Waistband type wearing fall protection early warning device for the elderly living alone

Abstract

The application discloses a waistband type worn fall protection early warning device for the elderly living alone and belongs to the technical field of intelligent wearable devices. The application solves the problem that the existing device has a single early warning mode and cannot effectively guarantee the safety of the elderly living alone. A convolutional neural network and a lightweight decision tree model are used to match corresponding action results for various data of the elderly. If the preliminary judgment result conforms to the normal mode, the judgment result is directly output. If the preliminary judgment result does not conform to the normal mode, a secondary analysis mechanism is triggered to further judge whether the elderly person has a fall behavior. Whether the same position has multiple fall behaviors is analyzed through a GPS locator. If a high-risk position is found, the elderly person is reminded to pay attention to action behavior through a warning light and an alarm when the elderly person approaches the high-risk position. The application not only improves the accuracy of fall detection and reduces the possibility of false positives and false negatives, but also actively prevents the elderly from falling again in a high-risk area, thereby providing more comprehensive safety protection for the elderly living alone.

Description

Technical Field

[0001] This invention relates to the field of smart wearable device technology, specifically a fall prevention and early warning device for elderly people living alone that is worn as a waist belt. Background Technology

[0002] Falls are a common accident among elderly people living alone, potentially leading to serious injury or even death. Currently, some fall detection devices exist on the market, primarily using motion sensors such as accelerometers and gyroscopes to monitor the elderly person's movement. However, most existing fall warning devices suffer from the following problems: 1. Most existing fall detection devices are in the form of wristbands or anklebands, which are worn on the wrist or ankle. This wearing method not only affects the normal activities of the elderly, but may also cause discomfort and reduce the elderly's willingness to use the devices. 2. Existing fall detection equipment can only issue alarms with simple beeps or vibrations after detecting a fall, and cannot promptly notify guardians or emergency rescue agencies, making it difficult to effectively protect the safety of the elderly.

[0003] Therefore, to address the inadequacy of existing needs, we proposed a waist-worn fall prevention and early warning device for elderly people living alone. Summary of the Invention

[0004] The purpose of this invention is to provide a waistband-style fall prevention and early warning device for elderly people living alone. It uses a convolutional neural network and a lightweight decision tree model to match various data points of the elderly person with corresponding action results. If the initial judgment result matches the normal pattern, the judgment result is directly output; if not, a secondary analysis mechanism is triggered to further determine whether the elderly person has committed a fall. Furthermore, it uses a GPS locator to analyze whether multiple falls have occurred at the same location. If a high-risk location is detected, warning lights and alarms are used to alert the elderly person when they approach it. This not only improves the accuracy of fall detection and reduces the possibility of false alarms and missed alarms, but also proactively prevents the elderly person from falling again in high-risk areas, providing more comprehensive safety protection for elderly people living alone and solving the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A waist belt-type fall prevention and early warning device for elderly people living alone includes: a waist belt body, with a warning light and an alarm on the outside of the waist belt body, a microprocessor A on the inside of the waist belt body, and a motion sensor B for real-time monitoring of the elderly person's behavior, including: an accelerometer and a gyroscope, which can accurately acquire the elderly person's motion acceleration, angular velocity and height change information; a barometric pressure sensor C for acquiring atmospheric pressure signals in the elderly person's exercise environment; a pressure sensor D for monitoring changes in the pressure distribution on the elderly person's waist; and a GPS locator E for monitoring the location of the elderly person. The microprocessor A includes: a primary analysis module and a secondary analysis module; The first-level analysis module is configured to build a convolutional neural network, a lightweight decision tree model, and a long short-term memory network. The convolutional neural network extracts features from the elderly's action data, pressure distribution data, and air pressure change data, and outputs the confidence score of the current action. Based on the predefined rules and features of the lightweight decision tree model, the corresponding action result is matched to the confidence score of the current action. The secondary analysis module is configured to learn the temporal dependencies of features through long short-term memory networks to further determine whether elderly people will exhibit fall behavior.

[0006] Furthermore, after matching the confidence level of the current action to the corresponding action result through the primary analysis module, it also includes: Based on the historical training results of the convolutional neural network and the lightweight decision tree model, a confidence threshold is preset. If the output confidence level is within the threshold range and the pressure distribution and air pressure changes conform to the normal pattern, then the current judgment result will be taken as the final confirmation result. If the output confidence level is below the threshold range, or if the pressure distribution and air pressure changes do not conform to the normal pattern, the secondary analysis mechanism is triggered. After analyzing the elderly's movement data, pressure distribution data, and air pressure change data through the secondary analysis module, if multiple models support fall detection, it is finally confirmed as a fall event and an alarm is triggered.

[0007] Furthermore, the secondary analysis module includes: The early warning analysis module is configured to collect the locations where elderly people fall via GPS locator E, analyze whether the same location has been the site of multiple falls, and if so, mark the location as a high-risk location and feed it back to microprocessor A. The early warning and protection module is configured to obtain the location information of the elderly in real time through GPS locator E, and to alert the elderly to their actions and behaviors by using warning lights and alarms when the elderly approach a high-risk location.

[0008] Furthermore, marking the location as a high-risk location includes: Fall event data is collected in real time via GPS locator E; For each fall event e i Represent it as a ternary element, i.e., e i =(t i lat i lon i ), where t i Indicates a fall event e i The corresponding event occurrence time; lat i Indicates a fall event e i Corresponding latitude coordinates; lon i Indicates a fall event e i Corresponding longitude coordinates; Real-time monitoring of the translational distance between each fall event and its previous fall event, as well as the walking distance between each fall event and the elderly person's current starting position; A dynamic risk threshold is set by utilizing the translational distance between each fall event and its previous fall event, and the walking distance between each fall event and the elderly person's current starting position. The dynamic risk threshold is obtained using the following formula: Where Q represents the dynamic risk threshold; L p L represents the average translation distance between each fall event and its previous fall event; n represents the number of historical fall events; L i+1 L represents the distance traveled between the (i+1)th fall event and the elderly person's current starting position; i This represents the distance traveled between the i-th fall event and the elderly person's current starting position. Using each fall event e i The corresponding three elements set the risk density for the current fall event; The risk density corresponding to the current fall event is compared with the preset dynamic risk threshold. When a fall event occurs multiple times at the same location and the corresponding risk density is not lower than a preset dynamic risk threshold, the location is marked as a high-risk location.

[0009] Furthermore, the use of each fall event e i The corresponding three elements set the risk density for the current fall event, including: Retrieve each fall event ei The corresponding three elements; Using each fall event e i The corresponding three elements are used to obtain the average risk density of historical fall events. The average risk density corresponding to the historical fall events is obtained using the following formula: Where Y(p,t) represents the average risk density corresponding to historical fall events; n represents the number of historical fall events; α i d(p, p) represents the confidence weight corresponding to the i-th historical fall event; β represents the spatial sensitivity parameter, with a value of 0.001-0.005, used to control the rate of risk decay with distance; i (p) represents the elderly person's home location p and the i-th historical fall event p. i The spherical distance between them, using each fall event e i Corresponding longitude coordinates lon i and latitude coordinates lat i The Haversine formula is used to calculate the time decay rate by combining the longitude coordinates lon and latitude coordinates lat of the elderly person's home location p; r represents the time decay rate parameter, with a value of 0.4-1.3; t i The time corresponding to the i-th historical fall event is represented by t; t represents the current time. Retrieve the three elements corresponding to the current fall event; The risk density corresponding to the current fall event is obtained by combining the three elements corresponding to the current fall event with the average risk density corresponding to historical fall events. The risk density corresponding to the current fall event is obtained using the following formula: Wherein, Y(p) d α(t) represents the risk density corresponding to the current fall event; d This represents the confidence weight corresponding to the current fall event; d(p, p) d (p) represents the elderly person's home location and the current fall event p. i The spherical distance between them is calculated using the longitude coordinates (lon) corresponding to the current fall event. d and latitude coordinates lat d The Haversine formula is used to calculate t, combining the longitude coordinates lon and latitude coordinates lat of the elderly person's home location p; d This indicates the moment corresponding to the current fall event.

[0010] Furthermore, the primary analysis module includes: a data collection module and a data clustering module; The data collection module is configured to collect volunteers' daily movement data, fall movement data, and pressure change data, and preset the fall pressure threshold and normal pressure threshold for the elderly. The data clustering module is configured to cluster and classify the fall pressure threshold and normal pressure threshold, as well as their corresponding daily action data, fall action data, and pressure change data based on a lightweight decision tree model.

[0011] Furthermore, the primary analysis module also includes: a data partitioning module and a model validation module; The data partitioning module is configured to divide the volunteers' data into training and testing sets, and to train and test convolutional neural networks and long short-term memory networks using the training and testing sets. The model validation module is configured to compare the test results of the convolutional neural network and the long short-term memory network with the actual action information of volunteers to verify whether the output results of the convolutional neural network and the long short-term memory network are accurate; if accurate, the convolutional neural network and the long short-term memory network are saved and put into practical use.

[0012] Furthermore, the waist belt body is composed of a rear belt body, a front strap, support members, and tension bands. The two ends of the rear belt body are integrated with the front strap. Multiple support members for supporting the waist muscles are arranged side by side on the inner side of the rear belt body. Two tension bands are symmetrically sewn between the support member in the middle and the rear belt body. The other ends of the two tension bands pass through the inside of the front strap and extend to the outside of the front strap. Velcro is provided on the inner and outer sides of the front strap. The Velcro on the inner side is used to attach the two ends of the front strap, and the Velcro on the outer side is used to attach the two tension bands.

[0013] Furthermore, the microprocessor A also includes: a device integration module, an information setting module, and a data storage module; The device integration module is configured to associate the microprocessor A built into the belt body with the guardian's mobile terminal via wireless communication technology; The information setting module is configured to input the personal information of the elderly person wearing the waist belt into the microprocessor A based on the guardian's mobile terminal; The data storage module is configured to store the elderly’s motion data, stress data, and air pressure change data in a time series, and is used to periodically optimize the learning performance of the convolutional neural network, lightweight decision tree model, and long short-term memory network.

[0014] Furthermore, the microprocessor A also includes: a remote assistance module and a power module; The remote assistance module is configured to send fall information and location information detected by GPS locator E to a preset contact person or emergency rescue organization via wireless communication technology after it is determined that an elderly person has fallen. The power module is configured to use a rechargeable battery to provide stable power to the belt body and is equipped with a low battery reminder function.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. In this invention, a convolutional neural network is used to extract features from the elderly's action data, pressure distribution data, and air pressure change data, outputting the confidence level of the current action. Combined with a lightweight decision tree model, the corresponding action result is quickly matched. If the preliminary judgment result meets the confidence threshold and the normal pattern, the judgment result is directly output. If the confidence level is low or the pressure distribution and air pressure change do not meet the normal pattern, a secondary analysis mechanism is triggered to further determine whether the elderly person has fallen. This ensures that fall events can be reliably detected in complex situations, improves the accuracy of fall detection for the elderly, reduces the possibility of false alarms and missed alarms, and provides strong protection for the safety of elderly people living alone.

[0016] 2. In this invention, GPS locators are used to collect information on the locations where elderly people fall, and the system analyzes whether multiple falls occur at the same location. If a high-risk location is identified, warning lights and alarms are used to alert elderly people to their movements when they approach the high-risk location. This proactively prevents elderly people from falling again in high-risk areas, effectively reduces the risk of falls, enhances the protective function of the early warning device, and provides more comprehensive safety protection for elderly people living alone. Attached Figure Description

[0017] Figure 1 This is a structural diagram of the fall protection and early warning device for elderly people living alone that is worn as a waistband according to the present invention. Figure 2 This is an inner view of the fall protection and early warning device for elderly people living alone that is worn as a belt according to the present invention. Figure 3 This is a flowchart of the fall protection early warning process of the present invention.

[0018] In the diagram: 1. Waist belt body; 11. Warning light; 12. Alarm; 2. Rear belt body; 3. Front strap; 4. Support component; 5. Tension band. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] To address the technical challenges of existing fall detection devices, which mostly use wristbands or ankle bracelets and are uncomfortable to wear, and which only provide simple alarms like beeps or vibrations after a fall is detected, failing to promptly notify caregivers or emergency response agencies and thus hindering the effective protection of elderly people living alone, please refer to [link to relevant documentation]. Figures 1-3 This embodiment provides the following technical solution: A fall protection and warning device for elderly people living alone, worn as a waist belt, includes: a waist belt body 1, which is the main part of the entire fall protection and warning device. To ensure comfort and functionality for the elderly, the waist belt body 1 consists of a rear strap 2, a front strap 3, a support 4, and a tension band 5, allowing the waist belt body 1 to fit snugly against the waist while providing necessary support and stability. Warning lights 11 and an alarm 12 are installed on the outer side of the waist belt body 1, emitting clear light and sound signals when a fall risk is detected or a fall event is confirmed, promptly reminding the elderly to pay attention to safety and notifying guardians or emergency rescue agencies. The rear strap 2 is the main support part of the waist belt, with its ends connected to the front strap 3. The straps 3 are connected as one piece to form a complete waist belt structure; multiple support pieces 4 are arranged side by side on the inner side of the rear belt body 2 to support the waist muscles. These support pieces are made of soft and elastic material to support the waist muscles, reduce the burden on the waist, and improve wearing comfort; two tension straps 5 are symmetrically sewn between the support piece 4 in the middle and the rear belt body 2. The other ends of the two tension straps 5 pass through the inside of the front strap 3 and extend to the outside of the front strap 3, which not only enhances the stability of the waist belt body 1, but also makes the waist belt body 1 more adaptable to elderly people of different body types; Velcro is provided on the inner and outer sides of the front strap 3. The Velcro on the inner side is used to fasten the two ends of the front strap 3. The belt body 1 is fastened to ensure a tight fit around the waist. The Velcro straps on the outside are used to attach two tension bands 5. Adjusting the tightness of the tension bands 5 further optimizes the fit of the belt body 1, adapting to different waist sizes of elderly individuals and improving wearing comfort and stability. The inner side of the belt body 1 houses a microprocessor A and a motion sensor B for real-time monitoring of the elderly person's behavior. Motion sensor B includes an accelerometer and a gyroscope, accurately acquiring information on the elderly person's acceleration, angular velocity, and altitude changes. A barometric pressure sensor C is used to acquire atmospheric pressure signals within the elderly person's environment; for example, when an elderly person falls from a height, the air pressure rises rapidly. The system includes several components: a pressure sensor (D) for monitoring changes in air pressure to detect abnormalities promptly; a pressure sensor (D) for monitoring changes in pressure distribution in the lower back of elderly individuals (e.g., a drastic change in pressure distribution when an elderly person falls, which allows for timely detection of falls and improves accuracy and reliability); a GPS locator (E) for monitoring the location of elderly individuals (by acquiring their location information in real time, it can promptly notify caregivers or emergency response agencies in the event of a fall and provide accurate geographic location information); and a GPS locator (E) for analyzing the activity range and behavioral patterns of elderly individuals to proactively plan high-risk locations and prevent falls.

[0021] Microprocessor A includes: a primary analysis module and a secondary analysis module; the primary analysis module includes: a data collection module, a data clustering module, a data partitioning module, and a model validation module; the secondary analysis module includes: an early warning analysis module and an early warning protection module.

[0022] The data collection module is configured to collect volunteers' daily movement data, fall movement data, and stress mutation data. It presets the fall pressure threshold and normal pressure threshold for the elderly as data samples for training convolutional neural networks and long short-term memory networks.

[0023] The data clustering module is configured to cluster the fall pressure threshold and normal pressure threshold, as well as the corresponding daily movement data, fall movement data, and pressure mutation data, based on a lightweight decision tree model. Through cluster analysis, the characteristic distribution of different movement patterns can be better understood.

[0024] The data partitioning module is configured to divide the volunteers' data into training and testing sets. The convolutional neural network and long short-term memory network are trained and tested using the training and testing sets to ensure that the convolutional neural network and long short-term memory network perform well on different datasets.

[0025] The model validation module is configured to compare the test results of the convolutional neural network and the long short-term memory network with the actual action information of volunteers to verify the accuracy of the output results of the convolutional neural network and the long short-term memory network. If accurate, the convolutional neural network and the long short-term memory network are saved and put into practical use to ensure that the waist belt-style fall prevention and early warning device for elderly people living alone can effectively detect the fall behavior of the elderly, while reducing false alarms and missed alarms, and providing strong protection for the safety of the elderly.

[0026] The first-level analysis module is configured to build a convolutional neural network, a lightweight decision tree model, and a long short-term memory network. It uses the convolutional neural network to extract features from the elderly's movement data, pressure distribution data, and air pressure change data, outputting the confidence level of the current action. Based on the historical training results of the convolutional neural network and the lightweight decision tree model, it presets a confidence threshold; for example, setting the confidence threshold range to ≤65%. Using predefined rules and features from the lightweight decision tree model, it matches the corresponding action result to the confidence level of the current action. For example, if the convolutional neural network outputs the action "sit down normally" with a confidence level of 85%, and the decision tree model matches the action result as "sit down normally," then the action is considered "sit down normally," and the second-level analysis module does not need to be activated. If the output confidence level is within the threshold range, and the pressure distribution and air pressure changes conform to a normal pattern, then the current judgment result is taken as the final confirmation result. If the output confidence level is below the threshold range, or the pressure distribution and air pressure changes do not conform to a normal pattern, then the second-level analysis mechanism is triggered.

[0027] The secondary analysis module is configured to learn the temporal dependencies of features through a long short-term memory network. For example, the long short-term memory network can capture the changing trends of acceleration and angular velocity over a period of time, thereby further determining whether the elderly person has fallen. If multiple models support the fall detection, it is finally confirmed as a fall event and an alarm is triggered. For example, the convolutional neural network outputs a "normal sitting" action with a confidence level of 60%, and the decision tree model matches the result as "suspicious action". The long short-term memory network outputs a fall behavior with a confidence level of 90%, and combined with the abnormal situation of pressure distribution and air pressure changes, it is finally confirmed as a fall event and an alarm is triggered to notify the preset contact person or emergency rescue agency.

[0028] The beneficial effects achieved by the above are as follows: Feature extraction is performed on the elderly's movement data, pressure distribution data, and air pressure change data using a convolutional neural network. The confidence level of the current movement is output, and a lightweight decision tree model is used to quickly match the corresponding movement result. If the preliminary judgment result meets the confidence threshold and the normal pattern, the judgment result is directly output. If the confidence level is low or the pressure distribution and air pressure changes do not conform to the normal pattern, a secondary analysis mechanism is triggered to further determine whether the elderly person has fallen. This ensures reliable detection of fall events in complex situations, improves the accuracy of fall detection for the elderly, reduces the possibility of false alarms and missed alarms, and provides strong protection for the safety of elderly people living alone.

[0029] The early warning analysis module is configured to collect the locations where elderly people fall via GPS locator E, analyze whether the same location has been the site of multiple falls, and if so, mark the location as a high-risk location and send the data back to microprocessor A. The early warning and protection module is configured to acquire the elderly person's location information in real time via GPS locator E, compare the real-time location information with stored high-risk location information, and when the elderly person approaches a high-risk location, the warning light 11 flashes a specific color light and the alarm 12 sounds an alarm to remind the elderly person to pay attention to the danger of the current location and to pay attention to their actions; and provide personalized early warning prompts based on the elderly person's personal behavior habits and historical fall data; for example, for elderly people who have fallen multiple times in the bathroom, the system can issue more frequent warning signals when they enter the bathroom.

[0030] The beneficial effects achieved by the above are as follows: GPS locators collect data on the locations where elderly people fall, and analyze whether multiple falls occur at the same location; if high-risk locations are identified, warning lights and alarms will alert elderly people to their movements when they approach these locations; thereby proactively preventing elderly people from falling again in high-risk areas, effectively reducing the risk of falls, enhancing the protective function of early warning devices, and providing more comprehensive safety protection for elderly people living alone.

[0031] Specifically, marking this location as a high-risk location includes: Fall event data is collected in real time via GPS locator E; For each fall event e i Represent it as a ternary element, i.e., e i =(t i lat i lon i ), where t i Indicates a fall event e i The corresponding event occurrence time; lat i Indicates a fall event e i Corresponding latitude coordinates; lon i Indicates a fall event e i Corresponding longitude coordinates; Real-time monitoring of the translational distance between each fall event and its previous fall event, as well as the walking distance between each fall event and the elderly person's current starting position; A dynamic risk threshold is set by utilizing the translational distance between each fall event and its previous fall event, and the walking distance between each fall event and the elderly person's current starting position. The dynamic risk threshold is obtained using the following formula: Where Q represents the dynamic risk threshold; L p L represents the average translation distance between each fall event and its previous fall event; n represents the number of historical fall events; L i+1 L represents the distance traveled between the (i+1)th fall event and the elderly person's current starting position; i This represents the distance traveled between the i-th fall event and the elderly person's current starting position. Using each fall event e i The corresponding three elements set the risk density for the current fall event; The risk density corresponding to the current fall event is compared with the preset dynamic risk threshold. When a fall event occurs multiple times at the same location and the corresponding risk density is not lower than a preset dynamic risk threshold, the location is marked as a high-risk location.

[0032] The technical advantages of the above solution are as follows: It collects data on fall events (time, latitude, longitude, etc., represented by three elements) using GPS locators, while also considering the translational distance between fall events and the distance traveled from the starting position, thus characterizing the fall scenario from multiple spatiotemporal dimensions. Based on this data, dynamic risk thresholds and risk density are calculated, enabling more accurate capture of location risk characteristics. Compared to single-dimensional judgment, this avoids misjudgments or omissions, making high-risk location identification more realistic. For example, it can distinguish between accidental fall locations and truly high-risk locations prone to repeated falls. The dynamic risk threshold formula combines the number of historical fall events, the average translational distance, and the difference in distance traveled from different fall events to the starting position, and can adaptively adjust as fall event data changes. Unlike fixed thresholds, which may lead to inaccurate judgments due to differences in scenarios and elderly people's walking habits, this solution better adapts to the fall risk location identification under different daily activities of elderly people (such as different walking routes and activity ranges), improving the rationality of high-risk location determination. After accurately marking high-risk locations, targeted preventative measures can be implemented based on this information. These measures could include adding handrails, laying anti-slip mats, and setting up warning signs to proactively eliminate or reduce fall hazards and decrease the probability of elderly people falling again in high-risk areas. This environmental intervention ensures the safety of elderly people during activities such as walking and reduces the likelihood of fall injuries. Because the calculation and judgment are based on real-time collected fall events, changes in location risk can be continuously tracked. If falls decrease after protection measures are implemented at high-risk locations, changes in risk density and other data will lead to corresponding adjustments in thresholds and judgments. This allows for dynamic monitoring and optimization of risk locations, forming a closed-loop management system of "identification-protection-reassessment-re-optimization," ensuring a safe environment for elderly activities in the long term. The continuously collected three-element data of fall events, as well as the translational distance and walking distance involved in the calculation process, can be accumulated into a fall risk-related database. With data accumulation, the parameters in the formula (such as the number of historical fall events n) and the calculation logic can be continuously optimized based on actual feedback. For example, if it is found that the threshold calculation is not well adapted to the risk density in certain scenarios, the weights of the elements in the formula can be adjusted or new influencing factors can be introduced to continuously iterate and upgrade the entire high-risk location identification algorithm, improve its intelligence level and application effect, and provide more efficient and accurate technical means for fall prevention in smart elderly care scenarios.

[0033] Specifically, the use of each fall event e i The corresponding three elements set the risk density for the current fall event, including: Retrieve each fall event e i The corresponding three elements; Using each fall event ei The corresponding three elements are used to obtain the average risk density of historical fall events. The average risk density corresponding to the historical fall events is obtained using the following formula: Where Y(p,t) represents the average risk density corresponding to historical fall events; n represents the number of historical fall events; α i d(p, p) represents the confidence weight corresponding to the i-th historical fall event; β represents the spatial sensitivity parameter, with a value of 0.001-0.005, used to control the rate of risk decay with distance; i (p) represents the elderly person's home location p and the i-th historical fall event p. i The spherical distance between them, using each fall event e i Corresponding longitude coordinates lon i and latitude coordinates lat i The Haversine formula is used to calculate the time decay rate by combining the longitude coordinates lon and latitude coordinates lat of the elderly person's home location p; r represents the time decay rate parameter, with a value of 0.4-1.3; t i This represents the time corresponding to the i-th historical fall event; t represents the current time; specifically, and This simulation addresses the propagation and attenuation characteristics of fall risk in real-world scenarios. Spatially, the greater the physical distance, the lower the "risk correlation" between past falls and the current location. For example, in a home environment, the risk impact of a fall at the bedroom doorway on the living room area decreases rapidly with increasing distance. This spatial correlation attenuation is quantitatively represented using an exponential decay function, ensuring that risk calculations align with actual spatial propagation patterns. Temporally, falls from too long ago have less reference value for current risk (e.g., a fall from several months ago has little guiding significance due to changes in environment and the elderly person's condition). A time decay function simulates the "expiration" of risk over time, making the average risk density more focused on the impact of recent, nearby falls, thus meeting the dynamic risk changes required in engineering scenarios.

[0034] Retrieve the three elements corresponding to the current fall event; The risk density corresponding to the current fall event is obtained by combining the three elements corresponding to the current fall event with the average risk density corresponding to historical fall events. The risk density corresponding to the current fall event is obtained using the following formula: Wherein, Y(p) d α(t) represents the risk density corresponding to the current fall event; d This represents the confidence weight corresponding to the current fall event; d(p, p) d(p) represents the elderly person's home location and the current fall event p. i The spherical distance between them is calculated using the longitude coordinates (lon) corresponding to the current fall event. d and latitude coordinates lat d The Haversine formula is used to calculate t, combining the longitude coordinates lon and latitude coordinates lat of the elderly person's home location p; d表示 The exact moment corresponding to the current fall event. Specifically, and This approach emphasizes the immediate impact of current falls, calculating the spatiotemporal correlation of each fall separately from the existing historical average risk density. This reflects the logic of "new events supplementing and updating risk perception." For example, if an elderly person has just fallen in a certain location, the impact of this event on the current risk assessment needs to be considered separately. This formula quantitatively calculates its immediate risk contribution in the spatial (distance from the current assessment location) and temporal (interval with the current moment) dimensions, providing a real-time response to dynamic risks.

[0035] The technical advantages of the above solution are as follows: Existing technologies often fail to adequately consider the spatiotemporal correlation of fall risk, mostly relying on simple statistical analysis of fall location or single-time-dimensional characteristics. This solution, however, introduces a decay function for spatial distance (spherical distance) and time difference, dynamically incorporating the spatiotemporal impact of historical fall events into the risk density calculation. For example, spatially, historical fall points closer to the current location have a greater impact on the current risk assessment (because...). (The weight of historical falls decreases rapidly with increasing distance); in terms of time, the closer a historical fall is to the current moment, the higher its weight. The risk density decreases as time difference increases, allowing for a more accurate reflection of the changing patterns of fall risk in real-world scenarios, improving the accuracy of risk identification and making the assessment of high-risk locations more scientific. Dynamic updates and adaptive adjustments are based on calculating the average risk density from historical fall events, then updating it to the current risk density using current fall events—a dynamic iterative process. Unlike existing static risk assessment models (such as fixed location risk levels remaining unchanged over time), this approach continuously updates risk perception based on changes in elderly activity and fall patterns. For example, if an elderly person frequents a certain area, new fall events continuously add to the risk density, reflecting the latest risk situation in that area in real time. This provides better adaptability to environmental changes (such as changes in home facilities altering fall risk), helping to promptly identify new high-risk points or adjust existing risk location assessments. An alpha is introduced to differentiate events using confidence weights. i (Confidence weights of historical fall events) and α d(Current fall event confidence weight) allows assigning different importance to different fall events. Existing technologies often treat all fall events equally, ignoring the differences in the events themselves (e.g., serious falls should be given more importance than minor trips). In practical applications, weights can be set based on the severity of the fall injury, the reliability of the monitoring equipment, etc. For example, a high value can be set for α of serious falls confirmed by hospitals, increasing the impact of these key events on risk density, making risk assessment more in line with the actual degree of harm, and optimizing the identification results of high-risk locations.

[0036] Microprocessor A also includes: a device integration module, an information setting module, a data storage module, a remote assistance module, and a power supply module.

[0037] The device integration module is configured to associate the dedicated application of the microprocessor A built into the waist belt body 1 with the guardian's mobile terminal via wireless communication technology. The microprocessor A synchronizes the sensor data collected in real time to the guardian's mobile terminal. The guardian can view the elderly person's real-time location, movement status, and fall warning information through the mobile terminal, so that the guardian can understand the elderly person's safety status in real time and take appropriate measures when necessary.

[0038] The information setting module is configured to input the personal information of the elderly person wearing the waist belt body 1 into the microprocessor A based on the guardian's mobile terminal, such as name, age, gender, health status, emergency contact information, etc.

[0039] The data storage module is configured to store the elderly’s motion data, stress data, and air pressure change data in a time series, and is used to periodically optimize the learning performance of the convolutional neural network, lightweight decision tree model, and long short-term memory network.

[0040] The remote assistance module is configured to send fall information and location information detected by the GPS locator E to a preset contact person or emergency rescue organization via wireless communication technology, such as Bluetooth, Wi-Fi or mobile communication network, after determining that an elderly person has fallen. This allows the contact person or emergency rescue organization to take swift action to ensure that the elderly person living alone receives timely assistance and reduces the severity of the accident.

[0041] The power module is configured to use a rechargeable battery, such as a lithium-ion battery, to provide stable power support for the belt body 1, and is equipped with a low battery reminder function.

[0042] Working principle: By using a convolutional neural network and a lightweight decision tree model, the system matches various data points of the elderly to corresponding action results. If the initial judgment result matches the normal pattern, the judgment result is directly output; if not, a secondary analysis mechanism is triggered to further determine whether the elderly person has committed a fall. It also uses a GPS locator to analyze whether there are multiple falls in the same location. If a high-risk location is found, warning lights and alarms are used to remind the elderly person to pay attention when they approach the high-risk location. This not only improves the accuracy of fall detection but also proactively prevents the elderly person from falling again in high-risk areas, providing more comprehensive safety protection for elderly people living alone.

[0043] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "possessing," or any other variations thereof are intended to cover non-exclusive possession, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0044] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that variations, modifications, substitutions and alterations may be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A waist-worn fall prevention and warning device for elderly people living alone, including: The waist belt body (1) is characterized in that: a warning light (11) and an alarm (12) are provided on the outer side of the waist belt body (1), a microprocessor A is provided on the inner side of the waist belt body (1), a motion sensor B for real-time monitoring of the behavior of the elderly, a barometric pressure sensor C for acquiring atmospheric pressure signals in the exercise environment of the elderly, a pressure sensor D for monitoring changes in the pressure distribution of the elderly waist, and a GPS locator E for monitoring the location of the elderly. The microprocessor A includes: a primary analysis module and a secondary analysis module; The first-level analysis module is configured to build a convolutional neural network, a lightweight decision tree model, and a long short-term memory network. The convolutional neural network extracts features from the elderly's action data, pressure distribution data, and air pressure change data, and outputs the confidence score of the current action. Based on the predefined rules and features of the lightweight decision tree model, the corresponding action result is matched to the confidence score of the current action. The secondary analysis module is configured to learn the temporal dependencies of features through a long short-term memory network to further determine whether the elderly person has fallen. The secondary analysis module includes: The early warning analysis module is configured to collect the locations where elderly people fall via GPS locator E, analyze whether the same location has been the site of multiple falls, and if so, mark the location as a high-risk location and send the data back to microprocessor A. The early warning and protection module is configured to obtain the location information of the elderly in real time through GPS locator E, and to remind the elderly to pay attention to their actions and behaviors through warning lights (11) and alarms (12) when the elderly approach a high-risk location. The step of marking this location as a high-risk location includes: Fall event data is collected in real time via GPS locator E; for each fall event e i Represent it as a ternary element, i.e., e i =(t i lat i lon i ), where t i Indicates a fall event e i The corresponding event occurrence time; lat i Indicates a fall event e i Corresponding latitude coordinates; lon i Indicates a fall event e i Corresponding longitude coordinates; Real-time monitoring of the translational distance between each fall event and its previous fall event, as well as the walking distance between each fall event and the elderly person's current starting position; A dynamic risk threshold is set by utilizing the translational distance between each fall event and its previous fall event, and the walking distance between each fall event and the elderly person's current starting position. The dynamic risk threshold is obtained using the following formula: Where Q represents the dynamic risk threshold; L p L represents the average translation distance between each fall event and its previous fall event; n represents the number of historical fall events; L i+1 L represents the distance traveled between the (i+1)th fall event and the elderly person's current starting position; i This represents the distance traveled between the i-th fall event and the elderly person's current starting position. Using each fall event e i The corresponding three elements set the risk density for the current fall event; The risk density corresponding to the current fall event is compared with a preset dynamic risk threshold. When the current fall event is multiple falls at the same location, and the corresponding risk density is not lower than the preset dynamic risk threshold, the location is marked as a high-risk location. The use of each fall event e i The corresponding three elements set the risk density for the current fall event, including: Retrieve each fall event e i The corresponding three elements; utilizing each fall event e i The corresponding three elements are used to obtain the average risk density of historical fall events. The average risk density corresponding to the historical fall events is obtained using the following formula: Where Y(p,t) represents the average risk density corresponding to historical fall events; n represents the number of historical fall events; α i d(p, p) represents the confidence weight corresponding to the i-th historical fall event; β represents the spatial sensitivity parameter, with a value of 0.001-0.005, used to control the rate of risk decay with distance; i (p) represents the elderly person's home location p and the i-th historical fall event p. i The spherical distance between them, using each fall event e i Corresponding longitude coordinates lon i and latitude coordinates lat i The Haversine formula is used to calculate the time decay rate by combining the longitude coordinates lon and latitude coordinates lat of the elderly person's home location p; r represents the time decay rate parameter, with a value of 0.4-1.3; t i The time corresponding to the i-th historical fall event is represented by t; t represents the current time. Retrieve the three elements corresponding to the current fall event; combine the three elements corresponding to the current fall event with the average risk density corresponding to historical fall events to obtain the risk density corresponding to the current fall event; The risk density corresponding to the current fall event is obtained using the following formula: Among them, Y(p d α(t) represents the risk density corresponding to the current fall event; d This represents the confidence weight corresponding to the current fall event; d(p, p) d (p) represents the elderly person's home location and the current fall event p. i The spherical distance between them is calculated using the longitude coordinates (lon) corresponding to the current fall event. d and latitude coordinates lat d The Haversine formula is used to calculate t, combining the longitude coordinates lon and latitude coordinates lat of the elderly person's home location p; d表示 The moment corresponding to the current fall event.

2. The fall prevention and early warning device for elderly people living alone, worn as a belt, as described in claim 1, is characterized in that... After matching the confidence level of the current action to the corresponding action result through the primary analysis module, it also includes: Based on the historical training results of the convolutional neural network and the lightweight decision tree model, a confidence threshold is preset. If the output confidence level is within the threshold range and the pressure distribution and air pressure changes conform to the normal pattern, then the current judgment result will be taken as the final confirmation result. If the output confidence level is below the threshold range, or if the pressure distribution and air pressure changes do not conform to the normal pattern, the secondary analysis mechanism is triggered. After analyzing the elderly's movement data, pressure distribution data, and air pressure change data through the secondary analysis module, if multiple models support fall detection, it is finally confirmed as a fall event and an alarm is triggered.

3. The fall prevention and early warning device for elderly people living alone, worn as a belt, as described in claim 2, is characterized in that... The primary analysis module includes: a data collection module and a data clustering module; The data collection module is configured to collect volunteers' daily movement data, fall movement data, and pressure change data, and preset the fall pressure threshold and normal pressure threshold for the elderly. The data clustering module is configured to cluster and classify the fall pressure threshold and normal pressure threshold, as well as their corresponding daily action data, fall action data, and pressure change data based on a lightweight decision tree model.

4. The fall prevention and early warning device for elderly people living alone, worn as a belt, as described in claim 3, is characterized in that... The primary analysis module also includes: a data partitioning module and a model validation module; The data partitioning module is configured to divide the volunteers' data into training and testing sets, and to train and test convolutional neural networks and long short-term memory networks using the training and testing sets. The model validation module is configured to compare the test results of the convolutional neural network and the long short-term memory network with the actual action information of volunteers to verify whether the output results of the convolutional neural network and the long short-term memory network are accurate; if accurate, the convolutional neural network and the long short-term memory network are saved and put into practical use.

5. The fall prevention and early warning device for elderly people living alone, worn as a belt, as described in claim 1, is characterized in that... The waist belt body (1) consists of a rear belt body (2), a front strap (3), a support member (4), and a tension band (5). The two ends of the rear belt body (2) are integrated with the front strap (3). Multiple support members (4) for supporting the waist muscles are arranged side by side on the inner side of the rear belt body (2). Two tension bands (5) are symmetrically sewn between the support member (4) in the middle and the rear belt body (2). The other ends of the two tension bands (5) pass through the inside of the front strap (3) and extend to the outside of the front strap (3). Velcro is provided on the inner and outer sides of the front strap (3). The Velcro on the inner side is used to attach the two ends of the front strap (3), and the Velcro on the outer side is used to attach the two tension bands (5).

6. The fall prevention and early warning device for elderly people living alone, worn as a belt, as described in claim 1, is characterized in that... The microprocessor A further includes: a device integration module, an information setting module, and a data storage module; The device integration module is configured to associate the microprocessor A built into the waist belt body (1) with the guardian's mobile terminal via wireless communication technology; The information setting module is configured to input the personal information of the elderly wearing the waist belt body (1) into the microprocessor A based on the guardian's mobile terminal; The data storage module is configured to store the elderly’s motion data, stress data, and air pressure change data in a time series, and is used to periodically optimize the learning performance of the convolutional neural network, lightweight decision tree model, and long short-term memory network.

7. The fall prevention and early warning device for elderly people living alone, worn as a belt, as described in claim 6, is characterized in that... The microprocessor A further includes: a remote assistance module and a power module; The remote assistance module is configured to send fall information and location information detected by GPS locator E to a preset contact person or emergency rescue organization via wireless communication technology after it is determined that an elderly person has fallen. The power module is configured to use a rechargeable battery to provide power to the belt body (1) and is equipped with a low battery reminder function.

Images