Intelligent fall detection method based on wireless signal

Abstract

The invention provides an intelligent fall detection method based on wireless signals, and belongs to the technical field of intelligent detection, and the method comprises the steps: arranging a plurality of wireless signal receiving and transmitting devices in a target detection region, continuously collecting each wireless signal in the target region, determining a reflection path and a refraction path of each wireless signal, and transmitting the reflection path and the refraction path to the target detection region; determining a sharp attenuation point and a thermal value of the sharp attenuation point; original signal data carrying human body behavior information are separated according to the thermodynamic value, and the collected original wireless signal data are preprocessed; and extracting an action expression capable of representing a human body activity state from the preprocessed wireless signal data, inputting the action expression into a tumble detection model to determine a tumble event, and carrying out tumble reminding. And a clear behavior basis is provided for fall detection.

Description

Technical Field

[0001] This invention relates to the field of intelligent detection technology, and in particular to an intelligent fall detection method based on wireless signals. Background Technology

[0002] With the accelerating global aging process and the increasing number of people living alone, falls among the elderly or those with mobility impairments are receiving increasing attention. Traditional fall detection methods mainly include wearable devices and vision-based detection. While wearable devices can detect falls relatively accurately, they require users to actively wear them, leading to inconvenience, poor comfort, and the risk of device loss. Vision-based detection methods, although not requiring users to wear devices, can infringe on user privacy, and their detection effectiveness is significantly affected in poor lighting or obstructed conditions. Wireless signal detection technology has gradually become a research hotspot due to its advantages such as non-contact operation and strong privacy protection. However, existing fall detection methods based on wireless signals still have some problems, such as low detection accuracy, susceptibility to environmental interference and daily human activities, resulting in a high false alarm rate; at the same time, they are not adaptable enough to fall detection in complex scenarios and cannot meet the diverse needs of practical applications.

[0003] Therefore, this invention proposes an intelligent fall detection method based on wireless signals. Summary of the Invention

[0004] This invention provides an intelligent fall detection method based on wireless signals to solve the aforementioned technical problems.

[0005] This invention provides an intelligent fall detection method based on wireless signals, comprising: Step 1: Deploy multiple wireless signal transceivers within the target detection area to continuously collect each wireless signal within the target area, determine the reflection path and refraction path of each wireless signal, and determine the point of rapid attenuation and the thermal value of the point of rapid attenuation. Step 2: Separate the raw signal data carrying human behavior information according to the thermal value, and preprocess the collected raw wireless signal data; Step 3: Extract action descriptions that can characterize human activity states from the preprocessed wireless signal data, wherein the action descriptions contain at least one complete behavioral action; Step 4: Input the action description into the fall detection model to determine the fall event and provide a fall reminder.

[0006] Preferably, multiple wireless signal transceivers are deployed within the target detection area, including: Laser scanning is performed on the target detection area to acquire point cloud data; Each point cloud data point is projected onto a 3D coordinate system to locate the block to be analyzed and the missing block. A baseline x-coordinate, baseline y-coordinate, and baseline y-coordinate are then constructed based on each block to be analyzed, resulting in a deviation array for each block. ,in, These represent the bias vectors in each subspace coordinate system, and the bias vectors include: the bias direction and the bias length; The deviation array is input into the coordinate normalization model to obtain the reconstructed coordinates of each block to be analyzed; Obtain the eight neighboring points that match the missing block, draw the edge contours of the eight neighboring points, and lock the coordinates of the center point as the reconstruction coordinates of the missing block. The projection results in the three-dimensional coordinate system are replaced by reconstructed coordinates to obtain the regional structure of the target detection area; The region structure is divided according to its regional function, and the detection accuracy is determined based on the regional function and structural morphology of the divided structure. Based on the detection accuracy and the wireless interference situation in adjacent division areas, wireless signal transceivers are deployed in the corresponding division areas.

[0007] Preferably, it further includes: obtaining the bias vector of each block to be analyzed in each subspace coordinate system, including: The first point with the largest deviation length is selected from all deviation points involved in each subspace coordinate. The deviation density is obtained by drawing a circle with the distance from the first point to the origin of the subspace coordinate as the radius, and a value coefficient is assigned to the first point. Based on the value coefficient and the original vector of the first point, the bias vector is obtained.

[0008] Preferably, the action descriptions that characterize human activity states are extracted from the preprocessed wireless signal data, including: Based on the preprocessed wireless signal, a behavior matrix representing the human activity state is extracted. Each detection time point corresponds to a behavior matrix, and the behavior matrix at consecutive time points is continuously analyzed to capture dynamic changes in the time dimension. Extract the key feature points and their positions from the dynamically changing actions to obtain a position vector, and determine the action description.

[0009] Preferably, after determining the reflection path and refraction path of each wireless signal, the process includes: Based on the reflection and refraction paths of each wireless signal, the first reflection map and the first refraction map are drawn. Based on the regional structure of the target detection area, the first reflection map and the first refraction map are respectively applied to the regional structure to determine the signal coverage surface of the empty area, wherein the signal coverage surface includes the signal strength of each coverage area point; Based on the signal transmission direction and signal transmission intensity, determine the first signal attenuation along the reflection direction and the second signal attenuation along the refraction direction; The first intermittent attenuation point in the first reflection diagram and the second intermittent attenuation point in the first refraction diagram are determined based on the first signal attenuation and the second signal attenuation, respectively.

[0010] Preferably, determining the point of rapid decay and its thermodynamic value includes: Obtain the attenuation level of each intermittent attenuation point. If the attenuation level is greater than the preset level, set the corresponding intermittent attenuation point as an abrupt attenuation point. Otherwise, set the corresponding intermittent decay point as the general decay point; A thermal map, including points of sharp decay, is obtained based on the sensing results of the infrared sensor, and a thermal profile is extracted based on the thermal map. When the thermal profile is a non-human profile, a foreign object safety warning will be issued; When the thermal profile is a human profile, capture and retain the thermal values ​​that match the abrupt decay points from the thermal map. Capture and preserve the thermal values ​​and profile types that match typical decay points; The original signal data is obtained based on the retained results and the wireless transceiver signals.

[0011] Preferably, the key feature points of the dynamically changing action and their positions are extracted to obtain a position vector, including: Determine the degree of difference between the sub-description of each feature point and the sub-description of the periodic region points, and retain feature points whose degree of difference is greater than a preset threshold. Determine the angle of change of the motion form of each feature point, and determine the displacement change significance coefficient in combination with the body part where the feature point is located, and retain the points whose displacement change significance coefficient is greater than the preset coefficient. The retained points are used as key feature points, and the position vector is obtained by combining the positions of the key feature points.

[0012] Preferably, the fall event includes the manner of the fall, the type of fall, and the obstacle to the fall.

[0013] Preferably, assigning a value coefficient to the first point includes:

[0014] in, This indicates the area occupied in the corresponding drawn circle; The radius is the distance from the first point to the origin. The area of ​​the circle obtained; This represents the sum of the lengths of all vectors within the circle being drawn; This represents the total number of vectors present in the corresponding drawn circle; This represents the minimum length among all the lengths of the vectors drawn in the corresponding circle; This represents the value coefficient.

[0015] Compared with the prior art, the beneficial effects of this application are: By determining the reflection and refraction paths of wireless signals, as well as the points of sharp attenuation and thermal values, the signal propagation characteristics within the target detection area can be understood. This provides a foundation for the subsequent separation of signal data carrying human behavior information. After preprocessing, a large amount of environmental interference and noise signals are removed, separating relatively pure raw signal data carrying human behavior information, thus improving signal quality. This lays the foundation for accurately extracting the action descriptions of human activity states. By accurately extracting the action descriptions of human activity states from wireless signal data, complex wireless signal changes are transformed into understandable human action information, providing clear behavioral evidence for fall detection. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a flowchart of an intelligent fall detection method based on wireless signals provided in an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0019] This invention provides an intelligent fall detection method based on wireless signals, such as... Figure 1 As shown, it includes: Step 1: Deploy multiple wireless signal transceivers within the target detection area to continuously collect each wireless signal within the target area, determine the reflection path and refraction path of each wireless signal, and determine the point of rapid attenuation and the thermal value of the point of rapid attenuation. Step 2: Separate the raw signal data carrying human behavior information according to the thermal value, and preprocess the collected raw wireless signal data; Step 3: Extract action descriptions that can characterize human activity states from the preprocessed wireless signal data, wherein the action descriptions contain at least one complete behavioral action; Step 4: Input the action description into the fall detection model to determine the fall event and provide a fall reminder.

[0020] Preferably, the fall event includes the manner of the fall, the type of fall, and the obstacle to the fall.

[0021] In this embodiment, wireless signal transceivers include Wi-Fi routers, Bluetooth beacons, etc.

[0022] In this embodiment, the reflection path is, for example, the path of a wireless signal emitted from the transmitter, reflected after hitting the wardrobe, and then to the receiver.

[0023] Refraction paths, such as when a wireless signal passes through a door, will create a refraction path.

[0024] A sharp attenuation point is a location where the signal strength of a wireless signal suddenly and significantly decreases during propagation. For example, in a corner of a room, due to factors such as furniture obstruction, signal reflection and refraction loss, the signal strength will drop sharply; this corner location is a sharp attenuation point.

[0025] Thermodynamic value is used to measure whether a person exists at the point of rapid decay.

[0026] In this embodiment, preprocessing refers to the preliminary processing of the original signal data, including operations such as denoising (removing interference signals such as environmental noise and equipment noise), filtering (preserving signals within a specific frequency range through bandpass filters, etc.), and normalization (adjusting parameters such as signal strength to a uniform range).

[0027] In this embodiment, the action description is a description of the human body's activity state, consisting of at least one complete behavioral action. For example, "walking" (including a series of foot movement actions) and "bending over to pick something up" (including a combination of actions such as bending over, reaching out, and grabbing).

[0028] In this embodiment, the fall detection model is a machine learning or deep learning model trained with a large amount of labeled fall and non-fall data. It is used to determine whether the input action description belongs to a fall event and can identify the fall method (such as falling forward or falling backward), fall type (accidental fall or fall caused by illness), and fall obstruction (whether there are obstacles affecting the fall process).

[0029] Fall alerts are systems that alert users to falls through various means (such as sending text messages to family members' mobile phones or triggering indoor alarms) so that timely rescue measures can be taken.

[0030] The beneficial effects of the above technical solution are as follows: by determining the reflection and refraction paths of wireless signals, as well as the points of sharp attenuation and thermal values, the signal propagation characteristics within the target detection area can be grasped, providing a basis for subsequent separation of signal data carrying human behavior information. After preprocessing, a large amount of environmental interference and noise signals are removed, and relatively pure original signal data carrying human behavior information is separated, improving signal quality and laying the foundation for accurately extracting the action description of human activity state. The action description of human activity state is accurately extracted from wireless signal data, transforming complex wireless signal changes into understandable human action information, providing clear behavioral basis for fall detection.

[0031] This invention provides an intelligent fall detection method based on wireless signals, comprising: multiple wireless signal transceivers deployed within a target detection area, including: Laser scanning is performed on the target detection area to acquire point cloud data; Each point cloud data point is projected onto a 3D coordinate system to locate the block to be analyzed and the missing block. A baseline x-coordinate, baseline y-coordinate, and baseline y-coordinate are then constructed based on each block to be analyzed, resulting in a deviation array for each block. ,in, These represent the bias vectors in each subspace coordinate system, and the bias vectors include: the bias direction and the bias length; The deviation array is input into the coordinate normalization model to obtain the reconstructed coordinates of each block to be analyzed; Obtain the eight neighboring points that match the missing block, draw the edge contours of the eight neighboring points, and lock the coordinates of the center point as the reconstruction coordinates of the missing block. The projection results in the three-dimensional coordinate system are replaced by reconstructed coordinates to obtain the regional structure of the target detection area; The region structure is divided according to its regional function, and the detection accuracy is determined based on the regional function and structural morphology of the divided structure. Based on the detection accuracy and the wireless interference situation in adjacent division areas, wireless signal transceivers are deployed in the corresponding division areas.

[0032] In this embodiment, the target detection area refers to the specific spatial range that needs to be detected and analyzed. A 3D laser scanner is used to scan the target detection area, such as a bedroom or living room. In the scenario of the target detection area, a corner point of the target area can be set as the origin of the coordinate system, and a 3D coordinate system can be established to describe the position of each point in the point cloud data.

[0033] In this embodiment, each block to be analyzed contains at least two consecutive points to be analyzed. That is, during the point cloud data acquisition process, there are non-overlapping situations where points are in the same real location. Therefore, analysis is required to ensure the uniqueness of the coordinates of the corresponding positions in the same real location.

[0034] A missing block refers to a block that is missing at least one coordinate point.

[0035] Based on the reference abscissa, reference ordinate, and reference ordinate drawn from the center point of the contour formed by the block to be analyzed, the bias vector of each subspace in the coordinate system is determined. The reference abscissa, reference ordinate, and reference ordinate are established to measure the positional deviation of the block to be analyzed and are set as the coordinate values ​​under the ideal state. The bias vector is a vector that describes the offset of the block to be analyzed relative to the reference coordinates, including the deviation direction (such as offset along the positive X-axis or offset along the negative Y-axis) and the deviation length (the distance of the offset).

[0036] In this embodiment, the deviation array is an array that records the deviation information of the block to be analyzed from the reference coordinates in each direction in the three-dimensional coordinate system. It contains the bias vector under each subspace coordinate (X, Y, Z direction), such as [ΔX, ΔY, ΔZ], where ΔX, ΔY, and ΔZ represent the deviation in the X, Y, and Z directions, respectively. For example, if the actual coordinates of a block to be analyzed are (10, 12, 8) and the reference coordinates are (10, 10, 10), then its deviation array is [0, 2, -2], which means that there is no deviation in the X direction, a positive offset of 2 units in the Y direction, and a negative offset of 2 units in the Z direction.

[0037] In this embodiment, the coordinate normalization model calculates corrected coordinate values ​​based on the input deviation array of the block to be analyzed, making the position of the point more consistent with the actual or ideal state. For example, a neural network model trained with a large amount of data containing known accurate coordinates and corresponding deviation arrays can be used as a coordinate normalization model.

[0038] Reconstructed coordinates are the corrected coordinate values ​​of the block to be analyzed obtained after processing by the coordinate normalization model, which are used to more accurately describe the position of the point in the target detection area.

[0039] In three-dimensional space, an eight-neighborhood point is a set of points that are adjacent to the missing block in three adjacent dimensions. For example, in a three-dimensional point cloud, the eight points (front, back, left, right, up, and down) that are adjacent to a missing block are its eight-neighborhood points.

[0040] Edge contour drawing involves connecting eight neighboring points to outline the approximate shape of the area where the missing block is located, with the geometric center of the contour being considered the center point.

[0041] The projection result is the initial distribution and representation of point cloud data in a three-dimensional coordinate system.

[0042] The region structure is a three-dimensional structural model that more accurately reflects the shape, position, and spatial relationship of objects within the target detection area after the coordinates of the blocks to be analyzed and the missing blocks in the projection results are corrected (replaced with reconstructed coordinates).

[0043] Wireless interference refers to the degree of mutual interference between wireless signals in adjacent areas.

[0044] The beneficial effects of the above technical solution are: by acquiring data through laser scanning, and after a series of processing and analysis, the wireless signal transceiver device can be rationally deployed, providing accurate structural information and an efficient wireless detection solution for the target detection area.

[0045] This invention provides an intelligent fall detection method based on wireless signals, further comprising: acquiring the bias vector of each block to be analyzed in each subspace coordinate, including: The first point with the largest deviation length is selected from all deviation points involved in each subspace coordinate. The deviation density is obtained by drawing a circle with the distance from the first point to the origin of the subspace coordinate as the radius, and a value coefficient is assigned to the first point. Based on the value coefficient and the original vector of the first point, the bias vector is obtained.

[0046] Preferably, assigning a value coefficient to the first point includes:

[0047] in, This indicates the area occupied in the corresponding drawn circle; The radius is the distance from the first point to the origin. The area of ​​the circle obtained; This represents the sum of the lengths of all vectors within the circle being drawn; This represents the total number of vectors present in the corresponding drawn circle; This represents the minimum length among all the lengths of the vectors drawn in the corresponding circle; This represents the value coefficient.

[0048] In this embodiment, the direction of the original vector is the direction of the bias vector, and the magnitude of the bias vector is: the magnitude of the original vector × the value coefficient.

[0049] In this embodiment, This section considers the relative relationship between vector length and the total number of vectors and radius. When the vector length is relatively large, this part becomes smaller, which will affect the value coefficient, reasonably reflecting the effect of the relative scale and strength between vectors on the result. For example, the overall situation of vectors is different under different subspace coordinates, and this calculation of relative relationships can accurately measure various scenarios; The properties of the logarithmic function can cause changes in distance to have different degrees of impact on the results at different scales. When the distance is small, changes in distance have a relatively large impact on the value coefficient; when the distance is large, the impact is relatively small.

[0050] The beneficial effects of the above technical solution are: by selecting the point with the largest deviation length to draw a circle to obtain the deviation density, and then using the value coefficient as the entry point to obtain the bias vector, a reliable foundation is provided for reconstructing the coordinates.

[0051] This invention provides an intelligent fall detection method based on wireless signals, which extracts action descriptions that characterize human activity states from preprocessed wireless signal data, including: Based on the preprocessed wireless signal, a behavior matrix representing the human activity state is extracted. Each detection time point corresponds to a behavior matrix, and the behavior matrix at consecutive time points is continuously analyzed to capture dynamic changes in the time dimension. Extract the key feature points and their positions from the dynamically changing actions to obtain a position vector, and determine the action description.

[0052] In this embodiment, the behavior matrix is ​​a mathematical structure that represents the activity state of a human body at a certain detection moment in matrix form. The elements in the matrix can be numerical values ​​related to wireless signal characteristics, used to quantify the activity information of the human body at that moment. Assuming the behavior matrix is ​​an m×n matrix, where m rows may correspond to different wireless signal characteristics (such as signal strength, frequency changes, etc.), and n columns correspond to different spatial locations or detection areas. For example, if multiple wireless signal receiving points are set up in a room, a row of the matrix represents the signal strength value at each receiving point, and a column represents the value of a receiving point under different wireless signal characteristics.

[0053] In this embodiment, the detection time point is a specific point in time for wireless signal detection and analysis. For example, one second is used as a detection time point, recording the wireless signal data at that moment and generating a corresponding behavior matrix. Dynamic changes in movement are continuous, changing movements of the human body over a period of time, such as walking, waving, and bending over. These movements manifest as a series of continuous posture changes in the time dimension.

[0054] In this embodiment, key feature points are those points that highlight the main morphological changes and movement trends of a dynamic action. For example, in human walking, the positional changes of joints such as the knee and ankle joints can well reflect the characteristics of walking, and these joints can be used as key feature points; in waving gestures, points on the wrist, fingers, and other parts are also key feature points.

[0055] A position vector is a vector composed of the coordinates of key feature points in space. For example, for walking, assuming five key feature points such as the knee and ankle joints are selected, their position vectors might be: Action descriptions are textual descriptions of the dynamic changes in human body movements, such as "walking," "waving," and "bending over," used to visually express the actions performed by the human body.

[0056] The beneficial effects of the above technical solution are: it converts wireless signals into a behavior matrix form that is easy to analyze; by analyzing the behavior matrix at continuous moments, it is possible to identify the dynamic changes of human body movements from the wireless signals, realize a preliminary understanding and analysis of human activity state, and lay the foundation for more accurate movement recognition in the future; it accurately extracts key feature points and their position information from dynamic changes to form a position vector; and it accurately determines the movement description through a movement classification model, thereby realizing accurate recognition and description of human body movements and providing accurate human body movement instructions for intelligent devices.

[0057] This invention provides an intelligent fall detection method based on wireless signals, which, after determining the reflection and refraction paths of each wireless signal, includes: Based on the reflection and refraction paths of each wireless signal, the first reflection map and the first refraction map are drawn. Based on the regional structure of the target detection area, the first reflection map and the first refraction map are respectively applied to the regional structure to determine the signal coverage surface of the empty area, wherein the signal coverage surface includes the signal strength of each coverage area point; Based on the signal transmission direction and signal transmission intensity, determine the first signal attenuation along the reflection direction and the second signal attenuation along the refraction direction; The first intermittent attenuation point in the first reflection diagram and the second intermittent attenuation point in the first refraction diagram are determined based on the first signal attenuation and the second signal attenuation, respectively.

[0058] In this embodiment, the first reflection map graphically presents the distribution of all reflection paths of the wireless signal within the target detection area. The map marks the signal source, reflection points, and propagation direction after reflection, resembling a "map" recording the signal reflection trajectory. The first refraction map, similar to the first reflection map, is a graphical representation of all refraction paths of the wireless signal within the target detection area, clearly showing the starting point of refraction, the boundary of the medium it passes through, and the propagation direction after refraction. By using the relationship between signal propagation time and speed (the signal propagation speed is approximately the speed of light), combined with the location information of the signal transceiver, the positions of the reflection and refraction points are calculated using algorithms such as triangulation and polygonal positioning, thereby determining the reflection and refraction paths. Finally, using professional plotting software or programming tools (such as Python's Matplotlib library or GIS plotting software), the calculated reflection and refraction paths are plotted as the first reflection map and the first refraction map.

[0059] In this embodiment, the regional structure of the target detection area refers to the spatial layout and shape structure of objects within the target detection area; the idle area refers to the spatial area within the target detection area that is not occupied by objects; the signal coverage area is the range of the idle area that the wireless signal can cover, and the signal strength received by each point within that range; and the coverage area point refers to a discrete point in the space that constitutes the signal coverage area, used to describe the specific location of the signal coverage. A three-dimensional structural model of the target detection area is generated using 3D modeling software (such as AutoCAD, SketchUp) or based on laser scanning and point cloud data processing to accurately represent the position and shape of objects within the area. Then, the first reflection map and the first refraction map are superimposed and fused with the 3D structural model. Using a signal propagation model (such as a ray tracing model or an empirical propagation model), combined with parameters such as the transmission power and frequency of the wireless signal transceiver, the signal strength of the wireless signal at each coverage area point within the idle area is calculated. In the specific calculation process, the effects of signal attenuation during propagation (such as free space attenuation, obstacle blockage attenuation), reflection, and refraction on the signal strength are considered. Finally, the calculated signal strength is marked on the coverage area point to determine the signal coverage area of ​​the idle area.

[0060] In this embodiment, the first signal attenuation is the degree to which the signal strength decreases relative to the transmission strength as the wireless signal propagates along the reflection direction. For example, if the transmitted signal strength is -30dBm, and the received signal strength at a certain point after one reflection is -60dBm, then the first signal attenuation along that reflection direction is 30dB. The second signal attenuation is similar to the first, representing the degree to which the signal strength decreases as the wireless signal propagates along the refraction direction. For example, if five detection points are selected along a reflection path, and the attenuation at each point is calculated to be 25dB, 28dB, 30dB, 27dB, and 26dB respectively, the average value yields a first signal attenuation of approximately 27.2dB along that reflection direction.

[0061] In this embodiment, in the first reflection map, the point where the signal intensity decreases intermittently (discontinuously, suddenly) along the reflection direction is the first intermittent attenuation point. In the first refraction map, the point where the signal intensity decreases intermittently along the refraction direction is the second intermittent attenuation point. The search and marking of these points are automatically completed using programming algorithms (such as loops and conditional statements written in Python).

[0062] The beneficial effects of the above technical solution are: it presents the abstract wireless signal reflection and refraction path in a visual graphic, making it easy to intuitively understand the propagation trajectory of the wireless signal in the target detection area, providing a clear visual basis for subsequent analysis of signal coverage and attenuation, accurately determining the coverage range of the wireless signal in the idle area and the signal strength at each point by combining the actual structure of the target detection area, quantifying the degree of intensity attenuation of the wireless signal during reflection and refraction, understanding the signal loss on different propagation paths, providing data basis for evaluating wireless signal quality and optimizing signal propagation paths, accurately identifying the locations where the wireless signal intermittently and significantly attenuates during reflection and refraction, and helping to discover potential factors affecting wireless signal quality.

[0063] This invention provides an intelligent fall detection method based on wireless signals, which determines the point of rapid decay and the thermal value of the point of rapid decay, including: Obtain the attenuation level of each intermittent attenuation point. If the attenuation level is greater than the preset level, set the corresponding intermittent attenuation point as an abrupt attenuation point. Otherwise, set the corresponding intermittent decay point as the general decay point; A thermal map, including points of sharp decay, is obtained based on the sensing results of the infrared sensor, and a thermal profile is extracted based on the thermal map. When the thermal profile is a non-human profile, a foreign object safety warning will be issued; When the thermal profile is a human profile, capture and retain the thermal values ​​that match the abrupt decay points from the thermal map. Capture and preserve the thermal values ​​and profile types that match typical decay points; The original signal data is obtained based on the retained results and the wireless transceiver signals.

[0064] In this embodiment, the attenuation level is used to quantify the degree of signal strength reduction at intermittent attenuation points of a wireless signal. The attenuation level can be divided into multiple levels, such as 1-5, where level 1 represents the minimum attenuation and level 5 represents the maximum attenuation. For example, in a living room, if the signal strength at a certain intermittent attenuation point of a wireless signal decreases by 15dB, according to the established grading rules, it might correspond to attenuation level 2.

[0065] The preset level is a threshold value set in advance by the user to distinguish between points of rapid attenuation and points of normal attenuation. For example, if the preset level is set to 3, an attenuation level greater than 3 is considered a point of rapid attenuation; otherwise, it is considered a point of normal attenuation.

[0066] A point of sharp attenuation is a location where the strength of a wireless signal decreases significantly and abruptly, which has a substantial impact on the quality of wireless signal transmission. For example, in a living room, if a large metal cabinet blocks the wireless signal, causing a sharp drop in signal strength at a certain location, that location is a point of sharp attenuation.

[0067] A typical attenuation point is an intermittent attenuation point where the signal strength weakens relatively little and has a weak impact on the quality of wireless signal transmission. For example, an intermittent attenuation point near a curtain is considered a typical attenuation point because the curtain obstructs the signal to some extent but the impact is minor.

[0068] Based on the previously calculated signal attenuation at each intermittent attenuation point, the attenuation level of each intermittent attenuation point is determined according to a predefined attenuation level classification rule (e.g., each level interval is 10dB). Then, the attenuation level of each intermittent attenuation point is compared with a preset level. If the attenuation level is greater than the preset level, the intermittent attenuation point is designated as a sharp attenuation point through programming or database recording; if the attenuation level is less than or equal to the preset level, it is designated as a normal attenuation point. For example, a program written in Python can iterate through all intermittent attenuation points and use conditional statements to classify sharp and normal attenuation points.

[0069] Heatmaps are visual representations that intuitively display the degree of signal attenuation or the presence of objects within a region, using different colors or shades of color. In this scenario, they are generated based on the sensing results of infrared sensors and information on points of sharp signal attenuation. The darker the color, the more severe the signal attenuation or the higher the probability of the presence of an object. For example, in a heatmap, red areas represent locations where the signal attenuates sharply or where people or objects are clustered, while light-colored areas represent locations where the signal attenuation is less or where no objects are present.

[0070] Thermal profiles are boundary lines extracted from heatmaps that reflect the shape of an object or region where a signal attenuates. Thermal profiles allow for a visual identification of the shape or area affecting signal attenuation. For example, in a heatmap, the boundary lines surrounding a human body represent the thermal profile of that person; the boundary lines surrounding large pieces of furniture represent the thermal profile of that furniture.

[0071] Non-human contours: Thermal contours that differ significantly in shape from human contours represent the contours of objects (such as furniture, appliances, and miscellaneous items). For example, contours that appear as regular shapes such as squares or rectangles on a thermal map may be the contours of furniture such as cabinets or tables, and are therefore considered non-human contours.

[0072] Foreign Object Safety Alert: This alert is issued when the system detects a non-human-shaped thermal profile that may affect wireless signal transmission or environmental safety. The alert can be sent via sound alarm, push notification, or flashing lights. For example, if a large metal object is detected obstructing the signal, the system will send a message to the user via a mobile app stating, "A foreign object that may affect the signal has been detected. Please check."

[0073] Human silhouette: A thermal silhouette that conforms to the shape of the human body indicates the presence of a human being in the area. For example, a silhouette that resembles a human figure suggests that someone is in that location.

[0074] Heatmap value: A numerical value used to quantify the degree of signal attenuation or the probability of an object's presence at a point or area on a heatmap. The higher the value, the more severe the signal attenuation or the higher the probability of an object's presence. For example, in a heatmap, a heatmap value of 80 (assuming a range of 0-100) corresponds to a point of sharp attenuation, indicating a high degree of signal attenuation at that point.

[0075] Contour Type: A classification identifier for the object type to which the thermal contour belongs, such as "furniture," "appliances," or "human body." It is used to distinguish the impact of different object types on signal attenuation.

[0076] The raw signal data contains an initial set of data containing various information about the wireless signal during its propagation within the target detection area (such as attenuation, the influence of objects, and the influence of human activities).

[0077] The beneficial effects of the above technical solution are as follows: It effectively classifies intermittent attenuation points, distinguishing between rapidly attenuating points that severely impact wireless signal transmission and moderately attenuating points that have a smaller impact. This provides a basis for subsequent targeted processing of points with different attenuation levels. It combines wireless signal attenuation with infrared sensor sensing results, presenting them in the form of heat maps and thermal profiles. This visually demonstrates the relationship between object distribution and signal attenuation within the target detection area, facilitating rapid identification of objects or areas affecting signal attenuation. Different processing is applied based on the type of thermal profile. Timely alerts are issued when non-human profiles are detected that may affect the signal or pose safety hazards, ensuring wireless signal transmission and environmental safety. When human profiles are detected, key data related to the attenuation point are retained, providing data support for subsequent analysis of the relationship between human activity and wireless signal changes, thus contributing to the realization of human behavior analysis and monitoring based on wireless signals.

[0078] This invention provides an intelligent fall detection method based on wireless signals, which extracts key feature points of the dynamically changing movements and the positions of these key feature points to obtain a position vector, including: Determine the degree of difference between the sub-description of each feature point and the sub-description of the periodic region points, and retain feature points whose degree of difference is greater than a preset threshold. Determine the angle of change of the motion form of each feature point, and determine the displacement change significance coefficient in combination with the body part where the feature point is located, and retain the points whose displacement change significance coefficient is greater than the preset coefficient. The retained points are used as key feature points, and the position vector is obtained by combining the positions of the key feature points.

[0079] In this embodiment, the sub-description is a feature description of the local region surrounding the feature point, typically including information such as gradient, texture, and color. For example, the 128-dimensional descriptor of a feature point in the SIFT algorithm is a type of sub-description used to characterize the gradient direction histogram distribution in the neighborhood of the feature point. A sub-description vector is generated for each feature point using feature extraction algorithms (such as SIFT and ORB), and corresponding sub-descriptions are also generated for points within the periodic region.

[0080] Periodic region points are points within a region that exhibits periodic changes over time, such as the movement area of ​​the knee joint during human walking. Assuming one cycle of walking motion (from the left foot landing to its next landing) is used as the unit of analysis, the points covered by the knee joint's movement trajectory within that cycle are the periodic region points.

[0081] The dissimilarity measure is the degree of dissimilarity between the sub-description of a feature point and the sub-description of a periodic region point. It can be calculated using metrics such as Euclidean distance and cosine similarity. For example, if the Euclidean distance between the sub-description of feature point A and the sub-description of periodic region point B is 3.5, this value is the dissimilarity measure.

[0082] The preset threshold is a manually set critical value for the difference between feature points. If the difference is greater than this threshold, the feature point is considered unique; otherwise, it is considered a redundant point. For example, if the preset threshold is set to 2.0, feature points with a difference greater than 2.0 will be retained.

[0083] The change angle of a movement's form is the amount of angular change of a feature point relative to its initial position during a movement, used to measure changes in posture. For example, the change in angle of the elbow joint from an extended to a bent position during a flexion movement is 90°, and this angle is the change angle of form.

[0084] Body part location refers to the anatomical position of a feature point on the human body or object, such as the "shoulder joint," "hip joint," and "wrist." The degree of displacement of feature points in different parts of the body varies during movement; for example, feature points at the extremities usually displace more than feature points in the torso.

[0085] The displacement change significance coefficient is a weighted coefficient that comprehensively considers the displacement of the feature point and the position of the body part, and is used to quantify the degree of dynamic change of the feature point during a movement. For example, the wrist has a large displacement in a waving motion and is located at the end of the limbs, so its displacement change significance coefficient will be higher than that of the trunk feature point.

[0086] The preset coefficient is a manually set threshold used to determine whether a feature point should be retained. If the coefficient is greater than the threshold, the feature point displacement is considered to have changed significantly; otherwise, the change is considered to be insignificant.

[0087] The beneficial effects of the above technical solution are: through a dual screening mechanism (difference screening and displacement coefficient screening), key feature points are accurately extracted from the original feature points and transformed into position vectors, thereby achieving efficient capture of the core dynamic features of the action.

[0088] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A smart fall detection method based on wireless signals, characterized in that, include: Step 1: Deploy multiple wireless signal transceivers within the target detection area to continuously collect each wireless signal within the target area, determine the reflection path and refraction path of each wireless signal, and determine the point of rapid attenuation and the thermal value of the point of rapid attenuation. Step 2: Separate the raw signal data carrying human behavior information according to the thermal value, and preprocess the collected raw wireless signal data; Step 3: Extract action descriptions that can characterize human activity states from the preprocessed wireless signal data, wherein the action descriptions contain at least one complete behavioral action; Step 4: Input the action description into the fall detection model to determine the fall event and provide a fall reminder.

2. The intelligent fall detection method based on wireless signals according to claim 1, characterized in that, Multiple wireless signal transceivers are deployed within the target detection area, including: Laser scanning is performed on the target detection area to acquire point cloud data; Each point cloud data point is projected onto a 3D coordinate system to locate the block to be analyzed and the missing block. A baseline x-coordinate, baseline y-coordinate, and baseline y-coordinate are then constructed based on each block to be analyzed, resulting in a deviation array for each block. ,in, These represent the bias vectors in each subspace coordinate system, and the bias vectors include: the bias direction and the bias length; The deviation array is input into the coordinate normalization model to obtain the reconstructed coordinates of each block to be analyzed; Obtain the eight neighboring points that match the missing block, draw the edge contours of the eight neighboring points, and lock the coordinates of the center point as the reconstruction coordinates of the missing block. The projection results in the three-dimensional coordinate system are replaced by reconstructed coordinates to obtain the regional structure of the target detection area; The region structure is divided according to its regional function, and the detection accuracy is determined based on the regional function and structural morphology of the divided structure. Based on the detection accuracy and the wireless interference situation in adjacent division areas, wireless signal transceivers are deployed in the corresponding division areas.

3. The intelligent fall detection method based on wireless signals according to claim 2, characterized in that, Also includes: Obtain the bias vector of each block to be analyzed in each subspace coordinate system, including: The first point with the largest deviation length is selected from all deviation points involved in each subspace coordinate. The deviation density is obtained by drawing a circle with the distance from the first point to the origin of the subspace coordinate as the radius, and a value coefficient is assigned to the first point. Based on the value coefficient and the original vector of the first point, the bias vector is obtained.

4. The intelligent fall detection method based on wireless signals according to claim 1, characterized in that, Extracting action representations that characterize human activity states from preprocessed wireless signal data, including: Based on the preprocessed wireless signal, a behavior matrix representing the human activity state is extracted. Each detection time point corresponds to a behavior matrix, and the behavior matrix at consecutive time points is continuously analyzed to capture dynamic changes in the time dimension. Extract the key feature points and their positions from the dynamically changing actions to obtain a position vector, and determine the action description.

5. The intelligent fall detection method based on wireless signals according to claim 1, characterized in that, After determining the reflection and refraction paths of each wireless signal, the following steps are included: Based on the reflection and refraction paths of each wireless signal, the first reflection map and the first refraction map are drawn. Based on the regional structure of the target detection area, the first reflection map and the first refraction map are respectively applied to the regional structure to determine the signal coverage surface of the empty area, wherein the signal coverage surface includes the signal strength of each coverage area point; Based on the signal transmission direction and signal transmission intensity, determine the first signal attenuation along the reflection direction and the second signal attenuation along the refraction direction; The first intermittent attenuation point in the first reflection diagram and the second intermittent attenuation point in the first refraction diagram are determined based on the first signal attenuation and the second signal attenuation, respectively.

6. The intelligent fall detection method based on wireless signals according to claim 6, characterized in that, Determine the point of rapid decay and its thermodynamic value, including: Obtain the attenuation level of each intermittent attenuation point. If the attenuation level is greater than the preset level, set the corresponding intermittent attenuation point as an abrupt attenuation point. Otherwise, set the corresponding intermittent decay point as the general decay point; A thermal map, including points of sharp decay, is obtained based on the sensing results of the infrared sensor, and a thermal profile is extracted based on the thermal map. When the thermal profile is a non-human profile, a foreign object safety warning will be issued; When the thermal profile is a human profile, capture and retain the thermal values ​​that match the abrupt decay points from the thermal map. Capture and preserve the thermal values ​​and profile types that match typical decay points; The original signal data is obtained based on the retained results and the wireless transceiver signals.

7. The intelligent fall detection method based on wireless signals according to claim 5, characterized in that, Extract the key feature points and their positions from the dynamically changing actions to obtain a position vector, including: Determine the degree of difference between the sub-description of each feature point and the sub-description of the periodic region points, and retain feature points whose degree of difference is greater than a preset threshold. Determine the angle of change of the motion form of each feature point, and determine the displacement change significance coefficient in combination with the body part where the feature point is located, and retain the points whose displacement change significance coefficient is greater than the preset coefficient. The retained points are used as key feature points, and the position vector is obtained by combining the positions of the key feature points.

8. The intelligent fall detection method based on wireless signals according to claim 1, characterized in that, The fall event includes the manner of the fall, the type of fall, and the obstacle to the fall.

9. The intelligent fall detection method based on wireless signals according to claim 3, characterized in that, Assigning a value coefficient to the first point includes: in, This indicates the area occupied in the corresponding drawn circle; The radius is the distance from the first point to the origin. The area of ​​the circle obtained; This represents the sum of the lengths of all vectors within the circle being drawn; This represents the total number of vectors present in the corresponding drawn circle; This represents the minimum length among all the lengths of the vectors drawn in the corresponding circle; This represents the value coefficient.

Images