CN115299937B - Intelligence detection platform that falls down - Google Patents
Intelligence detection platform that falls down Download PDFInfo
- Publication number
- CN115299937B CN115299937B CN202211130235.3A CN202211130235A CN115299937B CN 115299937 B CN115299937 B CN 115299937B CN 202211130235 A CN202211130235 A CN 202211130235A CN 115299937 B CN115299937 B CN 115299937B
- Authority
- CN
- China
- Prior art keywords
- data
- frequency domain
- fall
- time domain
- gesture data
- 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.)
- Active
Links
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
- A61B5/1116—Determining posture transitions
- A61B5/1117—Fall detection
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/68—Arrangements of detecting, measuring or recording means, e.g. sensors, in relation to patient
- A61B5/6801—Arrangements of detecting, measuring or recording means, e.g. sensors, in relation to patient specially adapted to be attached to or worn on the body surface
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
-
- 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
- G08B21/0438—Sensor means for detecting
- G08B21/0446—Sensor means for detecting worn on the body to detect changes of posture, e.g. a fall, inclination, acceleration, gait
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Animal Behavior & Ethology (AREA)
- Artificial Intelligence (AREA)
- Pathology (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Veterinary Medicine (AREA)
- Public Health (AREA)
- Physiology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Evolutionary Computation (AREA)
- Fuzzy Systems (AREA)
- Mathematical Physics (AREA)
- Dentistry (AREA)
- Psychiatry (AREA)
- Signal Processing (AREA)
- Gerontology & Geriatric Medicine (AREA)
- Business, Economics & Management (AREA)
- Emergency Management (AREA)
- General Physics & Mathematics (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The application relates to the technical field of abnormal condition data detection and analysis, in particular to an intelligent fall detection platform which comprises an intelligent terminal used for being worn by a user, wherein the intelligent terminal is used for acquiring and analyzing gesture data in a standby mode, and entering a working mode when the gesture data are abnormal; the system is also used for sending abnormal gesture data in the working mode and receiving a tumbling judgment result fed back according to the abnormal gesture data; the system also comprises a remote server, wherein the remote server is used for receiving abnormal gesture data, carrying out frequency domain analysis and time domain analysis on the gesture data, and generating a fall judgment result according to the frequency domain analysis result and the time domain analysis result. By adopting the scheme, the technical problems of high power consumption and low precision of the fall detection equipment in the prior art can be solved.
Description
Technical Field
The application relates to the technical field of abnormal condition data detection and analysis, in particular to an intelligent fall detection platform.
Background
The biggest accidental injury and disease trouble of the old in daily life comes from falling, and the falling is an accident which is very easy to happen when the old is alone at home or one person goes out, and the old is difficult to timely obtain rescue in the happening, so that the equipment capable of protecting the old for safe driving is needed.
Currently, many smart devices for the elderly are available on the market, such as smart digital watches, smart bracelets, smart foot rings, etc., which can provide functions of positioning, emergency calling and automatic alarming, and some of them can also provide simple fall detection functions. However, as the motion amplitude and the motion variety on the hands and feet of the person are large, the detection of the gesture behavior is troublesome, and the precision of the existing equipment for detecting the falling is low. In order to improve the detection precision, and fewer waist movements, intelligent devices worn in the waist are appeared, and a distributed wearable device is appeared in consideration of a multi-point detection mode, but the following disadvantages exist in the devices: the equipment integrates a complex gesture resolving algorithm, so that the power consumption of the equipment is high, frequent charging is needed when the equipment is used, the equipment is complex to wear and easy to forget, and the defects bring great difficulty to the use of the old.
Disclosure of Invention
The application aims to provide an intelligent falling detection platform for solving the technical problems of high power consumption and low precision of falling detection equipment in the prior art.
The application provides the following basic scheme:
an intelligent fall detection platform comprises an intelligent terminal used for being worn by a user, wherein the intelligent terminal is used for acquiring and analyzing gesture data in a standby mode, and entering a working mode when the gesture data are abnormal; and the system is also used for sending abnormal gesture data in the working mode and receiving a tumbling judgment result fed back according to the abnormal gesture data.
The basic scheme has the beneficial effects that:
1. in the scheme, the intelligent terminal comprises multiple modes, and in the standby mode, the intelligent terminal only collects and analyzes gesture data of a user, and other functions are in a dormant state, so that the power consumption of the intelligent terminal is reduced.
2. In the scheme, the intelligent terminal only judges the abnormality of the gesture data, when the abnormality occurs in the gesture data, the abnormal data is sent to the far end, and the remote end equipment further analyzes the abnormal gesture data. And the tumbling analysis of the gesture data is carried out at the far end, the intelligent terminal does not need to carry out complex gesture calculation, and the power consumption of the intelligent terminal is further reduced.
3. In the scheme, the gesture of the user is resolved and analyzed at the far end, more calculation resources can be provided, the requirements on the complexity, occupied space and the like of the gesture resolving algorithm are lower, and algorithms with higher precision, such as an artificial intelligence algorithm, can be optimized, so that the detection precision of the tumbling detection is improved.
Further, the intelligent terminal comprises a main control module, the main control module is preset with a judging threshold value, the main control module is used for judging whether the gesture data is abnormal according to the judging threshold value, and when the gesture data is located outside the judging threshold value, the gesture data is judged to be abnormal.
The beneficial effects are that: in the scheme, the intelligent terminal performs pre-analysis judgment on the gesture of a user through setting of a judgment threshold value, and screens out a large-amplitude action; and secondly, the non-falling gesture is removed, the workload of subsequent gesture calculation is reduced, and the detection efficiency is improved.
Further, the system comprises a remote server, wherein the remote server is used for receiving abnormal gesture data, carrying out frequency domain analysis and time domain analysis on the gesture data, and generating a fall judgment result according to the frequency domain analysis result and the time domain analysis result.
The beneficial effects are that: in this scheme, through the analysis of further resolving of remote server to unusual gesture data, carry out the behavior detection that falls from frequency domain and time domain respectively during the resolving, compare with the direct gesture data analysis of following among the prior art, the analysis dimension is more, and the degree of accuracy of final falling decision result is higher, improves the detection precision that falls the detection promptly.
Further, the remote server is preset with a frequency domain SVM classifier and a time domain LSTM classifier, and is used for carrying out frequency domain analysis according to the frequency domain SVM classifier and carrying out time domain analysis according to the time domain LSTM classifier.
The beneficial effects are that: in the scheme, different classifiers are adopted for frequency domain analysis and time domain analysis, the SVM classifier supports the division boundary of the motion vector learned, the rapid detection of the falling behavior can be realized, the LSTM classifier supports the long-short-term memory neural network learning, the relation between data time sequences can be better analyzed, the defects of the SVM classifier are overcome, and the detection precision of the falling behavior is improved.
Further, the intelligent terminal is also used for acquiring current positioning information and uploading the current positioning information according to the falling judgment result.
The beneficial effects are that: in this scheme, obtain current location information through intelligent terminal, learn user's position, when user takes place to fall down the action, can in time rescue it.
Further, the intelligent terminal is also used for acquiring and storing the strong positioning information under the strong signal; the inertial data acquisition module is also used for acquiring and storing inertial data; and the method is also used for uploading strong positioning information and inertial data when uploading the current positioning information.
The beneficial effects are that: the user's action of falling down probably takes place in the weaker region of signal, and intelligent terminal's positioning accuracy is not high this moment, probably appears the inaccurate condition of location, so in this scheme, when uploading current location information, still upload the strong location information that gathers when this signal is stronger before to and the inertial data after gathering this strong location information, can predict user's current position through strong location information and inertial data, thereby obtain more accurate user position information, make the user can obtain timely rescue.
Further, the fall determination results include fall types including general falls and loss of consciousness falls, and the remote server is further configured to trigger a corresponding alarm according to the fall types.
The beneficial effects are that: the user is the old person, and two situations are likely to occur after the old person falls down, namely the old person can freely move after falling down, namely general falling down; secondly, the patient is in a static state after falling, namely, the patient falls in a consciousness loss type, the consciousness loss type falls in a general falling state, and the patient is more dangerous, so that the patient needs to be rescued more urgently. Therefore, the remote server can trigger corresponding alarm according to the type of the fall, so that corresponding feedback can be timely made.
Further, the remote server comprises a classification training module, wherein the classification training module is used for acquiring sample data, extracting positive samples according to labels in the sample data, and extracting data except the labels as negative samples; the method is also used for dividing a training set and a testing set for positive and negative samples, training a frequency domain SVM classifier and a time domain LSTM classifier according to the positive and negative samples in the training set, and testing the frequency domain SVM classifier and the time domain LSTM classifier according to the positive and negative samples in the testing set; the remote server is used for calling the trained frequency domain SVM classifier and the trained time domain LSTM classifier to conduct frequency domain analysis and time domain analysis.
The beneficial effects are that: in the scheme, sample data are acquired through a remote server, training is carried out according to the sample data, and a final high-precision frequency domain SVM classifier and a final high-precision time domain LSTM classifier are obtained. When positive and negative samples are divided, labeling is carried out on the sample data, for example, labeling time points of beginning and ending of tumbling action in the sample data, positive samples are extracted from the sample data according to labeling, negative samples are extracted from the sample data which are not extracted, positive sample data and negative sample data do not need to be collected respectively, and the workload of sample data collection is reduced.
Further, the classification training module is further used for establishing a dictionary according to the sample data, wherein the dictionary comprises a plurality of words; the classification training module is also used for searching corresponding words in the dictionary according to the data in the training set, the number of times of searching each word is counted to obtain counting characteristics of the corresponding data, normalization is used as frequency domain characteristics, and the frequency domain SVM classifier is trained according to the frequency domain characteristics.
The beneficial effects are that: in the scheme, a dictionary is established through cluster analysis, frequency domain characteristics are obtained through normalization analysis of frequencies of words in the dictionary in data in a training set, and the frequency domain characteristics are used as samples to train a frequency domain SVM classifier.
Further, the classification training module is also used for expanding the positive samples, including random sampling and random lifting sampling rate.
The beneficial effects are that: random sampling is to perform random sampling within a time window covering positive samples, and all sampling results of which the starting time point and the ending time point contain the positive samples are regarded as positive samples; the random collection rate is improved to sample data with longer duration of tumbling action, the random collection rate is improved, double-speed sampling is achieved on the sample data through linear interpolation, and a tumbling process with a quicker occurrence process is simulated. And the number of positive samples is increased through random sampling and random increasing sampling rate, and the frequency domain SVM classifier and the time domain LSTM classifier are fully trained.
Drawings
FIG. 1 is a flow chart of an embodiment of an intelligent fall detection platform of the present application;
FIG. 2 is a schematic diagram of a fall detection of an embodiment of an intelligent fall detection platform of the present application;
FIG. 3 is a schematic diagram of a portion of raw data of an embodiment of an intelligent fall detection platform according to the present application;
fig. 4 is a diagram showing a comparison between original data and a labeling time window of an embodiment of an intelligent fall detection platform according to the present application.
Detailed Description
The following is a further detailed description of the embodiments:
examples
An intelligent fall detection platform comprises a remote server and an intelligent terminal used for being worn by a user, wherein the intelligent terminal is used for being worn at the waist of the user.
The intelligent terminal comprises three modes, namely a standby mode, wherein the intelligent terminal is in the standby mode in daily use of the intelligent terminal, and at the moment, the intelligent terminal only collects and analyzes gesture data of a user, and other functions are in a dormant state; secondly, when the intelligent terminal analyzes that the gesture data is abnormal, the intelligent terminal enters the working mode, and all functions of the intelligent terminal are in an operating state; thirdly, an active alarm mode is adopted, an alarm key is arranged on the intelligent terminal, a user presses the alarm key, the intelligent terminal enters the active alarm mode, at the moment, the intelligent terminal actively and circularly transmits alarm signals, and uploads the position information of the user until receiving processing signals fed back by the remote server, and the user can cancel the processing signals by pressing a cancel alarm key on the intelligent terminal.
As shown in fig. 1, the intelligent terminal is used for acquiring and analyzing gesture data in a standby mode, and specifically, the intelligent terminal comprises an inertial sensor module, wherein the inertial sensor module is used for acquiring and analyzing gesture data of a user, and the gesture data comprise position, speed, angle, angular speed and acceleration. In this embodiment, the inertial sensor module employs a gyroscope and an accelerometer, and the acquired attitude data includes angular velocity and acceleration.
The intelligent terminal is used for entering a working mode when the gesture data are abnormal, specifically, the intelligent terminal comprises a main control module, a judging threshold value is preset in the main control module, the main control module is used for judging whether the gesture data are abnormal according to the judging threshold value, and when the gesture data are located outside the judging threshold value, the gesture data are judged to be abnormal, and the working mode is entered at the moment. And the gesture data sequence of suspected tumbling behaviors is screened out through judging the threshold value, so that the subsequent remote server sending is convenient to calculate. The main control module is also used for configuring and correcting the initial state of the inertial sensor module, and is also used for initializing and configuring the subsequent positioning module and confirming the star searching.
The intelligent terminal is also used for sending abnormal gesture data in the working mode, the remote server is used for receiving the abnormal gesture data, carrying out frequency domain analysis and time domain analysis on the gesture data, and generating a tumbling judgment result according to the frequency domain analysis result and the time domain analysis result. In this embodiment, the remote server is preset with a frequency domain SVM classifier and a time domain LSTM classifier, and the remote server is configured to perform frequency domain analysis according to the frequency domain SVM classifier and perform time domain analysis according to the time domain LSTM classifier. As shown in fig. 2.
Specifically, the remote server comprises a frequency domain analysis module, a time domain analysis module and a comprehensive analysis module, wherein the frequency domain analysis module is used for calling a frequency domain SVM classifier; the method is also used for acquiring frequency domain signals from abnormal gesture data, inputting the frequency domain signals into a frequency domain SVM classifier, and acquiring frequency domain analysis results output by the frequency domain SVM classifier according to the frequency domain signals. By analyzing the behavior corresponding to the abnormal gesture data through the frequency domain, the scoring is given for regular large-amplitude movement (such as exercise and the like) or sporadic large-amplitude movement (such as falling and the like), if the behavior is classified as sporadic, the scoring is given for falling tendency.
The time domain analysis module is used for calling a time domain LSTM classifier; and the time domain analysis method is also used for acquiring time domain signals from the abnormal gesture data, inputting the time domain signals into the time domain LSTM classifier and acquiring time domain analysis results output by the time domain LSTM classifier according to the time domain signals. For example to obtain as input the timing signal X of the positioning and sensing module during a time period t (t=5s),，X 1 -X 3 corresponding to triaxial acceleration, X 4 Output Y corresponding to the angular velocity amplitude 1 -Y 4 All the time sequence information are prediction results ('tumbling' or 'non-tumbling'), and whether the current action is a tumbling action is predicted through nonlinear transformation by extracting fusion characteristics of the time sequence information of each shaft.
The comprehensive analysis module is used for generating a fall judgment result according to the frequency domain analysis result and the time domain analysis result, wherein the fall judgment result comprises whether fall and fall types, and the fall types comprise general fall and consciousness loss type fall. Specifically, the comprehensive analysis module is used for generating a fall judgment result according to the frequency domain analysis result and the time domain analysis result, and judging whether the fall type is a consciousness loss fall according to the rest time after the fall occurs. General falls refer to free movement after a fall, and more dangerous loss of consciousness falls refer to resting after a fall.
The remote server is also used for triggering corresponding alarm according to the tumbling type, analyzing the tumbling degree through the remote server and sending corresponding alarm information to the intelligent terminal or the associated guardian terminal.
The intelligent terminal is also used for receiving a tumbling judgment result fed back according to the abnormal gesture data; the intelligent terminal is also used for acquiring current positioning information, acquiring and storing strong positioning information under a strong signal, and acquiring and storing inertial data. In this embodiment, the posture data is inertial data. Specifically, the intelligent terminal further comprises a positioning module and a storage module, wherein the positioning module is further used for monitoring the position of a user and acquiring current positioning information, if the current positioning information is acquired under a strong signal, the current positioning information is stored as the strong positioning information, and after the strong positioning information is acquired, the acquired inertial data is stored. In this embodiment, the positioning module uses WIFI and bluetooth positioning indoors, and uses navigation services such as beidou/GPS outdoors. The storage module is used for storing and updating the strong positioning information and the inertia data. The weak region of signal, intelligent terminal's positioning accuracy is not high this moment, probably appears the inaccurate condition of location, so in this scheme, still acquire and store the inertial data after the strong positioning information to the strong positioning information that gathers when this former signal is stronger.
The intelligent terminal is also used for uploading current positioning information according to the tumbling judgment result, and uploading strong positioning information and inertial data when uploading the current positioning information. Specifically, the intelligent terminal further comprises a communication module, wherein the communication module is used for uploading current positioning information, strong positioning information and inertial data when the falling judgment result is falling. The current position of the user can be predicted through the strong positioning information and the inertia data, and more accurate user position information can be obtained through the predicted position information and the current positioning information, so that the user can be timely helped.
Example two
The present embodiment is different from the first embodiment in that: training of the frequency domain SVM classifier and the time domain LSTM classifier is also included.
The remote server comprises a classification training module, wherein the classification training module is used for acquiring sample data, specifically, selecting testers with different heights and weights to wear an inertial sensor, carrying out various real falling behavior simulations under protection, carrying out daily behavior simulations at various places, including land leveling, seats, stairwells and the like, and finally acquiring original data comprising required positive and negative samples. In this embodiment, the sample data only uses 9 sets of data with numerical variation to describe the inertial characteristics of the time point, namely, three-axis acceleration (AccX, accY, accZ), three-axis angular velocity (GyrX, gyrY, gyrZ), and three-axis rotation angle (Roll, pitch, yaw), the sampling frequency of the inertial sensor is 100Hz, that is, a sequence length of 5 seconds of the sequence of raw data is 5×100×9, and a part of the raw data is shown in fig. 3.
In this example, positive samples were data containing fall behavior, including 1-1 fall to lie sideways (side body on ground), 1-2 fall to lie sideways (side body part on ground with a little support), 1-3 fall to face downwards (lie flat after fall), 1-4 fall to face upwards (back on ground fall), 1-5 fall to buttocks squat (buttocks on ground fall), 1-6 chair recline fall, 1-7 fall off the chair on the ground sitting, 1-8 buttress slip (knee-soft fall), 1-9 fall to kneel pose, 1-10 fall to kneel pose. The negative examples are daily data including 2-1 walking (straight and turning), 2-2 running (different speeds), 2-3 other walking (abnormal walking such as walking forward, lifting legs, twisting waist, bending waist, walking slowly, pulling one leg one round, etc.), 2-4 going up and down stairs, 2-5 squatting and tying laces, 2-6 squatting (mimicking a toilet), 2-7 squatting and sitting on a small stool, 2-8 sitting and sofa (sitting more random), 2-9 sitting and sitting (front sitting), 2-10 sitting and sitting, 2-11 lying down (slow lying), 2-12 lying down, 2-13 lying back, 2-14 bending waist for washing hands, 2-15 stretching hands, 2-16 stretching front, back, left and right sides lower waist for pulling, 2-17 bending waist for sweeping, and 2-18 bending waist for table.
The classification training module is also used for extracting positive samples according to labels in the sample data and extracting data except the labels as negative samples. Labeling the starting and ending time points of the tumbling action in the sample data, wherein the data in the labeling time window are positive samples, the rest are negative samples, as shown in fig. 4, a left graph in fig. 4 is an original 30s sequence, and a right graph is a labeling result of the tumbling action.
The classification training module is also used for expanding the positive samples, including random sampling and random lifting sampling rate. The random sampling is to perform random sampling in a labeling time window covering the positive sample, and all sampling results of which the starting time point and the ending time point contain the positive sample are regarded as the positive sample; the random collection rate is improved to sample data with longer duration of tumbling action, such as a sequence larger than 3s, the random collection rate is improved, multiple speed sampling, such as 1.1-1.5 multiple sampling, is realized on the sample data through linear interpolation, and a tumbling process with quicker occurrence process is simulated. And expanding the positive samples in two forms through random sampling and random lifting sampling rate, and lifting the number of the positive samples.
The classification training module is further configured to divide the positive and negative samples into a training set and a testing set, specifically, divide the positive and negative samples according to a ratio of 80% training and 20% testing, in this embodiment, the tester is N people, randomly select positive and negative samples corresponding to n×20% people as the testing set, and the positive and negative samples corresponding to the rest are the training sets.
The classification training module is also used for training the frequency domain SVM classifier and the time domain LSTM classifier according to the positive and negative samples in the training set, and testing the frequency domain SVM classifier and the time domain LSTM classifier according to the positive and negative samples in the testing set.
When training and testing the frequency domain SVM classifier, the classification training module is also used for establishing a dictionary according to sample data, wherein the dictionary comprises a plurality of words; the classification training module is also used for searching corresponding words in the dictionary according to the data in the training set, the number of times of searching each word is counted to obtain counting characteristics of the corresponding data, normalization is used as frequency domain characteristics, and the SVM classifier is trained according to the frequency domain characteristics. Specifically, the dictionary is built according to the sample data: cutting the frequency domain data of the positive and negative samples into equal-length fragments, extracting the amplitude values of the fragments through discrete Fourier transform, sorting according to the amplitude values, and screening the frequency domain data of N% before sorting according to a preset value N; acquiring cluster features of the screened frequency domain data to form three-dimensional data features; and acquiring all three-dimensional data features, clustering a plurality of clustering center points according to a Kmeans clustering method, and arranging the clustering center points according to frequency to generate a dictionary.
In this embodiment, when training and testing the frequency domain SVM classifier, the following are included:
(1) Establishing a dictionary: in a training set, cutting acceleration amplitude signals of an inertial sensor into fragments with equal length of 500, extracting amplitude values of the fragments through discrete Fourier transform, only preserving frequency domain signals of which the size is 10% before ranking of each fragment, and then forming three-dimensional data features by amplitude, phase and frequency corresponding to each preserved point; and integrating three-dimensional features of all data in the training set, clustering by a Kmeans clustering method to obtain a plurality of clustering center points, and arranging the clustering center points according to frequency to serve as dictionary features.
(2) Extracting characteristics: for the test set and the training set, the test set and the training set are cut into 500-length fragments and discrete Fourier transform is extracted, points with the amplitude of 40% are reserved, three-dimensional data features of the amplitude, the phase and the frequency of the points are extracted, counting features with the size of 200 are extracted, words closest to each feature of each reserved point (each feature in a dictionary is a word) are searched, one is added to the number of the positions of the corresponding words, the counting features of the whole fragments are obtained, and normalization is carried out to obtain frequency domain features.
(3) Training a classifier: and executing feature extraction operation on each data of the training set to obtain frequency domain features with the size of 200 x 1, training an SVM classifier by using labels corresponding to the frequency domain features and the data, wherein the labels corresponding to positive samples are tumblers, and the labels corresponding to negative samples are non-tumblers.
(4) Testing classifier: and performing feature extraction operation on each data in the test set, and inputting the data into a trained SVM classifier to obtain a score for predicting the tendency of falling.
When the time domain LSTM classifier is trained and tested, the triaxial acceleration signal and the angular velocity amplitude signal of the inertial sensor are directly fused to be used as input, and the long-term and short-term memory neural network LSTM classifier is used, so that the performance of the time domain LSTM classifier is superior to that of the traditional neural network and SVM classifier.
The remote server is used for calling the trained frequency domain SVM classifier and the trained time domain LSTM classifier to conduct frequency domain analysis and time domain analysis.
By adopting the scheme, the performance of the frequency domain SVM classifier and the time domain LSTM classifier after training is shown in the table I and the table III, the detection rate in the table I is the number of falling behavior samples/the total number of falling behavior samples judged to be falling behaviors, and the false alarm rate is the number of non-falling behavior samples/the total number of non-falling behavior samples judged to be falling behaviors.
Table-frequency domain SVM classifier Performance
Table two time domain LSTM classifier performance
As shown in a first table, the detection rate of the frequency domain SVM classifier is 96.0% and the false alarm rate is 5.6%, and as shown in a second table, the detection rate of the time domain LSTM classifier is 96.7% and the false alarm rate is 5.2%. After the frequency domain SVM classifier and the time domain LSTM classifier are fused, the overall detection rate of fall detection is 98.9%, and the false alarm rate is 1.7%. The detection precision of the tumbling behavior detection is far superior to that of products on the current market.
The foregoing is merely an embodiment of the present application, and a specific structure and characteristics of common knowledge in the art, which are well known in the scheme, are not described herein, so that a person of ordinary skill in the art knows all the prior art in the application date or before the priority date, can know all the prior art in the field, and has the capability of applying the conventional experimental means before the date, and a person of ordinary skill in the art can complete and implement the present embodiment in combination with his own capability in the light of the present application, and some typical known structures or known methods should not be an obstacle for a person of ordinary skill in the art to implement the present application. It should be noted that modifications and improvements can be made by those skilled in the art without departing from the structure of the present application, and these should also be considered as the scope of the present application, which does not affect the effect of the implementation of the present application and the utility of the patent. The protection scope of the present application is subject to the content of the claims, and the description of the specific embodiments and the like in the specification can be used for explaining the content of the claims.
Claims (7)
1. An intelligent fall detection platform which is characterized in that: the intelligent terminal is used for acquiring and analyzing gesture data in a standby mode, and enters a working mode when the gesture data are abnormal; the system is also used for sending abnormal gesture data in the working mode and receiving a tumbling judgment result fed back according to the abnormal gesture data;
the remote server is used for receiving abnormal gesture data, carrying out frequency domain analysis and time domain analysis on the gesture data, and generating a tumbling judgment result according to the frequency domain analysis result and the time domain analysis result;
the remote server is preset with a frequency domain SVM classifier and a time domain LSTM classifier, and is used for carrying out frequency domain analysis according to the frequency domain SVM classifier and carrying out time domain analysis according to the time domain LSTM classifier;
the remote server comprises a classification training module, wherein the classification training module is used for acquiring sample data, extracting positive samples according to labels in the sample data, and extracting data except the labels as negative samples, and the sample data are triaxial acceleration, triaxial angular velocity and triaxial rotation angle; the method is also used for dividing a training set and a testing set for the positive and negative samples; the dictionary is also used for establishing a dictionary according to the sample data, and the dictionary comprises a plurality of words; the classification training module is also used for searching corresponding words in the dictionary according to the data in the training set, counting the number of times of searching each word to obtain counting characteristics of the corresponding data, normalizing the counting characteristics to be used as frequency domain characteristics, and training a frequency domain SVM classifier according to the frequency domain characteristics;
wherein the counting feature of the corresponding data is obtained, and the normalization is used as a frequency domain feature, comprising: for the test set and the training set, cutting the test set and the training set into fragments with equal length, reserving N% points before the amplitude value through discrete Fourier transform, wherein N is a preset value, extracting three-dimensional data characteristics of the amplitude value, the phase position and the frequency of the points, taking the three-dimensional data characteristics as counting characteristics, extracting counting characteristics, searching words closest to the characteristics of each point, wherein each characteristic in a dictionary is a word, adding one to the position number of the corresponding word, obtaining the counting characteristics of the whole fragment, and normalizing the counting characteristics to be frequency domain characteristics.
2. The intelligent fall detection platform according to claim 1, wherein: the intelligent terminal comprises a main control module, wherein the main control module is preset with a judging threshold value, and is used for judging whether the gesture data is abnormal according to the judging threshold value, and judging that the gesture data is abnormal when the gesture data is located outside the judging threshold value.
3. The intelligent fall detection platform according to claim 1, wherein: the intelligent terminal is also used for acquiring current positioning information and uploading the current positioning information according to the falling judgment result.
4. The intelligent fall detection platform according to claim 3, wherein: the intelligent terminal is also used for acquiring and storing the strong positioning information under the strong signal; the inertial data acquisition module is also used for acquiring and storing inertial data; and the method is also used for uploading strong positioning information and inertial data when uploading the current positioning information.
5. The intelligent fall detection platform according to claim 1, wherein: the fall judgment result comprises fall types, wherein the fall types comprise general fall and consciousness loss fall, and the remote server is further used for triggering corresponding alarm according to the fall types.
6. The intelligent fall detection platform according to claim 1, wherein: the remote server is also used for training the frequency domain SVM classifier and the time domain LSTM classifier according to the positive and negative samples in the training set and testing the frequency domain SVM classifier and the time domain LSTM classifier according to the positive and negative samples in the testing set; the remote server is used for calling the trained frequency domain SVM classifier and the trained time domain LSTM classifier to conduct frequency domain analysis and time domain analysis.
7. The intelligent fall detection platform according to claim 6, wherein: the classification training module is also used for expanding the positive samples, including random sampling and random lifting sampling rate.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211130235.3A CN115299937B (en) | 2022-09-16 | 2022-09-16 | Intelligence detection platform that falls down |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211130235.3A CN115299937B (en) | 2022-09-16 | 2022-09-16 | Intelligence detection platform that falls down |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115299937A CN115299937A (en) | 2022-11-08 |
| CN115299937B true CN115299937B (en) | 2023-08-11 |
Family
ID=83866645
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211130235.3A Active CN115299937B (en) | 2022-09-16 | 2022-09-16 | Intelligence detection platform that falls down |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115299937B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117333929B (en) * | 2023-12-01 | 2024-02-09 | 贵州省公路建设养护集团有限公司 | Method and system for identifying abnormal personnel under road construction based on deep learning |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103976739A (en) * | 2014-05-04 | 2014-08-13 | 宁波麦思电子科技有限公司 | Wearing type dynamic real-time fall detection method and device |
| WO2014172671A1 (en) * | 2013-04-18 | 2014-10-23 | Digimarc Corporation | Physiologic data acquisition and analysis |
| CN106503672A (en) * | 2016-11-03 | 2017-03-15 | 河北工业大学 | A kind of recognition methods of the elderly's abnormal behaviour |
| CN107688790A (en) * | 2017-09-01 | 2018-02-13 | 东软集团股份有限公司 | Human bodys' response method, apparatus, storage medium and electronic equipment |
| CN111582146A (en) * | 2020-05-06 | 2020-08-25 | 宁波大学 | High-resolution remote sensing image city function partitioning method based on multi-feature fusion |
| CN113178192A (en) * | 2021-04-30 | 2021-07-27 | 平安科技（深圳）有限公司 | Training method, device and equipment of speech recognition model and storage medium |
| CN114469078A (en) * | 2022-01-28 | 2022-05-13 | 北京航空航天大学 | Human motion detection method based on optical-inertial fusion |
| CN114818952A (en) * | 2022-05-07 | 2022-07-29 | 南开大学 | Intelligent falling posture classification and identification method based on mobile phone sensor |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11282363B2 (en) * | 2017-09-29 | 2022-03-22 | Apple Inc. | Detecting falls using a mobile device |
| EP3828854A1 (en) * | 2019-11-29 | 2021-06-02 | Koninklijke Philips N.V. | Fall detection method and system |
-
2022
- 2022-09-16 CN CN202211130235.3A patent/CN115299937B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2014172671A1 (en) * | 2013-04-18 | 2014-10-23 | Digimarc Corporation | Physiologic data acquisition and analysis |
| CN103976739A (en) * | 2014-05-04 | 2014-08-13 | 宁波麦思电子科技有限公司 | Wearing type dynamic real-time fall detection method and device |
| CN106503672A (en) * | 2016-11-03 | 2017-03-15 | 河北工业大学 | A kind of recognition methods of the elderly's abnormal behaviour |
| CN107688790A (en) * | 2017-09-01 | 2018-02-13 | 东软集团股份有限公司 | Human bodys' response method, apparatus, storage medium and electronic equipment |
| CN111582146A (en) * | 2020-05-06 | 2020-08-25 | 宁波大学 | High-resolution remote sensing image city function partitioning method based on multi-feature fusion |
| CN113178192A (en) * | 2021-04-30 | 2021-07-27 | 平安科技（深圳）有限公司 | Training method, device and equipment of speech recognition model and storage medium |
| CN114469078A (en) * | 2022-01-28 | 2022-05-13 | 北京航空航天大学 | Human motion detection method based on optical-inertial fusion |
| CN114818952A (en) * | 2022-05-07 | 2022-07-29 | 南开大学 | Intelligent falling posture classification and identification method based on mobile phone sensor |
Non-Patent Citations (1)
| Title |
|---|
| 基于智能手机的人体跌倒监控系统的研究;朱彤昆;《中国优秀硕士学位论文全文数据库 (信息科技辑)》;I140-150 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115299937A (en) | 2022-11-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107212890B (en) | A kind of movement identification and fatigue detection method and system based on gait information | |
| CN106095099B (en) | A kind of user behavior motion detection recognition methods | |
| CN101194273B (en) | Spatio-temporal self organising map | |
| CN109363659B (en) | Heart rate monitoring method and device based on deep learning and storage medium | |
| CN109166275B (en) | Human body falling detection method based on acceleration sensor | |
| Jian et al. | A portable fall detection and alerting system based on k-NN algorithm and remote medicine | |
| CN108831527A (en) | A kind of user movement condition detection method, device and wearable device | |
| CN105310696A (en) | Fall detection model construction method as well as corresponding fall detection method and apparatus | |
| CN102368297A (en) | Equipment, system and method for recognizing actions of detected object | |
| CN105590409A (en) | Human body tumble detection method and human body tumble detection system based on big data | |
| CN106725445B (en) | A kind of the portable body injury gained in sports monitor system and method for brain wave control | |
| CN105726034B (en) | Based on Wristwatch type of the alarm with track and localization function is fallen down, intelligently platform is nursed in endowment | |
| CN115299937B (en) | Intelligence detection platform that falls down | |
| CN105023022A (en) | Tumble detection method and system | |
| CN104539726A (en) | System and method used for concerning behavior rules of the old by children | |
| Zhao et al. | Recognition of human fall events based on single tri-axial gyroscope | |
| CN106650300B (en) | Old man monitoring system and method based on extreme learning machine | |
| CN107019501B (en) | Remote tumble detection method and system based on genetic algorithm and probabilistic neural network | |
| CN108717548B (en) | Behavior recognition model updating method and system for dynamic increase of sensors | |
| CN203931101U (en) | A kind of wearable human paralysis device of falling detection alarm | |
| CN114343574A (en) | Sleep quality evaluation method, client, server and storage medium | |
| CN114601431A (en) | Wearable health monitoring system based on cloud edge cooperation | |
| Li | Fall detection with wrist-worn watch by observations in statistics of acceleration | |
| CN114267430A (en) | Route and environment-based tourist physical ability and consumption estimation method and system | |
| Silva et al. | Fusion of clinical, self-reported, and multisensor data for predicting falls |
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 | ||
| CB03 | Change of inventor or designer information | ||
| CB03 | Change of inventor or designer information |
Inventor after: Liu Yutong Inventor after: Wu Linchun Inventor after: Cheng Jingchun Inventor before: Wu Linchun Inventor before: Liu Yutong Inventor before: Cheng Jingchun |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |