CN112346050A - Fall detection method and system based on Wi-Fi equipment - Google Patents

Fall detection method and system based on Wi-Fi equipment Download PDF

Info

Publication number
CN112346050A
CN112346050A CN202011148763.2A CN202011148763A CN112346050A CN 112346050 A CN112346050 A CN 112346050A CN 202011148763 A CN202011148763 A CN 202011148763A CN 112346050 A CN112346050 A CN 112346050A
Authority
CN
China
Prior art keywords
falling
fall
samples
human body
detection model
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202011148763.2A
Other languages
Chinese (zh)
Other versions
CN112346050B (en
Inventor
杨铮
张驿
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tsinghua University
Original Assignee
Tsinghua University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Tsinghua University filed Critical Tsinghua University
Priority to CN202011148763.2A priority Critical patent/CN112346050B/en
Priority claimed from CN202011148763.2A external-priority patent/CN112346050B/en
Publication of CN112346050A publication Critical patent/CN112346050A/en
Application granted granted Critical
Publication of CN112346050B publication Critical patent/CN112346050B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
    • G01S13/88—Radar or analogous systems specially adapted for specific applications
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/103—Detecting, measuring or recording devices for testing the shape, pattern, colour, size or movement of the body or parts thereof, for diagnostic purposes
    • A61B5/11—Measuring movement of the entire body or parts thereof, e.g. head or hand tremor, mobility of a limb
    • A61B5/1116—Determining posture transitions
    • A61B5/1117—Fall detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Remote Sensing (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Radar, Positioning & Navigation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Medical Informatics (AREA)
  • Pathology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • Public Health (AREA)
  • Veterinary Medicine (AREA)
  • Electromagnetism (AREA)
  • Dentistry (AREA)
  • Physiology (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Alarm Systems (AREA)

Abstract

The embodiment of the invention provides a fall detection method based on Wi-Fi equipment, which comprises the following steps: determining human body speed characteristics based on channel state information of wireless signals acquired in a target area; inputting the human body speed characteristics into a falling detection model to obtain a human body falling identification result output by the falling detection model; the falling detection model is obtained by training based on a falling/non-falling sample data set; the non-falling samples in the falling/non-falling sample data set are from the samples of the daily activities of the persons which are really collected, and the falling samples in the falling/non-falling sample data set are obtained by preprocessing the input real falling samples based on a data capacity expansion model. The fall detection method based on the Wi-Fi equipment, provided by the embodiment of the invention, is convenient for large-scale popularization, and can improve the effect of training a fall detection model, so that the fall situation of people in a target area can be detected more accurately and efficiently.

Description

Fall detection method and system based on Wi-Fi equipment
Technical Field
The invention relates to the technical field of wireless perception, in particular to a falling detection method and system based on Wi-Fi equipment.
Background
According to statistics, the falling-down has become the leading cause of death due to injury of the old, and especially for the old living alone, how to timely detect the falling-down condition is very important. In the past decade, along with the development of internet of things technology, people fall detection technology based on cameras, wearable sensors and radar equipment emerges.
At present, a fall detection method based on a camera mainly includes deploying the camera in an environment to acquire image information, and then analyzing a person moving picture captured by the camera to judge whether a fall action occurs. However, due to the limited field of view of the camera, the system fails due to the occlusion of walls and objects, and the system performance is significantly reduced in weak light scenes. In addition, deployment of a camera in a home scene can affect privacy of a user inevitably, and commercial popularization of the technology is limited.
The fall detection method based on the wearable sensor mainly detects the fall by detecting the motion mode of a user by using sensors such as an acceleration sensor and a gyroscope. But the use of the sensor can bring inconvenience to the activities of the personnel, and the requirements of timely charging, continuous wearing and the like of the equipment can also bring troubles to the lives of the old.
The falling detection method based on the radar equipment mainly utilizes electromagnetic wave signals emitted by the radar to sense the movement of people. However, large-scale commercial use is difficult due to high cost and complex arrangement of the radar device.
Disclosure of Invention
The embodiment of the invention provides a fall detection method and system based on Wi-Fi equipment, which are used for overcoming the defects that the fall detection in the prior art is too limited and difficult to popularize on a large scale and apply daily, are convenient to popularize on a large scale, are wider in applicable scenes, improve the effect of training a fall detection model, and enable the fall detection of people in a target area to be more accurate and efficient.
The embodiment of the invention provides a fall detection method based on Wi-Fi equipment, which comprises the following steps: determining human body speed characteristics based on channel state information of wireless signals acquired in a target area; inputting the human body speed characteristics into a falling detection model to obtain a human body falling identification result output by the falling detection model; wherein the fall detection model is obtained by training based on a fall/non-fall sample data set; the non-falling samples in the falling/non-falling sample data set are from samples of the daily activities of the really acquired personnel, and the falling samples in the falling/non-falling sample data set are obtained by preprocessing the input real falling samples based on a data capacity expansion model.
According to the fall detection method based on the Wi-Fi equipment, the data expansion model is used for carrying out simulation expansion on the input real fall sample based on the principle of the differential automatic encoder to obtain the fall sample.
According to the fall detection method based on the Wi-Fi device, the simulation expansion is performed on the input real fall sample based on the principle of the differential automatic encoder to obtain the fall sample, and the method comprises the following steps: mapping the real fall sample to a low-dimensional feature; randomly sampling the low-dimensional features to obtain a plurality of simulated fall samples, and taking the real fall samples and the simulated fall samples as the fall samples.
According to a Wi-Fi device-based fall detection method of an embodiment of the invention, the mapping of the real fall samples to low-dimensional features comprises: extracting a mean vector and a standard deviation vector from the real falling sample; randomly extracting a plurality of sampling data from N (0,1) obeying normal distribution, respectively adding the plurality of sampling data with the mean vector, and respectively multiplying with the standard deviation vector to obtain the low-dimensional feature.
According to the fall detection method based on the Wi-Fi device, the method for inputting the human body speed characteristics into the fall detection model to obtain the human body fall recognition result output by the fall detection model comprises the following steps: inputting the human body speed features into a feature extraction layer of the falling detection model, removing noise components in the human body speed features and identity information of a user, and reserving specific falling activity mode information as high-order features; and inputting the high-order features into a falling detection layer of the falling detection model to obtain the human body falling identification result.
According to a fall detection method based on a Wi-Fi device, the identity information includes: at least one of height, weight, posture, and fall habit characteristics.
According to the fall detection method based on the Wi-Fi device, the method for determining the human body speed characteristics based on the channel state information of the wireless signals acquired in the target area includes: based on a preset time threshold, dividing the channel state information into a plurality of channel state sub-segments; obtaining autocorrelation functions of a plurality of the channel state sub-segments; obtaining the position of the first extreme point of any one of the autocorrelation functions, wherein the human body speed characteristic is based on a formula
Figure BDA0002740510970000031
Determining where τ is0Representing the position of a first extreme point of any of said autocorrelation functions; x is the number of0Representing the position of the first extreme point of the sinc (x) function; λ represents the wavelength of the wireless signal.
The embodiment of the invention also provides a fall detection method based on the Wi-Fi equipment, which comprises the following steps: the extraction module is used for determining human body speed characteristics based on the channel state information of the wireless signals acquired in the target area; the detection module is used for inputting the human body speed characteristics into a falling detection model and obtaining a human body falling identification result output by the falling detection model; wherein the fall detection model is obtained by training based on a fall/non-fall sample data set; the non-falling samples in the falling/non-falling sample data set are from samples of the daily activities of the really acquired personnel, and the falling samples in the falling/non-falling sample data set are obtained by preprocessing the input real falling samples based on a data capacity expansion model.
An embodiment of the present invention further provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the program, the processor implements the steps of any one of the above-mentioned fall detection methods based on a Wi-Fi device.
Embodiments of the present invention also provide a non-transitory computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of a method for fall detection based on a Wi-Fi device as described in any one of the above.
According to the method for detecting falling based on the Wi-Fi equipment, provided by the embodiment of the invention, the falling detection is carried out by adopting the wireless network signals, so that the large-scale popularization is facilitated, the applicable scene is wider, the real falling samples are subjected to capacity expansion and screening by utilizing the data capacity expansion model, a falling/non-falling sample data set is formed, the training effect of the falling detection model is improved, and the falling condition of the personnel in the target area is more accurately and efficiently detected.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and those skilled in the art can also obtain other drawings according to the drawings without creative efforts.
Fig. 1 is a schematic flowchart of a fall detection method based on a Wi-Fi device according to an embodiment of the present invention;
fig. 2 is a schematic diagram of a fall detection method based on a Wi-Fi device according to an embodiment of the present invention;
fig. 3 is a schematic propagation diagram of a wireless signal of a fall detection method based on a Wi-Fi device provided by an embodiment of the present invention;
fig. 4 is a schematic diagram of changes in human body speed characteristics corresponding to a fall from normal walking according to the fall detection method based on the Wi-Fi device provided by the embodiment of the present invention.
Fig. 5 is a schematic diagram of changes in body velocity characteristics of a real fall sample and a simulated fall sample of a fall detection method based on a Wi-Fi device according to an embodiment of the present invention;
fig. 6 is a schematic diagram of a relationship between a core speed and a residual speed of a fall detection method based on a Wi-Fi device according to an embodiment of the present invention;
fig. 7 is a schematic diagram of a data capacity expansion model of a fall detection method based on a Wi-Fi device according to an embodiment of the present invention;
fig. 8 is a schematic structural diagram of a fall detection model of a fall detection method based on a Wi-Fi device according to an embodiment of the present invention;
fig. 9 is a schematic structural diagram of a fall detection method based on a Wi-Fi device according to an embodiment of the present invention;
fig. 10 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In recent years, some researchers have looked at the wide deployment of commercial Wi-Fi equipment. Wi-Fi equipment can emit electromagnetic wave signals in the communication process, and the signals can be reflected by objects and people in the environment and received by a receiver, so that the Wi-Fi equipment can be regarded as low-cost radar equipment. The relevant work is to establish a relation model of CSI (Channel State Information) and personnel activity by analyzing the CSI measured by the Wi-Fi equipment, so as to detect the falling of the personnel.
However, these tasks have limitations in three dimensions, environment, personnel and data, and are difficult to generalize to real business scenarios. From the environmental point of view, the signal characteristics extracted by the related work are greatly influenced by the environment, so that the system deployment in different environments can influence the accuracy of detection, and particularly when people are shielded by objects or walls and the like, the related work is difficult to make accurate detection. From a personnel perspective, the related work ignores the inter-personnel variability, such as: different people have different heights, weights, postures and the like, and the factors can influence the falling motion mode, so that the detection accuracy is influenced. From a data point of view, the relevant fall detection work only considers a simple fall pattern, and only a small amount of fall data is collected for system training. However, the fall activity involves complex movements in all parts of the body, resulting in various forms of fall habit features, such as: the user can feel dizzy, slippery, stumble, kneel, fall from a chair and the like, so that the related work can not identify fall modes which are not considered in the training stage in the actual deployment process, and the alarm missing rate is high.
A method and system for fall detection based on Wi-Fi devices according to embodiments of the present invention are described below with reference to fig. 1 to 10.
As shown in fig. 1, an embodiment of the present invention provides a fall detection method based on a Wi-Fi device, which includes the following steps 100-200.
Step 100, determining human body speed characteristics based on the channel state information of the wireless signals acquired in the target area.
In the field of wireless communications, Channel State Information (CSI) of a wireless signal is a channel property of a communication link. It describes the fading factor of the signal on each transmission path, i.e. the value of each element in the channel gain matrix H, such as signal Scattering (Scattering), fading or fading, distance fading (power fading) and other information. The CSI may adapt the communication system to the current channel conditions, providing a guarantee for high reliability and high rate communication in a multi-antenna system.
As shown in fig. 2 and 3, in some terminal devices, such as a router, a smart phone, or a laptop, a wireless signal collected in a Wi-Fi signal coverage area may be affected by an obstacle, and CSI may reflect a situation that the wireless signal is affected by the obstacle.
When the CSI is processed, the phase information or the amplitude information is often directly extracted by the conventional method, but the phase information or the amplitude information cannot accurately reflect the falling condition of the human body because the wireless signal may be affected by environmental interference, the position of the person and the orientation of the person.
As shown in fig. 4, the change of the body speed characteristics of the human body from normal walking to falling is reflected, the body speed characteristics are extracted from the CSI and used as a basis for falling judgment, the body speed characteristics can eliminate environmental interference, and the falling condition of the human body can be most directly depicted.
Step 200, inputting the human body speed characteristics into a falling detection model to obtain a human body falling identification result output by the falling detection model; the fall detection model is obtained by training based on a fall/non-fall sample data set; the fall detection model is obtained by training based on a fall/non-fall sample data set; the non-falling samples in the falling/non-falling sample data set are from the samples of the daily activities of the persons which are really collected, and the falling samples in the falling/non-falling sample data set are obtained by preprocessing the input real falling samples based on a data capacity expansion model.
It can be understood that the body velocity features extracted from the CSI in step 100 are input into a fall detection model, and the fall detection model can output a body fall identification result. The human body falling identification result can represent whether a person falls in the target area.
The fall detection model is obtained by training a fall/non-fall sample dataset, wherein the fall/non-fall sample dataset can comprise a fall sample and a normal walking sample, the sample is a human body speed characteristic, and the sample labels are non-fall and fall, wherein the non-fall sample is a sample of the daily activities of the really acquired personnel.
It is worth noting that in the research process, the inventor finds that a real fall sample can be used to train a fall detection model, but for normal people, the fall occurrence is rare, a large number of fall samples are difficult to collect, and the collected samples are difficult to cover most fall situations.
In order to enable the fall detection model to detect more fall situations, the fall/non-fall sample data set needs more fall samples, and the embodiment of the invention abandons the traditional way of manually acquiring sample data, and utilizes the data expansion model to generate the fall/non-fall sample data set.
A small amount of real falling samples can be collected, the real falling samples are input into the data expansion model, the data expansion model can simulate according to the real falling samples, and therefore a plurality of simulated falling samples are generated, sample expansion is achieved, the real falling samples are expanded by the data expansion model before the falling detection model is trained, a richer falling/non-falling sample data set is formed, and the robustness of the falling detection model to falling diversity can be improved.
According to the fall detection method based on the Wi-Fi equipment, provided by the embodiment of the invention, the fall detection is carried out by adopting the wireless network signals, so that the large-scale popularization is facilitated, the applicable scene is wider, the real fall sample is expanded by utilizing the data expansion model to form the fall/non-fall sample data set, the training effect of the fall detection model is improved, and the fall situation of the personnel in the target area is more accurately and efficiently detected.
As shown in fig. 8, in some embodiments, inputting the human body speed characteristics into the fall detection model, and obtaining the human body fall recognition result output by the fall detection model, the method includes: inputting the human body speed characteristics into a characteristic extraction layer of a falling detection model, removing noise components in the speed characteristics and identity information of a user, and reserving special falling activity mode information as high-order characteristics; and inputting the high-order features into a falling detection layer of the falling detection model to obtain a human body falling recognition result.
It is understood that the fall detection model comprises a feature extraction layer and a fall detection layer. The feature extraction layer is used for removing noise information and identity information of people from the speed sequence, acquiring high-order features for describing falling activities of the people, and the falling detection layer is used for detecting falling states of the people from the high-order features.
The high-order characteristics remove noise components and identity information in human body speed characteristics, and retain speed change characteristics in the falling process of people, including trend characteristics in an acceleration stage and a deceleration stage and maximum and minimum values in the speed change process.
It should be noted that the traditional fall detection model assumes that fall situations of different users are similar, but based on the foregoing analysis, the user may cause a reduction in accuracy when the debugged fall detection model is applied to other users due to differences in height, weight, posture and fall habit characteristics. In this regard, conventional fall detection models can only be used for retraining the system by re-acquiring data samples of a new user.
In the embodiment of the invention, a more general application scenario is considered: a debugged fall detection model can be easily generalized to new users without the need to collect the user's fall data for model training. To solve the problem, only the identity information can be removed from the human body speed features input by the model, and meanwhile, the features related to falling judgment are kept as the features of falling detection.
As shown in fig. 8, in some embodiments, in the Fall detection method based on a Wi-Fi device provided by the embodiments of the present invention, an identity recognition layer (User Discriminator) may be fused in a training phase to perform joint training, and the identity recognition layer is deleted in an application and deployment phase, and only a Feature extraction layer (Feature Extractor) and a Fall detection layer (Fall Detector) are used.
It can be understood that the fall detection model is combined with the identity recognition module for joint training in a training stage, the input of the identity recognition module is the high-order features extracted by the feature extraction layer of the fall detection model, and the fuzzy personnel identity information is forced to be output by constraining the output result of the feature extraction layer, so that the high-order features extracted by the feature extraction layer of the fall detection model are forced to contain the personnel identity information as little as possible, and finally the detection result of the fall detection layer of the fall detection model is unrelated to the identity information of the personnel.
The Loss function (Loss) of the fall detection model and the identity recognition model for training is designed as follows:
Loss＝LTall--αLuser
wherein L isTallIs a Binary Cross-entropy (L) of the fall situation of the simulated fall sample and the fall situation of the real fall sampleuserThe binary cross entropy of the identity information of the simulated falling sample and the identity information of the real falling sample is adopted, alpha is a positive constant, alpha is generally selected to be 1, but the artificial dynamic adjustment can be carried out according to the system performance, and the adjustment range is suggested to be 0.01 to 100]An interval.
The loss function is used for training the falling detection model, the training target is equivalent to enabling the falling detection model to output a correct falling detection result, and meanwhile, the identity recognition module cannot recognize the identity of people, so that the falling detection model is forced to extract high-order features irrelevant to the people, and finally the detection performance of the detection module is not influenced by the difference of the people.
By removing the identity information, the influence of the identity of the person can be eliminated, which is considered that walking occupies most of activity cycles in the daily life of the person, and related researches show that the gait information of the person is unique and contains a large amount of identity information, and the removal of the identity information can enable the fall detection model to be unrelated with the identity information of the person, thereby enhancing the generalization capability of the fall detection model among different persons.
In some embodiments, the data expansion model is used for performing simulation expansion on an input real fall sample based on a differential automatic encoder principle to obtain the fall sample.
It should be noted that the fall of a person involves coordinated movements of various parts of the whole body, which also creates a variety of falls. Relevant biological studies have shown that a person spontaneously swings limbs during a fall to seek balance or prevent injury, and this protective limb movement affects the speed of movement of the trunk, which is similar to when jumping up to shoot a basketball, with the trunk leaning backwards. Thus, the diversity of falls is present in the movement of the limbs. The traditional fall detection method based on Wi-Fi equipment only considers the free fall of the trunk and neglects the participation of limbs in the fall process, so that the real fall behavior is difficult to detect correctly.
In generating a fall/non-fall sample dataset, it is necessary to both simulate the diverse movements of the limbs during a fall and follow the general movement trends of the trunk. As shown in fig. 5, a fall case of one dummy and a fall case of 50 real persons are shown. It can be seen that a dummy without self-protection awareness has two phases of Acceleration (Acceleration) in which the trunk speed is gradually increased from 0m/s to about 4m/s for about 1000 milliseconds, and Deceleration (decelaration) in which the speed is abruptly decreased from 4m/s to 0m/s for about 300 milliseconds, typical for a fall. In the case of a real person, the deviation occurs to a different extent. In order to ensure that the generated simulated fall sample obeys this pattern.
As shown in fig. 6, the relationship between the nuclear velocity and the residual velocity in a fall is shown. The movement velocity of the dummy when the dummy falls was defined as the nuclear velocity (Kernel velocity), the Kernel velocity (Kernel velocity) was subtracted from the fall velocity of the real person, and the remaining Residual velocity (Residual velocity) was used as the simulation target.
On the basis of obtaining the residual speed from the human body speed characteristics in the input real falling sample, the input real falling sample can be subjected to simulation capacity expansion by using a differential automatic Encoder (VAE) principle to obtain a capacity expansion sample data set.
In some embodiments, performing simulation capacity expansion on an input real fall sample based on a principle of a differential automatic encoder to obtain a capacity expansion sample data set, including: mapping the real fall sample to a low-dimensional feature; and randomly sampling the low-dimensional features to obtain a plurality of simulated falling samples, and taking the real falling samples and the simulated falling samples as expansion sample data sets.
It will be appreciated that the VAE network structure of the capacity expansion layer is shown in FIG. 7, and the VAE network may comprise two main parts: an Encoder (Encoder) and a Decoder (Decoder), the input real fall sample x may be a vector of length 1500, representing a residual velocity sequence lasting 1500 milliseconds.
Mapping the real fall sample to low-dimensional features may include: extracting a mean vector and a standard deviation vector from a real falling sample; and randomly extracting a plurality of sampling data from N (0,1) which obeys normal distribution, adding the plurality of sampling data with the mean vector respectively, and multiplying the plurality of sampling data with the standard deviation vector respectively to obtain the low-dimensional feature.
It is understood that the encoder first extracts the high-order features using two Fully-connected layers (FC), and then extracts the mean (μ) vector and the standard deviation (σ) vector, representing the normal distribution parameters to which the hidden layer vector obeys, using the two Fully-connected layers. The encoder randomly extracts a sample from a standard normal distribution N (0,1), and the sample is added to mu and multiplied by sigma, so that the low-dimensional feature Z obtained obeys normal distribution with the mean value of mu and the standard deviation of sigma.
Decoder mapping Z to output vector using two fully-connected layers
Figure BDA0002740510970000111
Representing the generated simulated fall sample.
In the VAE training phase, the training goal is to let the output go
Figure BDA0002740510970000112
Approximating x, and Z approximates a standard normal distribution, so the Loss function (Loss) of VAE can be:
Figure BDA0002740510970000121
wherein, KL [, ] is]Is the KL Divergence of the two variables (Kullback-Leibler Divergence),
Figure BDA0002740510970000122
is the Euclidean distance of two vectors, C is a positive constant, and is generally constructedThe value of the constant is artificially and dynamically adjusted according to the system performance, and the adjustment range is recommended to be in [0.01,100 ]]An interval.
In the actual use process, the residual speed in the real falling sample is used for training the VAE network, after the training is finished, the encoder of the VAE network is discarded and the decoder is reserved, different random values are given to the low-dimensional characteristic Z and input into the decoder, then the input of the decoder is superposed with the kernel speed, and finally a new falling sample is obtained. Because the sampling data is randomly selected, the generated new samples have randomness, and theoretically, most of falling situations can be covered when the simulation falling samples are enough.
In some embodiments, determining the human body velocity characteristics based on channel state information of the wireless signals acquired in the target area comprises: based on a preset time threshold, dividing the channel state information into a plurality of channel state sub-segments; obtaining autocorrelation functions of a plurality of channel state sub-segments; obtaining the position of the first extreme point of any autocorrelation function, and the human body speed characteristic is based on a formula
Figure BDA0002740510970000123
Determining where τ is0Representing the position of the first extreme point of any autocorrelation function; x is the number of0Representing the position of the first extreme point of the sinc (x) function; x represents a positive real number; λ represents the wavelength of the wireless signal.
As shown in fig. 3, in a common home scene, a wireless signal may be shielded by a wall and reflected by a plurality of objects, so that a signal reflected by a human body can be received by a terminal device after being reflected for a plurality of times, and thus a signal propagation path cannot be analyzed geometrically, and a human body movement speed cannot be calculated by using a traditional signal propagation analysis method based on geometry.
The embodiment of the invention does not adopt the traditional signal propagation analysis method based on geometry, adopts the signal analysis method based on statistics,wherein each object in the environment, such as a table, a wall, a human body, a ceiling, etc., is regarded as a potential signal scatterer (scatter), and the contribution component of the ith scatterer to the CSI is denoted as Hi(f，t)。
By the superposition principle of electromagnetic wave signals, the total CSI can be expressed as formula one:
Figure BDA0002740510970000131
wherein omegad(t) represents all static scatterers within the target region, Ωs(t) represents all dynamic scatterers within the target region.
For the ith scatterer, if a coordinate system is established with its motion direction as the Z axis, its contribution component H to CSIi(f, t) can be further decomposed into equation two:
Figure BDA0002740510970000132
where k is 2 pi/λ, λ represents a signal wavelength, and may be obtained indirectly through a frequency band in which the terminal device operates: λ c/f, c denotes the speed of light, f denotes the frequency band in which the terminal operates, hi(α, β, f, t) represents the contribution of the ith scatterer to the CSI in the directions of a pitch angle α and a yaw angle β, viRepresenting the motion velocity of the ith scatterer, the formula adopts double integration, and the contribution of the scatterer to the CSI in all spatial directions is considered.
Based on the theory of relevant electromagnetism, h can be approximatedi(α, β, f, t) is considered a gaussian random variable and has three attributes: (1) for arbitrary α and β, hi(α, β, f, t) are circularly symmetric gaussian random variables and have the same variance; (2) for any two different scatterers i and j, hiAnd hjAre independent of each other; (3) for any two directions (alpha) of the scatterer1，β1) And (alpha)2，β2)，h(α1，β1F, t) and h (. alpha.) (a)2，β2F, t) are independent of each other.
Therefore, in combination with the first formula and the second formula, the CSI is expressed as the sum of a set of gaussian random variables, and the relationship between the Auto-correlation function (ACF) of the derived CSI and the human body velocity characteristic v is represented as the third formula:
Figure BDA0002740510970000133
wherein, Cov [, ] is]Representing the covariance, σ, of two random variablesi 2(f) Represents hiThe last step of the approximation process is that the trunk of the body of the person during the fall contributes most of the reflected signals, other parts can be ignored, and the speeds of all parts of the trunk are approximately the same, so that the variance of (i) represents the time lag (time lag) of the autocorrelation function
Figure BDA0002740510970000141
vi≈v。
Through the derivation of the formula three, the relationship between the ACF function of the CSI and the human body velocity characteristic is established, the human body velocity characteristic can be calculated by comparing the ACF with the extreme point of the sinc (x) function, and the relationship can be based on the formula four:
Figure BDA0002740510970000142
determining where τ is0Representing the position of the first extreme point of any autocorrelation function; x is the number of0Representing the position of the first extreme point of the sinc (x) function; x represents a positive real number; λ represents the wavelength of the wireless signal.
As shown in fig. 4, speed change from walking to falling of a person and an ACF function are shown, in an actual use process, a CSI sequence acquired by a terminal device is cut into sub-segments according to a preset time threshold, the time threshold may be 100ms, each segment is self-correlated with itself to obtain an ACF, and a first ACF is foundPosition of extreme point0Then, the human body velocity characteristic is calculated using formula four.
In the following, the method 30 for fall detection based on a Wi-Fi device according to the embodiment of the present invention is described, and the fall detection apparatus described below and the method for fall detection based on a Wi-Fi device described above may be referred to correspondingly.
As shown in fig. 9, an embodiment of the present invention further provides a fall detection method 30 based on a Wi-Fi device, including: an extraction module 10 and a detection module.
And the extraction module 10 is configured to determine a human body velocity characteristic based on the channel state information of the wireless signal acquired in the target area. The detection module 20 is used for inputting the human body speed characteristics into the fall detection model to obtain a human body fall identification result output by the fall detection model; the fall detection model is obtained by training based on a fall/non-fall sample data set; the non-falling samples in the falling/non-falling sample data set are from the really acquired samples of the daily activities of the people, and the falling samples in the falling/non-falling sample data set are from the falling samples generated by the data expansion model simulation.
The fall detection method 30 based on the Wi-Fi device provided by the embodiment of the present invention is used for executing the fall detection method based on the Wi-Fi device, and the specific implementation manner thereof is consistent with the implementation manner of the method, and is not described herein again.
Fig. 10 illustrates a physical structure diagram of an electronic device, which may include, as shown in fig. 8: a processor (processor)101, a communication Interface (communication Interface)102, a memory (memory)103 and a communication bus 104, wherein the processor 101, the communication Interface 102 and the memory 103 complete communication with each other through the communication bus 104. The processor 101 may invoke logic instructions in the memory 103 to perform a Wi-Fi device based fall detection method comprising: determining human body speed characteristics based on channel state information of wireless signals acquired in a target area; inputting the human body speed characteristics into a falling detection model to obtain a human body falling identification result output by the falling detection model; wherein the fall detection model is obtained by training based on a fall/non-fall sample data set; the non-falling samples in the falling/non-falling sample data set are from samples of the daily activities of the really acquired personnel, and the falling samples in the falling/non-falling sample data set are obtained by preprocessing the input real falling samples based on a data capacity expansion model.
In addition, the logic instructions in the memory 103 may be implemented in the form of software functional units and stored in a computer readable storage medium when the logic instructions are sold or used as independent products. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
In another aspect, embodiments of the present invention also provide a computer program product, where the computer program product includes a computer program stored on a non-transitory computer-readable storage medium, where the computer program includes program instructions, and when the program instructions are executed by a computer, the computer can execute the method for fall detection based on a Wi-Fi device provided in the above-mentioned method embodiments, where the method includes: determining human body speed characteristics based on channel state information of wireless signals acquired in a target area; inputting the human body speed characteristics into a falling detection model to obtain a human body falling identification result output by the falling detection model; wherein the fall detection model is obtained by training based on a fall/non-fall sample data set; the non-falling samples in the falling/non-falling sample data set are from samples of the daily activities of the really acquired personnel, and the falling samples in the falling/non-falling sample data set are obtained by preprocessing the input real falling samples based on a data capacity expansion model.
In yet another aspect, embodiments of the present invention further provide a non-transitory computer-readable storage medium, on which a computer program is stored, where the computer program is implemented by a processor to execute the method for fall detection based on a Wi-Fi device provided in the foregoing embodiments, where the method includes: determining human body speed characteristics based on channel state information of wireless signals acquired in a target area; inputting the human body speed characteristics into a falling detection model to obtain a human body falling identification result output by the falling detection model; wherein the fall detection model is obtained by training based on a fall/non-fall sample data set; the non-falling samples in the falling/non-falling sample data set are from samples of the daily activities of the really acquired personnel, and the falling samples in the falling/non-falling sample data set are obtained by preprocessing the input real falling samples based on a data capacity expansion model.
The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.
Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims (10)

1. A method for fall detection based on a Wi-Fi device, comprising:
determining human body speed characteristics based on channel state information of wireless signals acquired in a target area;
inputting the human body speed characteristics into a falling detection model to obtain a human body falling identification result output by the falling detection model;
wherein the fall detection model is obtained by training based on a fall/non-fall sample data set; the non-falling samples in the falling/non-falling sample data set are from samples of the daily activities of the really acquired personnel, and the falling samples in the falling/non-falling sample data set are obtained by preprocessing the input real falling samples based on a data capacity expansion model.
2. The Wi-Fi device-based fall detection method according to claim 1, wherein the data expansion model is configured to perform simulated expansion on an input real fall sample based on a differential automatic encoder principle to obtain a fall sample.
3. The Wi-Fi apparatus-based fall detection method according to claim 2, wherein the artificial extension of the input real fall sample based on the differential automatic encoder principle to obtain the fall sample comprises:
mapping the real fall sample to a low-dimensional feature;
randomly sampling the low-dimensional features to obtain a plurality of simulated fall samples, and taking the real fall samples and the simulated fall samples as the fall samples.
4. The Wi-Fi device-based fall detection method of claim 3, wherein the mapping the real fall samples to low-dimensional features comprises:
extracting a mean vector and a standard deviation vector from the real falling sample;
randomly extracting a plurality of sampling data from N (0,1) obeying normal distribution, respectively adding the plurality of sampling data with the mean vector, and respectively multiplying with the standard deviation vector to obtain the low-dimensional feature.
5. The Wi-Fi apparatus-based fall detection method according to any one of claims 1 to 4, wherein the inputting the body velocity characteristics into a fall detection model to obtain the body fall recognition result output by the fall detection model comprises:
inputting the human body speed features into a feature extraction layer of the falling detection model, removing noise components in the human body speed features and identity information of a user, and reserving specific falling activity mode information as high-order features;
and inputting the high-order features into a falling detection layer of the falling detection model to obtain the human body falling identification result.
6. The Wi-Fi device-based fall detection method of claim 5, wherein the identity information comprises: at least one of height, weight, posture, and fall habit characteristics.
7. The Wi-Fi apparatus-based fall detection method according to any one of claims 1 to 4, wherein the determining the body velocity characteristics based on channel state information of the wireless signals acquired at the target area comprises:
based on a preset time threshold, dividing the channel state information into a plurality of channel state sub-segments;
obtaining autocorrelation functions of a plurality of the channel state sub-segments;
obtaining the position of the first extreme point of any one of the autocorrelation functions, wherein the human body speed characteristic is based on a formula
Figure FDA0002740510960000021
Determining where τ is0Representing the position of a first extreme point of any of said autocorrelation functions; x is the number of0Representing the position of the first extreme point of the sinc (x) function; λ represents the wavelength of the wireless signal.
8. A method for fall detection based on a Wi-Fi device, comprising:
the extraction module is used for determining human body speed characteristics based on the channel state information of the wireless signals acquired in the target area;
the detection module is used for inputting the human body speed characteristics into a falling detection model and obtaining a human body falling identification result output by the falling detection model;
wherein the fall detection model is obtained by training based on a fall/non-fall sample data set; the non-falling samples in the falling/non-falling sample data set are from samples of the daily activities of the really acquired personnel, and the falling samples in the falling/non-falling sample data set are obtained by preprocessing the input real falling samples based on a data capacity expansion model.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor when executing the program realizes the steps of the Wi-Fi device based fall detection method according to any of claims 1 to 7.
10. A non-transitory computer-readable storage medium, having stored thereon a computer program, which, when being executed by a processor, carries out the steps of the method for fall detection based on a Wi-Fi device of any one of claims 1 to 7.
CN202011148763.2A 2020-10-23 Fall detection method and system based on Wi-Fi equipment Active CN112346050B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011148763.2A CN112346050B (en) 2020-10-23 Fall detection method and system based on Wi-Fi equipment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011148763.2A CN112346050B (en) 2020-10-23 Fall detection method and system based on Wi-Fi equipment

Publications (2)

Publication Number Publication Date
CN112346050A true CN112346050A (en) 2021-02-09
CN112346050B CN112346050B (en) 2024-06-28

Family

ID=

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113499064A (en) * 2021-07-07 2021-10-15 郑州大学 Wi-Fi perception human body tumbling detection method and system in bathroom scene
CN113712538A (en) * 2021-08-30 2021-11-30 平安科技（深圳）有限公司 Fall detection method, device, equipment and storage medium based on WIFI signal
CN114692683A (en) * 2022-03-23 2022-07-01 南京邮电大学 CSI-based fall detection method and device and storage medium
CN115457732A (en) * 2022-08-24 2022-12-09 电子科技大学 Fall detection method based on sample generation and feature separation
CN116027324A (en) * 2023-03-24 2023-04-28 德心智能科技(常州)有限公司 Fall detection method and device based on millimeter wave radar and millimeter wave radar equipment
CN116310734A (en) * 2023-04-25 2023-06-23 慧铁科技有限公司 Fault detection method and system for railway wagon running part based on deep learning
CN116840835A (en) * 2022-05-05 2023-10-03 南方科技大学 Fall detection method, system and equipment based on millimeter wave radar
CN117471421A (en) * 2023-12-25 2024-01-30 中国科学技术大学 Training method of object falling detection model and falling detection method
CN117556333A (en) * 2023-12-18 2024-02-13 国网江苏省电力有限公司双创中心 Falling detection method and device, electronic equipment and storage medium
CN114692683B (en) * 2022-03-23 2024-07-05 南京邮电大学 Fall detection method and device based on CSI and storage medium

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130197856A1 (en) * 2011-06-17 2013-08-01 James R. Barfield Method and system for discerning a false positive in a fall detection signal
CN105590409A (en) * 2016-02-26 2016-05-18 江苏大学 Human body tumble detection method and human body tumble detection system based on big data
CN109635837A (en) * 2018-11-10 2019-04-16 天津大学 A kind of carefree fall detection system of scene based on commercial wireless Wi-Fi
CN110555368A (en) * 2019-06-28 2019-12-10 西安理工大学 Fall-down behavior identification method based on three-dimensional convolutional neural network
CN111310647A (en) * 2020-02-12 2020-06-19 北京云住养科技有限公司 Generation method and device for automatic identification falling model
CN111597877A (en) * 2020-04-02 2020-08-28 浙江工业大学 Fall detection method based on wireless signals

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130197856A1 (en) * 2011-06-17 2013-08-01 James R. Barfield Method and system for discerning a false positive in a fall detection signal
CN105590409A (en) * 2016-02-26 2016-05-18 江苏大学 Human body tumble detection method and human body tumble detection system based on big data
CN109635837A (en) * 2018-11-10 2019-04-16 天津大学 A kind of carefree fall detection system of scene based on commercial wireless Wi-Fi
CN110555368A (en) * 2019-06-28 2019-12-10 西安理工大学 Fall-down behavior identification method based on three-dimensional convolutional neural network
CN111310647A (en) * 2020-02-12 2020-06-19 北京云住养科技有限公司 Generation method and device for automatic identification falling model
CN111597877A (en) * 2020-04-02 2020-08-28 浙江工业大学 Fall detection method based on wireless signals

Cited By (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113499064A (en) * 2021-07-07 2021-10-15 郑州大学 Wi-Fi perception human body tumbling detection method and system in bathroom scene
CN113712538A (en) * 2021-08-30 2021-11-30 平安科技（深圳）有限公司 Fall detection method, device, equipment and storage medium based on WIFI signal
CN114692683A (en) * 2022-03-23 2022-07-01 南京邮电大学 CSI-based fall detection method and device and storage medium
CN114692683B (en) * 2022-03-23 2024-07-05 南京邮电大学 Fall detection method and device based on CSI and storage medium
CN116840835B (en) * 2022-05-05 2024-05-10 南方科技大学 Fall detection method, system and equipment based on millimeter wave radar
CN116840835A (en) * 2022-05-05 2023-10-03 南方科技大学 Fall detection method, system and equipment based on millimeter wave radar
CN115457732B (en) * 2022-08-24 2023-09-01 电子科技大学 Fall detection method based on sample generation and feature separation
CN115457732A (en) * 2022-08-24 2022-12-09 电子科技大学 Fall detection method based on sample generation and feature separation
CN116027324A (en) * 2023-03-24 2023-04-28 德心智能科技(常州)有限公司 Fall detection method and device based on millimeter wave radar and millimeter wave radar equipment
CN116310734A (en) * 2023-04-25 2023-06-23 慧铁科技有限公司 Fault detection method and system for railway wagon running part based on deep learning
CN116310734B (en) * 2023-04-25 2023-12-15 慧铁科技股份有限公司 Fault detection method and system for railway wagon running part based on deep learning
CN117556333A (en) * 2023-12-18 2024-02-13 国网江苏省电力有限公司双创中心 Falling detection method and device, electronic equipment and storage medium
CN117471421A (en) * 2023-12-25 2024-01-30 中国科学技术大学 Training method of object falling detection model and falling detection method
CN117471421B (en) * 2023-12-25 2024-03-12 中国科学技术大学 Training method of object falling detection model and falling detection method

Similar Documents

Publication Publication Date Title
Wang et al. Fall detection based on dual-channel feature integration
Jiang et al. Towards 3D human pose construction using WiFi
Cippitelli et al. Radar and RGB-depth sensors for fall detection: A review
CN108226892B (en) Deep learning-based radar signal recovery method in complex noise environment
Gu et al. Paws: Passive human activity recognition based on wifi ambient signals
CN108875708A (en) Behavior analysis method, device, equipment, system and storage medium based on video
CN106462725A (en) Systems and methods of monitoring activities at a gaming venue
CN108135469A (en) Estimated using the eyelid shape of eyes attitude measurement
CN106796656A (en) Away from the depth of time-of-flight camera
CN108683724A (en) A kind of intelligence children's safety and gait health monitoring system
WO2008007471A1 (en) Walker tracking method and walker tracking device
CN109840467A (en) A kind of in-vivo detection method and system
CN107577451A (en) More Kinect human skeletons coordinate transformation methods and processing equipment, readable storage medium storing program for executing
CN110070029A (en) A kind of gait recognition method and device
CN106960473B (en) behavior perception system and method
Cai et al. GBDT‐Based Fall Detection with Comprehensive Data from Posture Sensor and Human Skeleton Extraction
CN110113116A (en) Human bodys' response method based on WIFI channel information
Mokhtari et al. Non-wearable UWB sensor to detect falls in smart home environment
CN113378649A (en) Identity, position and action recognition method, system, electronic equipment and storage medium
CN107122711A (en) A kind of night vision video gait recognition method based on angle radial transformation and barycenter
Janakaraj et al. STAR: Simultaneous tracking and recognition through millimeter waves and deep learning
CN115861915A (en) Fire fighting access monitoring method, fire fighting access monitoring device and storage medium
CN115116127A (en) Fall detection method based on computer vision and artificial intelligence
Yan et al. Human-object interaction recognition using multitask neural network
CN111507301A (en) Video processing method, video processing device, computer equipment and storage medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant