Device and method for predicting underwater activity risks based on multi-sensor fusion and machine learning

Abstract

The present invention relates to a drowning risk detection and automatic flotation system based on the fusion of biosignal and inertial sensor data. Specifically, it relates to a technology that quantitatively distinguishes between regular swimming movements and critical drowning situations by fusion-analyzing a user's electrocardiogram, optical heart rate data, and movement data in real time, and safely provides buoyancy by variably controlling the opening time of a flotation valve according to the detected risk level and the pressure status of a gas cartridge.

Description

Technology Field

[0001] The present invention relates to a drowning risk detection and automatic flotation system based on the fusion of biosignal and inertial sensor data. Specifically, it relates to a technology that quantitatively distinguishes between regular swimming movements and critical drowning situations by fusion-analyzing a user's electrocardiogram, optical heart rate data, and movement data in real time, and safely provides buoyancy by variably controlling the opening time of a flotation valve according to the detected risk level and the pressure status of a gas cartridge. Background Technology

[0002] With the recent increase in marine leisure activities, drowning accidents are occurring frequently, and the importance of life vests and wearable devices to prevent them is emerging. Conventional life vests have limitations, such as being bulky and hindering mobility, or requiring manual inflation, which renders them useless if the user loses consciousness or panics and is unable to respond.

[0003] Accordingly, technologies that automatically detect danger by embedding acceleration or heart rate sensors have been proposed, but the following technical problems remain unresolved.

[0004] First, the existing method, which simply sets the magnitude of acceleration as the threshold, frequently suffers from a 'false positive' problem where high-intensity swimming movements, such as butterfly or sprinting, are mistaken for the flailing motions of drowning, causing the flotation device to operate incorrectly.

[0005] Second, in underwater environments, contact failures inevitably occur where the sensor temporarily detaches from the skin due to water currents or violent movements. Conventional technology fails to distinguish this noise from actual heart abnormality signals, thereby reducing the reliability of the alarm.

[0006] Third, drowning accidents include not only 'dynamic drowning' involving significant movement but also 'static drowning' where the victim loses consciousness and sinks quietly; however, methods relying on a single sensor (acceleration or heart rate) have a blind spot that fails to cover these two contrasting situations simultaneously.

[0007] Fourth, in the case of a gas expansion flotation device, the valve is kept open for a fixed period of time despite the internal gas pressure fluctuating according to the external temperature, so there is a risk of tube breakage due to overpressure in high temperatures in summer and incomplete flotation due to insufficient gas in low temperatures in winter.

[0008] Therefore, there is a need to develop an integrated control system capable of logically correcting errors in sensor data, actively responding to various drowning scenarios, and ensuring mechanical safety by adapting to environmental variables. The problem to be solved

[0009] The present invention is intended to solve the aforementioned problems,

[0010] The first objective is to provide a dynamic motion confusion calculation logic that clearly distinguishes between high-intensity swimming and actual drowning panic by reflecting regular swimming stroke movements as suppression factors to correct the confusion level, even when rapid acceleration and directional changes are detected in inertial sensor data.

[0011] The second task aims to provide a logic for calculating biosignal severity that increases sensitivity by exponentially weighting the ratio of abnormal waveforms deviating from the normal range during biosignal analysis, while physically blocking malfunctions caused by noise by reflecting the number of electrode contact failures as a reliability reduction factor.

[0012] The third task aims to detect static drowning cases resulting from loss of consciousness without blind spots, by utilizing an integrated judgment logic that forcibly increases the risk probability by considering a motionless stationary state exceeding a threshold time as a dominant trigger, in addition to dynamic drowning cases involving high movement.

[0013] The fourth task aims to provide a variable control method that prevents airbag failure and provides quantitative buoyancy regardless of temperature or pressure fluctuations by adjusting the valve opening time inversely proportional to the internal pressure of the gas cartridge based on Bernoulli's principle and dynamically allocating a safety margin according to the risk level.

[0014] The problems of the present invention are not limited to those mentioned above, and other problems not mentioned will be clearly understood by those skilled in the art from the description below. means of solving the problem

[0016] In a system for fusion analysis of biosignals and movement data generated during underwater activities, worn on a user's body according to an embodiment of the present invention for solving the above problem,

[0017] A biosignal measurement unit that measures biosignals including a user's electrocardiogram (ECG) and optical heart rate (PPG) to generate heart rate interval variation data;

[0018] An inertial measurement unit that measures the user's underwater movement vector and rotation amount using a 3-axis accelerometer and gyroscope;

[0019] A situation analysis module that calculates a dynamic movement disturbance index by extracting the number of sudden acceleration events and direction changes from the data of the inertial measurement unit, performing calculations with a positive correlation, and reflecting regular swimming motions as a suppression factor, and calculates a biosignal urgency index by extracting the heart rate ratio deviating from the normal range from the data of the biosignal measurement unit, weighting exponentially, and reflecting the number of electrode contact failures as a damping factor;

[0020] A risk judgment module that calculates the probability of drowning risk by integrating and analyzing the dynamic movement confusion index and the biosignal urgency index, and generates a flotation command when preset trigger conditions are satisfied; and

[0021] A flotation unit that opens a gas cartridge to inflate an airbag upon receiving the above flotation command;

[0022] Includes,

[0023] The above situation analysis module is,

[0024] In calculating the above dynamic motion disturbance index, a first operation logic that generates a raw risk value having a mutual synergistic effect by multiplying the number of rapid acceleration events measured per unit time and the number of changes in the direction of the acceleration axis to quantify the irregularity of movement rather than the simple magnitude of acceleration;

[0025] A second operation logic that corrects the confusion index calculated as strokes are detected to converge to 0 by setting the number of regular strokes identified through frequency analysis as the denominator term dividing the raw risk value to prevent mistaking high-intensity swimming movements, such as butterfly or sprinting, for dangerous situations; and

[0026] A third operation logic that prevents data divergence by applying a logarithmic function that gradually slows down the rate of increase of the result value as the total amount of input motion data increases;

[0027] It is characterized by performing this sequentially to exclude regular swimming strokes and selectively extract only irregular flailing, and

[0028] The above situation analysis module is,

[0029] As a result of executing the first and second operation logic above, even if the number of rapid acceleration events and the number of direction changes are both measured to be higher than a preset threshold, if the number of strokes is detected one or more times, variable power control is performed to determine this as a high-intensity exercise state, block the generation of the buoyancy command, and increase the sampling period of the biosignal measurement unit.

[0030] Characterized by determining the dynamic movement confusion index as a dangerous level and identifying the drowning panic state only when both the number of sudden acceleration events and the number of direction changes are high, and simultaneously the number of strokes is not detected, so that no inhibition action by the denominator term occurs.

[0031] The above situation analysis module is,

[0032] A first processing process for calculating the above-mentioned vital signal urgency index, which calculates the ratio of abnormal heart rate samples deviating from the normal range relative to the total number of valid waveforms to reflect the lethality of the vital signal, and performs an exponential weighting logic that applies the power of the natural constant (e) to the ratio value to rapidly increase the risk score in an initial emergency situation; and

[0033] A second processing process to prevent malfunction caused by sensor shaking in an underwater environment, wherein the number of contact failures detected through changes in electrode impedance is set as a reliability reduction factor that divides the risk score, thereby processing such that the final urgency index decreases as contact failures become more frequent;

[0034] Characterized by performing operations including,

[0035] The above situation analysis module is,

[0036] By the interaction of the above first processing process and second processing process,

[0037] Even if the ratio of the above abnormal heart rate samples exceeds a preset threshold, if the number of contact failures is also measured to be high, this is determined to be sensor noise or sensor detachment caused by underwater activity, the generation of the above flotation command is suspended, and a wearing status check notification is generated.

[0038] The method is characterized by determining the above-mentioned abnormal heart rate sample ratio as an actual heart abnormality or loss of consciousness situation and establishing the above-mentioned vital sign urgency index at a dangerous level only when the above-mentioned abnormal heart rate sample ratio is measured as high while the above-mentioned number of contact failures is maintained at a low level below a preset threshold so that no attenuation due to the reliability reduction factor occurs.

[0039] The above risk judgment module is,

[0040] The dynamic movement confusion index and the biosignal urgency index calculated from the above situation analysis module are each summed using preset weights,

[0041] When a stationary state is detected in which the change in acceleration vector of the above-mentioned inertial measurement unit remains below the noise level, a dominant stop trigger logic is applied to measure the duration of the said stationary state and grant additional points as the duration increases, and

[0042] It can be characterized by integrating dynamic drowning water with a lot of movement and static drowning water with no movement into a single probability indicator by inputting the final summed score into an S-shaped sigmoid activation function and converting it into a normalized probability value between 0 and 1. Effects of the invention

[0044] According to the present invention, the following effects can be expected.

[0045] By introducing a stroke inhibition factor when calculating dynamic movement disturbance, malfunctions of the flotation device that may occur during vigorous swimming activities such as butterfly strokes or turns are significantly reduced, thereby ensuring the user's activity and convenience.

[0046] By applying an inverse attenuation logic with the number of contact failures as the denominator when calculating the criticality of biological signals, sensor shaking or detachment situations unique to underwater environments are quantitatively distinguished from actual cardiac arrest situations, thereby improving the reliability of the alarm system.

[0047] By weighted summing dynamic confusion, vitality, and duration of arrest, and converting this into a probability using an S-shaped activation function, the system integrates and detects contrasting danger situations—flailing (dynamic) and loss of consciousness (static)—into a single system, thereby maximizing the life-saving rate.

[0048] By controlling the valve opening time inversely proportional to the square root of the cartridge pressure through physical compensation logic, airbag rupture accidents caused by overpressure in summer and buoyancy insufficient due to low pressure in winter are prevented, thereby ensuring the durability and safety of a product that can be used in all four seasons and all weather conditions.

[0049] The effects according to the present invention are not limited to those exemplified above, and a wider variety of effects are included within the present invention. Brief explanation of the drawing

[0051] Figure 1 illustrates an overall relationship diagram according to the present invention. Figure 2 illustrates the overall system operation flowchart according to the present invention. Figure 3 illustrates a flowchart for calculating a dynamic motion confusion index according to the present invention. Figure 4 illustrates a flowchart for calculating the biological signal urgency index according to the present invention. Figure 5 illustrates a flowchart for calculating the integrated risk of drowning according to the present invention. Figure 6 illustrates a flowchart for calculating the control time for opening a floating valve according to the present invention. Specific details for implementing the invention

[0052] Hereinafter, various embodiments are described in more detail with reference to the attached drawings. The embodiments described in this specification may be modified in various ways. Specific embodiments may be depicted in the drawings and described in detail in the detailed description. However, specific embodiments disclosed in the attached drawings are intended only to facilitate understanding of various embodiments. Accordingly, the technical concept is not limited by specific embodiments disclosed in the attached drawings, and it should be understood that it includes all equivalents or substitutions that fall within the spirit and scope of the invention.

[0053] Terms including ordinal numbers, such as first, second, etc., may be used to describe various components, but these components are not limited by the aforementioned terms. The aforementioned terms are used solely for the purpose of distinguishing one component from another.

[0054] The functions related to artificial intelligence according to the present disclosure are operated through a processor and memory. The processor may be composed of one or more processors. In this case, the one or more processors may be general-purpose processors such as CPUs, APs, and DSPs (Digital Signal Processors), graphics-dedicated processors such as GPUs and VPUs (Vision Processing Units), or artificial intelligence-dedicated processors such as NPUs. The one or more processors control the processing of input data according to predefined operation rules or artificial intelligence models stored in memory. Alternatively, if the one or more processors are artificial intelligence-dedicated processors, the artificial intelligence-dedicated processors may be designed with a hardware structure specialized for processing a specific artificial intelligence model.

[0055] The predefined rules of operation or artificial intelligence models are characterized by being created through learning. Here, being created through learning means that a basic artificial intelligence model is trained using a number of training data by a learning algorithm, thereby creating predefined rules of operation or artificial intelligence models configured to perform desired characteristics (or objectives). Such learning may be performed on the device itself where the artificial intelligence according to the present disclosure is executed, or it may be performed through a separate server and / or system. Examples of learning algorithms include supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning, but are not limited to the examples described above.

[0056] An artificial intelligence model can be composed of multiple neural network layers. Each of the multiple neural network layers has multiple nodes and weight values, and performs neural network operations through calculations between the results of previous layers and the multiple weights. The multiple weights possessed by the multiple neural network layers can be optimized based on the learning results of the artificial intelligence model. For example, multiple weights can be updated so that the loss value or cost value obtained by the artificial intelligence model during the learning process is reduced or minimized. Additionally, to minimize the loss value or cost value, multiple weights can be updated in a direction that minimizes the gradient associated with the loss value or cost value. Artificial neural networks may include deep neural networks (DNNs), such as Convolutional Neural Networks (CNNs), Deep Neural Networks (DNNs), Recurrent Neural Networks (RNNs), Restricted Boltzmann Machines (RBMs), Deep Belief Networks (DBNs), Bidirectional Recurrent Deep Neural Networks (BRDNNs), or Deep Q-Networks, but are not limited to the examples mentioned above.

[0057] A network is a network that serves as a transmission path for web pages; it may be a closed network such as a LAN (Local Area Network) or WAN (Wide Area Network), but it is desirable for it to be an open network such as the Internet. The Internet refers to a global open computer network structure that provides the TCP / IP protocol and various services existing at its upper layers, namely HTTP (HyperText Transfer Protocol), Telnet, FTP (File Transfer Protocol), DNS (Domain Name System), SMTP (Simple Mail Transfer Protocol), SNMP (Simple Network Management Protocol), NFS (Network File Service), and NIS (Network Information Service).

[0058] Terminals can be implemented in various forms. For example, the terminals described in this specification may include mobile terminals such as smartphones, tablet PCs, PDAs, portable multimedia players, and MP3 players, as well as fixed terminals such as smart TVs and desktop computers.

[0059] In this specification, terms such as “comprising” or “having” are intended to specify the existence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof. When a component is described as being “connected” or “connected” to another component, it should be understood that it may be directly connected to or connected to that other component, or that there may be other components in between. On the other hand, when a component is described as being “directly connected” or “directly connected” to another component, it should be understood that there are no other components in between.

[0060] Meanwhile, a "module" or "part" for a component as used in this specification performs at least one function or operation. Furthermore, a "module" or "part" may perform a function or operation by hardware, software, or a combination of hardware and software. Additionally, a plurality of "modules" or a plurality of "parts," excluding a "module" or "part" that must be performed on specific hardware or on at least one processor, may be integrated into at least one module. A singular expression includes a plural expression unless the context clearly indicates otherwise.

[0061] In addition, power, power transmission, and control therefor for the following assembly configurations and embodiments, including "by control," follow conventional technology including terminals, applications, hardware control modules, etc., so they are omitted to avoid redundancy.

[0062] In addition, the operation embodiments and configurations described in a general manner without being explained in detail below follow the prior art and are omitted in order to focus on describing the purpose of the present invention and the resulting effects.

[0063] Furthermore, in describing the present invention, if it is determined that a detailed description of related known functions or configurations may unnecessarily obscure the essence of the invention, such detailed description is abbreviated or omitted.

[0064] The drowning risk detection and automatic flotation system based on the fusion of biosignals and inertial sensor data according to the present invention includes a biosignal measurement unit (100), an inertial measurement unit (200), a data collection module (300), a situation analysis module (400), a risk judgment module (500), a flotation unit (600), a communication unit (700), and a control unit (800).

[0065] The above biosignal measurement unit (100) is configured to detect electrical and optical signals indicating a physiological state by contacting the surface of a user's body, and includes an electrocardiogram sensor that measures the electrical action potential of the heart and an optical heart rate sensor that optically detects changes in blood flow. The unit obtains mutually complementary data by simultaneously operating the electrocardiogram sensor and the optical heart rate sensor to compensate for the possibility that the reliability of a single sensor may be reduced due to contact with water or detachment of the electrode in an underwater environment. If a contact failure condition is detected where the contact resistance of the electrocardiogram electrode exceeds a threshold, the unit increases the data weight of the optical heart rate sensor or generates a reliability reduction indicator and transmits the data to the situation analysis module (400).

[0066] The above-mentioned inertial measurement unit (200) is a sensor assembly for measuring the user's movement, posture, and exercise intensity, and is implemented as an inertial measurement device including a 3-axis accelerometer and a 3-axis gyroscope. The unit generates basic data that can determine whether the user is performing a regular swimming stroke, is in a stationary state, or is flailing irregularly in the water, and maintains the precision of the data by performing a zero-point reset function in the background to correct the drift error of the sensor if the stationary state persists for longer than a preset time.

[0067] The data collection module (300) is configured to convert analog signals output from the biosignal measurement unit (100) and the inertial measurement unit (200) into digital data and perform preprocessing. This module performs time synchronization of biosignal data and motion data having different sampling periods, applies a band-pass filter of a specific band to remove underwater motion-inducing high-frequency noise and baseline fluctuations included in the collected raw data, and packetizes the data into units of a sliding window of a preset length for real-time analysis and transmits it to an upper module.

[0068] The above-mentioned situation analysis module (400) is a computational module that infers the user's current activity status based on refined data and calculates a dynamic movement confusion index and a biosignal urgency index. This module performs a calculation that increases the confusion index when rapid acceleration and direction change are high simultaneously, but corrects the value by reflecting regular strokes as suppression factors when they are detected. In the case of biosignals, it prevents false detection by applying an inversely proportional calculation that exponentially increases the urgency when the abnormal waveform ratio is high, but lowers the reliability when the number of electrode contact failures is high.

[0069] The above-mentioned risk judgment module (500) calculates the integrated risk of drowning based on the analysis results of the above-mentioned situation analysis module (400) and determines whether a final risk situation will occur. The module performs a calculation to forcibly increase the risk probability by weighting the confusion and urgency levels, and if a stationary state without movement exceeds a threshold time, it considers this as a dominant trigger. The calculated sum score is converted into a probability value between 0 and 1 through an S-shaped activation function, and if the probability exceeds a preset trigger threshold, a command to operate a flotation device is generated.

[0070] The above-described buoyancy unit (600) is a device that physically inflates to provide buoyancy to a user when it receives a trigger signal from the above-described risk judgment module (500), and includes a cartridge in which compressed gas is stored, an electronic valve that controls the discharge of gas, and an inflation tube. The control unit of the above-described unit measures the internal pressure of the cartridge and drives the valve through a variable control logic that shortens the valve opening time inversely proportional to the square root as the pressure increases and extends the time by providing a safety margin as the risk level increases, thereby preventing airbag damage in any environment while simultaneously achieving reliable buoyancy.

[0071] The communication unit (700) is an interface that transmits and receives data with an external device or a control server, and generates an emergency rescue signal simultaneously with the danger judgment module (500) detecting a dangerous situation. Considering the limitations of underwater communication, the unit attempts ultrasonic communication or logs data in internal memory while submerged, and immediately after the buoyancy unit (600) detects that the user has surfaced, it activates a satellite positioning system and long-distance wireless communication to transmit the current location coordinates.

[0072] The control unit (800) is a central processing unit that controls the overall operation of each unit and module described above and distributes power. Normally, this unit operates in a low-power mode, but performs variable power control by switching the sensor data collection cycle to high speed when the level of confusion calculated by the situation analysis module (400) rises above the warning level, and periodically self-diagnoses the remaining battery level and system errors to provide feedback to the user.

[0073] Each of the aforementioned hardware components does not merely collect data and execute commands, but performs the four-stage organic computation and control process described below to prevent malfunctions and maximize detection accuracy. The control logic of the present invention is designed by mathematically modeling the physical characteristics of sensor data and the physiological characteristics of the drowning situation.

[0074] First, in the process of calculating the dynamic motion disturbance index, abrupt acceleration and directional changes are extracted from inertial sensor data to increase the risk level; however, if regular swimming movements (stroke) are identified, they are reflected as suppression factors to correct the value, thereby clearly distinguishing between high-intensity swimming exercises and actual drowning panic situations.

[0075] Next, in the process of calculating the vital sign severity index, exponential weighting is applied to the heart rate deviation rate to increase the sensitivity of initial risk detection, while the number of frequent electrode contact failures in an underwater environment is reflected as a reliability reduction factor to attenuate the score, thereby significantly reducing the occurrence of false positive alarms caused by sensor detachment or shaking.

[0076] Next, in the process of calculating the integrated drowning risk probability, the above-calculated confusion and urgency are integrated into a single probability value through an S-shaped activation function, but if a motionless stationary state exceeds a threshold time, this is considered a dominant trigger and the risk probability is forcibly increased, thereby identifying not only flailing (dynamic drowning) but also loss of consciousness (static drowning) situations without blind spots.

[0077] Finally, in the process of calculating the buoyancy valve opening control time, when deploying the airbag in response to a confirmed hazardous situation, the valve opening time is controlled inversely proportional to the internal pressure of the gas cartridge based on Bernoulli's principle, and a safety margin is dynamically allocated according to the level of risk; this prevents airbag failure and provides quantitative buoyancy regardless of external temperature or pressure changes.

[0078] The specific operation logic and operation methods of the four core processes mentioned above will be explained in detail below.

[0079] The above-mentioned situation analysis module (400) calculates a dynamic movement disorder index, which is a key indicator for quantitatively distinguishing whether the user's underwater movement is a regular swimming stroke or an irregular flailing for survival, based on data obtained from the inertial measurement unit (200). The dynamic movement disorder index defined in the present invention is set with the purpose of suppressing the risk by maximally reflecting the irregularity of movement and the disorder of direction changes, while considering it a normal state when regular periodicity is detected, in order to solve the problem where high acceleration values ​​occur even during high-intensity exercises such as butterfly stroke or freestyle sprinting when only the magnitude of acceleration is measured, which could be mistaken for a dangerous situation.

[0080] The above situation analysis module (400) determines the level of confusion through a logical chain process based on data obtained from the inertia measurement unit (200). Since the panic state immediately before drowning is characterized by the simultaneous occurrence of acceleration, which signifies a strong force, and uncontrolled changes in direction, the number of rapid acceleration events per unit time and the number of changes in direction of the acceleration axis are set to have a positive correlation with each other, so that when both elements are high simultaneously, the level of confusion increases exponentially. On the other hand, since swimming exhibits a repetitive stroke pattern according to a specific swimming stroke, the number of strokes identified through pattern matching is defined as a suppression factor that offsets the above-mentioned synergistic effect, and a correction operation is performed so that the calculated level of confusion converges to 0 in inverse proportion as the number of strokes increases. In addition, since motion data can explode instantaneously due to external impact, to prevent the data scale from becoming excessively large and causing calculation errors or distorting judgments, a logarithmic damping logic is applied so that the rate of increase of the result value gradually slows down as the total amount of input values ​​increases, thereby ensuring the stability of the data.

[0081] The system of the present invention acquires physical variables through a sliding window in 1-second intervals, and the number of rapid acceleration events is defined as the number of occurrences of peaks where the vector sum of the 3-axis accelerometer exceeds a preset threshold, and the number of direction changes is defined as the number of times the data of each axis of the gyroscope or accelerometer crosses the zero point to indicate how vigorously the user is swinging their arms. In addition, the number of strokes is defined as the number of movements classified as normal swimming strokes when a periodic sinusoidal acceleration pattern is detected, and data is collected.

[0082] The above situation analysis module (400) utilizes the calculated dynamic movement disturbance as a primary physical risk index to be combined with the biosignal urgency described later, rather than as a trigger for the operation of the final flotation device, and if the calculated disturbance is below the first threshold, it is determined to be in a normal swimming or resting state and maintains the biosignal analysis cycle in normal mode. On the other hand, if the calculated disturbance is above the first threshold, it is classified as a caution stage and the sampling rate of the biosignal sensor is doubled, and if it is above the second threshold, it is classified as a danger stage and immediately transmits data to the danger judgment module (500) and generates a flotation preparation signal.

[0083] The above situation analysis module (400) performs sequential data processing steps every second. First, it collects acceleration and gyroscope data from an inertial sensor and performs a preprocessing step of applying a band-pass filter of a specific band to remove high-frequency noise. Afterward, it extracts features by counting the number of sudden acceleration events, the number of direction changes, and the number of strokes that occurred during one second from the filtered data. Next, it generates a raw risk level by multiplying the number of sudden acceleration events and the number of direction changes; for example, if there are 3 sudden accelerations and 4 direction changes, the raw risk level is calculated as 12. In the next step, it performs a normal motion correction operation by dividing the calculated raw risk level by the value obtained by adding the number of strokes to 1. If a stroke is detected once because it is a butterfly stroke, the raw risk level of 12 is reduced to 6, whereas if there is a drowning situation and there are 0 strokes, it is processed so that 12 remains unchanged. Finally, it performs a natural logarithm operation on the corrected value to determine the final confusion index.

[0084] The above situation analysis module (400) filters out malfunctions by designing it so that in the case of a single impact occurring during diving entry or a turning motion of kicking a wall, the number of rapid acceleration events is high but the number of direction changes is low or the duration is short, thereby filtering out malfunctions. In addition, to prevent bias drift of the sensor, if a stationary state with a direction change count close to 0 persists for more than a preset time, a logic to recalibrate the sensor's zero point is performed in the background.

[0085] When comparing this logic with prior art, prior art using a simple threshold comparison method has a problem in that it is falsely detected as dangerous because acceleration exceeds the threshold during butterfly sprinting, and it is also falsely detected as dangerous because impact exceeds the threshold during diving entry. However, the present invention determines it as normal when acceleration is high but a stroke is detected, by activating a suppression factor, and determines it as normal when diving entry, because acceleration is high but the change of direction is small and mutual synergy is insufficient, and determines it as high risk only when both acceleration and change of direction are high and there is no stroke.

[0086] Looking at a specific simulation example, in a stationary or resting state, when the number of sudden acceleration events is 0, the number of direction changes is 1, and the number of strokes is 0, the input value is negligible, so the confusion level is calculated as 0 and it is judged as safe. In a high-speed swimming state such as butterfly, the number of sudden acceleration events is very high at 5 and the number of direction changes is 6, but since the number of strokes is detected as 2, a strong suppression action occurs by dividing the value of 30 by 3, correcting it to 10, and after final log transformation, it is judged as a caution stage. On the other hand, in a drowning panic state, the mutual synergy is maximized with the number of sudden acceleration events at 6 and the number of direction changes at 12, resulting in a value of 72; at this time, since the number of strokes is 0, the suppression factor does not operate, so 72 is reflected as is, and as a result of the final log transformation, it is judged as a danger stage.

[0087] Accordingly, the above-mentioned situation analysis module (400) determines that it is safe when both the number of sudden acceleration events and the number of direction changes are low, and caution when only one of them is high. If both values ​​are high and a stroke is present, it is monitored as safe or caution, considering it as high-intensity exercise. Furthermore, a rule is applied to make a final judgment of danger by considering it a drowning panic state only when both values ​​are high and a stroke is absent.

[0088] These judgment rules possess objectivity, as the variables used in this process are numerical values ​​that can be objectively measured through physical sensors such as accelerometers and gyroscopes, leaving no room for the operator's subjectivity to intervene. Furthermore, they are an inevitable configuration based on the kinematic principles that, physically, swimming inevitably involves regular stroke patterns for energy efficiency, while drowning involves inefficient and disorderly flailing due to survival instincts.

[0089] Consequently, the present invention provides the effect of significantly reducing the false detection problem of mistaking vigorous swimming strokes, such as butterfly stroke, for drowning, which is inherent in conventional simple acceleration threshold methods, by introducing a stroke suppression factor. In addition, by stabilizing the range of data fluctuation through log scaling, the system does not overreact to noise generated during various underwater leisure activities, such as splashing, diving, and turning movements, thereby achieving the technical effect of maximizing usability.

[0090] In this specification, the number of rapid acceleration events refers to a phenomenon in which the magnitude of the 3-axis composite acceleration vector is greater than a preset multiple compared to normal conditions and persists for a certain period of time or longer; the number of direction changes refers to zero crossing points where the angular velocity value of the gyroscope changes from positive to negative or from negative to positive; and the number of strokes refers to periodic operation in which a dominant frequency component is detected in a specific band as a result of frequency analysis of the 3-axis acceleration data.

[0091] The present invention applies the following [Logical Formula 1] to calculate the dynamic movement confusion index (I_chaos). This is based on the Weber-Fechner law and aims to prevent data divergence by slowing down the increase in sensation (risk) on a logarithmic scale as the intensity of the stimulus (movement) increases.

[0092] [Logical Formula 1]

[0093]

[0094] N_jerk is the number of sudden acceleration events greater than 2.5 times the normal gravitational acceleration (1g), N_zc is the number of zero crossings (direction changes) of the gyroscope, N_stroke is the number of regular swimming strokes identified through frequency analysis, β is the inhibition weighting constant for skilled swimming strokes (set to 2.0 in this simulation), and k_1 is the sensor sensitivity correction constant (set to 1.0 in this simulation).

[0095] The results of simulating various underwater activity scenarios 10 times to derive an appropriate 'risk judgment threshold' are as follows.

[0096] [Table 1]

[0097]

[0098] Analyzing the table above, it can be seen that for regular swimming (sessions 2–5), the level of confusion remains below 2.5, but for drowning situations (sessions 8–10), the value increases rapidly to 3.5 or higher due to the absence of inhibitory factors (N_stroke) and mutual synergy (N_jerk x N_zc).

[0099] Therefore, the present invention has demonstrated that setting the 'caution level threshold = 2.0' and the 'danger level (trigger) threshold = 3.5' is statistically significant.

[0100] The verification simulation below shows the difference in performance compared to the prior art (simple acceleration threshold 3.0g) when the set threshold (3.5) is applied.

[0101] [Table 2]

[0102]

[0103] The above-mentioned situation analysis module (400) determines whether the user's heart rate and electrocardiogram waveform are outside the physiologically normal range based on data collected from the biosignal measurement unit (100), and calculates a biosignal urgency index, which is a quantitative indicator for distinguishing between noise caused by poor electrode contact unique to the underwater environment and actual cardiac abnormalities. The biosignal urgency index defined in the present invention is set for the purpose of dual filtering to maximize sensitivity for the heart rate deviation rate while forcibly reducing reliability when sensor contact instability is detected, in order to solve the problem of causing false positives by maximizing sensitivity for the heart rate deviation rate, in order to resolve the irregular signal that occurs frequently when the sensor is temporarily lifted from the skin due to water currents or vigorous movements underwater, which is similar to arrhythmia.

[0104] The above-mentioned situation analysis module (400) analyzes data collected from the biosignal measurement unit (100) and performs a logical weighting and attenuation procedure. Since biosignals, unlike motion signals, can cause life-threatening fatal consequences even with slight deviations, it applies an exponential urgency weighting logic that increases the risk score exponentially rather than linearly as the proportion of abnormal waveforms outside the normal heart rate range increases, thereby immediately triggering an initial emergency situation. Additionally, since the number of times electrode contact failures are detected is proportional to the probability that the data is noise, it performs an operation to physically suppress misjudgment caused by sensor malfunctions by applying an inverse attenuation logic in which the number of contact failures is set as the denominator and the total urgency score is divided and reduced as the number of failures increases.

[0105] The system of the present invention acquires biological variables through an analysis window in 5-second intervals, and the number of samples deviating from the normal heart rate range is defined as the number of samples in which the measured heart rate interval deviates from a preset ratio relative to the user's usual resting heart rate, and the total number of valid waveforms is defined as the total number of valid heartbeats identified as not being noise within the window, and is used as the denominator value for calculating the deviation ratio. In addition, data is collected by defining the number of electrode contact failures as the number of times the sensor is considered to have detached from the skin because the hardware-detected lead-off signal or impedance value exceeds a preset threshold.

[0106] The above-mentioned situation analysis module (400) uses the calculated vital sign severity level alone or combines it with the previously calculated dynamic movement disorder level to make a final judgment. If the severity level is below the first threshold, it is determined to be in a normal state. On the other hand, if the severity level is above the second threshold and the movement disorder level is high, it is determined to be in a panic state due to drowning. If the severity level is above the second threshold and there is no movement, it is determined to be loss of consciousness or cardiac arrest, and a command is transmitted to immediately activate the flotation trigger.

[0107] The above-mentioned situation analysis module (400) performs sequential processing steps by analyzing biometric data in real time. First, it performs a quality inspection by measuring the impedance of signals input from electrocardiogram and optical heart rate sensors to primarily count whether there is a contact failure. Subsequently, it detects valid waveforms within valid signals and determines whether each beat interval is outside the normal range, classifying them into abnormal samples and normal samples. Next, it calculates the heart rate deviation rate by dividing the number of abnormal samples by the total number of valid waveforms; for example, if 3 out of a total of 10 beats were abnormal, the rate is calculated as 0.3. In the next step, it performs an operation to explosively increase the raw critical score by applying the natural constant to the calculated rate or by raising a preset weight to a power. Finally, it performs a reliability-based correction operation by dividing the raw critical score by the sum of the value obtained by multiplying 1 by the number of electrode contact failures and the sensitivity coefficient, so that if it is an actual heart attack, the score is maintained, but if the sensor is loose, the score is drastically reduced.

[0108] The above situation analysis module (400) performs dual sensor cross-verification in which the system automatically switches the main data to optical heart rate sensor data for analysis when the contact failure of the electrocardiogram electrode is excessively high. Additionally, if the contact failure condition persists for the entire window time, it is assumed that the user has removed or damaged the device, so the urgency calculation is stopped and a warning to check the wearing status is sent.

[0109] When comparing this logic with prior art, prior art using simple heart rate thresholds suffers from problems such as misidentifying shaking noise during vigorous swimming as high heart rate, leading to false alarms of danger, and misidentifying signal loss when the sensor is not worn as cardiac arrest, also leading to false alarms of danger. However, the present invention determines the level as safe or caution by attenuating the score if the number of contact failures is high, even if the abnormal waveform ratio is high; it prevents malfunctions by stopping analysis or sending a wearing warning if the duration of contact failures exceeds a threshold, and determines the level as high risk only when the abnormal waveform ratio is high and contact failures are low.

[0110] Looking at a specific simulation example, in a situation where the sensor is loose, the abnormality rate appears high at 0.6, but as the number of contact failures is detected as 8, a strong damping effect occurs by dividing the high raw score by 9, resulting in a final urgency of 0.8 and a determination of safety. On the other hand, in a situation immediately preceding cardiac arrest, the abnormality rate is very high at 0.8 and the number of contact failures is low at 1, so almost no damping occurs; thus, exponential weighting is reflected as is, resulting in a final urgency of 4.8 and a determination of danger. Additionally, in a complete asystole state, there is no signal, so the abnormality rate is 1.0, and since contact is good, the maximum value is maintained, resulting in a determination of danger.

[0111] Accordingly, the above-mentioned situation analysis module (400) determines that it is safe when both the heart rate deviation rate and the number of contact failures are low, and that it is safe when the heart rate deviation rate is low but the number of contact failures is high, considering it to be a simple contact failure. Furthermore, if both values ​​are high, it monitors it as safe or caution, considering it to be vigorous exercise or sensor detachment, and applies a rule to finally determine it as dangerous only when the heart rate deviation rate is high and the number of contact failures is low, considering it to be an actual heart abnormality.

[0112] This judgment rule is an inevitable configuration because it has objectivity in that the heart rate deviation rate is a statistical figure based on the user's usual data and the number of contact failures is based on the hardware's impedance value, thus having a physically clear standard, and it is based on the physiological principle that the heart is a core organ for maintaining life and requires a rapid warning of abnormal signs, and the electrical engineering principle that electrical noise inevitably occurs in water, a conductive medium.

[0113] Consequently, the present invention provides the effect of ensuring the reliability of the alarm by preventing safety complacency and the boy-in-the-worship effect, which arise from mistaking a high or fluctuating heart rate for danger, through contact failure filtering. Furthermore, since contact failure rarely occurs in actual emergency situations such as cardiac arrest, the exponential weighting logic operates to activate the flotation device as quickly as possible, thereby achieving the technical effect of securing the golden time.

[0114] In this specification, the heart rate deviation rate is defined as the ratio of beats that exceed a certain level or the normal range when converted to heart rates per minute among the total number of valid waveforms within the analysis window, the rate of variation of adjacent waveform intervals is defined as the ratio of beats that exceed the normal range, the contact failure is defined as a state in which the contact resistance of an electrode remains above a preset threshold for a certain period of time or the signal-to-noise ratio falls below a reference value, and the exponential weighting is defined as a data processing method that applies a transfer function in the form of an exponential power of the natural constant to the input value to convert it so that the output value increases with acceleration as the input value increases.

[0115] The vital signal urgency index ($I_{vital}$) calculates the deviation rate based on the target heart rate range according to the Karvonen formula and applies the following [Logical Equation 2]. This is intended to exponentially warn of even minor vital signal abnormalities, which are life-threatening, while quantitatively suppressing the effects of underwater noise.

[0116] [Logical Formula 2]

[0117]

[0118] Here, R_dev is the ratio of beats outside the normal heart rate range (e.g., 60–180 bpm) (0.0–1.0), C_loss is the number of lead-off detections, λ is the risk sensitivity constant (set to 2.5 in this simulation - sensitive to small deviations), α is the noise suppression coefficient (set to 0.5 in this simulation), and A is the default scale constant (1.0).

[0119] [Table 3]

[0120]

[0121] As a result of the simulation, simple sensor noise (cycles 3–5) is suppressed to a criticality level of 1.5 or less due to the increase in the denominator term, even if the detachment rate is high. On the other hand, in actual critical situations (cycles 8–10), the criticality level increases rapidly to 5.0 or higher due to the low number of contact failures.

[0122] Therefore, it is reasonable for the present invention to set the 'caution level threshold = 2.0' and the 'danger level (trigger) threshold = 4.5'.

[0123] [Table 4]

[0124]

[0125] The above-mentioned risk judgment module (500) converts the dynamic movement disturbance and the biological signal urgency, which are data of different dimensions, into probability values ​​that serve as a single decision-making criterion, and calculates an integrated drowning risk probability, which is an indicator for finally determining whether to operate the flotation device. The integrated drowning risk probability defined in the present invention is designed to encompass two opposing risk scenarios—dynamic drowning represented by flailing and static drowning represented by silent sinking—in order to solve the problem that using an arithmetic mean method of simply adding two indices may result in errors when one side of the data is omitted or when the situation is extreme. To achieve this, a stationary state where no movement is detected is introduced as a dominant trigger variable, thereby aiming to eliminate blind spots in emergency situations.

[0126] The above-mentioned risk judgment module (500) performs a probabilistic integration procedure based on input variables. It applies a dominant stop trigger logic based on the observation that while general emergency situations have a high degree of confusion, drowning caused by fainting has a movement that converges to 0. This involves the system monitoring the duration of a stop state, during which no movement is detected by an inertial sensor, as an independent risk factor. If this time exceeds a preset threshold, it performs an addition logic that forcibly increases the final risk probability, even if the preceding dynamic confusion value is low. Additionally, since the sum of the confusion, emergency, and stop time sums can theoretically have an infinite range, to convert this into a clear trigger signal, it applies an S-shaped activation function logic that causes the result value to rapidly converge to 1 when the sum of the input values ​​exceeds a specific threshold and to converge to 0 when it falls short, thereby excluding ambiguous intermediate values ​​and processing to make a definite decision.

[0127] The system of the present invention acquires an integrated variable that is updated at a 1-second interval, and processes data by using the previously calculated dynamic motion disturbance, biological signal urgency, and the duration of suspension during which the change in acceleration vector is below the noise level as input variables, and assigning preset weights to the dynamic disturbance, biological urgency, and excess duration of suspension, respectively, and summing them up.

[0128] The above-mentioned risk judgment module (500) performs a step-by-step response using the characteristic that the calculated integrated drowning risk probability has a value between 0 and 1. If the probability is below a first threshold, it is classified as a safe state and continuously monitored. On the other hand, if the probability is above the first threshold but below the second threshold, it is classified as a caution state, a haptic feedback warning is sent, and a preliminary notification is sent to the guardian. If the probability is above the second threshold, it is classified as a danger state, and a command is sent to immediately deploy a flotation device and automatically transmit satellite positioning system coordinates and a rescue signal.

[0129] The above-mentioned risk judgment module (500) performs a sequential processing step for every frame. First, it receives the level of confusion and the level of urgency from the situation analysis module (400) and determines whether the current user is in a stopped state. If the user is in a stopped state for more than a certain amount of time, it generates a stop risk score proportional to the time. Subsequently, it calculates a weighted sum by multiplying the level of confusion, the level of urgency, and the stop risk score by their respective weights and adding them together to calculate an integrated risk score. In the next step, it undergoes a probability conversion process in which the calculated integrated risk score is applied to an S-shaped function logic. For example, if the integrated score is 0, the probability becomes 0.5, and if the score exceeds a positive threshold, the probability converges to 1. However, to ensure safety bias, the reference point is adjusted so that if the score is low, the probability is set to be close to 0. Finally, the moment the calculated probability exceeds the trigger threshold, it generates a solenoid valve opening command.

[0130] The above-mentioned risk judgment module (500) performs voluntary apnea training exception handling by balancing the final probability so that it does not exceed the trigger threshold even if the cessation risk score increases, taking into account that when a professional diver or swimmer performs breath-holding training, the cessation state is prolonged but the biological criticality remains stable, and the biological criticality is low. In addition, since resting while floating on the water may be recognized as a cessation state, it performs a surface misidentification prevention logic that detects surface delegation through a barometric pressure sensor and an inertial measurement unit (200) to block the addition of the cessation risk score.

[0131] When comparing this logic with prior art, the prior art, which relies on a single sensor, has a critical flaw in that while it can detect dynamic drowning using an accelerometer, it misidentifies static drowning as safe; it also has the limitation of determining safety during breath-holding training simply because there is no movement. However, the present invention eliminates blind spots through an integrated probability logic that activates when the confusion index is high, detects a static drowning situation if the stop time is exceeded and a biological abnormality is detected despite the absence of movement, and does not activate if the stop time is exceeded but the biological signals are normal, considering it a training situation.

[0132] Looking at a specific simulation example, in a situation where a 1-minute breath-holding training is being conducted, the stopping time is 60 seconds, which creates a risk factor; however, since the biological urgency is stable at 0.3, the weights are offset, resulting in a final probability of 0.4, which is determined as safe or caution. On the other hand, in a dynamic drowning situation, the confusion and urgency levels rise together, causing the sum score to surge, and the probability is calculated as 0.95, so the device operates. Additionally, in a static drowning situation, there is no movement but the urgency is very high, or the urgency is at an intermediate level but the stopping time exceeds a threshold, so the addition logic operates, and the probabilities are calculated as 0.92 and 0.88, respectively, and the flotation device operates.

[0133] Accordingly, the above-mentioned risk judgment module (500) determines it as safe if both dynamic confusion and biological criticality are low and it is not in a stationary state, and as safe if the confusion is high but the criticality is low, considering it as high-intensity exercise. Additionally, if there is no movement and the stationary time is long but the biological criticality is low, it determines it as safe or caution if it is considered breath-holding training. On the other hand, it applies a rule that determines it as dangerous if there is no movement but the criticality increases rapidly within a short period of time, as acute cardiac arrest if there is no movement and the stationary time is long and the criticality rises, as loss of consciousness if there is no movement and the criticality is long, and as flailing if both confusion and criticality are high.

[0134] These judgment rules are a decision-making method verified in the fields of artificial intelligence and control engineering through probabilistic transformation via an S-shaped activation function, and possess objectivity in that the output value for the input value is mathematically fixed and reproducible. Since drowning accidents occur in a mixture of noisy and quiet situations, a variable weight integration logic that varies the main variable depending on the presence or absence of movement is a technically necessary configuration to detect these two contradictory situations.

[0135] Consequently, the present invention provides an all-around detection effect that dramatically increases the probability of life rescue by detecting silent drowning water, which conventional technology could not detect, through the combination of a stationary state trigger and a bio-signal. In addition, by clearly and probabilistically separating the training of professional swimmers or situations of intense exercise from dangerous situations, it achieves a technical effect that eliminates economic losses and psychological anxiety caused by unnecessary airbag deployment and minimizes malfunctions.

[0136] In this specification, the integrated drowning risk probability refers to a numerical value between 0 and 1 representing the probability that the current user is in a drowning state, calculated by weighting and summing heterogeneous sensor data and undergoing a normalization process; the stationary state refers to a state where the change in the acceleration vector of the inertial measurement unit (200) is less than the noise level and the change in the gyroscope is minute; the dominant trigger refers to a high-priority judgment logic that forcibly pulls the final risk probability to a risk level regardless of the value of other variables when a specific condition is satisfied; and normalization refers to a process of ensuring consistency in judgment by mapping a sum score, which can have an infinite range, to a probability value between 0 and 1 through an S-shaped function logic.

[0137] The integrated risk probability (P_risk) applies the following [Logical Equation 3], which is an application of the activation function of an artificial intelligence perceptron, to convert variables of different units into probabilities.

[0138] [Logical Formula 3]

[0139] [Logical Formula 3]

[0140]

[0141] T_score is the score for the duration of the stationary state (seconds) (add 1.0 for every second exceeding 10 seconds), w_1, w_2, and w_3 are the weights of each factor (in this simulation, w_1=0.5, w_2=0.8, and w_3=1.0 are set - Bias is the threshold bias value for the integrated score input to the sigmoid function, and in this simulation, Bias is set to 5.0 so that the sigmoid input value exceeds 0 only when a sufficient risk score is accumulated, and the probability exceeds 0.5.

[0142] [Table 5]

[0143]

[0144] Simulation results show that parameter tuning is possible to increase w_1 or lower the Bias to 4.0 to increase the dynamic drowning detection rate.

[0145] Due to the characteristics of the sigmoid function, the output probability becomes 0.5 when the input value z is 0. In this simulation, z is defined as z = w_1·I_{chaos} + w_2·I_{vital} + w_3·T_{score} - Bias, and by setting Bias to 5.0, the design ensures that z remains below 0 unless a definite risk factor accumulates, thereby suppressing the output probability to less than 0.5.

[0146] Data analysis confirmed that high-risk groups can be identified without malfunction when set to 'operation trigger probability = 0.80'.

[0147] [Table 6]

[0148]

[0149] The control unit of the above-mentioned flotation unit (600) calculates a flotation valve opening control time, which is the duration for which power is supplied to the solenoid valve to physically open and maintain the gas passage. If the valve is opened for a fixed time as in the conventional method, there is a risk that the airbag will be damaged due to excessive gas being injected as a result of the rise in internal pressure of the cartridge in high-temperature summer environments, and conversely, there is a risk of incomplete flotation due to insufficient pressure in winter or in the event of a gas leak. Therefore, the flotation valve opening control time defined in the present invention is set with the purpose of physically compensating for changes in flow rate according to cartridge pressure and dynamically allocating a safety margin according to the level of urgency so that the airbag can be inflated to a target volume without damage in any environment.

[0150] The control unit of the above-mentioned flotation unit (600) performs fluid dynamic and situational control procedures based on input sensor data. Based on the principle of Bernoulli's theorem and fluid dynamics, which states that the flow velocity of gas passing through a nozzle is proportional to the square root of the pressure difference, and that the higher the pressure, the faster the gas is ejected, pressure compensation inverse proportional control is performed. This maintains a constant total amount of gas injected by applying a logic that shortens the valve opening time inversely proportional to the square root of the pressure as the measured cartridge pressure is higher than the reference value. In addition, risk-linked acceleration control is performed, considering that a high integrated drowning risk probability calculated earlier implies a life-threatening condition, and that certain and rapid flotation is the top priority even if it sacrifices some of the airbag's durability. This reflects a safety-first design intended to prevent the possibility of drowning due to incomplete inflation by applying a preset increase factor to the calculated opening time to inject gas for a longer period as the risk probability approaches 100%.

[0151] The system of the present invention acquires variables in milliseconds immediately before flotation and utilizes the cartridge current pressure, which is the internal pressure of the cartridge measured by a pressure sensor; the reference design pressure, which is the standard atmospheric pressure used as a reference for airbag design; the integrated probability of drowning; and the target injection time constant, which is the theoretical time required to fill the target volume at the reference pressure, as input data. The calculated flotation valve opening control time is used as the timer interrupt setting value of the microcontroller, and when the control time is calculated, a driving signal is applied to the solenoid valve driving circuit, and when the timer count reaches the setting value, the signal is immediately cut off to control the valve to close.

[0152] The control unit of the above-mentioned flotation unit (600) performs real-time calculations as soon as a trigger signal is received. First, it measures the cartridge pressure and, if the pressure is below the minimum operating pressure, performs a status check to set the valve opening time to a maximum limit and simultaneously store a gas shortage error code. Subsequently, it calculates the physical base time by taking the square root of the value obtained by dividing the reference pressure by the current pressure and multiplying it by the target injection time constant. For example, if the current pressure is twice as high as the reference, the flow rate increases, so the time is calculated to be shortened. In the next step, a safety margin coefficient is generated by applying a weight in the form of a logarithmic function to the risk probability value, and the coefficient increases as the risk level increases. Finally, it determines the final opening time by multiplying the physical base time by the safety margin coefficient and performs the procedure of driving the valve.

[0153] The control unit of the above-mentioned buoyancy unit (600) performs a temperature compensation function by using a temperature sensor as an alternative in the case of a model without a pressure sensor, estimating the internal pressure through the temperature according to Boyle-Charles' law, and applying the same inverse logic. In addition, if no expansion shock is detected by the inertia measurement unit (200) while the valve is open, it performs a blockage detection logic that automatically switches to a shock opening mode in which the valve is driven by cutting off the gas outlet in a pulse form, considering it a blockage of the gas outlet.

[0154] When comparing this logic with prior art, the prior art using a fixed time has limitations: there is a risk of airbag rupture due to over-injection caused by high flow rates during high temperatures in summer, and a risk of incomplete buoyancy due to insufficient gas caused by slow flow rates during low temperatures in winter, and it operates in the same way as in normal situations even in ultra-high-risk situations. However, the present invention induces quantitative injection by shortening the time inversely proportional to the square root of the pressure, secures buoyancy by extending the time in low-temperature environments, and guarantees reliable buoyancy by further extending the time with an added safety margin in ultra-high-risk situations.

[0155] Looking at a specific simulation example, under standard conditions, both physical compensation and risk compensation are applied at a factor of 1, allowing it to open for a standard time of 2,000 milliseconds. In a high-temperature environment, if the pressure rises to 90 bar, the time is shortened by a factor of 0.82 due to physical compensation, calculated as 1,640 milliseconds, thereby preventing rupture. Conversely, in a low-temperature environment, if the pressure drops to 40 bar, the time is extended by a factor of 1.22, calculated as 2,440 milliseconds, thereby ensuring buoyancy. Additionally, in a severe emergency situation where the probability of danger reaches 0.99, a safety margin factor of 1.20 is applied, extending the time to 2,400 milliseconds and allowing overcharging for survival.

[0156] Accordingly, the control unit of the above-mentioned flotation unit (600) shortens the time to prevent overpressure when the cartridge pressure is high, extends the time to prevent insufficient buoyancy when the pressure is low, and extends the time to maximize survival rate when the probability of danger is high. In addition, when the pressure is high and the probability of danger is also high, the shortening factor and the extension factor mutually offset each other to balance it, and when the pressure is lower than a threshold, a rule is applied to perform error processing.

[0157] Since the relationship between pressure and flow velocity follows the established physical law known as Bernoulli's theorem, the control logic possesses objectivity as it is the result of mathematical modeling of physical phenomena. Furthermore, because failure to adjust the injection amount according to environmental variables leads to fatal defects such as the tube bursting or insufficient buoyancy, variable control logic is a technically essential component for the safety and reliability of the product.

[0158] Consequently, the present invention provides the effect of enabling all-weather use in all four seasons by ensuring operational consistency of the product through constant buoyancy, regardless of whether the external temperature is high in summer or low in winter. In addition, it achieves the technical effect of extending the durability life of the equipment even during repeated use by minimizing stress applied to the airbag joints through the prevention of unnecessary overpressure injection.

[0159] In this specification, the flotation valve opening control time refers to the duration of an electrical signal during which the solenoid valve must be open to inflate the airbag to a target volume; pressure compensation refers to a control technique that inversely adjusts the opening time to offset fluctuations in flow rate caused by changes in internal pressure of the cartridge; safety margin factor refers to a weighting factor that is multiplied to inject additional gas beyond the theoretically calculated value in preparation for unexpected variables in emergency situations; and impact opening mode refers to an emergency driving method that attempts to clear the blockage by physical impact by repeatedly applying a short, strong pulse signal instead of continuous opening when the gas outlet is blocked.

[0160] The valve opening time (T_open) is determined by the following [Logic Equation 4], which applies the flow velocity-pressure relationship according to Bernoulli's theorem and Boyle-Charles's law. This is a control based on physical laws.

[0161] [Logical Formula 4]

[0162]

[0163] P_std is the reference pressure (60 bar), P_curr is the current cartridge pressure measured by the sensor, T_base is the target injection time at the reference pressure (2000 ms), γ is the safety margin factor (set to 0.5 in this simulation), and P_risk is the integrated risk probability (0.8 ~ 1.0).

[0164] [Table 7]

[0165]

[0166] Simulation results show that if a fixed time (2000ms) is used, excessive gas may be injected due to increased flow velocity at high temperatures (90 bar); however, this logic maintains a constant total injection amount by applying a pressure compensation factor to adjust the time to 1640ms (approximately 0.82 times, shortened by about 18%). Additionally, as the risk level increases, a safety margin factor is added to reliably guarantee buoyancy.

[0167] [Table 8]

[0168]

[0169] In the following, the process of a system according to one embodiment of the present invention responding to complex situations that may occur in an actual swimming environment, wherein a data collection module (300), a situation analysis module (400), and a risk judgment module (500) sequentially perform calculations with a control unit (800) as the main axis and finally control a flotation unit (600) and a communication unit (700), is described by organically combining them according to a time-series flow.

[0170] First, in the initial stage where the user performs high-intensity swimming such as butterfly stroke, the inertial measurement unit (200) detects signals of high acceleration and frequent direction changes, and the biosignal measurement unit (100) measures a high heart rate, and the data collection module (300) performs the process of synchronizing the timing of these data and transmitting them to the situation analysis module (400). The situation analysis module (400) calculates the dynamic movement disturbance index as a caution stage below the risk threshold by performing a division operation, reflecting the regular stroke pattern as a suppression factor, as both the number of rapid acceleration events and the number of direction changes appear high and the raw risk level increases rapidly due to mutual synergy, but as a regular stroke pattern is clearly detected through frequency analysis. Accordingly, the control unit (800) recognizes this as a high-intensity exercise state and performs variable power control by not operating the flotation device but instead increasing the sampling cycle of the sensor to switch to a precision monitoring mode.

[0171] Subsequently, when the user experiences a leg cramp while swimming and becomes flustered and flailing about, the regular strokes are lost and acceleration and rotation occur disorderly; therefore, the situation analysis module (400) calculates the dynamic movement confusion index as a danger level by multiplying the rapid acceleration and direction change in a state where the inhibitory factor has disappeared. At the same time, if the heart rate changes rapidly or arrhythmia occurs due to the user's fear, the situation analysis module (400) calculates the heart rate deviation rate. At this time, even if electrode contact failure occurs intermittently due to vigorous movement, the criticality score is partially attenuated by the reliability reduction logic, and a correction operation is performed so that it is not triggered alone. Since the confusion level is high but the reliability of the biological criticality is low, the danger judgment module (500) generates a command to send a vibration warning to the user via a haptic motor and send a caution notification to the guardian terminal instead of immediately buoyant.

[0172] If the situation deteriorates and a static drowning situation occurs where the user loses consciousness and sinks below the surface, the flailing stops and the change in the acceleration vector of the inertial measurement unit (200) drops below the noise level, and a stop state is detected. The risk judgment module (500) activates a dominant trigger that operates the addition logic from the moment the duration of the stop state exceeds the threshold, even though the confusion index has dropped to 0. In addition, as movement stops, the electrode contact failure is resolved, and the actual heart abnormality signal that had been attenuated is explosively reflected by the exponential weighting logic, causing the vital signal urgency index to surge. The risk judgment module (500) calculates the integrated drowning risk probability through an S-shaped activation function by combining the stop time addition point and the vital urgency score, and finally generates a flotation command as this value exceeds the preset trigger threshold.

[0173] Finally, assuming the environment in which the system operates is a cold sea during winter, the sensor of the flotation unit (600) detects that the cartridge pressure is lower than the standard, and the control unit (800) applies Bernoulli inverse proportional control logic to calculate an extension of the valve opening time longer than the standard to compensate for the slow flow rate caused by the low pressure. Additionally, since the probability of danger is very high, a safety margin factor is additionally applied to open the solenoid valve for the final determined time, thereby controlling the airbag to receive sufficient gas even in a low-temperature environment and fully inflate it to bring the user's head to the surface of the water. Furthermore, as soon as the communication unit (700) detects surfacing via the pressure sensor, it switches from underwater ultrasonic mode to long-range wireless communication mode and performs a procedure to transmit the current location coordinates and a drowning accident occurrence message to the control server.

[0174] As such, the present invention organically combines a series of processes that collect sensor data, calculate individual physical and biological indices, probabilistically integrate and judge them, and then precisely control physical devices by considering environmental variables, thereby operating as an intelligent safety system equipped with a logical flow of data and a feedback structure, rather than being a simple sum of components.

[0175] Although preferred embodiments of the present invention have been illustrated and described above, the present invention is not limited to the specific embodiments described above. Various modifications are possible by those skilled in the art without departing from the essence of the invention as claimed in the claims, and such modifications should not be understood individually from the technical spirit or perspective of the present invention. Explanation of the symbols

[0177] Biosignal measurement unit (100) Inertial measurement unit (200) Data collection module (300) Situation analysis module (400) Risk judgment module (500) Floating unit (600) Communication unit (700) Control unit (800)

Claims

Claim 1 A multi-sensor fusion and machine learning-based underwater activity risk prediction device, which is worn on a user's body to collect underwater activity data and predict dangerous situations through machine learning-based risk judgment logic, comprising: a biosignal measurement unit that measures biosignals including the user's electrocardiogram and optical heart rate; an inertial measurement unit that measures the user's underwater movement vector and rotation amount using a 3-axis accelerometer and gyroscope; extracting the number of sudden acceleration events (N_jerk) and the number of direction changes (N_zc) based on data received from the inertial measurement unit, generating a raw risk level by multiplying the number of sudden acceleration events (N_jerk) and the number of direction changes (N_zc), performing a normal motion correction operation by dividing the raw risk level by the sum of the value obtained by multiplying 1 by a inhibition weighting constant (β) and the number of strokes (N_stroke), and performing a natural logarithm operation reflecting a sensor sensitivity correction constant (k_1) on the result of the normal motion correction operation to calculate a dynamic motion confusion index (I_chaos) according to the following [Logic Equation 1], wherein the biosignal measurement A situation analysis module that calculates a heart rate deviation ratio (R_dev) by dividing the number of abnormal samples by the total number of valid waveforms based on data received from the unit, calculates a raw critical score by reflecting the natural constant raised to the power of the value obtained by multiplying the basic scale constant (A) by the risk sensitivity constant (λ) and the heart rate deviation ratio (R_dev), and then performs a reliability-based correction operation by dividing the raw critical score by the sum of the value obtained by multiplying 1 by the number of electrode contact failures (C_loss) and the noise suppression coefficient (α) to calculate a vital signal criticality index (I_vital) according to the following [Logical Formula 2];...receives the dynamic motion confusion index (I_chaos) and the vital signal urgency index (I_vital) from the situation analysis module; if a stop state is detected where the change in the acceleration vector of the inertial measurement unit is below a preset noise level and the change in the gyroscope is minute, calculates a score for the duration of the stop state (T_score) based on the duration of the stop state and reflects it as an independent risk factor; calculates an integrated risk score by multiplying the dynamic motion confusion index (I_chaos), the vital signal urgency index (I_vital), and the score for the duration of the stop state (T_score) by their respective weights (w_1, w_2, w_3), adding them, and subtracting the threshold bias value (Bias); performs a probability transformation by substituting the integrated risk score into a sigmoid function to calculate an integrated drowning risk probability (P_risk) according to [Logical Formula 3] below; and if the integrated drowning risk probability (P_risk) is greater than or equal to a preset judgment threshold, determines it as a risk of drowning. A multi-sensor fusion and machine learning-based underwater activity risk prediction device comprising a judgment module. [Logic Equation 1]; [Logical Formula 2] [Logical Formula 3] Claim 2 delete

Images