A personnel fall alarm device and method based on a wireless sensor network

Abstract

This application relates to the field of personnel fall alarm, and specifically discloses a personnel fall alarm device and method based on a wireless sensor network. It collects the time queue of real-time posture data and static standing posture data of users through an acceleration sensor module and an indoor positioning module, calculates the posture deviation degree between each piece of real-time posture data of users and the static data, uses artificial intelligence-based data coding and processing technologies to perform context time series coding on each posture deviation degree, and at the same time performs embedding coding on each piece of real-time posture data. Then, it performs posture jump degree time series transfer coding on each piece of real-time posture embedding feature of users, so as to automatically generate an identification result according to the multi-dimensional representation between the implicit features of the posture change time series pattern and the posture deviation time series feature after transfer coding, and generate corresponding fall warning prompts.

Description

Technical Field

[0001] This application relates to the field of personnel fall alarm, and specifically relates to a personnel fall alarm device and method based on a wireless sensor network. Background Art

[0002] In today's society, the degree of population aging is increasing, and the number of elderly people living alone is growing. With the decline of physical functions, balance ability and reaction speed, falling has become one of the serious safety hazards they face. This not only brings great pain to themselves, but also increases the medical burden on families and society.

[0003] In response to this, the invention with the publication number CN107909771A proposes a personnel fall alarm system based on a wireless sensor network and its implementation method. It monitors the posture change through the acceleration sensor worn by the user, and records the standing posture as a reference when starting. Calculate and filter the posture deviation degree, count the deviation probability and calculate the acceleration variance to distinguish falls in strenuous exercise and static states. And use the indoor positioning module to confirm the personnel position. Once an accidental fall is confirmed, trigger a local alarm, take a photo of the scene and push information.

[0004] This invention mainly relies on traditional methods such as threshold comparison, variance calculation and statistical analysis to judge falls. This method has limitations in dealing with non-linear or complex posture change patterns, and it is difficult to capture and mine potential subtle changes. Especially in an environment with diverse user behaviors, it may not be able to effectively identify real fall events. Moreover, this method requires manual setting of thresholds and other parameters, which may need to be recalibrated for different users or environments, lacking adaptability, and may lead to an increase in false alarm rate or missed alarm rate in specific situations.

[0005] Therefore, an optimized personnel fall alarm solution based on a wireless sensor network is expected. Summary of the Invention

[0006] This application is made in consideration of the above problems. An object of this application is to provide a personnel fall alarm device and method based on a wireless sensor network.

[0007] This application provides a personnel fall alarm method based on a wireless sensor network, which includes:

[0008] S1: Wear an acceleration sensor module and place an indoor positioning module;

[0009] S2: Prompt the user to stand stably at startup and obtain static standing posture data;

[0010] S3: Collect the user's real-time posture data to obtain a time queue of the user's real-time posture data, and calculate a time queue of the posture deviation degree based on the static standing posture data;

[0011] S4: Perform time series encoding on the time queue of the attitude deviation degree to obtain an implicit encoding vector of attitude deviation time series features;

[0012] S5: Perform embedding encoding on each piece of user real-time attitude data in the time queue of the user real-time attitude data to obtain a time queue of user real-time attitude embedding encoding vectors;

[0013] S6: Perform temporal context encoding of the attitude jump degree on the time queue of the user real-time attitude embedding encoding vectors to obtain an implicit encoding vector of attitude change time series patterns;

[0014] S7: Concatenate the implicit encoding vector of the attitude change time series pattern and the implicit encoding vector of the attitude deviation time series features to obtain a multi-dimensional implicit encoding vector of user attitude time series;

[0015] S8: Obtain an identification result based on the multi-dimensional implicit encoding vector of user attitude time series, and generate a fall warning prompt based on the identification result.

[0016] For example, in the method for personnel fall alarm based on a wireless sensor network according to the present application, wherein calculating the time queue of the attitude deviation degree based on the static standing attitude data includes: calculating the attitude deviation degree of each piece of user real-time attitude data at each predetermined time point in the time queue of the user real-time attitude data relative to the static standing attitude data to obtain the time queue of the attitude deviation degree.

[0017] For example, in the method for personnel fall alarm based on a wireless sensor network according to the present application, wherein S4 includes: using a sequence encoder based on forward LSTM to perform time series encoding on the time queue of the attitude deviation degree to obtain the implicit encoding vector of the attitude deviation time series features.

[0018] For example, in the method for personnel fall alarm based on a wireless sensor network according to the embodiments of the present application, wherein S5 includes: using an attitude embedding encoder based on a fully connected layer to perform fully connected encoding on each piece of user real-time attitude data in the time queue of the user real-time attitude data to obtain the time queue of the user real-time attitude embedding encoding vectors.

[0019] For example, in the method for personnel fall alarm based on a wireless sensor network according to the present application, wherein S6 includes:

[0020] Calculating the message propagation significant factor of each user real-time attitude embedding encoding vector in the time queue of the user real-time attitude embedding encoding vectors to obtain a time queue of user real-time attitude message propagation significant factors;

[0021] Gate the time queue of the significant factors for the real-time user posture message propagation to obtain the time queue of the significant weights for the real-time user posture message propagation;

[0022] Using the time queue of the significant weights for the real-time user posture message propagation, calculate the weighted sum of the time queue of the user real-time posture embedding coding vectors to obtain the implicit coding vector of the posture change time series pattern.

[0023] For example, in the method for personnel fall alarm based on wireless sensor network according to the present application, wherein calculating the significant factors of message propagation for each user real-time posture embedding coding vector in the time queue of the user real-time posture embedding coding vectors to obtain the time queue of the significant factors of user real-time posture message propagation includes:

[0024] Calculate the feature jump degree of each user real-time posture embedding coding vector in the time queue of the user real-time posture embedding coding vectors to obtain the time queue of the user real-time posture feature jump degree;

[0025] Calculate the message propagation space span of each user real-time posture embedding coding vector in the time queue of the user real-time posture embedding coding vectors to obtain the time queue of the user real-time posture message propagation space span;

[0026] Based on the feature jump degree and the message propagation space span of each user real-time posture embedding coding vector in the time queue of the user real-time posture embedding coding vectors, calculate the significant factors of message propagation for each user real-time posture embedding coding vector to obtain the time queue of the significant factors of user real-time posture message propagation.

[0027] For example, in the method for personnel fall alarm based on wireless sensor network according to the present application, wherein gating the time queue of the significant factors for the real-time user posture message propagation to obtain the time queue of the significant weights for the real-time user posture message propagation includes:

[0028] Perform normalization processing on the time queue of the significant factors for the real-time user posture message propagation based on the softmax function to obtain the time queue of the normalized significant factors for the real-time user posture message propagation;

[0029] Perform screening processing on the time queue of the normalized significant factors for the real-time user posture message propagation based on the mask function to obtain the time queue of the significant weights for the real-time user posture message propagation.

[0030] For example, in the method for personnel fall alarm based on wireless sensor network according to the present application, wherein the S8 includes:

[0031] Input the multi - dimensional implicit encoding vector of the user posture time series into a state recognizer based on a classifier to obtain the recognition result;

[0032] In response to the recognition result indicating that the user has fallen, generate a fall warning prompt.

[0033] This application also provides a personnel fall alarm device based on a wireless sensor network, which includes:

[0034] A data acquisition module, configured to prompt the user to stand stably at startup and acquire static standing posture data;

[0035] An acquisition and calculation module, configured to acquire the user's real - time posture data to obtain a time queue of the user's real - time posture data, and calculate a time queue of the posture deviation degree based on the static standing posture data;

[0036] A time - series encoding module, configured to perform time - series encoding on the time queue of the posture deviation degree to obtain a posture deviation time - series feature implicit encoding vector;

[0037] An embedding encoding module, configured to perform embedding encoding on each piece of the user's real - time posture data in the time queue of the user's real - time posture data to obtain a time queue of the user's real - time posture embedding encoding vectors;

[0038] A time - series context encoding module, configured to perform time - series context encoding of the posture jump degree on the time queue of the user's real - time posture embedding encoding vectors to obtain a posture change time - series pattern implicit encoding vector;

[0039] A vector splicing module, configured to splice the posture change time - series pattern implicit encoding vector and the posture deviation time - series feature implicit encoding vector to obtain a multi - dimensional implicit encoding vector of the user posture time series;

[0040] A fall warning prompt module, configured to obtain a recognition result based on the multi - dimensional implicit encoding vector of the user posture time series, and generate a fall warning prompt based on the recognition result.

[0041] For example, in the personnel fall alarm device based on a wireless sensor network according to this application, the acquisition and calculation module is configured to: calculate the posture deviation degree of the user's real - time posture data at each predetermined time point in the time queue of the user's real - time posture data relative to the static standing posture data to obtain the time queue of the posture deviation degree.

[0042] The personnel fall alarm device and method based on a wireless sensor network according to the present application collect the time queue of real-time posture data and static standing posture data of a user through an acceleration sensor module and an indoor positioning module, calculate the posture deviation degree between each piece of real-time posture data of the user and the static data, and use artificial intelligence-based data encoding and processing technologies to perform context time series encoding on each posture deviation degree, and at the same time perform embedding encoding on each piece of real-time posture data. Then, perform posture jump degree time series transfer encoding on each piece of real-time posture embedding feature of the user, so as to automatically generate an identification result according to the multi-dimensional representation between the implicit features of the posture change time series pattern and the posture deviation time series feature after transfer encoding, and generate a corresponding fall warning prompt. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments of the present application will be briefly introduced below. Obviously, the drawings described below only relate to some embodiments of the present application and do not limit the present application.

[0044] Figure 1 Shows a flowchart of a personnel fall alarm method based on a wireless sensor network in an embodiment of the present application;

[0045] Figure 2 Shows a flowchart of sub-step S6 of the personnel fall alarm method based on a wireless sensor network in an embodiment of the present application;

[0046] Figure 3 Shows a flowchart of sub-step S61 of the personnel fall alarm method based on a wireless sensor network in an embodiment of the present application;

[0047] Figure 4 Shows a flowchart of sub-step S62 of the personnel fall alarm method based on a wireless sensor network in an embodiment of the present application;

[0048] Figure 5 Shows a schematic structural diagram of a personnel fall alarm device based on a wireless sensor network in an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0049] The terms used in this specification are those general terms that are currently widely used in the art in consideration of the functions of the present application, but these terms may change according to the intentions of those of ordinary skill in the art, precedents, or new technologies in the art. In addition, specific terms can be selected, and in this case, their detailed meanings will be described in the detailed description of the present application. Therefore, the terms used in the specification should not be understood as simple names, but based on the meanings of the terms and the overall description of the present application.

[0050] Although the present application makes various references to certain modules in the system according to embodiments of the present application, any number of different modules may be used and run on the user terminal and / or the server. The modules are merely illustrative, and different aspects of the system and method may use different modules.

[0051] Flowcharts are used in the present application to illustrate the operations performed by the system according to embodiments of the present application. It should be understood that the operations above or below do not necessarily have to be performed precisely in order. Instead, various steps may be processed in reverse order or simultaneously as needed. At the same time, other operations may also be added to these processes, or one or more steps may be removed from these processes.

[0052] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall also fall within the scope of protection of the present application.

[0053] Based on this, the present application proposes a method for personnel fall alarm based on a wireless sensor network. Its technical concept is to collect the time queue of real-time posture data and static standing posture data of users through an acceleration sensor module and an indoor positioning module, and calculate the posture deviation degree between each piece of real-time posture data of the user and the static data. The data encoding and processing technology based on artificial intelligence is used to perform context time series encoding on each posture deviation degree, and at the same time, embedding encoding is performed on each piece of real-time posture data. Then, posture jump degree time series transfer encoding is performed on each piece of real-time posture embedding feature of the user, so as to automatically generate an identification result according to the multi-dimensional representation between the posture change time series pattern implicit feature and the posture deviation time series feature after the transfer encoding, and generate a corresponding fall warning prompt. The present application can accurately capture potential small changes in posture data and efficiently identify non-linear and complex posture patterns. Moreover, it can automatically adapt to the behavior habits of different users without frequently manually calibrating thresholds and other parameters, ensuring consistency and reliability in diverse environments.

[0054] Figure 1 The flowchart of the method for personnel fall alarm based on a wireless sensor network in the embodiments of the present application is shown. As Figure 1As shown, the method for personnel fall alarm based on wireless sensor network according to the embodiment of the present application includes the steps of: S1: Wear an acceleration sensor module and place an indoor positioning module; S2: Prompt the user to stand stably at startup and obtain static standing posture data; S3: Collect the real-time posture data of the user to obtain a time queue of the real-time posture data of the user, and calculate a time queue of the posture deviation degree based on the static standing posture data; S4: Perform time series encoding on the time queue of the posture deviation degree to obtain a posture deviation time series feature implicit encoding vector; S5: Perform embedding encoding on each piece of real-time posture data in the time queue of the real-time posture data of the user to obtain a time queue of real-time posture embedding encoding vectors of the user; S6: Perform time series context encoding of the posture jump degree on the time queue of the real-time posture embedding encoding vectors of the user to obtain a posture change time series pattern implicit encoding vector; S7: Concatenate the posture change time series pattern implicit encoding vector and the posture deviation time series feature implicit encoding vector to obtain a multi-dimensional implicit encoding vector of the user posture time series; S8: Based on the multi-dimensional implicit encoding vector of the user posture time series, obtain an identification result, and generate a fall warning prompt based on the identification result.

[0055] Specifically, in the technical solution of the present application, first, wear an acceleration sensor module and place an indoor positioning module.

[0056] Next, prompt the user to stand stably at startup and obtain static standing posture data. (The system prompts the user to maintain a standing posture and remain stationary for a period of time (e.g., 5 seconds). During this period, the acceleration sensor collects the components of the gravitational acceleration on the X, Y, and Z axes at a high frequency (e.g., 50Hz). The collected data will be used to calculate the average value, so as to obtain the distribution of the gravitational acceleration in the three-dimensional coordinate system in the current state. This set of data is regarded as "static standing data", which represents the body posture characteristics of the user when standing normally.)

[0057] Then, collect the user's real-time posture data to obtain a time queue of the user's real-time posture data. Specifically, in the method for personnel fall alarm based on wireless sensor network, data collection can be completed by the acceleration sensor module. This module is worn at the waist position of the person. When the system starts and completes the initial static standing posture recognition, it enters the real-time monitoring stage. In this stage, the acceleration sensor continuously collects the triaxial acceleration information of the wearer at a frequency of 50Hz. These information reflect the distribution of gravitational acceleration in the coordinate system of the acceleration sensor itself and change with the change of the wearer's body posture. Each sampled acceleration value is decomposed into specific values in the three dimensions of X, Y, and Z, respectively represented as X(i), Y(i), and Z(i), where i represents the index of the i-th sampling. These data will then be sent by the acceleration sensor module to the local server for processing through the ZigBee protocol. As the data continuously flows in, the local server stores the acceleration components at each moment received to form a time queue containing the user's real-time posture data. To ensure the validity and accuracy of the data, the server also performs mean filtering on the posture deviation degree in the recent period to reduce the influence of slight jitter caused by background noise interference. For this purpose, the server uses a circular queue array H with a length of 150 to store the filtered posture deviation degree data in the past 3 seconds. Every time a new sample is received, this array is updated, replacing the oldest data with the latest data, so as to ensure that all elements in the array are the measurement results in the latest 3 seconds. In this way, through continuous real-time posture data collection and processing, the system can construct a time series that accurately reflects the user's posture changes, laying a foundation for the calculation of the implicit coding vector of the posture deviation time series feature and the implicit coding vector of the posture change time series pattern, and then realizing the timely detection and early warning of potential fall events.

[0058] Next, considering that the static standing posture data represents the posture of the user in the normal static state, while the user's real-time posture data reflects the posture in the actual activity. Since the human body's posture will continuously change during activities, such as from standing to walking, turning, etc. Therefore, in the technical solution of this application, based on the static standing posture data, calculating the time queue of the posture deviation degree includes: calculating the posture deviation degree of the user's real-time posture data at each predetermined time point in the time queue of the user's real-time posture data relative to the static standing posture data to obtain the time queue of the posture deviation degree. That is to say, the posture deviation degree quantifies the difference between the user's posture and the reference at each time point. In this way, these changes in different user behavior patterns can be effectively captured, especially the subtle but continuous changes, which may be the early signals of abnormal situations such as falls.

[0059] It should be understood that the process of calculating the posture deviation degree begins with obtaining static standing posture data, which are collected within the first 5 seconds after the user activates the acceleration sensor module. During this period, the system will prompt the user to stand still and stably. The 250 acceleration data within these 5 seconds (because the sampling frequency is 50Hz) are processed to generate the average acceleration values on each axis, and these average values represent the posture reference of the user when standing still. Once the data of the static standing posture are available as a reference, the system can start to monitor the posture changes in the user's activities in real time. In the real-time monitoring stage, the acceleration sensor sends new three-axis acceleration data to the local server at a frequency of 50Hz every second. For the real-time acceleration data collected at each time point, the system will compare it with the average value when standing still to evaluate the current posture change. Specifically, the system will compare the difference between the real-time data and the previously obtained static standing average value to obtain the differences on each axis. These differences reflect the posture changes experienced by the user from the standing state to the current state. In this way, the deviation degree between the user's posture and the reference standing posture at each time point can be quantified, and this deviation degree can also be called the posture deviation degree. To ensure the accuracy and stability of the posture deviation degree, considering that the acceleration sensor may be affected by environmental noise, the system will smooth the posture deviation degree over a period of time. This processing method is similar to taking an average value, but it is more complex: instead of simply averaging all the data, it selects the posture deviation degrees at the last 10 time points and calculates their average value. This helps to reduce the impact of instantaneous jitters or short-term anomalies on the results. After that, this smoothed posture deviation degree will be stored in a special queue, and this queue only stores the data within the latest 3 seconds. Whenever a new posture deviation degree is calculated, it will replace the oldest data in the queue to ensure that the information in the queue is always the latest. When the system has accumulated enough time series of posture deviation degrees, it can start to analyze whether there are abnormal patterns in these data. For example, if it is found that the posture deviation degree increases significantly within a period of time and this state persists, it may mean that the user is performing abnormal actions, such as suddenly falling. However, some normal activities in daily life, such as squatting and standing up, will also cause a temporary increase in the posture deviation degree. Therefore, the system not only pays attention to the posture deviation degree itself, but also considers its change trend over time. If the increase in the posture deviation degree is accompanied by a low acceleration fluctuation, that is, the user's posture becomes relatively stable, then the system will further use the indoor positioning module to confirm the user's location to determine whether a fall has really occurred. In this way, the system can distinguish the common posture changes in daily activities from real fall events and thus decide whether to trigger an alarm.

[0060] Then, considering that the time queue of the attitude deviation degree is time series data, and there is a chronological order and mutual dependence relationship among its various data points. Therefore, in order to capture and mine the correlation effects between the data at each time point, in the technical solution of this application, a sequence encoder based on forward LSTM is used to perform time series encoding on the time queue of the attitude deviation degree to better capture the behavior patterns within a long time span, and obtain the implicit encoding vector of the attitude offset time series features. That is to say, forward LSTM is particularly suitable for processing this kind of time series data with long-term dependence relationships. By introducing structures such as cell state, input gate, forget gate, and output gate, it can better control the flow of information in the time series, remember long-term information, and thus accurately model the complex patterns and dependence relationships in the attitude deviation degree time series to better characterize the dynamic patterns and change trends of the attitude offset features. Correspondingly, S4 includes: using a sequence encoder based on forward LSTM to perform time series encoding on the time queue of the attitude deviation degree to obtain the implicit encoding vector of the attitude offset time series features.

[0061] Subsequently, each user's real-time attitude data contains information in multiple dimensions (such as the components of acceleration in different directions, etc.), and these components usually have a high dimension, and there is a certain correlation relationship between them. Based on this, in the technical solution of this application, a posture embedding encoder based on a fully connected layer is used to perform fully connected encoding on each user's real-time attitude data in the time queue of the user's real-time attitude data to map these high-dimensional data to a lower-dimensional space, and obtain a time queue of user real-time attitude embedding encoding vectors. That is, through fully connected encoding, the potential relationships between different dimensions can be mined, and richer and more detailed features in the attitude data can be captured. Correspondingly, S5 includes: using a posture embedding encoder based on a fully connected layer to perform fully connected encoding on each user's real-time attitude data in the time queue of the user's real-time attitude data to obtain a time queue of user real-time attitude embedding encoding vectors.

[0062] Furthermore, considering that there is a correlation influence relationship in the time scale among the real-time attitude embedding encoding vectors of each user. For example, when walking normally and falling, the patterns such as the rhythm and amplitude of attitude changes are different. Therefore, in order to capture the relationship between the attitude embedding vectors at adjacent and non-adjacent time points, discover the potential patterns of attitude changes, and thus better understand the dynamic changes of the entire attitude, this application performs temporal context encoding of attitude jump degree on the time queue of the real-time attitude embedding encoding vectors of the user to obtain an implicit encoding vector of the attitude change time series pattern. In particular, the temporal context encoding of attitude jump degree can transmit information across time steps, capture long-term and short-term dependence relationships, comprehensively analyze the attitude change process, and avoid ignoring important information due to only considering short-term dependence.

[0063] Next, as Figure 2 shown, in step S6, the S6 includes: S61, calculating the message propagation significant factor of each user real-time pose embedding coding vector in the time queue of the user real-time pose embedding coding vector to obtain the time queue of the user real-time pose message propagation significant factor; S62, performing gated propagation on the time queue of the user real-time pose message propagation significant factor to obtain the time queue of the user real-time pose message propagation significant weight; S63, using the time queue of the user real-time pose message propagation significant weight to calculate the weighted sum of the time queue of the user real-time pose embedding coding vector to obtain the implicit coding vector of the pose change time series pattern.

[0064] Among them, as Figure 3 shown, in step S61, calculating the message propagation significant factor of each user real-time pose embedding coding vector in the time queue of the user real-time pose embedding coding vector to obtain the time queue of the user real-time pose message propagation significant factor includes: S611, calculating the feature jump degree of each user real-time pose embedding coding vector in the time queue of the user real-time pose embedding coding vector to obtain the time queue of the user real-time pose feature jump degree; S612, calculating the message propagation spatial span of each user real-time pose embedding coding vector in the time queue of the user real-time pose embedding coding vector to obtain the time queue of the user real-time pose message propagation spatial span; S613, based on the feature jump degree and the message propagation spatial span of each user real-time pose embedding coding vector in the time queue of the user real-time pose embedding coding vector, calculating the message propagation significant factor of each user real-time pose embedding coding vector to obtain the time queue of the user real-time pose message propagation significant factor.

[0065] Among them, as Figure 4 shown, in step S62, performing gated propagation on the time queue of the user real-time pose message propagation significant factor to obtain the time queue of the user real-time pose message propagation significant weight includes: S621, performing normalization processing based on the softmax function on the time queue of the user real-time pose message propagation significant factor to obtain the time queue of the normalized user real-time pose message propagation significant factor; S622, performing screening processing based on the mask function on the time queue of the normalized user real-time pose message propagation significant factor to obtain the time queue of the user real-time pose message propagation significant weight.

[0066] Specifically, first, the degree of feature change between nodes is quantified. This step is achieved by calculating the user real-time pose feature jump degree of each user real-time pose embedding encoding vector in the time queue of the user real-time pose embedding encoding vectors. The user real-time pose feature jump degree reflects the distance or difference in the node feature space, which helps to capture local structural information and the mutation characteristics between nodes. Through the user real-time pose feature jump degree, important characteristics of each node feature in the feature sequence distribution can be revealed, such as boundaries, outliers, etc., so that not only can it identify which node features have large changes in the network, but also emphasizes the nodes that may have an important impact on the propagation dynamics.

[0067] This process can be expressed by the formula:

[0068] ;

[0069] ;

[0070] ;

[0071] Among them, is the time queue of the user real-time pose embedding encoding vectors, , , and are respectively the 1st, 2nd, th, and th user real-time pose embedding encoding vectors in the time queue of the user real-time pose embedding encoding vectors, is the eigenvalue at the th position in , is the number of eigenvalues in , and are respectively the th and th user real-time pose embedding feature weighted average eigenvalues in the time queue of the user real-time pose embedding feature weighted average eigenvalues, is the th user real-time pose feature jump degree in the time queue of the user real-time pose feature jump degree.

[0072] Next, when evaluating the propagation ability of information in the network, the system calculates the user real-time pose message propagation space span of each user real-time pose embedding encoding vector in the time queue of the user real-time pose embedding encoding vectors, and obtains the user real-time pose message propagation space span. This step goes beyond simple adjacency relationships and takes into account broader connectivity and path lengths. The user real-time pose message propagation space span helps to identify long-distance dependencies and key paths of information flow.

[0073] This process can be expressed by the formula as follows: ;

[0074] Wherein, and respectively represent the th and the th user real-time pose embedding encoding vectors in the time queue of the user real-time pose embedding encoding vectors, represents to the number of message propagation times, is the th user real-time pose message propagation spatial span in the time queue of the user real-time pose message propagation spatial span.

[0075] Combining the information obtained from the above two steps, the next step is to calculate the user real-time pose message propagation significance factor for each user real-time pose embedding encoding vector, forming a time queue of the user real-time pose message propagation significance factor. This stage comprehensively considers the user real-time pose feature jump degree and the user real-time pose message propagation spatial span to determine which nodes play a more important role in the information propagation process. The user real-time pose message propagation significance factor is similar to the attention mechanism, where nodes adjust the way of information exchange according to their own and their neighbors' importance. In this way, while ensuring that important nodes are highlighted and the overall structural information is retained, a score reflecting the importance of each node in the information propagation process is assigned to each node.

[0076] This process can be expressed by the formula as follows: ;

[0077] Wherein, is the th user real-time pose feature jump degree in the time queue of the user real-time pose feature jump degree, is the th user real-time pose message propagation spatial span in the time queue of the user real-time pose message propagation spatial span, and are modulation parameters, is the th user real-time pose message propagation significance factor in the time queue of the user real-time pose message propagation significance factor.

[0078] Subsequently, the time queue of the significant factors of the user's real-time pose message propagation is input into the gated transfer unit to obtain the time queue of the significant weights of the user's real-time pose message propagation. The gated transfer unit plays the role of regulating the information flow here, controlling which information should be emphasized or suppressed through the learned significant weights of the user's real-time pose message propagation. The gating mechanism draws on the concepts of the forget gate and input gate in the recurrent neural network, allowing the model to selectively remember or ignore part of the input. Accordingly, the gated transfer unit can enhance the attention to the features of different nodes by introducing additional parameters, thereby improving the expressiveness of the model. Such a design can not only adjust the information flow but also ensure that only the most relevant information passes through, while the less relevant features are appropriately weakened. Finally, based on the significant weights of the user's real-time pose message propagation obtained in the previous steps, the weighted sum of the time queue of the user's real-time pose embedding coding vectors is calculated to obtain the implicit coding vector of the pose change time series pattern.

[0079] This process can be expressed by the formula:

[0080] ;

[0081] ;

[0082] where, is the nth significant factor of the user's real-time pose message propagation in the time queue of the significant factors of the user's real-time pose message propagation, is the normalization function, is the masking operation, is the predetermined threshold, is the mth significant weight of the user's real-time pose message propagation in the time queue of the significant weights of the user's real-time pose message propagation, represents the kth user's real-time pose embedding coding vector in the time queue of the user's real-time pose embedding coding vectors, is the number of vectors in the time queue of the user's real-time pose embedding coding vectors, is the implicit coding vector of the pose change time series pattern. is the number of vectors in the time queue of the user's real-time pose embedding coding vectors, is the number of vectors in the time queue of the user's real-time pose embedding coding vectors, is the implicit coding vector of the pose change time series pattern.

[0083] Then, considering that the implicit coding vector of the posture change time series pattern mainly reflects the dynamic change pattern of the posture in the time series, reflecting the continuity and evolution trend of the action. And the implicit coding vector of the posture offset time series feature focuses on reflecting the change of the deviation degree of the real-time posture relative to the static standing posture over time. These two vectors describe the user's posture information from different angles. If only relying on one feature, it may not be possible to accurately judge the fall situation. For example, the posture change pattern may be similar during some strenuous exercises and falls, but the posture offset degree may be significantly different. Based on this, in the technical solution of this application, the implicit coding vector of the posture change time series pattern and the implicit coding vector of the posture offset time series feature are spliced to obtain a multi-dimensional implicit coding vector of the user's posture time series. In this way, the obtained multi-dimensional implicit coding vector of the user's posture time series combines the time pattern of the posture change and the features of the posture offset, forming a more comprehensive and detailed feature representation. This representation not only includes the dynamic characteristics of the user's behavior but also reflects its change relative to the reference posture, providing a solid foundation for subsequent analysis.

[0084] After that, the multi-dimensional implicit coding vector of the user's posture time series is input into a state recognizer based on a classifier to obtain a recognition result. That is, classification processing is performed using the multi-dimensional implicit coding vector of the user's posture time series obtained by splicing the implicit coding vector of the posture change time series pattern and the implicit coding vector of the posture offset time series feature. In this way, the classifier learns the feature combinations corresponding to different posture patterns according to the features represented by each dimension in the vector, so as to accurately classify it into different state categories and obtain a recognition result.

[0085] Correspondingly, step S8 includes: inputting the multi-dimensional implicit coding vector of the user's posture time series into a state recognizer based on a classifier to obtain the recognition result; in response to the recognition result indicating that the user has fallen, generating a fall warning prompt.

[0086] Preferably, inputting the multi-dimensional implicit coding vector of the user's posture time series into a state recognizer based on a classifier to obtain the recognition result includes:

[0087] Determining the multi-dimensional implicit probability value of the user's posture time series obtained by inputting the multi-dimensional implicit coding vector of the user's posture time series into a state recognizer based on a classifier , and the multi-dimensional implicit probability value of the user's posture time series represents the probability of the user falling;

[0088] Multiplying the feature mean value of the multi-dimensional implicit coding vector of the user's posture time series by the multi-dimensional implicit probability value of the user's posture time series to obtain the multi-dimensional implicit statistical field value of the user's posture time series ; where represents the feature mean value of the multi-dimensional implicit coding vector of the user's posture time series, represents the multi-dimensional implicit probability value of the user posture time series, represents the multi-dimensional implicit statistical field value of the user posture time series;

[0089] Subtract one from the multi-dimensional implicit statistical field value of the user posture time series and then divide it by the multi-dimensional implicit statistical field value of the user posture time series to obtain the multi-dimensional implicit partial probability value of the user posture time series ; where, represents the multi-dimensional implicit statistical field value of the user posture time series, represents the multi-dimensional implicit partial probability value of the user posture time series;

[0090] Calculate the power function with the multi-dimensional implicit coding vector of the user posture time series as the exponent with the multi-dimensional implicit partial probability value of the user posture time series , and multiply it point by point with the multi-dimensional implicit partial probability value of the user posture time series to obtain the multi-dimensional implicit microscopic representation vector of the user posture time series ; where, represents the multi-dimensional implicit coding vector of the user posture time series, represents the multi-dimensional implicit partial probability value of the user posture time series, represents the power function with the multi-dimensional implicit coding vector of the user posture time series as the exponent with the multi-dimensional implicit partial probability value of the user posture time series, represents point multiplication, represents the multi-dimensional implicit microscopic representation vector of the user posture time series;

[0091] After multiplying the multi-dimensional implicit coding vector of the user posture time series and the multi-dimensional implicit partial probability value of the user posture time series point by point, calculate the exponential function with the natural constant as the base to obtain the multi-dimensional implicit macroscopic mapping vector of the user posture time series ; where, represents the multi-dimensional implicit coding vector of the user posture time series, represents the multi-dimensional implicit partial probability value of the user posture time series, represents point multiplication, represents calculating the exponential function with the natural constant as the base, represents the multi-dimensional implicit macroscopic mapping vector of the user posture time series;

[0092] After calculating the logarithm with base 2 of the multi-dimensional implicit microscopic representation vector of the user posture time series, perform weighted summation with the multi-dimensional implicit macroscopic mapping vector of the user posture time series to obtain the optimized multi-dimensional implicit coding vector of the user posture time series ; where, represents the multi-dimensional implicit microscopic representation vector of the user posture time series, denotes the multi-dimensional implicit macro mapping vector of the user posture time series; denotes dot product; denotes dot addition; and is the weighted modulation parameter; denotes the optimized multi-dimensional implicit coding vector of the user posture time series;

[0093] Input the optimized multi-dimensional implicit coding vector of the user posture time series into the state recognizer based on the classifier to obtain the recognition result.

[0094] Since each user real-time posture embedding coding vector in the time queue of the user real-time posture embedding coding vector respectively represents the embedding coding features of the user real-time posture data in the local time domain, when performing feature sequence message passing based on the feature jump degree, there will be a micro jump degree - macro sequence message passing deviation of the embedding coding features in the local time domain relative to the global time domain, and when splicing the posture change time series pattern implicit coding vector and the posture offset time series feature implicit coding vector, a local - global feature space deviation in the feature space of the posture deviation degree time domain is further introduced. Thus, when inputting the multi-dimensional implicit coding vector of the user posture time series into the state recognizer based on the classifier, it causes a dynamic deviation of the feature mapping to the class target probability, reducing the accuracy of the obtained recognition result.

[0095] Therefore, through the partial low-order derivative of the statistical distribution field corresponding to the multi-dimensional implicit coding vector of the user posture time series, as the non-overlapping macro feature representation behavior patch of the multi-dimensional implicit coding vector of the user posture time series, based on the different macro behavior patch organization spaces under the non-isotropic backbone structure of the multi-dimensional implicit coding vector of the user posture time series, to strengthen the dynamic sensitivity of the long sequence micro complex information distribution of the multi-dimensional implicit coding vector of the user posture time series to the macro representation behavior of the class probability, thereby promoting the class target iterative dynamic consistency between the classification target and the extracted features during the feature space - class probability mapping, so as to improve the accuracy of the recognition result obtained by inputting the multi-dimensional implicit coding vector of the user posture time series into the state recognizer based on the classifier.

[0096] Subsequently, in response to the recognition result that the user has fallen, a fall warning prompt is generated. It should be understood that a person falling may cause serious physical injuries, especially for specific groups such as the elderly and patients. Timely detection and response to fall events can minimize the risk of injury. Generating a fall warning prompt is to quickly take measures to ensure the life, health and safety of the user when a fall situation is detected.

[0097] It is worth mentioning that after the system confirms a fall, it immediately takes action. First is the local alarm: the server sends a signal to the acceleration sensor worn by the elderly person, activating the built-in buzzer to emit an alarm sound to alert people nearby. At the same time, the server also notifies the wireless image sensor module installed at home. The wireless image sensor module has a camera with the function of sound source localization. It can locate the position of the elderly person according to the sound of the buzzer and take a photo, which is then transmitted to the local server as an important basis for subsequent situation assessment. To further provide help information, the environmental perception module is also activated to collect environmental parameters such as indoor temperature and humidity, carbon monoxide concentration, smoke concentration, and PM2.5 concentration. These data are very important for doctors and can help them initially judge the cause of the fall, such as whether there is gas leakage or air quality problems. All the collected information - including the time of the fall, the taken photo, and the environmental conditions - will be integrated by the local server into a detailed report. Finally, this report containing key information is pushed to the preset remote client via the Internet, such as the mobile phone or computer of the family members or medical staff. In this way, even people not on the scene can learn about the situation of the elderly person in a timely manner and can make timely responses accordingly, such as contacting the emergency service or going to check. Throughout the process, from detecting the fall to generating a warning prompt and notifying relevant personnel, the response speed of the system is very fast, ensuring that necessary rescue measures can be taken quickly in case of an emergency, thus guaranteeing the health and safety of the user.

[0098] Based on the above embodiments, refer to Figure 5As shown in the figure, it is a schematic structural diagram of a personnel fall alarm device 100 based on a wireless sensor network in an embodiment of the present application. The personnel fall alarm device 100 based on a wireless sensor network includes: a data acquisition module 110, configured to prompt the user to stand stably at startup and acquire static standing posture data; a collection and calculation module 120, configured to collect the user's real-time posture data to obtain a time queue of the user's real-time posture data, and calculate a time queue of the posture deviation degree based on the static standing posture data; a time series encoding module 130, configured to perform time series encoding on the time queue of the posture deviation degree to obtain an implicit encoding vector of the posture deviation time series feature; an embedding encoding module 140, configured to perform embedding encoding on each piece of the user's real-time posture data in the time queue of the user's real-time posture data to obtain a time queue of the user's real-time posture embedding encoding vectors; a time series context encoding module 150, configured to perform time series context encoding of the posture jump degree on the time queue of the user's real-time posture embedding encoding vectors to obtain an implicit encoding vector of the posture change time series pattern; a vector splicing module 160, configured to splice the implicit encoding vector of the posture change time series pattern and the implicit encoding vector of the posture deviation time series feature to obtain a multi-dimensional implicit encoding vector of the user's posture time series; a fall warning prompt module 170, configured to obtain an identification result based on the multi-dimensional implicit encoding vector of the user's posture time series, and generate a fall warning prompt based on the identification result. Wherein, the collection and calculation module 120 is configured to: calculate the posture deviation degree of each piece of the user's real-time posture data at each predetermined time point in the time queue of the user's real-time posture data relative to the static standing posture data to obtain the time queue of the posture deviation degree.

[0099] Here, those skilled in the art can understand that the specific functions and operations of each module in the above-mentioned personnel fall alarm device 100 based on a wireless sensor network have been introduced in detail in the description of the Figures 1 to 4 personnel fall alarm method based on a wireless sensor network above, and therefore, the repeated description thereof will be omitted.

[0100] Those of ordinary skill in the art can understand that all or part of the steps in the above method can be instructed by a program to complete the relevant hardware, and the program can be stored in a computer-readable storage medium, such as a read-only memory, a magnetic disk, or an optical disc, etc. Optionally, all or part of the steps in the above embodiments can also be implemented using one or more integrated circuits. Correspondingly, each module / unit in the above embodiments can be implemented in the form of hardware or in the form of a software functional module. The present application is not limited to any specific form of combination of hardware and software.

[0101] Unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. It should also be understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and should not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

[0102] The foregoing is a description of the present application and should not be taken as a limitation thereof. Although several exemplary embodiments of the present application have been described, those skilled in the art will readily appreciate that many modifications can be made to the exemplary embodiments without departing from the novel teachings and advantages of the present application.

Claims

1. A method for alarming a person falling down based on a wireless sensor network, characterized in that: include: S1: Wear the acceleration sensor module and place the indoor positioning module; S2: When starting, prompt the user to stand steadily and obtain static standing posture data; S3: collecting the user's real-time posture data to obtain a time queue of the user's real-time posture data, and calculating a time queue of posture deviation based on the static standing posture data; S4: performing time series encoding on the time queue of the posture deviation to obtain an implicit encoding vector of the posture deviation time series feature; S5: embedding and encoding each user real-time posture data in the time queue of the user real-time posture data to obtain a time queue of user real-time posture embedding coding vectors; S6: performing temporal context coding of the gesture jump degree on the time queue of the user's real-time gesture embedded coding vector to obtain a gesture change temporal pattern implicit coding vector; S7: concatenating the posture change timing pattern implicit coding vector and the posture offset timing feature implicit coding vector to obtain a user posture timing multi-dimensional implicit coding vector; S8: obtaining a recognition result based on the multi-dimensional implicit coding vector of the user posture time series, and generating a fall warning prompt based on the recognition result; Wherein, said S6 comprises: Calculate the message propagation significance factor of each user real-time posture embedding coding vector in the time queue of the user real-time posture embedding coding vector to obtain the time queue of the user real-time posture message propagation significance factor; Perform gated propagation on the time queue of the user's real-time posture message propagation significance factor to obtain the time queue of the user's real-time posture message propagation significance weight; Using the time queue of the significant weight of the user's real-time posture message propagation, calculate the weighted sum of the time queue of the user's real-time posture embedded coding vector to obtain the implicit coding vector of the posture change timing pattern; The process of calculating the message propagation significance factor of each user real-time posture embedding coding vector in the time queue of the user real-time posture embedding coding vector to obtain the time queue of the user real-time posture message propagation significance factor includes: Calculate the feature jump degree of each user real-time posture embedding coding vector in the time queue of the user real-time posture embedding coding vector to obtain the time queue of the user real-time posture feature jump degree; Calculating the message propagation space span of each user real-time posture embedding coding vector in the time queue of the user real-time posture embedding coding vector to obtain the time queue of the user real-time posture message propagation space span; Based on the characteristic jump degree and message propagation space span of each user real-time posture embedding coding vector in the time queue of the user real-time posture embedding coding vector, the message propagation significance factor of each user real-time posture embedding coding vector is calculated to obtain the time queue of the user real-time posture message propagation significance factor.

2. The method for alarming a person falling down based on a wireless sensor network according to claim 1 is characterized in that: Based on the static standing posture data, a time queue of posture deviation is calculated, including: calculating the posture deviation of the user's real-time posture data at each predetermined time point in the time queue of the user's real-time posture data relative to the static standing posture data to obtain the time queue of the posture deviation.

3. The method for alarming a person falling down based on a wireless sensor network according to claim 2 is characterized in that: The S4 includes: using a forward LSTM-based sequence encoder to perform time series encoding on the time queue of the posture deviation to obtain the implicit encoding vector of the posture deviation time series feature.

4. The method for alarming a person falling down based on a wireless sensor network according to claim 3 is characterized in that: The S5 includes: using a posture embedding encoder based on a fully connected layer to perform fully connected encoding on each user real-time posture data in the time queue of the user real-time posture data to obtain the time queue of the user real-time posture embedding coding vector.

5. The method for alarming a person falling down based on a wireless sensor network according to claim 4 is characterized in that: Performing gated propagation on the time queue of the user's real-time posture message propagation significance factor to obtain the time queue of the user's real-time posture message propagation significance weight, including: Performing a normalization process based on a softmax function on the time queue of the user's real-time posture message propagation significance factor to obtain a normalized time queue of the user's real-time posture message propagation significance factor; The time queue of the normalized user real-time posture message propagation significance factor is screened based on a mask function to obtain the time queue of the user real-time posture message propagation significance weight.

6. The method for alarming a person falling down based on a wireless sensor network according to claim 5 is characterized in that: The S8 comprises: Inputting the user posture temporal multi-dimensional implicit coding vector into a classifier-based state recognizer to obtain the recognition result; In response to the recognition result that the user has fallen, a fall warning prompt is generated.

7. A personnel fall alarm device based on a wireless sensor network according to the personnel fall alarm method based on a wireless sensor network according to claim 1, characterized in that: include: A data acquisition module, used to prompt the user to stand steadily and obtain static standing posture data when starting; A collection and calculation module, used for collecting the user's real-time posture data to obtain a time queue of the user's real-time posture data, and calculating a time queue of posture deviation based on the static standing posture data; A time series encoding module, used for performing time series encoding on the time queue of the posture deviation to obtain an implicit encoding vector of the posture deviation time series feature; An embedding coding module, used for embedding coding each user real-time posture data in the time queue of the user real-time posture data to obtain a time queue of user real-time posture embedding coding vectors; A temporal context coding module, used for performing temporal context coding of the gesture jump degree on the time queue of the user's real-time gesture embedding coding vector to obtain an implicit coding vector of the gesture change temporal pattern; A vector concatenation module, used for concatenating the posture change timing pattern implicit coding vector and the posture offset timing feature implicit coding vector to obtain a user posture timing multi-dimensional implicit coding vector; The fall warning prompt module is used to obtain a recognition result based on the multi-dimensional implicit coding vector of the user posture time sequence, and generate a fall warning prompt based on the recognition result.

8. The personnel fall alarm device based on wireless sensor network according to claim 7 is characterized in that: The collection and calculation module is used to calculate the posture deviation of the user's real-time posture data at each predetermined time point in the time queue of the user's real-time posture data relative to the static standing posture data to obtain the time queue of the posture deviation.

Images