Fall detection method and system and electronic equipment
By combining a portable ECG patch with a depthwise separable convolutional model and posture detection, the problems of short battery life and low accuracy of wearable devices are solved, enabling long-term continuous monitoring and high-precision fall detection, especially timely identification and early warning of severe falls.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-28
- Publication Date
- 2026-04-14
AI Technical Summary
Existing wearable devices have short battery life and low detection accuracy in fall detection, making it difficult to detect users' falls in a timely manner, especially severe falls where users cannot get up in a short time.
A portable ECG patch is used to collect ECG data and acceleration signals in real time. Fall detection is performed by combining it with a depthwise separable convolutional model. False positives are filtered out by multi-branch feature fusion and posture detection to improve detection accuracy.
It enables continuous monitoring for extended periods without frequent charging, improving the accuracy of fall detection, especially for the timely identification and early warning of severe falls, thus ensuring user safety.
Smart Images
Figure CN121845563A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of fall detection, and in particular to a fall detection method, system and electronic device. Background Technology
[0002] With the increasing aging of the population, falls are gradually becoming a major threat to the health and safety of the elderly. Statistics show that approximately 40 million elderly people experience at least one fall each year, with about 50% of these occurring at home. The mortality rate within one year of a fall for those over 65 is as high as 20%, and the mortality rate for hip fracture patients reaches 30%. Therefore, fall detection is of great significance for the health of the elderly.
[0003] Current wearable devices (such as watches) incorporate fall detection algorithms, but because falls can easily be confused with hand movements, the algorithms' sensitivity is relatively low, and the triggering conditions are quite stringent, making it difficult to detect falls promptly. For example, it's difficult to trigger a fall alarm if the person can get up immediately after a fall. Furthermore, the accuracy of detection still has room for improvement in cases of severe falls, where the person cannot get up quickly. Additionally, these devices have limited battery life and require frequent charging, which many elderly people find inconvenient.
[0004] Therefore, existing fall detection equipment has a short battery life and low detection accuracy. Summary of the Invention
[0005] This application provides a fall detection method, system, and electronic device to at least solve the problems of short battery life and low detection accuracy of fall detection devices in related technologies.
[0006] In a first aspect, embodiments of this application provide a fall detection method, the method being used in conjunction with an electrocardiogram patch, the method comprising: The monitoring parameters output by the ECG patch during the target time period are obtained, including acceleration signals, activity state characteristics, and posture detection characteristics. The system determines whether a user has fallen based on the acceleration signal and the fall detection model. The fall detection model is a binary classification model, which is trained based on a depthwise separable convolutional model. In response to a user falling, it is determined whether the activity state feature is static and whether the posture detection feature meets a preset posture. If both are met, the user is output as having fallen. Meeting the preset posture includes being in a non-standing or non-sitting posture.
[0007] In one embodiment, the model structure of the fall detection model includes: The input layer is used to receive input of multi-dimensional time-series monitoring data; The feature extraction module includes an up-dimensional convolutional layer and a multi-branch depthwise separable convolutional layer; The feature fusion module includes a feature fusion layer, a Dropout layer, and a dimension reduction convolutional layer. The output module includes a feature compression layer and a fully connected layer.
[0008] In one embodiment, the fall detection model is configured to use human posture rules as the basis for fall detection.
[0009] In one embodiment, determining whether a user has fallen based on the acceleration signal and the fall detection model includes: determining whether a user has fallen based on the acceleration signal within a first preset time period using the fall detection model.
[0010] In one embodiment, determining whether the activity state feature is static and whether the posture detection feature satisfies a preset posture includes: Determine whether the activity status characteristics within the second preset time period are static; the first preset time period and the second preset time period are mutually exclusive. Determine whether the posture detection features within the target time period meet the preset posture, wherein meeting the preset posture includes the presence of a non-standing posture or a non-sitting posture.
[0011] In one embodiment, the method further includes: in response to the failure of either the activity state feature or the posture detection feature, outputting that the user has not fallen, and detecting the monitoring parameters for the next target time period.
[0012] Secondly, embodiments of this application provide a fall detection system, which is used in conjunction with an electrocardiogram patch. The system includes: Acquisition module: used to acquire monitoring parameters output by the ECG patch within the target time period, the monitoring parameters including acceleration signal, activity state characteristics and posture detection characteristics; Model judgment module: used to determine whether the user has fallen based on the acceleration signal and the fall detection model. The fall detection model is a binary classification model, which is trained based on a depthwise separable convolutional model. Output module: In response to a user falling, it determines whether the activity state feature is static and whether the posture detection feature meets a preset posture. If both are met, it outputs that the user has fallen. Meeting the preset posture includes being in a non-standing or non-sitting posture.
[0013] In one embodiment, the model structure of the fall detection model includes: The input layer is used to receive input of multi-dimensional time-series monitoring data; The feature extraction module includes an up-dimensional convolutional layer and a multi-branch depthwise separable convolutional layer; The feature fusion module includes a feature fusion layer, a Dropout layer, and a dimension reduction convolutional layer. The output module includes a feature compression layer and a fully connected layer.
[0014] Thirdly, embodiments of this application provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a fall detection method as described in the first aspect above.
[0015] Fourthly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a fall detection method as described in the first aspect above.
[0016] The fall detection method, system, and electronic device provided in this application have at least the following technical effects.
[0017] This application utilizes a portable ECG monitoring patch to collect ECG data in real time, continuously capturing dynamic ECG signals and accelerometer features. The patch is small, comfortable to wear, and has a long battery life. By acquiring monitoring parameters through the patch and combining them with subsequent detection algorithms, it supports continuous monitoring 24 / 7 without frequent charging or donning / removing. It can promptly detect user anomalies and provide timely warnings. The fall detection model replaces ordinary convolution with depthwise separable convolution, significantly reducing the number of parameters and facilitating deployment in embedded devices. False positives are filtered out by detecting the user's activity state and posture, improving the accuracy of fall detection.
[0018] Details of one or more embodiments of this application are set forth in the following drawings and description to make other features, objects and advantages of this application more readily apparent. Attached Figure Description
[0019] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart illustrating a fall detection method according to an embodiment of this application; Figure 2 This is a schematic diagram illustrating the structure of a fall detection model according to an exemplary embodiment; Figure 3 This is a structural block diagram of a fall detection system according to an embodiment of this application; Figure 4This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation
[0020] To make the objectives, technical solutions, and advantages of this application clearer, the application is described and illustrated below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. All other embodiments obtained by those skilled in the art based on the embodiments provided in this application without inventive effort are within the scope of protection of this application.
[0021] Obviously, the accompanying drawings described below are merely some examples or embodiments of this application. Those skilled in the art can apply this application to other similar scenarios based on these drawings without any inventive effort. Furthermore, it is understood that although the efforts made in this development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, any changes to design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as insufficient disclosure of the content of this application.
[0022] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that is mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments without conflict.
[0023] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms “a,” “an,” “an,” “the,” and similar words used in this application do not indicate quantity limitation and may indicate singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units not listed, or may include other steps or units inherent to these processes, methods, products, or devices. The terms “connected,” “linked,” “coupled,” and similar words used in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. “Multiple” used in this application refers to two or more. “And / or” describes the relationship between related objects, indicating that three relationships may exist; for example, “A and / or B” can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following objects are in an "or" relationship. The terms "first," "second," and "third" used in this application are merely to distinguish similar objects and do not represent a specific ordering of the objects.
[0024] In a first aspect, embodiments of this application provide a fall detection method, which is used in conjunction with an electrocardiogram patch.
[0025] Optionally, a portable ECG monitoring patch is a minimally invasive, long-term, real-time ECG data acquisition device. The advantages of the ECG patch include its small size, comfortable wear, and long battery life. It can continuously capture dynamic ECG signals and accelerometer characteristics and is compatible with AMPs systems. The ECG patch can be directly attached to the chest using flexible electrode patches, transmitting ECG data wirelessly to a mobile phone / cloud. It supports continuous monitoring 24 / 7 without frequent charging or donning / removal, and can promptly detect abnormalities in the user. It should be noted that the fall detection method used in this application is mainly for situations where a user cannot get up quickly after a fall. The purpose is to accurately identify situations where the user loses mobility after a fall and to provide timely warnings to monitor the user's safety and health.
[0026] Figure 1 This is a flowchart illustrating a fall detection method according to an embodiment of this application, such as... Figure 1 As shown, the method includes: Step S101: Obtain the monitoring parameters output by the ECG patch within the target time period. The monitoring parameters include acceleration signal, activity state characteristics, and posture detection characteristics.
[0027] Optionally, the monitoring parameters can be directly output by the ECG patch, or they can be features calculated using an accelerometer or ECG signal. The monitoring data within the target time period is typically continuous, such as 10s, 20s, or 30s. Preferably, the continuous time period is 20s. By acquiring monitoring parameters through the ECG patch and combining them with subsequent detection algorithms, continuous monitoring can be achieved 24 / 7 without frequent charging or donning / removal, allowing for timely detection and alerts for any abnormalities in the user.
[0028] Step S102: Determine whether the user has fallen based on the acceleration signal and the fall detection model. The fall detection model is a binary classification model, which is trained based on a depthwise separable convolutional model.
[0029] Optionally, the fall detection model takes an acceleration signal as input and outputs the probability of falling or not falling. If the output probability exceeds a preset threshold, the user is determined to have fallen. The sampling frequency of the acceleration signal can be 25Hz, 50Hz, 100Hz, 125Hz, 250Hz, etc. Preferably, based on experimental analysis, considering factors such as accuracy, computational power consumption, and data transmission consumption, a sampling frequency of 50Hz yields the best results. The fall detection model is a lightweight 2D convolutional network. Through a multi-branch depthwise separable convolution and feature fusion structure, the model adapts to temporal and multi-dimensional inputs to achieve classification tasks. The fall detection model replaces ordinary convolution with depthwise separable convolution, significantly reducing the number of parameters, which is beneficial for deployment in embedded devices (such as ECG patches); the multi-branch feature fusion enhances feature representation capabilities, improving the model's detection accuracy.
[0030] In one example, step S102 includes: determining whether the user has fallen based on the acceleration signal within a first preset time period using a fall detection model. Optionally, determining whether the user has fallen based on the acceleration signal within the first two seconds of the target time period. The above method is only one example of the first preset time period and is not limited thereto.
[0031] In one example, the model structure of the fall detection model includes: The input layer is used to receive input of multi-dimensional time-series monitoring data.
[0032] The feature extraction module includes an up-dimensional convolutional layer and a multi-branch depthwise separable convolutional layer.
[0033] The feature fusion module includes a feature fusion layer, a Dropout layer, and a dimensionality reduction convolutional layer.
[0034] The output module includes a feature compression layer and a fully connected layer.
[0035] Optional, Figure 2This is a schematic diagram illustrating the structure of a fall detection model according to an exemplary embodiment, such as... Figure 2 As shown, the model input is input_1, and the shape is [None, 100, 3, 1]. None represents the batch size; 100 represents the sampling point sequence, which is determined by the product of the sampling frequency and the time step. For example, if the time step is 2 seconds and the sampling frequency is 50Hz, the sampling point sequence is 2*50=100; 3 represents the multi-channel / multi-lead dimension (such as the 3-lead data of ECG); and 1 represents the single input channel (raw signal dimension).
[0036] Then, the input is increased to the initial dimensionality-enhancing convolutional layer (conv2d). The input of this 2D convolutional layer is [None, 100, 3, 1], and the output is [None, 100, 3, 16]. This dimensionality-enhancing convolutional layer increases the input channels from 1 dimension to 16 dimensions, initially extracting multi-dimensional features. The multi-branch depthwise separable convolution consists of three parallel depthwise separable convolutional layers (DepthwiseConv2D), each with an input of [None, 100, 3, 16], and an output shape that is exactly the same as the input. By performing convolution on each channel individually (without crossing channels), the number of parameters is only about 1 / 16 of that of ordinary convolution, while maintaining feature expressiveness.
[0037] The feature fusion layer element-wise adds the outputs of three depthwise separable convolutions to achieve complementary fusion of multi-branch features, similar to a simplified form of residual connections, thus avoiding feature information loss. The Dropout layer randomly deactivates some neurons to prevent overfitting and improve the model's generalization ability. Dimensionality-reducing convolutions (conv2d_1) reduce the number of channels, compressing the feature dimension and reducing subsequent computation.
[0038] Overfitting is further prevented by randomly deactivating some neurons through the Dropout layer. A 2D result is output through a fully connected (Dense) layer.
[0039] In this way, the fall detection model replaces ordinary convolution with depthwise separable convolution, significantly reducing the number of parameters and facilitating deployment on embedded devices. Enhanced feature representation through multi-branch feature fusion improves the model's detection accuracy. The process of dimensionality enhancement, multi-branch extraction, dimensionality reduction, and classification balances model performance and speed.
[0040] Continue to refer to Figure 1 Step S103 is executed after step S102, as follows.
[0041] Step S103: In response to the user falling, determine whether the activity state feature is static and whether the posture detection feature meets the preset posture. If both are met, output that the user has fallen. Meeting the preset posture includes being in a non-standing or non-sitting posture.
[0042] Optionally, for situations where a user cannot get up quickly after falling, after the fall detection model determines that the user has fallen, it is also necessary to determine the user's activity state and posture. This filters out false positives or cases where the user can recover their mobility quickly after a fall, accurately identifying situations where the user has lost mobility after a fall and providing timely warnings, which is beneficial for monitoring user safety and health. The order of activity state determination and posture determination is not limited. Posture rules are used for posture detection, and the user's current posture is determined for secondary verification to eliminate some false positives from the fall detection model. Based on changes in posture before and after the fall, it is determined whether the user is likely to have fallen. This improves the final detection accuracy in the application environment of this application.
[0043] In one example, step S103 includes: Step S1031: Determine whether the activity status characteristics within the second preset time period are static, and the first preset time period and the second preset time period are mutually exclusive.
[0044] Optionally, the method in this application primarily targets severe falls, one definition of which is the inability to get up independently after a fall. If the user can get up immediately after a fall, it is not considered a severe fall, and in this case, an alarm is unlikely to be triggered. The activity status mentioned above indicates whether the user has been active recently; 0 indicates static activity, with no activity detected; 1 indicates significant activity. As an example, it determines whether the activity status within a target time period of 3-20 seconds is static. If so, it is highly likely that the user has difficulty moving or has fainted after a fall, which may trigger an alarm.
[0045] Step S1032: Determine whether the posture detection features within the target time period meet the preset posture. Meeting the preset posture includes the presence of a non-standing posture or a non-sitting posture.
[0046] Optionally, the system determines whether the user's posture within the target time period of 0-20 seconds is either non-standing or non-sitting. As an example, postures are defined as follows: 0-2 represent calibration or calculation and are generally not considered; 3 represents standing and sitting postures, which are indistinguishable for ECG patches; 4-7 represent different sleeping positions, such as supine, supine, and lateral, and can also represent different directions after a fall. This application determines whether the posture is consistently standing / sitting. If so, it indicates that the user has not experienced a serious fall. If non-standing or non-sitting postures exist, combined with the fall detection model analysis results and the above activity state analysis showing the user remains static, it can be determined that the user has fallen, and a timely warning is issued. In this way, by detecting the user's activity state and posture, false positives and non-serious fall situations are filtered out, improving the detection accuracy of serious falls and providing timely warnings, which is beneficial for monitoring user safety and health.
[0047] In one example, the method further includes: in response to the failure of either the activity state feature or the posture detection feature, outputting that the user has not fallen, and detecting the monitoring parameters for the next target time period.
[0048] Optionally, as an example, if the user's activity state within 3-20 seconds is not entirely static, or if the user's posture within 0-20 seconds is entirely either sitting or standing, the user is determined not to be in a serious fall state. Fall detection is then performed on the monitoring parameters for the next target time period. In this way, by detecting the user's activity state and posture, false positives and non-serious fall situations are filtered out, improving the accuracy of serious fall detection.
[0049] The fall detection method described above was tested below. In the first test, a relatively vigorous fall was conducted, with subjects wearing both ECG patches and a certain brand of watch for comparison. To ensure the safety of the subjects, during the vigorous fall, which sometimes involved a running start and resulted in a faster fall speed, subjects wore full protective gear and landed on a prepared cushioning mat. Table 1 compares the results of the first test between the ECG patches and the watch.
[0050] Table 1
[0051] As shown in Table 1, there were a total of 192 fall samples and 198 non-fall samples. The watch triggered the test 54 times, with 50 true positives and 4 false positives, resulting in a sensitivity of 28.1% and a specificity of 98.0%. The method used in this application triggered the test 143 times, with 123 true positives and 20 false positives, resulting in a sensitivity of 64.1% and a specificity of 89.9%. Table 1 shows that the method used in this application has higher sensitivity.
[0052] In the second test, a gentler fall was used, with subjects wearing both ECG patches and watches for comparison. In contrast, in gentle falls, subjects tend to fall slowly, perhaps by grabbing onto furniture or bracing themselves with their hands or elbows during the fall. This means that even if they were in motion before the fall, the descent is relatively slow. Table 2 compares the results of the second test using ECG patches and watches.
[0053] Table 2
[0054] As shown in Table 2, the experiment involved 116 fall samples and 194 non-fall samples. The watch triggered the test 3 times, all of which were true positives, with a sensitivity of 2.59% and a specificity of 100%. The method used in this application triggered the test 57 times, all of which were true positives, with a sensitivity of 49.14% and a specificity of 100%.
[0055] Table 3 shows the detection process of the fall detection algorithm used in this application.
[0056] Table 3
[0057] As shown in Table 3, A-False represents the number of events detected as falls by the fall detection model but filtered out by the activity state detection rules. In this experiment, this method filtered out 8 events, of which 4 were false positives and the other 4 were true falls. S-False represents the number of events detected as falls by the fall detection model, with the activity state being static, but the posture being entirely seated. In this experiment, 6 false positive events and 17 true falls were eliminated. NF represents the number of events detected as non-falls by the fall detection model. AL1-False represents the number of fall events not detected, of which 59 were missed. Based on the above data, the sensitivity of the detection method used in this application is 49.14%, and the accuracy is 80.97%. The error mainly comes from the fall model, which missed 38 falls. The human posture rules eliminated 6 false positive events but also filtered out 17 true fall events. Without using human posture rules, the sensitivity can reach 63.8%, the accuracy can reach 84.5%, and the corresponding false alarm rate is 3.1%.
[0058] In summary, this application utilizes a portable ECG monitoring patch to collect ECG data in real time, continuously capturing dynamic ECG signals and accelerometer features. The ECG patch is small, comfortable to wear, and has a long battery life, supporting continuous monitoring 24 / 7 without frequent charging or donning / removal. It can promptly detect user anomalies and provide timely warnings. The fall detection model replaces ordinary convolution with depthwise separable convolution, significantly reducing the number of parameters and facilitating deployment on embedded devices. Multi-branch feature fusion enhances feature representation capabilities and improves the model's detection accuracy. The process of dimensionality enhancement, multi-branch extraction, dimensionality reduction, and classification balances model performance and speed. For situations where a user cannot get up quickly after a fall, the fall detection model, after determining a fall, filters out false positives or cases where the user can quickly recover mobility after a fall by detecting the user's activity state and posture. This accurately identifies situations where the user loses mobility after a fall and provides timely warnings, further improving the accuracy of severe fall detection and monitoring user safety and health.
[0059] Secondly, embodiments of this application provide a fall detection system for use in conjunction with an electrocardiogram patch. Figure 3 This is a structural block diagram of a fall detection system according to an embodiment of this application, such as... Figure 3 As shown, the system includes: Acquisition module 100: Used to acquire monitoring parameters output by the ECG patch during the target time period. The monitoring parameters include acceleration signals, activity state characteristics, and posture detection characteristics.
[0060] Model Judgment Module 200: Used to determine whether a user has fallen based on acceleration signals and a fall detection model. The fall detection model is a binary classification model, which is trained based on a depthwise separable convolutional model.
[0061] Output module 300: In response to a user falling, it determines whether the activity state features are static and whether the posture detection features meet the preset posture. If both are met, it outputs "User has fallen". Meeting the preset posture includes being in a non-standing or non-sitting posture.
[0062] In one example, the model structure of the fall detection model in the model judgment module 200 includes: The input layer is used to receive input of multi-dimensional time-series monitoring data.
[0063] The feature extraction module includes an up-dimensional convolutional layer and a multi-branch depthwise separable convolutional layer.
[0064] The feature fusion module includes a feature fusion layer, a Dropout layer, and a dimensionality reduction convolutional layer.
[0065] The output module includes a feature compression layer and a fully connected layer.
[0066] In one example, the fall detection model is configured to use human posture rules as the basis for fall detection.
[0067] In one example, the model judgment module 200 includes: a method for determining whether a user has fallen based on an acceleration signal within a first preset time period using a fall detection model.
[0068] In one example, the output module 300 includes: determining whether the activity state features within a second preset time period are static, wherein the first preset time period and the second preset time period are mutually exclusive; and determining whether the posture detection features within a target time period satisfy a preset posture, wherein satisfying the preset posture includes the presence of a non-standing or non-sitting posture.
[0069] In one example, the system also includes: in response to the failure of either the activity state feature or the posture detection feature, outputting that the user has not fallen, and detecting the monitoring parameters for the next target time period.
[0070] In summary, this application utilizes a portable ECG monitoring patch to collect ECG data in real time, continuously capturing dynamic ECG signals and accelerometer features. The ECG patch is small, comfortable to wear, and has a long battery life, supporting continuous monitoring 24 / 7 without frequent charging or donning / removal. It can promptly detect user anomalies and provide timely warnings. The fall detection model replaces ordinary convolution with depthwise separable convolution, significantly reducing the number of parameters and facilitating deployment on embedded devices. Multi-branch feature fusion enhances feature representation capabilities and improves the model's detection accuracy. The process of dimensionality enhancement, multi-branch extraction, dimensionality reduction, and classification balances model performance and speed. For situations where a user cannot get up quickly after a fall, the fall detection model, after determining a fall, filters out false positives or cases where the user can quickly recover mobility after a fall by detecting the user's activity state and posture. This accurately identifies situations where the user loses mobility after a fall and provides timely warnings, further improving the accuracy of severe fall detection and monitoring user safety and health.
[0071] Thirdly, embodiments of this application provide an electronic device, Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements a fall detection method provided in the first aspect. Figure 4 The electronic device 60 shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.
[0072] Electronic device 60 may be in the form of a general-purpose computing device, such as a server device. Components of electronic device 60 may include, but are not limited to: at least one processor 61, at least one memory 62, and a bus 63 connecting different system components (including memory 62 and processor 61).
[0073] Bus 63 includes a data bus, an address bus, and a control bus.
[0074] The memory 62 may include volatile memory, such as random access memory (RAM) 621 and / or cache memory 622, and may further include read-only memory (ROM) 623.
[0075] The memory 62 may also include a program / utility 625 having a set (at least one) of program modules 624, including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.
[0076] The processor 61 executes various functional applications and data processing by running computer programs stored in the memory 62, such as the fall detection method provided in the first aspect of this application.
[0077] Electronic device 60 can also communicate with one or more external devices 64 (e.g., keyboard, pointing device, etc.). This communication can be performed via input / output (I / O) interface 65. Furthermore, the model-generated electronic device 60 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 66. Figure 4 As shown, network adapter 66 communicates with other modules of the model-generated electronic device 60 via bus 63. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in conjunction with the model-generated electronic device 60, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems.
[0078] It should be noted that although several units / modules or sub-units / modules of the electronic device have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of the present invention, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided and embodied by multiple units / modules.
[0079] Fourthly, embodiments of this application provide a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements a fall detection method provided in the first aspect.
[0080] The readable storage medium may be more specifically adopted, including but not limited to: portable disk, hard disk, random access memory, read-only memory, erasable programmable read-only memory, optical storage device, magnetic storage device, or any suitable combination thereof.
[0081] In a possible implementation, the present invention can also be implemented as a program product comprising program code, which, when the program product is run on a terminal device, causes the terminal device to perform steps implementing a fall detection method provided in the first aspect.
[0082] The program code for executing the present invention can be written in any combination of one or more programming languages. The program code can be executed entirely on the user device, partially on the user device, as a standalone software package, partially on the user device and partially on a remote device, or entirely on a remote device.
[0083] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0084] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.
Claims
1. A fall detection method, characterized in that, The method is used in conjunction with an ECG patch, and the method includes: The monitoring parameters output by the ECG patch during the target time period are obtained, including acceleration signals, activity state characteristics, and posture detection characteristics. The system determines whether a user has fallen based on the acceleration signal and the fall detection model. The fall detection model is a binary classification model, which is trained based on a depthwise separable convolutional model. In response to a user falling, it is determined whether the activity state feature is static and whether the posture detection feature meets a preset posture. If both are met, the user is output as having fallen. Meeting the preset posture includes being in a non-standing or non-sitting posture.
2. The fall detection method according to claim 1, characterized in that, The fall detection model structure includes: The input layer is used to receive input of multi-dimensional time-series monitoring data; The feature extraction module includes an up-dimensional convolutional layer and a multi-branch depthwise separable convolutional layer; The feature fusion module includes a feature fusion layer, a Dropout layer, and a dimension reduction convolutional layer. The output module includes a feature compression layer and a fully connected layer.
3. The fall detection method according to claim 1, characterized in that, The step of determining whether a user has fallen based on the acceleration signal and the fall detection model includes: The fall detection model determines whether a user has fallen based on the acceleration signal within a first preset time period.
4. The fall detection method according to claim 3, characterized in that, The step of determining whether the activity state feature is static and whether the posture detection feature meets a preset posture includes: Determine whether the activity status characteristics within the second preset time period are static; the first preset time period and the second preset time period are mutually exclusive. Determine whether the posture detection features within the target time period meet the preset posture, wherein meeting the preset posture includes the presence of a non-standing posture or a non-sitting posture.
5. The fall detection method according to claim 1, characterized in that, The method further includes: If either the activity state feature or the posture detection feature is not satisfied, the system outputs that the user has not fallen and then detects the monitoring parameters for the next target time period.
6. A fall detection system, characterized in that, The system is used in conjunction with an ECG patch, and the system includes: Acquisition module: used to acquire monitoring parameters output by the ECG patch within the target time period, the monitoring parameters including acceleration signal, activity state characteristics and posture detection characteristics; Model judgment module: used to determine whether the user has fallen based on the acceleration signal and the fall detection model. The fall detection model is a binary classification model, which is trained based on a depthwise separable convolutional model. Output module: In response to a user falling, it determines whether the activity state feature is static and whether the posture detection feature meets a preset posture. If both are met, it outputs that the user has fallen. Meeting the preset posture includes being in a non-standing or non-sitting posture.
7. A fall detection system according to claim 6, characterized in that, The fall detection model structure includes: The input layer is used to receive input of multi-dimensional time-series monitoring data; The feature extraction module includes an up-dimensional convolutional layer and a multi-branch depthwise separable convolutional layer; The feature fusion module includes a feature fusion layer, a Dropout layer, and a dimension reduction convolutional layer. The output module includes a feature compression layer and a fully connected layer.
8. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements a fall detection method as described in any one of claims 1 to 5.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements a fall detection method as described in any one of claims 1 to 5.