction and calculation period accordingly when the data sampling density is reduced, to reduce unnecessary computational overhead. For example, when the system is in a "low-risk activity area" or detects that the user is stationary for a long time, the processing period will be automatically lengthened, reducing the CPU active time, thereby significantly reducing the overall energy consumption of the device.

[0076] In addition, in order to avoid the waste of the battery caused by the continuous running of the system in the risk-free state, the risk assessment module and the alarm module both support a low-power standby mechanism. In the non-trigger state, these modules will be in a sleep mode, only retaining a basic monitoring thread to perceive the wake-up event. Once the data perception module detects a potential risk signal or reaches the preset timing detection point, the system will quickly wake up the corresponding functional module, restore the running state and perform real-time analysis and response, ensuring that the safety response speed is guaranteed while the power consumption control is taken into account.

[0077] Through the above optimization strategies, the safety monitoring and alarm system of the present application can achieve high real-time performance and high accuracy while significantly improving the overall endurance. Under the condition of a typical 200mAh battery capacity, it can achieve continuous running for up to 7 days, and is particularly suitable for use in portable electronic devices such as children's smart watches that are worn for a long time and are inconvenient to charge.

[0078] In some embodiments of the present application, the safety monitoring and alarm system can be provided with a three-level alarm system corresponding to the primary response, secondary response and ultimate response, respectively. The three-level alarm system can also be distinguished by setting thresholds, for example, when the risk coefficient is greater than 0.9, the ultimate response is triggered; when the risk coefficient is between 0.7 and 0.9, the secondary response is triggered; when the risk coefficient is between 0.5 and 0.7, the primary response is triggered; and when the risk coefficient is less than 0.5, no response is triggered.

[0079] The operations of various responses are also different for different types of risks. For example, when the risk type is falling, the primary response is a vibration prompt, the secondary response is sending the location to the parent's APP, and the ultimate response is automatically dialing the emergency contact; when the risk type is drowning, the primary response is to activate the GPS high-frequency positioning, the secondary response is to start the waterproof microphone recording, and the ultimate response is to link the nearby rescue equipment; when the risk type is abnormal dragging, the primary response is to display a normal screen, the secondary response is to upload the real-time trajectory silently, and the ultimate response is to trigger the sound and light alarm deterrent. In this way, the system can achieve differentiated response strategies under different risk levels and different risk types, improving the accuracy and effectiveness of the overall response. This is because different types of risk situations differ in the degree of emergency, evolution speed, on-site features and controllability. For example, drowning is a high-fatal event that develops rapidly and often occurs in an environment with weak network signal, so it requires stronger positioning and sound collection response. Abnormal dragging, on the other hand, is often covert, so silent tracking and deterrent methods are more suitable. Therefore, if a uniform and fixed response strategy is adopted, not only the best disposal window may be missed, but unnecessary interference or misjudgment may also be caused.

[0080] By dynamically matching the hazard type and the response strategy, the system can make more targeted processing according to the actual situation, making the response logic more intelligent, detailed and practical, not only improving the response efficiency and user trust of the system, but also providing more valuable information support for subsequent manual intervention and rescue decision-making.

[0081] In some embodiments of the application, in order to improve the response speed and real-time processing capability of the system, a real-time optimization scheme for resource-constrained devices is proposed, mainly including edge computing acceleration and data pipeline design. In terms of edge computing, the system makes full use of the machine learning core (MLC) integrated in the IMU to sink part of the feature extraction task to the hardware layer for processing, thereby significantly reducing the computing pressure and response delay of the host chip. At the same time, the inference model used in the risk assessment process adopts a lightweight architecture (such as TensorFlow Lite Micro), with a model size controlled within 20KB, suitable for microcontroller-level operation resources, ensuring that the model runs quickly on the terminal side, achieving millisecond-level danger discrimination.

[0082] In terms of data processing flow, the system designs an efficient data pipeline architecture. Heart rate, motion and positioning raw data from multiple sensors are uniformly sent to a ring buffer management, and various data are processed in the buffer according to modules: heart rate data is used for physiological abnormality detection, motion data is used for behavior pattern recognition, and positioning data is used for trajectory analysis and offset correction. Each processing module adopts a parallel processing strategy to avoid serial blocking and improve overall data throughput. This structure not only speeds up the whole process from collection to judgment, but also enhances the system's processing capability for multiple high-frequency events.

[0083] Through the above real-time optimization measures, the safety monitoring and alarm system of the application can maintain high accuracy while compressing the delay of identifying dangerous events to a very short time range, thereby completing analysis and response in the first time in emergency situations such as children falling, drowning or abnormal dragging. This optimization strategy significantly improves the emergency response capability and effectiveness of the system in actual use, providing more timely and reliable technical support for child safety protection.

[0084] Based on similar technical concepts, the application discloses an intelligent watch comprising a safety monitoring and alarm system as disclosed in the above embodiments.

[0085] As described in the above embodiments, the safety monitoring and alarm system can be a system specially designed for children, which is different from other systems on the market. The related parameters in this system are set according to the characteristics of children. Therefore, the intelligent watch comprising this system is a children's intelligent watch.

[0086] As described in the above embodiments, the safety monitoring and alarming system can be a system specially designed for children, which is different from other general safety monitoring systems on the market. The various determination thresholds, fusion strategies, and data processing models in the system are all customized based on the physiological characteristics and behavior characteristics unique to children. Therefore, the smart watch including the system is a smart watch for children, which is designed differently in structure and functional logic for children.

[0087] It should be noted that the safety monitoring needs of children are significantly different from those of adults, and even differ essentially from other common monitoring objects (such as the elderly). First, in terms of physiological characteristics, the heart rate baseline of children is higher, the fluctuation range is large, and the resting and active states are switched frequently. If the thresholds in the adult or elderly model are directly used, frequent false positives or false negatives will occur. Second, the behavior pattern of children is highly active and lacks regularity, and there are a large number of violent actions and high-frequency displacements in their daily activities, and conventional motion recognition models are difficult to accurately distinguish between normal activities and dangerous behaviors. In addition, children generally lack self-protection ability and danger perception ability, so the requirements for real-time and system response are higher, and the system needs to have shorter recognition delay and stronger adaptability.

[0088] In the above embodiments, the description of each embodiment has its own focus, and the parts not described or recorded in detail in a certain embodiment can be referred to the relevant description of other embodiments. In addition, the above embodiments can be freely combined as needed.

Claims

1. A safety monitoring and alarm system, characterized in that: Applied to portable electronic devices, the system includes: A data sensing module is used to obtain real-time data of the user or his / her surrounding environment, wherein the real-time data includes several of the following data: heart rate data, motion data, positioning data, light intensity data, temperature data, and humidity data; A data processing module, configured to perform edge computing on the real-time data acquired by the data sensing module to obtain feature parameters, wherein the edge computing includes filtering out motion artifacts, performing deviation correction and feature extraction on the positioning data; A risk assessment module, configured to identify dangerous behaviors and determine status based on the characteristic parameters obtained by the data processing module using a lightweight machine learning model that integrates multiple data sources, and to assess the risk level based on a preset individual dynamic baseline threshold; The alarm module is used to trigger different levels of response actions based on the assessment results of the risk assessment module and the type of danger and risk level of the user.

2. The safety monitoring and alarm system according to claim 1, characterized in that: Also includes: An anomaly identification enhancement module is used to assist in determining the authenticity of abnormal conditions and reduce the false alarm rate during the risk assessment process. The anomaly identification enhancement module performs processing operations based on at least one of the following mechanisms: Context-aware mechanism, used to dynamically adjust risk assessment thresholds or alert levels based on the user's current external environment or behavioral scenario information; A multi-source information cross-validation mechanism is used to combine real-time data from multiple sensors for consistency verification after detecting preliminary anomalies to improve the accuracy of anomaly identification; The individual dynamic baseline adaptation mechanism is used to dynamically update the reference baseline of feature parameters based on the user's historical health data to enhance the model's adaptability to individual differences.

3. The safety monitoring and alarm system according to claim 1, characterized in that: The data processing module includes: A submodule for performing motion artifact filtering processing, wherein the motion artifact filtering processing uses an adaptive filter to eliminate motion interference; A submodule for performing positioning data correction processing, wherein the positioning data correction processing fuses multi-source position information through Kalman filtering to correct positioning errors; A submodule for performing feature parameter extraction and rapid risk assessment, which is based on a lightweight random forest model constructed by fusing heart rate, motion and positioning data.

4. The safety monitoring and alarm system according to claim 1, characterized in that: The data processing module and the risk assessment module collaborate to identify a variety of dangerous situations, including: The data processing module extracts impact force peak parameters and body position mutation parameters respectively through dynamic threshold segmentation detection and quaternion posture solution, and the risk assessment module compares the impact force peak parameters and body position mutation parameters with corresponding parameters in the preset threshold to determine whether the user has fallen; The data processing module extracts a humidity surge parameter and a heart rate drop parameter respectively through first-order differential mutation detection and trend slope calculation, and the risk assessment module compares the humidity surge parameter and the heart rate drop parameter with corresponding parameters in the preset threshold value to determine whether the user is in a drowning state; The data processing module extracts the continuous rotation parameter and the abnormal gait frequency parameter respectively through fast Fourier transform spectrum analysis and gait cycle correlation matching, and the risk assessment module compares the continuous rotation parameter and the abnormal gait frequency parameter with the corresponding parameters in the preset threshold to determine whether the user has been dragged; The data processing module extracts heart rate variability parameters and motion-still parameters through Poincaré scatter plot analysis and zero-speed detection, respectively. The risk assessment module compares the heart rate variability parameters and motion-still parameters with corresponding parameters in the preset thresholds to determine whether the user has a risk of sudden illness.

5. The safety monitoring and alarm system according to claim 1, characterized in that: The risk assessment module performs weighted fusion based on the heart rate anomaly detection result, the motion pattern recognition result and the positioning trajectory analysis result to output the assessment result.

6. The safety monitoring and alarm system according to claim 5, characterized in that: The risk assessment module dynamically adjusts the weighted parameters of the heart rate anomaly detection, motion pattern recognition, and positioning trajectory analysis according to the user's environmental scenario, which includes daily activities, nighttime sleep, outdoor exercise, and unfamiliar areas.

7. The safety monitoring and alarm system according to claim 1, characterized in that: When the portable electronic device is in an online state, the alarm module is used to transmit the alarm information to the preset guardian's electronic mobile terminal through network communication; when the portable electronic device is in an offline state, the alarm module is used to directly establish a voice call connection with the guardian's electronic mobile terminal through the local communication module to realize the transmission of the alarm information.

8. The safety monitoring and alarm system according to claim 1 or 7, characterized in that: The response action includes an emergency processing response; when the alarm module triggers the emergency processing response, the following operations are performed: Activate high-frequency positioning function to improve positioning accuracy; Start collecting ambient sound for several seconds; Generate a compressed and encrypted data packet containing current physiological parameters, motion data, and environmental information. The compressed and encrypted data packet is encrypted using the national secret SM4 algorithm and a temporary session key is used for encryption key negotiation. The audio data is desensitized on the device to filter out frequency band information not related to dangerous events. Take appropriate action based on the current network status and risk level; When it is detected that the network is available, if the risk level is low, the encrypted data packet is sent to the guardian's electronic terminal device, and a voice call connection is established after a preset delay; if the risk level is high, the delay is skipped and the voice call connection is directly established, and the encrypted data packet is sent at the same time; When it is detected that the network is unavailable, the encrypted data packet is temporarily stored in a local non-volatile storage unit of the device, and a voice call connection with the guardian is immediately established through the local communication module.

9. The safety monitoring and alarm system according to claim 1, characterized in that: The data perception module and the data processing module dynamically adjust the data sampling frequency and calculation cycle according to different application scenarios; the risk assessment module and the alarm module are in a low-power standby mode when not triggered, and enter an active state only when a warning signal is detected or a timed wake-up is performed.

10. A smart watch, characterized in that: It comprises the safety monitoring and alarm system as described in any one of claims 1 to 9.

Images