WO2018107374A1 - 跌倒检测设备、跌倒检测方法及装置 - Google Patents

跌倒检测设备、跌倒检测方法及装置 Download PDF

Info

Publication number
WO2018107374A1
WO2018107374A1 PCT/CN2016/109844 CN2016109844W WO2018107374A1 WO 2018107374 A1 WO2018107374 A1 WO 2018107374A1 CN 2016109844 W CN2016109844 W CN 2016109844W WO 2018107374 A1 WO2018107374 A1 WO 2018107374A1
Authority
WO
WIPO (PCT)
Prior art keywords
fall
fall detection
value
feature
human body
Prior art date
Application number
PCT/CN2016/109844
Other languages
English (en)
French (fr)
Inventor
李景振
聂泽东
刘宇航
Original Assignee
深圳先进技术研究院
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 深圳先进技术研究院 filed Critical 深圳先进技术研究院
Priority to PCT/CN2016/109844 priority Critical patent/WO2018107374A1/zh
Publication of WO2018107374A1 publication Critical patent/WO2018107374A1/zh

Links

Images

Classifications

    • 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
    • G—PHYSICS
    • G08—SIGNALLING
    • G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
    • G08B21/00—Alarms responsive to a single specified undesired or abnormal condition and not otherwise provided for
    • G08B21/02—Alarms for ensuring the safety of persons
    • G08B21/04—Alarms for ensuring the safety of persons responsive to non-activity, e.g. of elderly persons

Definitions

  • the present disclosure relates to communication and signal processing techniques, for example, to a fall detection device, a fall detection method, and apparatus.
  • fall detection technology mainly includes fall detection based on image video sensor, fall detection based on environment-mounted sensor, and fall detection based on wearable sensor.
  • the image video sensor-based fall detection method requires one or more cameras to be installed in the surrounding environment, so that the cost of the fall detection system is high, and the detection range is also limited to a fixed area; based on the environment-mounted sensor
  • the fall detection method mainly captures the motion information of the human body and collects the body characteristic data related to the human body through pressure sensors, microphones, etc.
  • the method is susceptible to the surrounding environment; the wearable sensor fall detection method mainly utilizes the acceleration sensor
  • the gyro sensor acquires information such as acceleration or angular acceleration of the human body.
  • the detection result of the method has a great relationship with the position of the wearing. If the wearing position is incorrect, the detection result has a large error.
  • the embodiment of the present invention provides a fall detection device, a fall detection method, and a device, which optimize the fall detection technology in the related art, and improve the versatility of the fall detection scheme.
  • an embodiment of the present invention provides a fall detection device, including: a transmitter and a receiver;
  • the transmitter and the receiver are configured to communicate by using a human body as a transmission medium by capacitive coupling, and a transmitting electrode of the transmitter and a receiving electrode of the receiver form a forward loop.
  • the ground electrode of the transmitter and the ground electrode of the receiver form a backward loop;
  • the transmitter is configured to generate a fall detection signal and couple the fall detection signal to a human body;
  • the receiver is configured to acquire a received measurement value of the fall detection signal transmitted through a human body; and, according to the received measurement value, extract a fall for characterizing a difference in channel characteristics of the backward loop when the human body falls and does not fall Describe the feature value of the feature; perform fall detection according to the feature value of the fall description feature.
  • the embodiment of the present invention further provides a fall detection method, which is applied to a receiver of a fall detection device according to the embodiment of the present invention, where the method includes:
  • a fall detection is performed according to the feature value of the fall description feature.
  • an embodiment of the present invention further provides a fall detection device, where the device includes:
  • Receiving a measurement value acquisition module configured to acquire a received measurement value of the fall detection signal by the receiving electrode
  • An eigenvalue extraction module configured to extract, according to the received measurement value, a feature value of a fall description feature for characterizing a difference in channel characteristics of a backward loop when the human body falls and does not fall;
  • the fall detection module is configured to perform a fall detection according to the feature value of the fall description feature.
  • an embodiment of the present invention further provides a non-transitory computer readable storage medium storing computer executable instructions for performing the fall detection method described above.
  • the fall detection device, the fall detection method and the device provided by the embodiment of the present invention form a human body communication signal path by using a fall detection device that uses the human body as a transmission medium to communicate by using a capacitive coupling method, and utilizes a channel of a backward loop when the human body falls and does not fall.
  • the feature difference is based on the signal received by the receiver for fall detection, which solves various defect problems in the multi-class fall detection technology in the related art, optimizes the fall detection technology in the related technology, and improves the versatility of the fall detection scheme.
  • the fall detection device has the advantages of small size, light weight, low power consumption, insensitivity to wearing position and high detection precision, and provides a new idea for the development of fall detection technology.
  • FIG. 1a is a structural diagram of a fall detection device according to Embodiment 1 of the present invention.
  • FIG. 1b is a structural diagram of a transmitter according to Embodiment 1 of the present invention.
  • FIG. 1c is a structural diagram of a receiver according to Embodiment 1 of the present invention.
  • FIG. 2 is a flowchart of a fall detection method according to Embodiment 2 of the present invention.
  • Embodiment 3 is a flowchart of a fall detection method according to Embodiment 3 of the present invention.
  • Embodiment 4 is a flowchart of a fall detection method according to Embodiment 4 of the present invention.
  • FIG. 5 is a structural diagram of a fall detection device according to Embodiment 5 of the present invention.
  • Human body communication is a new communication technology with "human body” as communication medium, which has the advantages of low power consumption, micro volume, convenience and speed.
  • human body communication can be divided into current coupling and capacitive coupling.
  • the propagation characteristics of the human body communication channel are determined by the forward loop and the backward loop.
  • the transmitting electrode of the transmitter of the human body communication and the receiving electrode of the receiver form a forward loop through the human body; the ground electrode of the transmitter and the ground electrode of the receiver form a backward loop through the ground plane (earth).
  • the inventor discovered through research that if the transmitter and the receiver of the above-mentioned capacitively coupled human body communication are placed in the human body with a certain height value from the ground when the human body is standing, the transmitter and the receiver are in the event of a fall of the human body.
  • the distance between the ground electrode and the ground plane will change abruptly, and the characteristics of the backward loop of the human body communication will also change, causing the channel gain to change. Therefore, Fall detection can be achieved by analyzing the difference between the channel gain curve when the human body falls and does not fall.
  • FIG. 1 is a structural diagram of a fall detection device according to Embodiment 1 of the present invention. As shown in FIG. 1a, the fall detection device includes a transmitter 11 and a receiver 12.
  • the transmitter 11 and the receiver 12 are configured to communicate by using a human body as a transmission medium by means of capacitive coupling, and the transmitting electrode 111 of the transmitter 11 and the receiving electrode 121 of the receiver 12 form a forward loop.
  • the ground electrode 112 of the transmitter 11 and the ground electrode 122 of the receiver 12 constitute a backward loop.
  • FIG. 1a a schematic representation of the placement of the transmitter 11 and the receiver 12 is shown in Figure 1a.
  • the ground electrode 112 of the transmitter 11 and the ground electrode 122 of the receiver 12 and the ground are ensured when the user sends a fall.
  • the technical effects of the embodiments of the present invention can be achieved by changing the distance between the present invention. Therefore, the placement positions of the transmitter 11 and the receiver 12 described above are optional in the embodiment of the present invention.
  • the transmitter 11 can be placed at the upper arm of the human body, wherein the transmitting electrode 111 is in close contact with the skin surface of the upper arm, and the ground electrode 112 is suspended in the air.
  • the receiver 12 is placed at the abdomen of the human body, wherein the receiving electrode 121 abuts against the skin surface of the abdomen, and the ground electrode 122 is also suspended in the air, i.e., the placement position as shown in Fig. 1a.
  • the overall shape of the transmitter 11 may be a cylindrical box.
  • the upper surface and the lower surface of the box are made of a metal material, and the upper surface of the box serves as the transmitting electrode 111 of the transmitter 11.
  • the lower surface of the box serves as the ground electrode 112 of the transmitter 11, and the side of the box is made of a plastic material for isolating the transmitting electrode 111 and the ground electrode 112 to avoid contact therebetween; similarly, the overall shape of the receiver 12
  • the same may be a cylindrical box, as shown in FIG. 1a, the upper surface and the lower surface of the box are made of a metal material, the upper surface of the box serves as the receiving electrode 121 of the receiver 12, and the lower surface of the box serves as the receiver 12.
  • the ground electrode 122 has a plastic material on its side for isolating the receiving electrode 121 and the ground electrode 122.
  • the transmitter 11 is configured to generate a fall detection signal and couple the fall detection signal to a human body;
  • the receiver 12 is configured to acquire a received measurement value of the fall detection signal transmitted through a human body, and extract, according to the received measurement value, a backward circuit for characterizing a human body falling and not falling
  • the fall of the channel characteristic difference describes the feature value of the feature; the fall detection is performed according to the feature value of the fall description feature.
  • FIG. 1b An optional structure diagram of a transmitter is shown in FIG. 1b. As shown in FIG. 1b, the transmitter includes:
  • a microprocessor 1101 a microprocessor 1101, a DDS (Direct Digital Synthesizer) 1102, a balun converter 1103, a low pass filter 1104, a transmitting electrode 111, and a ground electrode 112;
  • DDS Direct Digital Synthesizer
  • the microprocessor 1101 is configured to control the DDS 1102 to generate a single-ended sine wave signal within a set frequency range;
  • the balun converter 1103 is configured to convert the single-ended sine wave signal output by the DDS 1102 into a double-ended sine wave signal, and output to the low-pass filter 1104;
  • the low-pass filter 1104 is configured to perform low-pass filtering on the double-ended sine wave signal, and couple the fall detection signal generated after filtering to the human body through the transmitting electrode 111;
  • the transmitting electrode 111 is disposed in contact with a skin surface of a human body, and the ground electrode 112 is disposed in an insulated connection with the transmitting electrode 111.
  • the channel gain refers to the attenuation and fading characteristics of the channel itself. Generally, it can be measured by the ratio of the received signal to the transmitted signal. Therefore, if it is sent.
  • the channel gain can be measured simply by receiving the signal, that is, the received signal change curve is used as the channel gain change curve, and the fall detection is performed by analyzing the signal characteristics of the received signal change curve.
  • the function of the transmitter is mainly to send a constant amplitude transmission signal as a fall detection signal.
  • the transmission signal may be a sine wave.
  • the sine wave sent by the transmitter can be a variable frequency sine wave.
  • the fall detection signal may be a periodic frequency conversion signal; wherein the frequency conversion signal starts from the first frequency in one cycle, and increases the value to the second frequency according to the set frequency span.
  • the first frequency may be 1 MHz
  • the second frequency may be 100 MHz
  • the set frequency span may be 0.5 MHz/ms.
  • the fall detection signal is a periodic sine wave signal, and starts from 1 MHz in one cycle, and increases by 0.5 MHz every 1 ms until the final frequency becomes 100 MHz.
  • the microprocessor 1101 is mainly configured to generate a control signal, and the DDS 1102 is controlled to output the fall detection signal.
  • the microprocessor 1101 can be an FPGA (Field- Programmable Gate Array, Field Programmable Gate Array, DSP (Digital Signal Processing) or microcontroller.
  • the cutoff frequency of the low pass filter 1104 is determined by the highest frequency of the fall detection signal, and is set to filter out a variety of clutter and interference, for example, if the frequency of the fall detection signal ranges from 1 MHz to 100 MHz, then The cutoff frequency of the low pass filter 1104 can be selected to be 120 MHz.
  • Figure 1c shows an optional structure of the receiver, as shown in Figure 1b, the receiver includes: a microprocessor 1201, a receiving electrode 121 and a ground electrode 122;
  • the receiving electrode 121 is configured to acquire the received measurement value of the fall detection signal transmitted through the human body, and send the received measurement value to the microprocessor 1201;
  • the microprocessor 1201 is configured to extract, according to the received measurement value, a feature value for characterizing a fall description feature of a channel feature difference of the backward loop when the human body falls and does not fall; according to the fall description feature Characteristic value for fall detection;
  • the receiving electrode 121 is disposed in contact with a skin surface of a human body, and the ground electrode 122 is disposed in an insulated connection with the receiving electrode 121.
  • the receiver 12 may further include a plurality of signal processing devices, such as an amplifier or a filter, for inputting the received measurement value acquired by the receiving electrode 121 to the micro.
  • the fall detection device uses the human body as a transmission medium to form a human body communication signal path by capacitive coupling, and utilizes a channel characteristic difference of a backward loop when the human body falls and does not fall, and performs a fall detection based on a signal received by the receiver.
  • the utility model solves the defects of the multi-type fall detection technology in the related art, optimizes the fall detection technology in the related technology, and improves the versatility of the fall detection scheme.
  • the fall detection device has small volume, light weight and work. The advantages of low consumption, insensitivity to wearing position and high detection accuracy provide a new idea for the development of fall detection technology.
  • Embodiment 2 is a flowchart of a fall detection method according to Embodiment 2 of the present invention, which may be performed by a fall detection device, which may be implemented by software and/or hardware, and generally integrated into a receiver of a fall detection device.
  • a fall detection device which may be implemented by software and/or hardware, and generally integrated into a receiver of a fall detection device.
  • the microprocessor in the receiver it is used in conjunction with the transmitter of the fall detection device.
  • the method in this embodiment includes:
  • the received measurement value of the fall detection signal is acquired by the receiving electrode.
  • the fall detection signal sent by the transmitter is periodically selectable as a variable frequency signal
  • variable frequency signal starts from the first frequency in one cycle, and increases the value to the second frequency according to the set frequency span.
  • a feature value of a fall description feature for characterizing a difference in channel characteristics of the backward loop when the human body falls and does not fall is extracted.
  • the received signal change curve will be different from the received signal change curve when it is not falling due to the channel characteristics of the backward loop. Therefore, considering the periodicity of the fall detection signal transmitted by the transmitter, the feature value of the fall description feature of the received signal variation curve in one cycle can be extracted in one cycle as a time unit.
  • the fall description feature may refer to a feature parameter used to characterize a difference in channel characteristics of a backward loop when the human body falls and does not fall, and may be, for example, a mean, a variance, or a mean square error.
  • the fall detection may be performed by extracting an extreme point, a variance value, a slope value, or a difference between the maximum value and the minimum value in the received signal curve in one cycle.
  • fall detection is performed according to the feature value of the fall description feature.
  • Fall detection can be achieved by comparing the extracted feature values of the fall description feature with the set feature values of the fall description feature when falling or not falling.
  • the maximum value in the signal change curve is received to determine the maximum fall threshold. For example, if it is determined that when the maximum value of the received signal variation curve exceeds -20 dB and the probability of sending a fall is 85%, the maximum fall threshold can be set to -20 dB, and 85% can be set to confidence. Therefore, whenever the maximum value of the received signal change curve of the human body that needs to perform the fall detection exceeds -20 dB, it is judged that the user has fallen, and thus a certain fall warning strategy can be adopted.
  • the fall threshold (for example, the variance value and the slope value) of the plurality of signal features when the fall occurs may be used at the same time, that is, when the received signal change curve in one cycle satisfies the variance drop threshold and the slope value simultaneously When the threshold is dropped, it is judged that the user has fallen. To improve the accuracy of the fall detection.
  • the above received signal variation curve when falling and not falling may be used as a training sample, training a prediction model, optionally, a decision tree model, a clustering model, a neural network model, etc., and using the trained above The predictive model performs fall detection.
  • the fall detection method provided by the embodiment of the present invention adopts a fall detection device that communicates by using a human body as a transmission medium by a capacitive coupling method to form a human body communication signal path, and utilizes a difference in channel characteristics of a backward loop when the human body falls and does not fall, based on the receiver
  • the received signal is subjected to fall detection, which solves various defect problems of the multi-type fall detection technology in the related art, optimizes the fall detection technology in the related technology, and improves the versatility of the fall detection scheme.
  • the fall detection device has The advantages of small size, light weight, low power consumption, insensitivity to wearing position and high detection accuracy provide a new idea for the development of fall detection technology.
  • Embodiment 3 is a flowchart of a fall detection method according to Embodiment 3 of the present invention.
  • the received measurement value is extracted for characterizing the human body.
  • the characteristic value of the fall description feature of the difference in channel characteristics of the backward loop when falling and not falling is changed by: updating the received signal variation curve according to the received measurement value; wherein the received signal variation curve is within one cycle
  • the fall detection signal corresponds to; calculating, according to the received signal change curve, a feature value of at least one fall description feature;
  • the fall detection is changed according to the feature value of the fall description feature to: input the feature value of the at least one fall description feature into the pre-trained fall detection prediction model, and detect the output of the prediction model according to the fall detection As a result, a fall detection is performed.
  • the method in this embodiment may include:
  • the received measurement value of the fall detection signal is acquired by the receiving electrode.
  • the fall detection signal sent by the transmitter is periodically selectable as a frequency conversion signal
  • variable frequency signal starts from the first frequency in one cycle, and increases the value to the second frequency according to the set frequency span.
  • the received signal variation curve is updated according to the received measurement value.
  • the received signal variation curve corresponds to the fall detection signal in one cycle.
  • the received signal variation curve obtained by the receiver acquisition is usually accompanied by various noises.
  • updating the received signal variation curve according to the received measurement value may include:
  • the Kalman filter According to the received measurement value of the fall detection signal acquired at the time K and the Kalman filter a method for calculating a pre-estimated optimal value of the fall detection signal at the K+1th time, wherein K is a positive integer greater than or equal to 1; and using the pre-estimated optimal value of the fall detection signal, updating the Receive signal curve.
  • the Kalman filter algorithm is used to filter the signal, and the process is as follows:
  • k-1) A(k,k-1) ⁇ X(k-1
  • k-1) is the estimated value of the k-time calculated by the received signal change curve at time k-1
  • k-1) is the optimal value at time k-1
  • u(k) is the control amount at time k
  • A(k, k-1) is the state transition matrix
  • B(k) is the control weight matrix.
  • k-1) is the mean square error of the pre-estimated optimal value X(k
  • k-1) is the optimal value X(k-1
  • the mean square error, U(k) is the dynamic noise of the received signal variation curve at time k.
  • K(k) is the Kalman gain value
  • N(k) is the observed noise of the received signal change curve at time k
  • H(k) is the observation matrix of the received signal change curve.
  • the optimal value equation is updated to obtain the pre-estimated optimal value of the time, as shown in equation (4):
  • Z(k) is the measured value of the received signal change curve at time Z(k).
  • the received signal change curve is filtered until the stop condition is satisfied, thereby implementing Kalman filter processing of the received signal change curve.
  • the micro-processing of the receiver can realize the optimal estimation of the received measurement value by implementing the Kalman filtering algorithm, and update the received signal variation curve in the current cycle according to the optimal estimation value.
  • a feature value of at least one fall description feature is calculated according to the received signal change curve.
  • one or more of the following five feature quantities in one cycle can be extracted as a fall description feature for fall analysis, respectively: (1) the maximum value of the gain variation curve; (2) The difference between the maximum and minimum values in the gain curve; (3) the variance of the gain curve; (4) the slope of the gain curve; and (5) the average of the gain curve after the fall.
  • the gain variation curve can be approximated by receiving a signal variation curve.
  • the fall description feature can include at least one of the following:
  • a maximum value of the received signal variation curve a difference between a maximum value and a minimum value of the received signal change curve, a variance of the received signal change curve, a slope of the received signal change curve, and a received signal change curve after the fall average value.
  • fall description feature can also be other types, for example, a mean square error of a received signal variation curve, and the like.
  • the feature value of the at least one fall description feature is input into the pre-trained fall detection prediction model, and the fall detection is performed according to the output result of the fall detection prediction model.
  • the fall detection prediction model may be a decision tree model, a neural network model, a machine learning model, or the like.
  • the technical solution of the embodiment of the invention achieves the optimal estimation of the received measurement value by using the Kalman filter algorithm, and can reduce the interference of the noise and the interference in the human communication system on the received signal, so as to finally improve the accuracy of the fall detection.
  • the fall prediction can be performed under a plurality of different fall description features, and the parameters in the fall detection prediction model can be updated and learned in real time to continuously improve the accuracy of the fall detection.
  • the training fall detection prediction model can be changed to:
  • the training example includes: a receiving signal change curve acquired by a receiver in a fall detecting device that communicates with the human body as a transmission medium when the human body does not fall; and when the human body falls, a received signal variation curve obtained by a receiver in a fall detection device that communicates the human body as a transmission medium;
  • Such a setting can train the fall detection prediction model under one or more fall description features, so that the prediction result of the training model is closer to the actual result.
  • Embodiment 4 is a flowchart of a fall detection method according to Embodiment 4 of the present invention.
  • the present embodiment is changed based on the foregoing embodiment.
  • the model construction algorithm is changed to a decision tree algorithm.
  • the fall detection prediction model is changed to a decision tree model; wherein different child nodes in the decision tree model correspond to different fall description features;
  • the algorithm is constructed according to the set model, and the feature value of each of the received signal variation curves under at least one of the fall description features, and the training generates the fall detection prediction model to: determine the decision tree algorithm The value of the category attribute is a fall and no fall occurs; determining a non-category attribute of the decision tree algorithm is the fall description feature, and setting a standard value of the fall description feature; calculating information of the category attribute Entropy; calculating an information entropy of each of the non-category attributes according to a feature value of each of the received signal variation curves under at least one of the fall description features; information entropy according to the category attribute, and the non-category An information entropy of the attribute, calculating an information delta value of each of the non-category attributes, and determining a target test attribute according to an information delta value of each of the non-category attributes; using the target test attribute as a current child node And repeating the iteration based on the current child node to
  • the training example includes: a received signal change curve obtained by a receiver in a fall detection device that communicates by using the human body as a transmission medium when the human body does not fall; and when the human body falls, by using the human body as a transmission medium Receiver signal variation curve acquired by the receiver in the communication fall detection device.
  • determining a non-category attribute of the decision tree algorithm is the fall description feature, and setting a standard value of the fall description feature.
  • At least one of the fall description features according to each of the received signal profiles The underlying feature value calculates the information entropy of each of the non-category attributes.
  • the target test attribute is regarded as a current child node, and based on the current child node, repeating the iteration to re-determine the new test attribute as a new child node until the decision tree model is generated as the fall Detect the prediction model.
  • a decision tree is a classification method of a tree structure in which each internal node represents a test on an attribute, each branch represents a test output, and each leaf node represents a category.
  • the principle of the decision tree is: starting from the root node, testing the data samples, and dividing the data samples into different subsets of data samples according to different results, and each subset of data samples constitutes a child node. Each child node is further divided to generate a new child node, which is repeated until a specific termination criterion is reached.
  • decision tree model The process of generating a decision tree model is described below by an optional example: wherein, in this example, the decision tree model is generated using ID3 (Iterative Dichotomies 3).
  • the process of the model training phase is:
  • the category attribute refers to determining whether a fall occurs according to relevant factors, so the output result of the category attribute is divided into two types: a fall occurs, and no fall occurs.
  • the non-category attribute refers to the fall description feature. In the fall detection, the non-category attribute and the standard value (the value of the standard value can be determined according to the actual measurement result) as shown in Table 1:
  • non-category attributes There are five types of non-category attributes, as shown in Table 1, which are the maximum value of the received signal change curve, the difference between the maximum value and the minimum value in the received signal change curve, the variance of the received signal change curve, and the slope of the received signal change curve. The average value of the received signal change curve after falling.
  • the information entropy for non-category attributes is calculated as follows:
  • is the number of the maximum value A instance set of the received signal variation curve in the non-category attribute
  • is the number of the i-th instance set, that is,
  • is the number of variance C instance sets of the received signal variation curve in the non-category attribute
  • is the number of the i-th instance set, that is,
  • is the number of instances of the slope D of the received signal variation curve in the non-category attribute
  • is the number of the i-th instance set
  • is the number of the average E instance set of the received signal change curve after the fall in the non-category attribute
  • is the number of the i-th instance set, that is,
  • the received signal variation curve is updated according to the received measurement value.
  • the received signal variation curve corresponds to the fall detection signal in one cycle
  • a feature value of at least one fall description feature is calculated according to the received signal variation curve.
  • the feature value of the at least one fall description feature is input into the pre-trained decision tree model, and the fall detection is performed according to the output result of the decision tree model.
  • the technical solution of the embodiment of the present invention can construct a fall detection prediction model by using a decision tree algorithm, and can implement a feasible and effective result for a large data source in a relatively short time, and can construct a decision on a data set having many attributes. tree.
  • the body-based fall detection system is used to collect and acquire volunteers of different ages, heights and weights in daily life and during the fall process. Receive signal change curve in the received signal and introduce the received signal change curve into the decision tree for fall detection.
  • the decision tree model is modified by comparing the actual measurement results with the results predicted by the decision tree simulation. If the error of the fall detection result obtained by the decision tree is large, the number of training instance sets can be increased, the training set is more universal, and the decision tree is recalculated until the requirement is met to improve the fall prediction accuracy.
  • FIG. 5 is a structural diagram of a fall detection device according to Embodiment 5 of the present invention.
  • the fall detection device provided by the embodiment of the present invention can be applied to the receiver of the fall detection device according to the embodiment of the present invention, optionally, the microprocessor of the receiver. As shown in FIG. 5, the device includes:
  • the received measurement value acquisition module 510 is configured to acquire a received measurement value of the fall detection signal through the receiving electrode.
  • the feature value extraction module 520 is configured to extract, according to the received measurement value, a feature value of a fall description feature for characterizing a difference in channel characteristics of a backward loop when the human body falls and does not fall.
  • the fall detection module 530 is configured to perform fall detection according to the feature value of the fall description feature.
  • the fall detection device forms a human body communication signal path by using a human body as a transmission medium fall detection device by using a capacitive coupling method, and utilizes a channel characteristic difference of a backward loop when the human body falls and does not fall, based on the receiver.
  • the detected signal is subjected to fall detection, which solves various defect problems of the multi-class fall detection technology in the related art, optimizes the fall detection technology in the related technology, and improves the versatility of the fall detection scheme.
  • the fall detection device has a volume. Small, light weight, low power consumption, insensitivity to wearing position and high detection accuracy provide a new idea for the development of fall detection technology.
  • the fall detection signal may be a periodic frequency conversion signal
  • variable frequency signal starts from the first frequency in one cycle, and increases the value to the second frequency according to the set frequency span.
  • the feature value extraction module may include:
  • a change curve updating unit configured to update a received signal change curve according to the received measurement value; wherein the received signal change curve corresponds to the fall detection signal in one cycle;
  • the feature update value calculation unit is configured to calculate a feature value of the at least one fall description feature according to the received signal change curve.
  • the change curve updating unit may be set to:
  • the received signal variation curve is updated using a pre-estimated optimal value of the fall detection signal.
  • the feature update value calculation unit may be configured as:
  • the feature value of the at least one fall description feature is input into the pre-trained fall detection prediction model, and the fall detection is performed according to the output result of the fall detection prediction model.
  • the fall detection prediction model training module may further include: the fall detection prediction model training module may include:
  • the training instance set obtaining unit is configured to obtain a fall detection training instance set, wherein the training example includes: a receiving signal variation curve acquired by a receiver in the fall detecting device that communicates by using the human body as a transmission medium when the human body does not fall. And a received signal variation curve obtained by a receiver in a fall detection device that communicates with the human body as a transmission medium when a human body falls;
  • a model training unit configured to construct an algorithm according to the set model and the feature value of each of the received signal variation curves under at least one of the fall description features to train the fall detection prediction model.
  • the model construction algorithm may be a decision tree algorithm, and the fall detection prediction model may be a decision tree model;
  • the different child nodes in the decision tree model correspond to different fall description features.
  • model training unit may be set to:
  • the target test attribute is taken as a current child node, and based on the current child node, the iterative iteration is repeated to re-determine the new test attribute as a new child node until the decision tree model is generated as the fall detection prediction model.
  • the fall description feature may include at least one of the following:
  • a maximum value of the received signal variation curve a difference between a maximum value and a minimum value of the received signal change curve, a variance of the received signal change curve, a slope of the received signal change curve, and a received signal change curve after the fall average value.
  • the fall detection device provided by the embodiment of the present invention can be used to perform the fall detection method provided by any embodiment of the present disclosure, and has corresponding functional modules to achieve the same beneficial effects.
  • the embodiment of the present invention further provides a non-transitory computer readable storage medium storing computer executable instructions for performing the fall detection method described in the foregoing embodiments.
  • the non-transitory computer readable storage medium may include: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.
  • ROM read-only memory
  • RAM random access memory
  • the fall detection device, the fall detection method and the device provided by the embodiment of the present invention form a human body communication signal path by using a fall detection device that uses the human body as a transmission medium to communicate by using a capacitive coupling method, and utilizes a channel of a backward loop when the human body falls and does not fall.
  • the feature difference is based on the signal received by the receiver for fall detection, which solves various defect problems in the multi-class fall detection technology in the related art, optimizes the fall detection technology in the related technology, and improves the versatility of the fall detection scheme.
  • the fall detection device has the advantages of small size, light weight, low power consumption, insensitivity to wearing position, and high detection precision.

Abstract

本发明实施例提供了一种跌倒检测设备、跌倒检测方法及装置。该跌倒检测设备包括：发送器及接收器；发送器与接收器设置为通过电容耦合方式利用人体作为传输介质进行通信，发送器的发送电极与接收器的接收电极构成前向回路，发送器的地电极与接收器的地电极构成后向回路；发送器，设置为产生跌倒检测信号，并将所述跌倒检测信号耦合至人体；接收器，设置为获取通过人体传输的跌倒检测信号的接收测量值；根据接收测量值，提取用于表征人体跌倒与未跌倒时后向回路的信道特征差异的跌倒描述特征的特征值；根据跌倒描述特征的特征值，进行跌倒检测。

Description

跌倒检测设备、跌倒检测方法及装置 技术领域
本公开涉及通信及信号处理技术，例如涉及一种跌倒检测设备、跌倒检测方法及装置。
背景技术
随着人口老龄化的加剧，老年人的安全已成为必须重视的问题。其中，跌倒是危害老年人健康的重要因素。据统计，在我国，65岁以上的居民中，有超过20％的男性、45％左右的女性曾跌倒过，而且年龄越大，发生跌倒的可能性也就越大。由于人体发生跌倒具有不确定性和不可预知性，当老年人跌倒后，如果长时间得不到及时有效的救治，可能会导致长期瘫痪甚至危及生命。因此，为了确保老人跌倒后能得到及时的救治，对老人进行跌倒检测是非常重要的。
目前，跌倒检测技术主要包括基于图像视频传感器的跌倒检测、基于环境布设式传感器的跌倒检测以及基于可穿戴式传感器跌倒检测等。
然而，基于图像视频传感器的跌倒检测方法需要在周围环境中安装一个或多个摄像头，使得跌倒检测系统的成本较高，并且检测范围也被限制在一个固定的区域内；基于环境布设式传感器的跌倒检测方法主要是通过压力传感器、麦克风等设备来捕捉人体的动作信息和采集人体相关的体态特征数据，然而该方法易受周围环境的影响；基于可穿戴式传感器跌倒检测方法主要是利用加速度传感器、陀螺仪传感器等获取人体的加速度或者角加速度等信息，然而该方法的检测结果与佩戴的位置有很大联系，如佩戴位置不正确，检测结果则存在较大的误差。
发明内容
有鉴于此，本发明实施例提供一种跌倒检测设备、跌倒检测方法及装置，优化了相关技术中的跌倒检测技术，提高跌倒检测方案的通用性。
第一方面，本发明实施例提供了一种跌倒检测设备，包括：发送器以及接收器；
所述发送器与所述接收器设置为通过电容耦合方式利用人体作为传输介质进行通信，所述发送器的发送电极与所述接收器的接收电极构成前向回路，所 述发送器的地电极与所述接收器的地电极构成后向回路；
所述发送器，设置为产生跌倒检测信号，并将所述跌倒检测信号耦合至人体；
所述接收器，设置为获取通过人体传输的所述跌倒检测信号的接收测量值；根据所述接收测量值，提取用于表征人体跌倒与未跌倒时所述后向回路的信道特征差异的跌倒描述特征的特征值；根据所述跌倒描述特征的特征值，进行跌倒检测。
第二方面，本发明实施例还提供了一种跌倒检测方法，应用于本发明实施例所述的跌倒检测设备的接收器中，所述方法包括：
通过接收电极获取跌倒检测信号的接收测量值；
根据所述接收测量值，提取用于表征人体跌倒与未跌倒时后向回路的信道特征差异的跌倒描述特征的特征值；以及
根据所述跌倒描述特征的特征值，进行跌倒检测。
第三方面，本发明实施例还提供了一种跌倒检测装置，所述装置包括：
接收测量值获取模块，设置为通过接收电极获取跌倒检测信号的接收测量值；
特征值提取模块，设置为根据所述接收测量值，提取用于表征人体跌倒与未跌倒时后向回路的信道特征差异的跌倒描述特征的特征值；以及
跌倒检测模块，设置为根据所述跌倒描述特征的特征值，进行跌倒检测。
第四方面，本发明实施例还提供了一种非暂态计算机可读存储介质，存储有计算机可执行指令，所述计算机可执行指令用于执行上述的跌倒检测方法。
本发明实施例提供的跌倒检测设备、跌倒检测方法及装置，通过采用电容耦合方式利用人体作为传输介质进行通信的跌倒检测设备构成人体通信信号通路，利用人体跌倒与未跌倒时后向回路的信道特征差异，基于接收器收到的信号进行跌倒检测，解决了相关技术中多类跌倒检测技术所存在的多种缺陷问题，优化相关技术中的跌倒检测技术，提高跌倒检测方案的通用性，同时，该跌倒检测设备具有体积小、重量轻、功耗低、对佩戴位置不敏感以及检测精度高等优点，为跌倒检测技术的发展提供一种新的思路。
附图概述
图1a是本发明实施例一提供的一种跌倒检测设备的结构图；
图1b是本发明实施例一提供的一种发送器的结构图；
图1c是本发明实施例一提供的一种接收器的结构图；
图2是本发明实施例二提供的一种跌倒检测方法的流程图；
图3是本发明实施例三提供的一种跌倒检测方法的流程图；
图4是本发明实施例四提供的一种跌倒检测方法的流程图；
图5是本发明实施例五提供的一种跌倒检测装置的结构图。
具体实施方式
下面结合附图和实施例对本公开作详细说明。此处所描述的具体实施例仅仅用于解释本公开，而非对本公开的限定。另外还需要说明的是，为了便于描述，附图中仅示出了与本公开相关的部分而非全部结构。
为了便于描述，附图中仅示出了与本公开相关的部分而非全部内容。在更加详细地讨论示例性实施例之前应当提到的是，一些示例性实施例被描述成作为流程图描绘的处理或方法。虽然流程图将每项操作(或步骤)描述成顺序的处理，但是其中的许多操作可以被并行地、并发地或者同时实施。此外，操作的顺序可以被重新安排。当操作完成时所述处理可以被终止，但是还可以具有未包括在附图中的附加步骤。所述处理可以对应于方法、函数、规程、子例程、子程序等等。
为了后文便于描述，首先对本发明实施例所使用的技术进行说明：
首先，本发明实施例主要使用人体通信技术。人体通信是一种以“人体”为通信媒介的新型通信技术，具有低功耗、微体积、方便快捷等优势。根据耦合方式的不同，人体通信可分为电流耦合和电容耦合。在电容耦合方式中，人体通信信道的传播特性由前向回路和后向回路共同决定。其中人体通信的发送器的发送电极以及接收器的接收电极通过人体形成前向回路；发送器的地电极以及接收器的地电极通过地平面(大地)形成后向回路。
发明人通过研究发现：如果保证人体站立时，上述电容耦合式人体通信的发送器以及接收器在人体中的放置位置距离地面具有一定的高度值，则在人体发生跌倒时，发送器和接收器的地电极与地平面的距离将发生急剧变化，导致人体通信的后向回路的特性也将发生变化，从而引起信道增益发生改变。因此， 可以通过分析人体跌倒与未跌倒时信道增益变化曲线的差异情况，实现跌倒检测。
实施例一
图1a是本发明实施例一提供的一种跌倒检测设备的结构图。如图1a所示，所述跌倒检测设备，包括：发送器11以及接收器12。
所述发送器11与所述接收器12设置为通过电容耦合方式利用人体作为传输介质进行通信，所述发送器11的发送电极111与所述接收器12的接收电极121构成前向回路，所述发送器11的地电极112与所述接收器12的地电极122构成后向回路。
为了示意的更加直观，在图1a中示出了一种发送器11与接收器12的放置位置示意图。但是，本领域技术人员可以理解的是，当用户佩戴上述发送器11以及接收器12之后，只要能够保证当用户发送跌倒时，发送器11的地电极112以及接收器12的地电极122与地面之间的距离发生改变，均可以达到本发明实施例的技术效果，因此，本发明实施例中上述发送器11以及接收器12的放置位置是可选的。
考虑到舒适性与便捷性要求，在本实施例中，可以将发送器11放置于人体的上臂处，其中发送电极111紧贴在上臂的皮肤表面，地电极112则悬空在空气中。接收器12放置在人体的腹部处，其中接收电极121紧贴在腹部的皮肤表面，地电极122同样悬空在空气中，即：如图1a所示的放置位置。
可选的，发送器11的整体形状可以为圆柱状盒子，如图1a所示，该盒子的上表面和下表面均为金属材料，所述盒子的上表面作为发送器11的发送电极111，所述盒子的下表面作为发送器11的地电极112，盒子的侧面为塑料材料，用于隔离发送电极111和地电极112，避免两者之间接触；相类似的，接收器12的整体形状同样可以为圆柱状盒子，如图1a所示，该盒子的上表面和下表面均为金属材料，该盒子的上表面作为接收器12的接收电极121，该盒子的下表面作为接收器12的地电极122，盒子的侧面为塑料材料，用于隔离接收电极121和地电极122。
所述发送器11，设置为产生跌倒检测信号，并将所述跌倒检测信号耦合至人体；
所述接收器12，设置为获取通过人体传输的所述跌倒检测信号的接收测量值；根据所述接收测量值，提取用于表征人体跌倒与未跌倒时所述后向回路的 信道特征差异的跌倒描述特征的特征值；根据所述跌倒描述特征的特征值，进行跌倒检测。
其中，在图1b中示出了一种发送器的可选结构图，如图1b所示，所述发送器包括：
微处理器1101、DDS(Direct Digital Synthesizer，直接数字式频率合成器)1102、巴伦转换器1103、低通滤波器1104、发送电极111以及地电极112；
其中，所述微处理器1101，设置为控制所述DDS1102产生设定频率范围内的单端正弦波信号；
所述巴伦转换器1103，设置为将所述DDS1102输出的所述单端正弦波信号转换为双端正弦波信号，并输出至所述低通滤波器1104；
所述低通滤波器1104，设置为对所述双端正弦波信号进行低通滤波，并将滤波后产生的跌倒检测信号通过所述发送电极111耦合至人体中；
所述发送电极111设置为与人体的皮肤表面相接触，所述地电极112设置为与所述发送电极111绝缘连接。
如前所述，当人体发生跌倒时，会引起信道增益发生变化，所谓信道增益，是指是信道本身的衰减及衰落特性，一般可以通过接收信号与发送信号的比值来衡量，因此，如果发送器发送等幅信号，则可以简单的通过接收信号来衡量信道增益，也即：将接收信号变化曲线作为信道增益变化曲线，通过分析该接收信号变化曲线的信号特征，进行跌倒检测。
基于此，所述发送器的作用主要是发送一个等幅的发送信号作为跌倒检测信号，可选的，该发送信号可以为正弦波。
可选的，考虑到如果发送一个单一频率的正弦波，在接收端不利于观测以及信号周期提取等。因此，发送器发送的正弦波可以为一个可变频率的正弦波。可选的，所述跌倒检测信号可以为周期性的变频信号；其中，所述变频信号在一个周期内以第一频率为起点，按照设定频率跨度递增值至第二频率。
作为示例，所述第一频率可以为1MHz，所述第二频率可以为100MHz，所述设定频率跨度可以为0.5MHz/ms。
也即，所述跌倒检测信号为周期性正弦波信号，在一个周期内，以1MHz为起点，每隔1ms频率增加0.5MHz，直至最终频率变为100MHz。
其中，所述微处理器1101主要设置为产生控制信号，控制所述DDS1102输出所述跌倒检测信号。可选的，所述微处理器1101可以为FPGA(Field- Programmable Gate Array，现场可编程门阵列)、DSP(Digital Signal Processing，数字信号处理)或者单片机等。
所述低通滤波器1104的截止频率由所述跌倒检测信号的最高频率确定，设置为滤除多种杂波以及干扰，例如，如果所述跌倒检测信号的频率范围为1MHz-100MHz，则所述低通滤波器1104的截止频率可以选取为120MHz。
其中，在图1c中示出了一种接收器的可选结构图，如图1b所示，所述接收器包括：微处理器1201、接收电极121以及地电极122；
所述接收电极121，设置为获取通过人体传输的所述跌倒检测信号的接收测量值，并将所述接收测量值发送至所述微处理器1201；
所述微处理器1201，设置为根据所述接收测量值，提取用于表征人体跌倒与未跌倒时所述后向回路的信道特征差异的跌倒描述特征的特征值；根据所述跌倒描述特征的特征值，进行跌倒检测；
所述接收电极121设置为与人体的皮肤表面相接触，所述地电极122设置为与所述接收电极121绝缘连接。
在本实施例中，所述接收器12还可以包括多种信号处理器件，例如，放大器或者滤波器等，用于将接收电极121获取的接收测量值进行一定的处理后，输入至所述微处理器1201。
其中，所述微处理器1201中执行的跌倒检测过程将在实施例二至实施例四中进行详述。
本发明实施例提供的跌倒检测设备，通过电容耦合方式利用人体作为传输介质构成人体通信信号通路，利用人体跌倒与未跌倒时后向回路的信道特征差异，基于接收器收到的信号进行跌倒检测，解决了相关技术中多类跌倒检测技术所存在的多种缺陷问题，优化相关技术中的跌倒检测技术，提高跌倒检测方案的通用性，同时，该跌倒检测设备具有体积小、重量轻、功耗低、对佩戴位置不敏感以及检测精度高等优点，为跌倒检测技术的发展提供一种新的思路。
实施例二
图2为本发明实施例二提供的一种跌倒检测方法的流程图，该方法可以由跌倒检测装置执行，该装置可由软件和/或硬件实现，并一般可集成于跌倒检测设备的接收器中，可选的，接收器中的微处理器中，与跌倒检测设备的发送器配合使用。如图2所示，本实施例的方法包括：
在S210中，通过接收电极获取跌倒检测信号的接收测量值。
如前所述，为了便于接收器中的信号周期提取，所述发送器发送的跌倒检测信号为周期性可选为变频信号；
其中，所述变频信号在一个周期内以第一频率为起点，按照设定频率跨度递增值至第二频率。
在S220中，根据所述接收测量值，提取用于表征人体跌倒与未跌倒时后向回路的信道特征差异的跌倒描述特征的特征值。
如前所述，人体跌倒后，由于后向回路的信道特性变化，接收信号变化曲线会与未跌倒时的接收信号变化曲线具有差异。因此，考虑到发送器发送的跌倒检测信号的周期性，可以以一个周期作为时间单位，提取接收信号变化曲线在一个周期内的跌倒描述特征的特征值。
其中，所述跌倒描述特征可以是指用于表征人体跌倒与未跌倒时后向回路的信道特征差异的特征参数，例如，可以是均值、方差或者均方差等。
可选的，可以提取一个周期内，接收信号曲线中的极值点、方差值、斜率值或者最大值与最小值之差等进行跌倒检测。
在S230中，根据所述跌倒描述特征的特征值，进行跌倒检测。
通过将提取的所述跌倒描述特征的特征值与跌倒或者未跌倒时所述跌倒描述特征的设定的特征值进行比对，可以实现跌倒检测。
在一个可选的例子中，可以通过多次试验，分析当人体发生跌倒时，在与跌倒发生时间对应的时间区间内，获取一个或者多个周期内接收信号变化曲线，通过分析上述一个多个接收信号变化曲线中的最大值，确定最大值跌倒阈值。例如，如果确定当接收信号变化曲线的最大值超过-20dB，发送跌倒的概率为85％，则可以将最大值跌倒阈值设置为-20dB，将85％设置为置信度。因此，每当发现需要进行跌倒检测的人体的接收信号变化曲线的最大值超过-20dB，判断用户发生了跌倒，进而可以采取一定的跌倒预警策略。
可选的，可以同时使用多个信号特征在发生跌倒时的跌倒阈值(例如，方差值以及斜率值)，即：当一个周期内的接收信号变化曲线同时满足方差值跌倒阈值以及斜率值跌倒阈值时，判断用户发生了跌倒。以提高跌倒检测的准确性。
可选的，还可以以跌倒时以及未跌倒时的上述接收信号变化曲线作为训练样本，训练预测模型，可选的，决策树模型、聚类模型以及神经网络模型等，并使用训练好的上述预测模型进行跌倒检测。
本发明实施例提供的跌倒检测方法，采用通过电容耦合方式利用人体作为传输介质进行通信的跌倒检测设备构成人体通信信号通路，利用人体跌倒与未跌倒时后向回路的信道特征差异，基于接收器收到的信号进行跌倒检测，解决了相关技术中多类跌倒检测技术所存在的多种缺陷问题，优化相关技术中的跌倒检测技术，提高跌倒检测方案的通用性，同时，该跌倒检测设备具有体积小、重量轻、功耗低、对佩戴位置不敏感以及检测精度高等优点，为跌倒检测技术的发展提供一种新的思路。
实施例三
图3是本发明实施例三提供的一种跌倒检测方法的流程图，本实施例以上述实施例为基础进行改变，在本实施例中，将根据所述接收测量值，提取用于表征人体跌倒与未跌倒时所述后向回路的信道特征差异的跌倒描述特征的特征值改变为：根据所述接收测量值，更新接收信号变化曲线；其中，所述接收信号变化曲线与一个周期内的所述跌倒检测信号相对应；根据所述接收信号变化曲线，计算至少一项跌倒描述特征的特征值；
同时，将根据所述跌倒描述特征的特征值，进行跌倒检测改变为：将至少一项跌倒描述特征的特征值输入至预先训练的跌倒检测预测模型中，并根据所述跌倒检测预测模型的输出结果，进行跌倒检测。相应的，本实施例的方法可以是包括：
在S310中，通过接收电极获取跌倒检测信号的接收测量值。
在本实施例中，所述发送器发送的跌倒检测信号为周期性可选为变频信号；
其中，所述变频信号在一个周期内以第一频率为起点，按照设定频率跨度递增值至第二频率。
在S320中，根据所述接收测量值，更新接收信号变化曲线。
其中，所述接收信号变化曲线与一个周期内的所述跌倒检测信号相对应。
由于跌倒检测装置的状态、周围环境等因素的影响，通过接收器采集获得的接收信号变化曲线通常都伴有多种多样的噪声。为了在进行后期处理时(特征提取和特征识别)能提取到精准的特征信息，提高跌倒检测的精度，对所采集的信号进行滤波处理是非常重要的。
可选的，根据所述接收测量值，更新接收信号变化曲线可以包括：
根据第K时刻下获取的所述跌倒检测信号的接收测量值以及卡尔曼滤波算 法，计算所述第K+1时刻下所述跌倒检测信号的预估计最优值，其中，K为大于等于1的正整数；使用所述跌倒检测信号的预估计最优值，更新所述接收信号变化曲线。
在本实施例中，根据在人体通信跌倒检测系统中噪声的特点，采用卡尔曼滤波算法对信号进行滤波处理，过程如下：
1、根据系统前一次的最优值，计算得到预估计最优值方程，如式(1)所示：
X(k|k-1)＝A(k，k-1)·X(k-1|k-1)+B(k)·u(k)       (1)
其中X(k|k-1)的值为接收信号变化曲线在k-1时刻计算出的k时刻的估计值，X(k-1|k-1)为k-1时刻的最优值，u(k)为k时刻的控制量，A(k，k-1)为状态转移矩阵，B(k)为控制加权矩阵。
2、根据预估计最优值方程，计算预估计最优值的均方误差，如式(2)所示：
P(k|k-1)＝A(k，k-1)·P(k-1|k-1)·A(k，k-1)+U(k)·U(k)      (2)
其中P(k|k-1)为预估计最优值X(k|k-1)的均方误差，P(k-1|k-1)为最优值X(k-1|k-1)的均方误差，U(k)为k时刻的接收信号变化曲线的动态噪声。
3、根据预估计最优值的均方误差，计算卡尔曼增益矩阵，如式(3)所示：
Figure PCTCN2016109844-appb-000001
其中K(k)为卡尔曼增益值，N(k)为k时刻的接收信号变化曲线的观测噪声，H(k)为接收信号变化曲线的观测矩阵。
4、根据所得到的卡尔曼增益矩阵，对最优值方程进行更新，得到时刻的预估计最优值，如式(4)所示：
X(k|k)＝X(k|k-1)+K(k)·(Z(k)-H(k)·X(k|k-1))       (4)
其中Z(k)为Z(k)时刻的接收信号变化曲线测量值。
5、根据步骤1至步骤4，对接收信号变化曲线进行滤波处理，直至满足停止条件，从而实现接收信号变化曲线的卡尔曼滤波处理。
接收器的微处理通过实现上述卡尔曼滤波算法，即可实现对所述接收测量值进行最优估计，并根据该最优估计值更新当前周期下的接收信号变化曲线。
在S330中，根据所述接收信号变化曲线，计算至少一项跌倒描述特征的特征值。
为了区分跌倒行为跟正常活动，就必须寻找能够区分跌倒和正常活动的物理量，即寻找增益变化曲线的特征量(即：跌倒描述特征)。由于在跌倒时，发送器和接收器的地电极与地平面的距离急剧减小，导致人体通信的后向回路的电容急剧增大，从而使信道增益大幅增加。在跌倒动作完成后，发送器和接收器的地电极与地平面的距离几乎保持不变，信道增益也逐渐趋于稳定。
因此，可以根据增益变化曲线的特点，提取一个周期内的以下五种特征量中的一种或者多种作为跌倒描述特征进行跌倒分析，分别为：(1)增益变化曲线的最大值；(2)增益变化曲线中最大值与最小值的差值；(3)增益变化曲线的方差；(4)增益变化曲线的斜率；(5)跌倒后增益变化曲线的平均值。
如前所述，可以通过接收信号变化曲线来近似替代增益变化曲线，相应的，所述跌倒描述特征可以包括下述至少一项：
所述接收信号变化曲线的最大值、所述接收信号变化曲线的最大值与最小值的差值、所述接收信号变化曲线的方差、所述接收信号变化曲线的斜率以及跌倒后接收信号变化曲线的平均值。
当然，可以理解的是，所述跌倒描述特征还可以为其他类型，例如：接收信号变化曲线的均方差等。
在S340中，将所述至少一项跌倒描述特征的特征值输入至预先训练的跌倒检测预测模型中，并根据所述跌倒检测预测模型的输出结果，进行跌倒检测。
其中，所述跌倒检测预测模型可以为决策树模型、神经网络模型以及机器学习模型等。
本发明实施例的技术方案通过使用卡尔曼滤波算法实现了对接收测量值进行最优估计，可以降低人体通信系统中存在的噪声和干扰对接收信号的干扰，以最终提高跌倒检测的精度。通过预先训练的跌倒检测预测模型完成跌倒检测，可以在多个不同的跌倒描述特征下进行跌倒预测，并可以实时对跌倒检测预测模型中的参数进行更新学习，以不断提高跌倒检测的精度。
在上述实施例的基础上，可以将训练跌倒检测预测模型改变为：
获取跌倒检测训练实例集，其中，训练实例包括：当人体未发生跌倒时，通过将人体作为传输介质进行通信的跌倒检测设备中的接收器获取的接收信号变化曲线；以及当人体发生跌倒时，通过将人体作为传输介质进行通信的跌倒检测设备中的接收器获取的接收信号变化曲线；
根据设定模型构建算法，以及每个所述接收信号变化曲线在至少一个所述 跌倒描述特征下的特征值，训练生成所述跌倒检测预测模型。
这样设置可以在一个或者多个跌倒描述特征下训练跌倒检测预测模型，使得训练模型的预测结果更加贴近实际结果。
实施例四
图4是本发明实施例四提供的一种跌倒检测方法的流程图，本实施例以上述实施例为基础进行改变，在本实施例中，将所述模型构建算法改变为决策树算法，将所述跌倒检测预测模型改变为决策树模型；其中，所述决策树模型中的不同子节点对应不同的跌倒描述特征；
同时，将根据设定模型构建算法，以及每个所述接收信号变化曲线在至少一个所述跌倒描述特征下的特征值，训练生成所述跌倒检测预测模型改变为：确定所述决策树算法的类别属性的取值为发生跌倒以及未发生跌倒；确定所述决策树算法的非类别属性为所述跌倒描述特征，并设定所述跌倒描述特征的标准取值；计算所述类别属性的信息熵；根据每个所述接收信号变化曲线在至少一个所述跌倒描述特征下的特征值，计算每个所述非类别属性的信息熵；根据所述类别属性的信息熵，以及所述非类别属性的信息熵，计算每个所述非类别属性的信息增量值，并根据每个所述非类别属性的信息增量值，确定目标测试属性；将所述目标测试属性作为一个当前子节点，并在所述当前子节点的基础上，重复迭代重新确定新的测试属性作为新的子节点，直至生成决策树模型作为所述跌倒检测预测模型。相应的，本实施例的方法包括：
在S410中，获取跌倒检测训练实例集。
其中，训练实例包括：当人体未发生跌倒时，通过利用人体作为传输介质进行通信的跌倒检测设备中的接收器获取的接收信号变化曲线；以及当人体发生跌倒时，通过利用人体作为传输介质进行通信的跌倒检测设备中的接收器获取的接收信号变化曲线。
在S420中，确定所述决策树算法的类别属性的取值为发生跌倒以及未发生跌倒。
在S430中，确定所述决策树算法的非类别属性为所述跌倒描述特征，并设定所述跌倒描述特征的标准取值。
在S440中，计算所述类别属性的信息熵。
在S450中，根据每个所述接收信号变化曲线在至少一个所述跌倒描述特征 下的特征值，计算每个所述非类别属性的信息熵。
在S460中，根据所述类别属性的信息熵，以及所述非类别属性的信息熵，计算每个所述非类别属性的信息增量值，并根据每个所述非类别属性的信息增量值，确定目标测试属性。
在S470中，将所述目标测试属性作为一个当前子节点，并在所述当前子节点的基础上，重复迭代重新确定新的测试属性作为新的子节点，直至生成决策树模型作为所述跌倒检测预测模型。
决策树是一种树形结构的分类方法，其中每个内部节点表示一个属性上的测试，每个分支代表一个测试输出，每个叶节点代表一种类别。决策树的原理是：从根节点开始，对数据样本进行测试，根据不同的结果将数据样本划分成不同的数据样本子集，每个数据样本子集构成一个子节点。对每个子节点再进行划分，生成新的子节点，不断反复，直至达到特定的终止准则。
下面通过一个可选实例对生成决策树模型的过程进行描述：其中，在本实例中，使用ID3(Iterative Dichotomies 3，迭代二分法3)生成所述决策树模型。
模型训练阶段的过程为：
1、确定训练实例集：假设训练实例集为X，则在X中训练实例总数目为|X|，其中第i类训练实例的数目是|Xi|。根据经验和实验结果，训练实例总个数|X|拟定为200个。
2、确定在基于ID3决策树中的类别属性和非类别属性：类别属性是指根据相关的因素来判断是否发生跌倒，因此类别属性的输出结果分为两种：发生跌倒，没有发生跌倒。非类别属性是指所述跌倒描述特征。在跌倒检测中，非类别属性以及标准取值(所述标准取值的值的大小可根据实际测量结果而确定)如表1所示：
表1 非类别属性以及标准取值
Figure PCTCN2016109844-appb-000002
Figure PCTCN2016109844-appb-000003
3、对类别属性的信息熵，即是否发生跌倒进行计算：
令类别属性的熵为T，在所有的训练实例集中，假设发生跌倒的实例集的数目为p1，没发生跌倒的实例集的数目为p2＝|X|-p1，则发生跌倒的概率和没发生跌倒的概率分别如公式(5)和公式(6)所示：
Figure PCTCN2016109844-appb-000004
Figure PCTCN2016109844-appb-000005
因此类别属性的信息熵如公式(7)所示：
INFO(T)＝-P(p1)log2P(p1)-P(p2)log2P(p2)       (7)
4、对非类别属性的信息熵进行计算：
非类别属性共有五种，如表1所示，分别为接收信号变化曲线的最大值、接收信号变化曲线中最大值与最小值的差值、接收信号变化曲线的方差、接收信号变化曲线的斜率、跌倒后接收信号变化曲线的平均值。对非类别属性的信息熵分别按如下方式计算：
(1)对于非类别属性中的接收信号变化曲线的最大值A的信息熵进行计算，如下所示：
Figure PCTCN2016109844-appb-000006
其中|A|为非类别属性中的接收信号变化曲线的最大值A实例集的数目，|Ai|为第i个实例集的数目，即|Ai|共有三种标准取值，分别对应为A＝-20dB，A＝-10dB，A＝-5dB时的实例集的数目，|Aik|为在|Ai|实例集中发生跌倒的数目。
(2)对于非类别属性中的接收信号变化曲线中最大值与最小值的差值B的信息熵进行计算，如下所示：
Figure PCTCN2016109844-appb-000007
其中|B|为非类别属性中的接收信号变化曲线中最大值与最小值的差值B实例集的数目，|Bi|为第i个实例集的数目，即|Bi|共有两种标准取值，分别对应为B＝15dB，B＝8dB的实例集的数目。
(3)对于非类别属性中的接收信号变化曲线的方差C的信息熵进行计算，如下 所示：
Figure PCTCN2016109844-appb-000008
其中|C|为非类别属性中的接收信号变化曲线的方差C实例集的数目，|Ci|为第i个实例集的数目，即|Ci|共有四种标准取值，分别对应为C＝5dB，C＝4dB，C＝3dB，C＝2dB的实例集的数目。
(4)对于非类别属性中的接收信号变化曲线的斜率D的信息熵进行计算，如下所示：
Figure PCTCN2016109844-appb-000009
其中|D|为非类别属性中的接收信号变化曲线的斜率D实例集的数目，|Di|为第i个实例集的数目，即|Di|共有两种标准取值，分别对应为D＝40dB/S，D＝20dB/S的实例集的数目。
(5)对于非类别属性中的跌倒后接收信号变化曲线的平均值E的信息熵进行计算，如下所示：
Figure PCTCN2016109844-appb-000010
其中|E|为非类别属性中的跌倒后接收信号变化曲线的平均值E实例集的数目，|Ei|为第i个实例集的数目，即|Ei|共有三种标准取值，分别对应为E＝-9dB，E＝-7dB，E＝-5dB的实例集的数目。
5、计算信息增益量：分别计算在每个非类别属性中的信息增益量，如式(13)所示：
Gain(X，T)＝INFO(T)-INFO(X，T)          (13)
6、确定测试属性：信息增益量越大，说明该非类别属性的信息对于实现跌倒检测的帮助越大，则将该非类别属性选择为测试属性。
7、构造决策树：应用上面的方法，通过递归算法建立决策树。递归下去，可得到通过算法构造出的一棵决策树，最终完成模型训练阶段。
在S480中，根据所述接收测量值，更新接收信号变化曲线。
其中，所述接收信号变化曲线与一个周期内的所述跌倒检测信号相对应；
在S490中，根据所述接收信号变化曲线，计算至少一项跌倒描述特征的特征值。
在S4100中，将所述至少一项跌倒描述特征的特征值输入至预先训练的所述决策树模型中，并根据所述决策树模型的输出结果，进行跌倒检测。
本发明实施例的技术方案通过使用决策树算法构造跌倒检测预测模型，可以实现在相对短的时间内能够对大型数据源做出可行且效果良好的结果，可以对有许多属性的数据集构造决策树。
在上述多个实施例的基础上，基于ID3决策树模型建立后，利用基于人体通信的跌倒检测系统，采集获取多个不同年龄、不同身高、不同体重的志愿者在日常生活中以及在跌倒过程中的接收信号变化曲线，并将接收信号变化曲线导入到决策树中进行跌倒检测。通过将实际测量结果与通过决策树模拟预测得到结果进行对比，对决策树模型进行修正。如果通过决策树得到的跌倒检测结果误差较大，则可以增加训练实例集的数目，使训练集更具普遍性，并重新计算决策树，直至满足要求，以提高跌倒预测精度。
实施例五
图5为本发明实施例五提供的一种跌倒检测装置的结构图。本发明实施例提供的跌倒检测装置可以应用于本发明实施例所述的跌倒检测设备的接收器中，可选的，接收器的微处理器中。如图5所示，所述装置包括：
接收测量值获取模块510，设置为通过接收电极获取跌倒检测信号的接收测量值。
特征值提取模块520，设置为根据所述接收测量值，提取用于表征人体跌倒与未跌倒时后向回路的信道特征差异的跌倒描述特征的特征值。
跌倒检测模块530，设置为根据所述跌倒描述特征的特征值，进行跌倒检测。
本发明实施例提供的跌倒检测装置，通过使用电容式耦合方式利用人体作为传输介质的跌倒检测设备构成人体通信信号通路，利用人体跌倒与未跌倒时后向回路的信道特征差异，基于接收器收到的信号进行跌倒检测，解决了相关技术中多类跌倒检测技术所存在的多种缺陷问题，优化相关技术中的跌倒检测技术，提高跌倒检测方案的通用性，同时，该跌倒检测设备具有体积小、重量轻、功耗低、对佩戴位置不敏感以及检测精度高等优点，为跌倒检测技术的发展提供一种新的思路。
在上述实施例的基础上，所述跌倒检测信号可以为周期性的变频信号；
其中，所述变频信号在一个周期内以第一频率为起点，按照设定频率跨度递增值至第二频率。
在上述实施例的基础上，所述特征值提取模块可以包括：
变化曲线更新单元，设置为根据所述接收测量值，更新接收信号变化曲线；其中，所述接收信号变化曲线与一个周期内的所述跌倒检测信号相对应；
特征更新值计算单元，设置为根据所述接收信号变化曲线，计算至少一项跌倒描述特征的特征值。
在上述实施例的基础上，所述变化曲线更新单元，可以是设置为：
根据第K时刻下获取的所述跌倒检测信号的接收测量值以及卡尔曼滤波算法，计算所述第K+1时刻下所述跌倒检测信号的预估计最优值，其中，K为大于等于1的正整数；
使用所述跌倒检测信号的预估计最优值，更新所述接收信号变化曲线。
在上述实施例的基础上，所述特征更新值计算单元可以是设置为：
将至少一项跌倒描述特征的特征值输入至预先训练的跌倒检测预测模型中，并根据所述跌倒检测预测模型的输出结果，进行跌倒检测。
在上述实施例的基础上，还可以包括，跌倒检测预测模型训练模块，其中，所述，跌倒检测预测模型训练模块可以包括：
训练实例集获取单元，设置为获取跌倒检测训练实例集，其中，训练实例包括：当人体未发生跌倒时，通过将人体作为传输介质进行通信的跌倒检测设备中的接收器获取的接收信号变化曲线；以及当人体发生跌倒时，通过将人体作为传输介质进行通信的跌倒检测设备中的接收器获取的接收信号变化曲线；
模型训练单元，设置为根据设定模型构建算法，以及每个所述接收信号变化曲线在至少一个所述跌倒描述特征下的特征值，训练生成所述跌倒检测预测模型。
在上述实施例的基础上，所述模型构建算法可以为决策树算法，所述跌倒检测预测模型可以为决策树模型；
其中，所述决策树模型中的不同子节点对应不同的跌倒描述特征。
在上述实施例的基础上，所述模型训练单元可以是设置为：
确定所述决策树算法的类别属性的取值为发生跌倒以及未发生跌倒；
确定所述决策树算法的非类别属性为所述跌倒描述特征，并设定所述跌倒描述特征的标准取值；
计算所述类别属性的信息熵；
根据每个所述接收信号变化曲线在至少一个所述跌倒描述特征下的特征值，计算每个所述非类别属性的信息熵；
根据所述类别属性的信息熵，以及所述非类别属性的信息熵，计算每个所述非类别属性的信息增量值，并根据每个所述非类别属性的信息增量值，确定目标测试属性；
将所述目标测试属性作为一个当前子节点，并在所述当前子节点的基础上，重复迭代重新确定新的测试属性作为新的子节点，直至生成决策树模型作为所述跌倒检测预测模型。
在上述实施例的基础上，所述跌倒描述特征可以包括下述至少一项：
所述接收信号变化曲线的最大值、所述接收信号变化曲线的最大值与最小值的差值、所述接收信号变化曲线的方差、所述接收信号变化曲线的斜率以及跌倒后接收信号变化曲线的平均值。
本发明实施例所提供的跌倒检测装置可用于执行本公开任意实施例提供的跌倒检测方法，具备相应的功能模块，实现相同的有益效果。
本发明实施例还提供了一种非暂态计算机可读存储介质，存储有计算机可执行指令，所述计算机可执行指令用于执行上述实施例所述的跌倒检测方法。
上述非暂态计算机可读存储介质可以是包括：U盘、移动硬盘、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、磁碟或者光盘等各种可以存储程序代码的介质，也可以是暂态存储介质。
工业实用性
本发明实施例提供的跌倒检测设备、跌倒检测方法及装置，通过采用电容耦合方式利用人体作为传输介质进行通信的跌倒检测设备构成人体通信信号通路，利用人体跌倒与未跌倒时后向回路的信道特征差异，基于接收器收到的信号进行跌倒检测，解决了相关技术中多类跌倒检测技术所存在的多种缺陷问题，优化相关技术中的跌倒检测技术，提高跌倒检测方案的通用性，同时，该跌倒检测设备具有体积小、重量轻、功耗低、对佩戴位置不敏感以及检测精度高等优点。

Claims (17)

  1. 一种跌倒检测设备，包括：发送器以及接收器；
    所述发送器与所述接收器设置为通过电容耦合方式利用人体作为传输介质进行通信，所述发送器的发送电极与所述接收器的接收电极构成前向回路，所述发送器的地电极与所述接收器的地电极构成后向回路；
    所述发送器，设置为产生跌倒检测信号，并将所述跌倒检测信号耦合至人体；
    所述接收器，设置为获取通过人体传输的所述跌倒检测信号的接收测量值；根据所述接收测量值，提取用于表征人体跌倒与未跌倒时所述后向回路的信道特征差异的跌倒描述特征的特征值；根据所述跌倒描述特征的特征值，进行跌倒检测。
  2. 根据权利要求1所述的跌倒检测设备，其中，所述发送器包括：
    微处理器、直接数字式频率合成器DDS、巴伦转换器、低通滤波器、发送电极以及地电极；
    其中，所述微处理器，设置为控制所述DDS产生设定频率范围内的单端正弦波信号；
    所述巴伦转换器，设置为将所述DDS输出的所述单端正弦波信号转换为双端正弦波信号，并输出至所述低通滤波器；
    所述低通滤波器，设置为对所述双端正弦波信号进行低通滤波，并将滤波后产生的跌倒检测信号通过所述发送电极耦合至人体中；
    所述发送电极设置为与人体的皮肤表面相接触，所述地电极设置为与所述发送电极绝缘连接。
  3. 根据权利要求1所述的跌倒检测设备，其中，所述接收器包括：微处理器、接收电极以及地电极；
    所述接收电极，设置为获取通过人体传输的所述跌倒检测信号的接收测量值，并将所述接收测量值发送至所述微处理器；
    所述微处理器，设置为根据所述接收测量值，提取用于表征人体跌倒与未跌倒时所述后向回路的信道特征差异的跌倒描述特征的特征值；根据所述跌倒描述特征的特征值，进行跌倒检测；
    所述接收电极设置为与人体的皮肤表面相接触，所述地电极设置为与所述接收电极绝缘连接。
  4. 一种跌倒检测方法，应用于如权利要求1-3任一项所述的跌倒检测设备的 接收器中，包括：
    通过接收电极获取跌倒检测信号的接收测量值；
    根据所述接收测量值，提取用于表征人体跌倒与未跌倒时后向回路的信道特征差异的跌倒描述特征的特征值；以及
    根据所述跌倒描述特征的特征值，进行跌倒检测。
  5. 根据权利要求4所述的方法，其中，所述跌倒检测信号为周期性的变频信号；
    其中，所述变频信号在一个周期内以第一频率为起点，按照设定频率跨度递增值至第二频率。
  6. 根据权利要求5所述的方法，其中，根据所述接收测量值，提取用于表征人体跌倒与未跌倒时所述后向回路的信道特征差异的跌倒描述特征的特征值包括：
    根据所述接收测量值，更新接收信号变化曲线；其中，所述接收信号变化曲线与一个周期内的所述跌倒检测信号相对应；
    根据所述接收信号变化曲线，计算至少一项跌倒描述特征的特征值。
  7. 根据权利要求6所述的方法，其中，根据所述接收测量值，更新接收信号变化曲线包括：
    根据第K时刻下获取的所述跌倒检测信号的接收测量值以及卡尔曼滤波算法，计算所述第K+1时刻下所述跌倒检测信号的预估计最优值，其中，K为大于等于1的正整数；
    使用所述跌倒检测信号的预估计最优值，更新所述接收信号变化曲线。
  8. 根据权利要求6所述的方法，其中，根据所述跌倒描述特征的特征值，进行跌倒检测包括：
    将至少一项跌倒描述特征的特征值输入至预先训练的跌倒检测预测模型中，并根据所述跌倒检测预测模型的输出结果，进行跌倒检测。
  9. 根据权利要求8所述的方法，其中，训练跌倒检测预测模型包括：
    获取跌倒检测训练实例集，其中，训练实例包括：当人体未发生跌倒时，通过将人体作为传输介质进行通信的跌倒检测设备中的接收器获取的接收信号变化曲线；以及当人体发生跌倒时，通过将人体作为传输介质进行通信的跌倒检测设备中的接收器获取的接收信号变化曲线；
    根据设定模型构建算法，以及每个所述接收信号变化曲线在至少一个所述 跌倒描述特征下的特征值，训练生成所述跌倒检测预测模型。
  10. 根据权利要求9所述的方法，其中，所述模型构建算法为决策树算法，所述跌倒检测预测模型为决策树模型；
    其中，所述决策树模型中的不同子节点对应不同的跌倒描述特征。
  11. 根据权利要求10所述的方法，其中，根据设定模型构建算法，以及每个所述接收信号变化曲线在至少一个所述跌倒描述特征下的特征值，训练生成所述跌倒检测预测模型包括：
    确定所述决策树算法的类别属性的取值为发生跌倒以及未发生跌倒；
    确定所述决策树算法的非类别属性为所述跌倒描述特征，并设定所述跌倒描述特征的标准取值；
    计算所述类别属性的信息熵；
    根据每个所述接收信号变化曲线在至少一个所述跌倒描述特征下的特征值，计算每个所述非类别属性的信息熵；
    根据所述类别属性的信息熵，以及所述非类别属性的信息熵，计算每个所述非类别属性的信息增量值，并根据每个所述非类别属性的信息增量值，确定目标测试属性；以及
    将所述目标测试属性作为一个当前子节点，并在所述当前子节点的基础上，重复迭代重新确定新的测试属性作为新的子节点，直至生成决策树模型作为所述跌倒检测预测模型。
  12. 根据权利要求5-11任一项所述的方法，其中，所述跌倒描述特征包括下述至少一项：
    所述接收信号变化曲线的最大值、所述接收信号变化曲线的最大值与最小值的差值、所述接收信号变化曲线的方差、所述接收信号变化曲线的斜率以及跌倒后接收信号变化曲线的平均值。
  13. 一种跌倒检测装置，包括：
    接收测量值获取模块，设置为通过接收电极获取跌倒检测信号的接收测量值；
    特征值提取模块，设置为根据所述接收测量值，提取用于表征人体跌倒与未跌倒时后向回路的信道特征差异的跌倒描述特征的特征值；以及
    跌倒检测模块，设置为根据所述跌倒描述特征的特征值，进行跌倒检测。
  14. 根据权利要求13所述的装置，其中，所述跌倒检测信号为周期性的变 频信号；
    其中，所述变频信号在一个周期内以第一频率为起点，按照设定频率跨度递增值至第二频率。
  15. 根据权利要求14所述的装置，其中，所述特征值提取模块包括：
    变化曲线更新单元，设置为根据所述接收测量值，更新接收信号变化曲线；其中，所述接收信号变化曲线与一个周期内的所述跌倒检测信号相对应；
    特征更新值计算单元，设置为根据所述接收信号变化曲线，计算至少一项跌倒描述特征的特征值。
  16. 根据权利要求15所述的装置，其中，所述跌倒检测模块据是设置为：
    将至少一项跌倒描述特征的特征值输入至预先训练的跌倒检测预测模型中，并根据所述跌倒检测预测模型的输出结果，进行跌倒检测。
  17. 一种非暂态计算机可读存储介质，存储有计算机可执行指令，所述计算机可执行指令用于执行权利要求4-12任一项所述的方法。
PCT/CN2016/109844 2016-12-14 2016-12-14 跌倒检测设备、跌倒检测方法及装置 WO2018107374A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2016/109844 WO2018107374A1 (zh) 2016-12-14 2016-12-14 跌倒检测设备、跌倒检测方法及装置

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2016/109844 WO2018107374A1 (zh) 2016-12-14 2016-12-14 跌倒检测设备、跌倒检测方法及装置

Publications (1)

Publication Number Publication Date
WO2018107374A1 true WO2018107374A1 (zh) 2018-06-21

Family

ID=62557672

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2016/109844 WO2018107374A1 (zh) 2016-12-14 2016-12-14 跌倒检测设备、跌倒检测方法及装置

Country Status (1)

Country Link
WO (1) WO2018107374A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117471421A (zh) * 2023-12-25 2024-01-30 中国科学技术大学 对象跌倒检测模型的训练方法及跌倒检测方法

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102469956A (zh) * 2009-07-22 2012-05-23 皇家飞利浦电子股份有限公司 跌倒检测器和检测跌倒的方法
US20120259577A1 (en) * 2011-04-11 2012-10-11 Transrex Ag Fall Detection Methods and Devices
CN104169680A (zh) * 2012-03-13 2014-11-26 皇家飞利浦有限公司 使用气压传感器监测设备的高度改变
WO2014195146A1 (en) * 2013-06-06 2014-12-11 Koninklijke Philips N.V. Fall detection system and method.
CN105493163A (zh) * 2013-08-26 2016-04-13 皇家飞利浦有限公司 用于检测跌倒的方法和跌倒检测系统
JP5997463B2 (ja) * 2012-03-07 2016-09-28 綜合警備保障株式会社 転倒検知装置および転倒検知方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102469956A (zh) * 2009-07-22 2012-05-23 皇家飞利浦电子股份有限公司 跌倒检测器和检测跌倒的方法
US20120259577A1 (en) * 2011-04-11 2012-10-11 Transrex Ag Fall Detection Methods and Devices
JP5997463B2 (ja) * 2012-03-07 2016-09-28 綜合警備保障株式会社 転倒検知装置および転倒検知方法
CN104169680A (zh) * 2012-03-13 2014-11-26 皇家飞利浦有限公司 使用气压传感器监测设备的高度改变
WO2014195146A1 (en) * 2013-06-06 2014-12-11 Koninklijke Philips N.V. Fall detection system and method.
CN105493163A (zh) * 2013-08-26 2016-04-13 皇家飞利浦有限公司 用于检测跌倒的方法和跌倒检测系统

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117471421A (zh) * 2023-12-25 2024-01-30 中国科学技术大学 对象跌倒检测模型的训练方法及跌倒检测方法
CN117471421B (zh) * 2023-12-25 2024-03-12 中国科学技术大学 对象跌倒检测模型的训练方法及跌倒检测方法

Similar Documents

Publication Publication Date Title
CN110364144B (zh) 一种语音识别模型训练方法及装置
US11551103B2 (en) Data-driven activity prediction
CN103989462B (zh) 一种脉搏波形第一特征点和第二特征点的提取方法
CN110213003B (zh) 一种无线信道大尺度衰落建模方法及装置
CN104021289B (zh) 一种非高斯非稳态噪声建模方法
EP3671441B1 (en) Application management method and apparatus, storage medium, and electronic device
CN103954295A (zh) 一种基于加速度传感器的计步方法
CN105302383A (zh) 一种电容触摸传感器抗干扰的方法和设备
CN106618499B (zh) 跌倒检测设备、跌倒检测方法及装置
CN103750820A (zh) 一种睡眠质量监测方法及装置
US11514315B2 (en) Deep neural network training method and apparatus, and computer device
CN110046633B (zh) 一种数据质量检测方法及装置
CN112690761B (zh) 睡眠状态检测方法、装置、设备及计算机可读介质
CN114038012A (zh) 基于毫米波雷达和机器学习的跌倒检测方法及系统
CN111862951A (zh) 语音端点检测方法及装置、存储介质、电子设备
CN112232486A (zh) 一种yolo脉冲神经网络的优化方法
CN106935038B (zh) 一种停车检测系统及检测方法
CN113049963A (zh) 一种基于局部离群因子的锂电池组一致性检测方法及装置
CN109190640A (zh) 一种基于大数据的浮游生物的拦截式采集方法及采集系统
WO2018107374A1 (zh) 跌倒检测设备、跌倒检测方法及装置
CN108245171B (zh) 获取参数模型的方法、疲劳检测方法及装置、介质及设备
CN113671489B (zh) 状态提醒方法及装置、电子设备和计算机可读存储介质
CN109965847B (zh) 服务器和信号分析系统
CN114971504A (zh) 一种实体类型确定方法和相关装置
CN105392106B (zh) 一种基于智能终端的情境与接近度感知方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 16924125

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205 DATED 21.11.2019)

122 Ep: pct application non-entry in european phase

Ref document number: 16924125

Country of ref document: EP

Kind code of ref document: A1