CN115089135A - Millimeter wave radar-based elderly health state detection method and system - Google Patents
Millimeter wave radar-based elderly health state detection method and system Download PDFInfo
- Publication number
- CN115089135A CN115089135A CN202210443414.6A CN202210443414A CN115089135A CN 115089135 A CN115089135 A CN 115089135A CN 202210443414 A CN202210443414 A CN 202210443414A CN 115089135 A CN115089135 A CN 115089135A
- Authority
- CN
- China
- Prior art keywords
- data
- point cloud
- user
- height
- millimeter wave
- 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.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 60
- 230000036541 health Effects 0.000 title claims abstract description 29
- 238000000034 method Methods 0.000 claims abstract description 44
- 230000002159 abnormal effect Effects 0.000 claims abstract description 11
- 238000012549 training Methods 0.000 claims description 35
- 238000012706 support-vector machine Methods 0.000 claims description 19
- 238000013145 classification model Methods 0.000 claims description 12
- 238000006243 chemical reaction Methods 0.000 claims description 10
- 230000036387 respiratory rate Effects 0.000 claims description 8
- 210000000115 thoracic cavity Anatomy 0.000 claims description 6
- 210000000038 chest Anatomy 0.000 claims description 5
- 230000001960 triggered effect Effects 0.000 claims description 5
- 238000012795 verification Methods 0.000 claims description 5
- 238000009499 grossing Methods 0.000 claims description 4
- 230000003862 health status Effects 0.000 claims description 4
- 238000010276 construction Methods 0.000 claims description 3
- 238000002790 cross-validation Methods 0.000 claims description 3
- 239000000284 extract Substances 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 230000035790 physiological processes and functions Effects 0.000 claims description 2
- 239000000126 substance Substances 0.000 claims description 2
- 230000008859 change Effects 0.000 abstract description 8
- 230000029058 respiratory gaseous exchange Effects 0.000 abstract description 3
- 230000006870 function Effects 0.000 description 19
- 238000012544 monitoring process Methods 0.000 description 16
- 230000006399 behavior Effects 0.000 description 11
- 230000009471 action Effects 0.000 description 8
- 238000010606 normalization Methods 0.000 description 8
- 238000000605 extraction Methods 0.000 description 7
- 238000001914 filtration Methods 0.000 description 7
- 208000027418 Wounds and injury Diseases 0.000 description 5
- 208000014674 injury Diseases 0.000 description 5
- 230000033001 locomotion Effects 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 4
- 230000006378 damage Effects 0.000 description 4
- 210000003128 head Anatomy 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 206010020772 Hypertension Diseases 0.000 description 3
- 230000005856 abnormality Effects 0.000 description 3
- 230000032683 aging Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 201000001421 hyperglycemia Diseases 0.000 description 3
- 230000007774 longterm Effects 0.000 description 3
- 238000005096 rolling process Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000002354 daily effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000009434 installation Methods 0.000 description 2
- 230000000474 nursing effect Effects 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 206010000117 Abnormal behaviour Diseases 0.000 description 1
- 206010063746 Accidental death Diseases 0.000 description 1
- 208000012260 Accidental injury Diseases 0.000 description 1
- 208000034656 Contusions Diseases 0.000 description 1
- 206010017076 Fracture Diseases 0.000 description 1
- 206010052428 Wound Diseases 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000002146 bilateral effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 239000003245 coal Substances 0.000 description 1
- 238000010924 continuous production Methods 0.000 description 1
- 238000010219 correlation analysis Methods 0.000 description 1
- 230000009193 crawling Effects 0.000 description 1
- 230000034994 death Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 229940079593 drug Drugs 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 210000001061 forehead Anatomy 0.000 description 1
- 230000012010 growth Effects 0.000 description 1
- 230000005802 health problem Effects 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000003340 mental effect Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 230000001681 protective effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000036391 respiratory frequency Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 201000002859 sleep apnea Diseases 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/02—Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
- A61B5/0205—Simultaneously evaluating both cardiovascular conditions and different types of body conditions, e.g. heart and respiratory condition
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/02—Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
- A61B5/024—Detecting, measuring or recording pulse rate or heart rate
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/05—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves
- A61B5/0507—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves using microwaves or terahertz waves
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/08—Detecting, measuring or recording devices for evaluating the respiratory organs
- A61B5/0816—Measuring devices for examining respiratory frequency
-
- 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
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
-
- 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
-
- 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
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/74—Details of notification to user or communication with user or patient ; user input means
- A61B5/746—Alarms related to a physiological condition, e.g. details of setting alarm thresholds or avoiding false alarms
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B2503/00—Evaluating a particular growth phase or type of persons or animals
- A61B2503/08—Elderly
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Public Health (AREA)
- Veterinary Medicine (AREA)
- Pathology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Physiology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- Artificial Intelligence (AREA)
- Signal Processing (AREA)
- Cardiology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Pulmonology (AREA)
- Psychiatry (AREA)
- Mathematical Physics (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- Radiology & Medical Imaging (AREA)
- Fuzzy Systems (AREA)
- Dentistry (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The invention discloses a method and a system for detecting the health state of an old man based on a millimeter wave radar, wherein the detection method comprises the following steps: acquiring three-dimensional point cloud data of a user, judging whether the user is in a bed area, if so, entering a second step, and if not, entering a third step; step two, acquiring physiological characteristic data of a user, judging whether the physiological characteristic data is abnormal or not, if so, entering the step four, otherwise, returning to the step one; step three, judging whether a falling behavior occurs according to the three-dimensional point cloud data, if so, entering the step four, and if not, returning to the step one; and step four, triggering an alarm when the physiological characteristic abnormity is detected in the bed area or the falling behavior is detected in the non-bed area. By adopting the body inclination angle, the change of local point cloud can be obviously found to be very strong, and even slight breathing activity can also cause the local part of the data to generate larger change.
Description
Technical Field
The invention relates to the technical field of health monitoring of old people, in particular to a method and a system for detecting the indoor health state of the old people based on a millimeter wave radar.
Background
With the increasing aging problem of the population, the elderly have more and more social attention to the monitoring problem of the aging safety. The main factors of causing injuries to the elderly are "falls" and "elderly sleep apnea syndrome".
"Fall" refers to an action in which a person falls on the ground or is in a low position for an unexpected reason, and is one of the most common abnormal actions of a human being. According to the statistics of the world health organization, a fall is the second leading cause of accidental injury and death worldwide, over 3730 thousands of people are injured by a fall each year, and 64.6 thousands of people die directly or indirectly by a fall. The frequency of falls increases with age, with 28-35% of the elderly over age 65 experiencing falls each year, and this proportion rises to around 45% in the population over 80 years of age and beyond. Health risks caused by falling show an exponential growth trend along with the age, for the old, the falling may cause serious consequences such as incised wounds, bruises or fractures, psychological injuries caused by falling cannot be ignored, and a lot of old people generate fear psychology to falling and affect the life quality of the old. However, the uncertainty of the fall time and the difficulty in preventing the fall cause practical difficulties for the rescue work of the old people after the fall. Therefore, finding that an elderly person falls in time has attracted much attention from researchers. In addition, slow response to help after a fall is also an important cause of reduced life expectancy of the elderly.
Falls have become a major economic and health problem worldwide, including hospitalization and long-term care resulting from falls, and physical, mental, and monetary burdens resulting from falls. From an economic perspective, any increase in medical burden would bring immediate economic costs to society, while also resulting in a loss of productivity. The aging of population pushes the medical care to provide updated services, and the development of the medical care from the traditional treatment that patients go to hospitals by themselves to the family care services is that the patients can be partially treated at home through the support of emerging technology. The long-term nursing method taking the family as the center not only improves the life quality of the patient, but also indirectly saves the medical nursing cost of the patient and the society. The elderly living alone are prone to fall down, whether due to home medical care or due to living alone for other reasons. In order to ensure the safety of the elderly in independent life, it is necessary to find out a fall event in a timely manner so as to notify medical staff or family members in an emergency.
The detection of the health state of the old people is to detect the health state of the old people through various technical means, and generally, after the health state is detected to be abnormal, warning information can be sent out in some modes, or intervention or protection actions can be taken. The existing health state detection methods mainly comprise 2 methods:
the first is based on wearable devices. However, since the wearable device requires the user to continuously wear the wearable device and charge the wearable device in time, the user may forget to wear or charge the wearable device, which affects the user experience, and most of the elderly are not used to wear such wearable devices, so that the scheme based on the wearable device is difficult to work.
The second is to use a non-wearable approach to detect for a fixed area. Therefore, for monitoring the fall of the elderly, a fall monitoring method based on a non-wearable device is more practical. There are mainly the following sensors: camera, WiFi, ultra wide band radar, millimeter wave radar. The image detection-based mode is that image video information is directly processed to obtain the behavior state of the old, and alarm information is sent to a terminal once an abnormal behavior state occurs. However, this method is not suitable for use in private places such as restrooms and bedrooms because of the risk of privacy disclosure, because it involves sensitive information such as images and videos, and also requires sufficient light, and is not suitable for use in night scenes. The WiFi-based method is easy to be interfered by other external signals, and meanwhile, the algorithm is based on traditional modeling and has poor generalization. The millimeter wave radar has unique advantages in the aspects of privacy protection, positioning accuracy, detection range, environment adaptability and the like. The method can realize the health and safety monitoring of the personnel in the room and timely judge and alarm the dangerous behaviors only by using the millimeter wave radar which is a non-contact monitoring technology, does not relate to any sound, image and video information in the monitoring method process, avoids the problem that the privacy of the user is possibly revealed, and provides a good choice for the user paying attention to the privacy protection. Therefore, a great number of researchers have intensively studied fall detection methods based on millimeter wave radars.
In the research of the implementation and detection method of the falling method based on the millimeter wave radar, such as Sunjiao, the position information of a target is obtained through the millimeter wave radar, the mean value filtering is carried out on the position coordinate sequence, then the approximate motion rule of the monitored object is obtained through piecewise linear fitting, the position change rules of the human body in the height direction and the horizontal direction and the corresponding relation between the position change rules are comprehensively considered when the human body falls, the falling mode is identified after the falling behavior is detected based on a plurality of threshold values, and if the falling is found, an alarm is sent immediately. The method described herein is implemented in an open area and does not involve collisions with other objects in the environment, whereas a person is in an uncontrolled state during a real fall.
The invention discloses a toilet fall detection method based on millimeter wave radars in Chinese patent with publication number CN112782664A and publication number 2021.05.11, and particularly relates to a method for collecting user data by using the arranged millimeter wave radars; carrying out coordinate conversion on the user data and determining three-dimensional point cloud data of the user data in an actual scene; removing the static object to obtain a point cloud picture of the dynamic object, and removing interference noise; and finally, carrying out falling detection according to comprehensive judgment of the aspect ratio of the user larger than a certain threshold, the highest point smaller than a certain threshold, the falling speed larger than a certain threshold and the like. But not all fall movements are fast movements and some non-fall movements are also fast. This is particularly so because in many cases people will grab a piece of furniture or lean against a wall when falling, which slows the fall and causes a reduction in rate. It is also common for the elderly to fall from a chair or wheelchair, in which the fall action reduces their speed and high frequency energy value. Meanwhile, the three-dimensional point cloud data are detected according to the method, the finally determined three-dimensional point cloud only records the positions of the point clouds corresponding to the reflection signals, and whether a certain action falls down or not can not be accurately judged by using the information.
In an integrated monitoring method for indoor personnel safety based on a millimeter wave radar, which is disclosed in the Chinese patent with the publication number of CN111887861A and the publication number of 2020.11.06, scanning data of a user are acquired by a millimeter wave radar for monitoring physiological characteristics of a bed area and a whole-house attitude monitoring millimeter wave radar, and the user is detected to fall down, fall down and the like. However, in the posture monitoring, only the height information is used for fall judgment, but when a person moves, a signal reflected by a body changes, and a signal source reflected by the body in two adjacent frames of images may be transferred from the waist of a first frame to the chest of a second frame, so that the false alarm and the false alarm are serious when the person judges that the person falls according to the change of the height information.
Therefore, how to reasonably extract features of various data information collected by the millimeter wave radar and perform accurate fall detection is a field worthy of research. A high-precision detection method is urgently needed, and the detection method can monitor the dynamic state of the old people in real time, alarm the old people falling down, rescue quickly and reduce the injury.
Disclosure of Invention
Aiming at the problems in the prior art, the invention aims to provide the millimeter wave radar-based method and the millimeter wave radar-based system for detecting the indoor health state of the old man, which can monitor the dynamic state of the old man in real time, alarm the old man when falling down, rescue the old man quickly and reduce injuries.
The technical scheme adopted by the invention is as follows:
an old man health state detection system based on millimeter wave radar comprises a millimeter wave data acquisition unit, a data conversion unit, a model training unit, a prediction unit and a falling judgment unit; wherein the content of the first and second substances,
the millimeter wave data acquisition unit comprises a posture detection millimeter wave radar and a physiological characteristic detection millimeter wave radar, three-dimensional point cloud data of a user is acquired through the posture detection millimeter wave radar, and physiological characteristic data of the user is acquired through the physiological characteristic detection millimeter wave radar;
the data conversion unit converts the three-dimensional point cloud data into human body posture information of a target, and extracts heart rate and respiratory rate from the physiological characteristic data;
the model training unit comprises two parts, wherein one part is to establish a physiological characteristic prediction model based on LSTM, the physiological characteristic prediction model obtains a prediction result of the physiological state of the user based on the physiological characteristic data, and the second part supports a vector machine model, and the vector machine model obtains a falling classification model based on three-dimensional point cloud data and body inclination data;
the prediction unit is used for inputting the data detected by the data conversion unit into a trained prediction model in real time and judging the state of the current user;
the alarm unit triggers an alarm to remind children and children to pay attention when the user is in the bed area and the physiological characteristic detection judges that the user is in a non-health state; when the user is in a non-bed area and the falling behavior is detected, the alarm is triggered to remind children to pay attention.
The application also provides a method for detecting the health state of the old based on the millimeter wave radar, wherein the detection method comprises the following steps:
acquiring three-dimensional point cloud data of a user, judging whether the user is in a bed area, if so, entering a second step, and if not, entering a third step;
step two, acquiring physiological characteristic data of a user, judging whether the physiological characteristic data is abnormal or not, if so, entering the step four, otherwise, returning to the step one;
step three, judging whether a falling behavior occurs according to the three-dimensional point cloud data, if so, entering the step four, and if not, returning to the step one;
and step four, triggering an alarm when the physiological characteristic abnormity is detected in the bed area or the falling behavior is detected in the non-bed area.
Further, the third step specifically comprises:
step 3.1, a training set made of the data generated in the data conversion unit is used for training a support vector machine model;
and 3.2, inputting the three-dimensional point cloud data of the object into the trained support vector machine model for falling judgment.
Further, the step 3.1 specifically comprises:
step 3.1.1, collecting training sample data, collecting three-dimensional point cloud data of a falling sample as a positive sample, and marking the positive sample as 1; collecting three-dimensional point cloud data characteristics of non-falling actions as a negative sample, and marking the characteristics as 0;
step 3.1.2, smoothing the three-dimensional point cloud data;
step 3.1.3, extracting the characteristics of the three-dimensional point cloud data to obtain the absolute height, height mean, height variance and height-to-width ratio data of a user;
step 3.1.4, recording the three-dimensional point cloud corresponding to the maximum reflection intensity in the three-dimensional point cloud data as a thoracic cavity position, and determining a body inclination angle according to the thoracic cavity position and the absolute height, height mean, height variance and aspect ratio data of the user;
and 3.1.5, obtaining basic data for training the support vector machine model based on the data obtained in the step 3.1.3 and the step 3.1.4.
Further, the step 3.2 specifically includes:
step 3.2.1, detecting the number of people, starting falling detection judgment when one person is detected, and entering step 3.2.2, otherwise, entering step one;
step 3.2.2, acquiring three-dimensional point cloud data of the object;
step 3.2.3, extracting the characteristics of the three-dimensional point cloud data;
and 3.2.4, using the trained fall detection model to predict the fall.
Further, the step 3.1.2 specifically comprises:
1) calculating absolute height of user
Based on the three-dimensional point cloud data of the users in the ith frame, acquiring the highest point z in the ith frame max [i]As the highest point in the z-axis direction in the three-dimensional point cloud picture of the ith frame user, the absolute height of the user at the ith frame is represented as height [ i [ ] i]＝z max [i]；
2) Calculating a height average value in a weighting mode;
wherein i represents the ith frame, N represents how many previous frames are used for calculating the weighted average value, N < 100, mean [ i ] represents the height average value of the ith frame, mean [0] is height [0], and height [ i ] represents the absolute height of the ith frame.
3) Calculating the height variance:
delta[i]＝mean[i]-meam[(i-10)％100]
4) calculating the aspect ratio:
calculating the aspect ratio of the user in the lateral direction in the ith frame as
Wherein x is max [i]Represents the maximum value in the x-axis direction, x, of the point cloud in the ith frame min [i]Representing the minimum value of the x-axis direction in the point cloud in the ith frame;
the aspect ratio of the user in the longitudinal direction in the ith frame is
Wherein, y max [i]Maximum value in y-axis direction in point cloud representing user in ith frame, y min [i]Representing the minimum value of the y-axis direction in the point cloud of the user in the ith frame;
the step 3.1.3 is specifically as follows:
the point cloud with the maximum emission power intensity is regarded as the characteristic point of the thorax part, and the velocity (v) of the point cloud x ，v y ，v z )
Velocity (v) x ，v y ，v z ) Can be respectively calculated as:
wherein x is max [t1]Is the x-axis coordinate value, x, corresponding to the point cloud with the maximum reflection intensity on the x-axis of the t1 th frame max [t2]The x-axis coordinate value, y, corresponding to the point cloud with the maximum reflection intensity on the x-axis of the t2 th frame max [t1]The coordinate value of the y axis corresponding to the point cloud with the maximum reflection intensity on the y axis of the t1 th frame max [t2]The coordinate value of the y axis corresponding to the point cloud with the maximum reflection intensity on the y axis of the t2 th frame, z max [t1]Z-axis coordinate value z corresponding to the point cloud with maximum reflection intensity on the z-axis of the t1 th frame max [t2]A Z-axis coordinate value corresponding to the point cloud with the maximum reflection intensity on the Z-axis of the t2 th frame; the body tilt angle θ is calculated by:
further, the step 3.1.5 specifically includes:
converting the sample text into a preset format by using a support vector machine:
<0> absolute height: eigenvalue, height mean: eigenvalue, height variance: eigenvalues, lateral aspect ratio: eigenvalues, longitudinal aspect ratio: eigenvalues, body dip: a characteristic value;
<1> absolute height: eigenvalue, height mean: eigenvalue, height variance: eigenvalues, lateral aspect ratio: eigenvalues, longitudinal aspect ratio: eigenvalues, body dip: a characteristic value;
and normalizing the characteristic value to a range of [0, 1 ].
Further, the support vector machine model is:
selecting a kernel function and determining parameters of the kernel function, and selecting a radial basis kernel function to solve the problem that the linearity of nonlinear data in a low-dimensional characteristic space is inseparable, wherein the formula of the kernel function is as follows:
wherein K (x, z) is a kernel function; z is the kernel function center; sigma is a width parameter; x is an indicated quantity;
selecting a punishment parameter C and a kernel function parameter g required by the construction of the model, and determining a final parameter pair C and g by a method of combining cross validation and network parameter optimization;
and inputting the basic data of the training sample into a support vector machine for training to obtain a classification model according to the kernel function of the determined parameter, and optimizing the classification model by using a verification sample to obtain an optimal classification model.
The beneficial results are that:
according to the invention, the health and safety of personnel in a room can be monitored only by using a millimeter wave radar which is a non-contact monitoring technology, and the judgment and alarm can be made on dangerous behaviors in time. Meanwhile, the safety problem of the old people without supervision is effectively solved, the harmony and stability of the society are enhanced, and the economic benefit and the social benefit are greater.
Drawings
FIG. 1 is a schematic structural diagram of an elderly people health status detection system based on millimeter wave radar according to the present application;
fig. 2 is a flowchart of the method for detecting the health state of the elderly based on the millimeter wave radar.
Detailed Description
The technical solution of the present invention will be explained with reference to the accompanying drawings. The described embodiments and their description serve to explain the invention without unduly limiting it.
Example 1
The embodiment discloses an old man health state detection system based on a millimeter wave radar, and a structural block diagram of the system is shown in fig. 1. The device comprises a millimeter wave data acquisition unit, a data conversion unit, a model training unit, a prediction unit and a falling judgment unit.
The method specifically comprises the following steps:
firstly, a data acquisition unit:
the gesture detection millimeter wave radar and the physiological characteristic detection millimeter wave radar are installed in a room to be monitored, the monitoring range of the gesture detection millimeter wave radar and the physiological characteristic detection millimeter wave radar is ensured to cover the whole room, the three-dimensional point cloud data of a user is obtained through the gesture detection millimeter wave radar, and the physiological characteristic data of the user is obtained through the physiological characteristic detection millimeter wave radar.
II, a feature extraction unit:
the three-dimensional point cloud data of the user acquired by the attitude detection millimeter wave radar is processed, noise is removed through a smoothing filtering algorithm, the absolute height, the height mean value, the height variance, the width-to-height ratio, the human body inclination angle information and the like of the target are acquired, and signals acquired by the physiological characteristic millimeter wave radar are filtered to extract the heart rate and the respiratory frequency.
Three, training unit
And counting data such as heart rate, respiratory rate, hypertension, hyperglycemia, height, weight, current temperature and the like of the user every minute for the past 1 month, and establishing a physiological characteristic prediction model based on the LSTM.
Collecting information such as absolute height, height mean, height variance, aspect ratio, human body inclination angle information and the like of a target of a user during falling and non-falling, and training by using a support vector machine to obtain a falling classification model;
fourth, prediction unit
And predicting the heart rate, respiratory rate and other information of the user in each next time period according to the historical data and the LSTM model, and entering a physiological characteristic abnormality detection unit when the actual value and the predicted value have large deviation.
Fifth, judge the unit
Judging whether the human body is in a bed area or not according to target information acquired by the current posture detection millimeter wave radar, if so, detecting the heart rate and the respiratory rate through the physiological characteristic millimeter wave radar, and judging the human body to be in an unhealthy state when the heart rate is abnormal; otherwise, information such as the absolute height, the height mean, the height variance, the aspect ratio, the human body inclination angle information and the like of the target acquired by the attitude detection millimeter wave radar is sent to a falling detection model, and when the target is judged to fall, falling behavior is shown.
Alarm unit
When the user is in the bed area and the physiological characteristic detection and judgment are in a non-healthy state, an alarm is triggered to remind children to pay attention; when the user is in a non-bed area and a falling behavior is detected, an alarm is triggered to remind children to pay attention.
Example 2
The invention also provides a method for detecting the health state of the old people based on the millimeter wave radar, and a flow chart of the method is shown in figure 2.
Millimeter waves are electromagnetic waves operating at 30GHZ-300GHZ, with wavelengths on the order of millimeters, 1mm-10 mm. One advantage of short wavelength millimeter waves is their high accuracy. In the embodiment of the invention, the millimeter wave radar used for acquiring physiological signals and measuring distance is an IWR6843 chip of TI company, and can emit frequency modulation continuous waves, so that the distance of a reflector can be measured, and the speed of the reflector in motion can be measured, therefore, the method can be used for monitoring information such as heart rate, respiratory rate, height, speed and the like of a user.
The installation method of the attitude monitoring millimeter wave radar comprises the following steps: and installing an attitude monitoring millimeter wave radar at one corner of the roof of each room. Install sign detection millimeter wave radar at the bedroom, sign detection millimeter wave radar is installed on the ceiling of bedroom or inside the bed of non-metallic structure, just the detection antenna of the sign detection millimeter wave radar of installation on the ceiling is just to the bed.
The method comprises the following specific steps:
acquiring three-dimensional point cloud data of a user through a posture detection millimeter wave radar, judging whether the user is in a bed area, if so, entering a second step, and otherwise, entering a third step;
secondly, data such as heart rate, respiratory rate and the like of a user are acquired through a physiological characteristic detection millimeter wave radar, whether the data are abnormal or not is judged, if the data are abnormal, the step four is carried out, otherwise, the step one is returned, and the method specifically comprises the following steps:
2.1, counting data such as the heart rate of a human body, the breathing rate, the month corresponding to the day to be predicted, the temperature of the corresponding day, the age, the height and the weight of a user, whether hypertension exists or not, whether hyperglycemia exists or not and the like of the human body every minute when the user is on a bed every day in the past month;
2.2 outlier identification and handling
And identifying abnormal values by using a linear regression and residual error elimination method, and replacing the abnormal values by using an average value method in two adjacent periods.
2.3, selecting the Primary dataset
And (2) carrying out grey correlation degree analysis on the p factor sequences of the physiological characteristic health state factors in the step (1) and the heart rate sequence of the user respectively, and selecting q factors of which the correlation degree is greater than a given value as main factors influencing the target.
(1) Gray correlation analysis selects main factors influencing targets
In the objective world, the relationship between many factors is grey, which is unclear and which is not so close, so that it is difficult to find the main contradiction and find the main characteristics and main relationships. The relevance analysis is a method for analyzing the relevance degree of each factor in the system, and the relevance coefficient is calculated firstly, and then the relevance degree is calculated.
The reference system is: x 0 ＝{X 0 (1),X 0 (2),...,X 0 (n)}
The sequences being compared are: x i ＝{X i (1),X i (2),...,X i (n)},i＝1,2,...,k
1) Initialization
For sequences with different units and different initial values, initialization is performed before calculating the correlation coefficient, that is, all data in the sequence are divided by the first data respectively.
2) Sequence of absolute differences
Δ i (t)＝|X 0 (t)-X i (t)|,t＝1,2,...,n
3) Calculating correlation coefficient
Representing a reference series X 0 With the ith series X being compared i The absolute difference at the t-th point, the correlation coefficient is defined as:
where ρ is a resolution of 0 to 1, and is generally 0.5.
4) Degree of association
Compared sequence X i And reference series X 0 The degree of association of (2) is defined as the average value of the association coefficients of the two points, namely the degree of association:
2.4 normalization of data
The characteristic refers to an influence factor in the invention.
And performing normalization processing before all data are trained, and performing scrambling processing before all data are trained. The original data is mapped between 0, 1 without loss by the following formula. Because the data magnitudes of different influence factors are different, if the same normalization method is adopted, for data with smaller data magnitudes, the normalized data becomes very small, and the influence on the prediction result is not large, so that the normalization method mentioned herein normalizes each feature to be between 0 and 1 respectively.
In this embodiment, the training set, the verification set, and the test set are adopted according to the following 8: 1: 1 to obtain corresponding data in a random distribution manner.
2.5 predicting future heart rate data
The heart rate data for the next hour of the person is predicted using the LSTM prediction model.
The BP neural network and the support vector machine both use a machine learning method to find the nonlinear mapping relation between human health state influencing factors and human health states, and ignore the mutual relation of sequence data among human physiological signs. In fact, as a typical time sequence, the human physiological sign data not only has nonlinearity, but also has a correlation, that is, for a given region, the change of the human physiological sign data is a continuous process, and today's physiological sign data and yesterday's physiological sign data are not independent of each other, and there is a strong correlation between the two. Therefore, the daily physiological sign data change depends not only on the daily input characteristics but also on past input characteristics. Therefore, the traditional method only establishes a nonlinear relation between the input characteristics and the output of a single sample, loses strong correlation among continuous sequence samples, and has limited prediction precision. Therefore, the invention adopts a physiological sign data prediction method based on a Long-Short Term Memory artificial neural network (LSTM).
LSTM is an improved time-cycled neural network. The LSTM can learn the long-term and short-term dependency information of the time sequence, and is suitable for processing and predicting interval and delay events in the time sequence due to the fact that a time memory unit is contained in the neural network.
LSTM is implemented using Keras-packaged libraries and Keras-based LSTM multivariate drug sales prediction is implemented using Python code. Converting the original data set into a data set suitable for multivariate time series prediction, constructing a training set of the LSTM network in a rolling mode by using a rolling prediction method, inputting the constructed LSTM network, updating network parameters and weight values through the characteristics of learning data of the LSTM, and outputting physiological sign data of the next day.
(1) Network design
Eigenvalues (9): the method comprises the following steps of human heart rate, respiratory rate, month corresponding to a day to be predicted, temperature of the corresponding day, age, height and weight of a user, hypertension and hyperglycemia.
Selecting the historical physiological characteristic data of the previous 30 days as network input, and sending the network input into the LSTM for training and prediction.
The maximum training times of the network is 1000;
the network learning rate is 0.01;
implicit number of layers: 2 are provided with
Hidden layer neuron number: 30*30
Input layer and hidden layer add dropout, which is 0.1
A training stage:
inputting: 30*8*60*9
And (3) outputting: 30*8*60
A prediction stage:
inputting: 1*9*30
And (3) outputting: 1*1.
(2) Roll prediction
The health state data of the underground coal mine power supply equipment of one day to be predicted are predicted in a rolling prediction mode one at a time, then the predicted human physiological characteristic data and other related information are added into an input value, the physiological characteristic data of the next day are predicted, and the like, so that the worse the situation is, the more inaccurate the situation is.
(3) Model incremental update
In order to ensure the timeliness of the model, the model needs to be updated regularly, the time interval is usually 1 month, and if the model is retrained by using all data each time, the time overhead is very large, so the model can be trained by using an incremental model updating mode.
After the model is trained by adopting the basic data set, the model is serialized, and then the model is imported again and the incremental training is carried out.
Abnormality determination
When the currently detected heart rate data is far higher or far lower than the predicted heart rate data, the heart rate is abnormal at the moment.
Acquiring three-dimensional point cloud data of a user to judge whether a falling behavior occurs, if so, entering a fourth step, otherwise, returning to the first step, specifically:
3.1 training fall detection model
(1) Collecting training sample data
Collecting three-dimensional point cloud data of a falling sample such as a falling sample of falling down when a chair tilts, falls down vertically, falls forward from the chair, falls in unbalance when carrying heavy objects, and the like, and taking the three-dimensional point cloud data as a positive sample and marking the data as 1;
collecting three-dimensional point cloud data characteristics of non-falling actions such as sitting, slightly forward leaning of the body, slightly backward leaning of the body, squatting for tying shoelaces, random walking, random actions (Tai Ji, body building, dancing, stretching and crawling) and the like, and marking the characteristics as a negative sample as 0;
the three-dimensional point cloud corresponding to the maximum reflection intensity in the point cloud set is basically the position of the thoracic cavity. The reflection intensity information is used as one dimension for identifying the trunk part of the human body, and the number of point clouds of the trunk part is normally more than that of four limbs and the head part, so that the point cloud group corresponding to the trunk of the human body can be detected based on the number of the point clouds of the point cloud group and the reflection intensity information.
(2) Data pre-processing
Because data acquired by the three-dimensional point cloud often contains noise and influences the feature extraction of the three-dimensional point cloud, the three-dimensional point cloud data needs to be subjected to smoothing treatment by a plurality of methods, such as gaussian filtering, bilateral filtering and the like;
in the embodiment of the invention, a mean filtering method is adopted, and the mean filtering is a method of replacing the value of point cloud data by the mean value of two adjacent frames of point cloud data.
(3) Feature extraction
1) Calculating absolute height of user
Based on the point cloud data of the user, acquiring the highest point z of the point cloud data of the current frame user max [i]As the highest point in the Z-axis direction in the three-dimensional point cloud picture of the ith frame user, the absolute height of the user in the ith frame is represented as height [ i [ ]]＝z max [i]。
2) Calculating a height average value in a weighting mode;
wherein i represents the frame, N represents how many previous frames are used to calculate the weighted average, N < 100, mean [ i ] represents the height average of the ith frame, mean [0] ═ height [0], height [ i ] represents the absolute height of the ith frame, and in the embodiment of the present invention, N ═ 20.
3) Calculating the height variance:
delta[i]＝mean[i]-mean[(i-10)％100]
4) calculating an aspect ratio;
calculating the aspect ratio of the user in the lateral direction in the ith frame as
Wherein x is max [i]Represents the maximum value in the x-axis direction, x, of the point cloud in the ith frame min [i]Represents the minimum value in the x-axis direction in the point cloud in the ith frame.
The aspect ratio of the user in the longitudinal direction in the ith frame is
Wherein, y max [i]Maximum value in y-axis direction in point cloud representing user in ith frame, y min [i]And the minimum value of the y-axis direction in the point cloud of the user in the ith frame is represented.
5) Calculating body inclination
The body inclination angle refers to the included angle between the human body column body and the zenith direction. When the millimeter wave radar collects data, the fact that the local point cloud changes very strongly can be obviously found, and even slight breathing activity can cause large local changes of the data. For example, the signal reflected by the head may have only 1 point, and the signal source reflected by the head in the two adjacent frames of images may be transferred from the forehead of the first frame to the cheek of the second frame. Aiming at the characteristic that the millimeter wave signal data collected by the millimeter wave radar has large change, the influence of instability of local information should be reduced by using the stability of the whole information as much as possible.
The height characteristic value is only based on the height information in the point cloud three-dimensional coordinate, and the dip angle characteristic value utilizes the three-dimensional information and the reflection intensity information, so that the stability is higher. On the one hand, the human body inclination angle series parameters have higher stability and can more accurately reflect the posture of a human body, and on the other hand, the human body inclination angle information relative height information can more effectively distinguish common misjudgment actions such as lacing and sitting, so that the accuracy of falling judgment is effectively improved.
In the embodiment of the invention, a human body inclination angle calculation method based on speed is adopted, the physical condition is calculated based on the speed of a chest part, and the speed corresponding to each direction is calculated through measured x, y and z coordinates.
The point cloud with the maximum emission power intensity is regarded as the characteristic point of the thorax part, and the velocity (v) of the point cloud x ，v y ，v z ) Velocity (v) x ，v y ，v z ) Can be respectively calculated as:
wherein x is max [t1]The x-axis coordinate value, x, corresponding to the point cloud with the maximum reflection intensity on the x-axis of the t1 th frame max [t2]The x-axis coordinate value, y, corresponding to the point cloud with the maximum reflection intensity on the x-axis of the t2 th frame max [t1]The coordinate value of the y axis corresponding to the point cloud with the maximum reflection intensity on the y axis of the t1 th frame max [t2]Is the y-axis coordinate value, z, corresponding to the point cloud with the maximum reflection intensity on the y-axis of the t2 th frame max [t1]Z-axis coordinate value z corresponding to the point cloud with maximum reflection intensity on the z-axis of the t1 th frame max [t2]A z-axis coordinate value corresponding to the point cloud with the maximum reflection intensity on the z-axis of the t2 th frame;
the body tilt angle θ is calculated by:
(4) training fall detection model
1) Sample normalization
The method comprises the steps of preprocessing three-dimensional point cloud data acquired by a millimeter wave radar, wherein the three-dimensional point cloud data comprises 6 features and 2 types, 1000 samples are selected for falling, 1000 samples are selected for non-falling for training, and 200 samples are reserved for verification.
Converting the sample text into a preset format by using a support vector machine:
"< 0> absolute height: eigenvalue, height mean: eigenvalue, height variance: eigenvalues, lateral aspect ratio: eigenvalues, longitudinal aspect ratio: eigenvalues, body dip: eigenvalue "
<1> absolute height: eigenvalue, height mean: eigenvalue, height variance: eigenvalues, lateral aspect ratio: characteristic value, longitudinal aspect ratio; eigenvalues, body dip: characteristic value ";
2) eigenvalue normalization
In general, different data types often have different dimensions, and the difference between numerical values may be large, and the data are directly input as original data, and relatively large weight occupied in the modeling process affects the performance of the model, and in order to effectively utilize the data, normalization processing needs to be performed on the data; the feature values are normalized to the [0, 1] range.
3) Selecting kernel functions
And selecting a kernel function, determining parameters of the kernel function, and selecting a radial basis kernel function to solve the problem that the linearity of nonlinear data in a low-dimensional characteristic space is inseparable.
The formula of the kernel function is:
wherein K (x, z) is a kernel function; z is the kernel function center; sigma is a width parameter; x is an indicated amount;
selecting a punishment parameter C and a kernel function parameter g required by the construction of the model, and determining a final parameter pair C and g by a method of combining cross validation and network parameter optimization;
4) inputting the feature vector of the training sample into a support vector machine to be trained to obtain a classification model according to the kernel function of the determined parameter, and optimizing the classification model by using a verification sample to obtain an optimal classification model;
in the embodiment of the invention, because the libsvm program is small, the application is flexible, the input parameters are less, the libsvm program is open-source and easy to expand, and the libsvm program becomes an SVM library most widely applied in China at present; therefore, the invention selects the libsvm library to input the converted training samples into the libsvm library for two-class training.
3.2 making Fall prediction
(1) Detecting the number of people;
(2) when a person is detected, starting falling detection judgment, and entering the step (3), otherwise, entering the step (1);
(3) obtaining three-dimensional point cloud data
Utilizing the three-dimensional point cloud corresponding to the maximum reflection intensity in the point cloud set as the body height of the monitored person, wherein half of the detected body height is the position of the thoracic cavity;
the reflection intensity information is used as one dimension for identifying the trunk part of the human body, and the number of point clouds of the trunk part is normally more than that of four limbs and the head part, so that the point cloud group corresponding to the trunk of the human body can be detected based on the number of the point clouds of the point cloud group and the reflection intensity information.
(4) According to the point cloud, calculating the characteristics of the user such as absolute height, height mean, height variance, aspect ratio, body inclination angle and the like according to the same characteristic extraction method in the steps (4) and (5) in the step (3.1), and performing smooth filtering, standardization and normalization to select the kernel function which is the same as the kernel function in the training stage.
(5) And carrying out fall prediction by using a trained fall detection model.
3.1 and 3.2, the training models and the fall detection are subjected to a feature extraction process.
(1) The positive and negative sample training set is subjected to feature extraction to obtain features such as absolute height, height mean, height variance, aspect ratio, body inclination angle and the like, and the features are sent to a support vector machine for training to obtain a support vector machine model;
(2) when in falling detection, the obtained three-dimensional point cloud data is used for obtaining absolute height, height mean, height variance, height-to-width ratio and body inclination angle through characteristic extraction, and then whether the person falls or not is judged through a trained support vector machine model;
fourth, trigger the alarm
When the physiological characteristic abnormality is detected in the bed area or the falling behavior is detected in the non-bed area, the alarm is triggered to remind children to pay attention.
The technical solution of the present invention is not limited to the limitations of the above-mentioned specific embodiments, and all technical modifications made according to the technical solution of the present invention fall within the protective scope of the present invention.
Claims (8)
1. An old man health state detection system based on a millimeter wave radar is characterized by comprising a millimeter wave data acquisition unit, a data conversion unit, a model training unit, a prediction unit and a falling judgment unit; wherein the content of the first and second substances,
the millimeter wave data acquisition unit comprises a posture detection millimeter wave radar and a physiological characteristic detection millimeter wave radar, three-dimensional point cloud data of a user is acquired through the posture detection millimeter wave radar, and physiological characteristic data of the user is acquired through the physiological characteristic detection millimeter wave radar;
the data conversion unit converts the three-dimensional point cloud data into human body posture information of a target and extracts heart rate and respiratory rate from the physiological characteristic data;
the model training unit comprises two parts, wherein one part is to establish a physiological characteristic prediction model based on LSTM, the physiological characteristic prediction model obtains a prediction result of the physiological state of the user based on the physiological characteristic data, and the second part supports a vector machine model, and the vector machine model obtains a falling classification model based on three-dimensional point cloud data and body inclination data;
the prediction unit is used for inputting the data detected by the data conversion unit into a trained prediction model in real time and judging the state of the current user;
the alarm unit triggers an alarm to remind children to pay attention when the user is in the bed area and the physiological characteristic detection and judgment are in a non-healthy state; when the user is in a non-bed area and a falling behavior is detected, an alarm is triggered to remind children to pay attention.
2. A method for detecting the health state of the old people based on a millimeter wave radar is characterized by comprising the following steps:
acquiring three-dimensional point cloud data of a user, judging whether the user is in a couch area, if so, entering a second step, and otherwise, entering a third step;
step two, acquiring physiological characteristic data of a user, judging whether the physiological characteristic data is abnormal or not, if so, entering step four, otherwise, returning to the step one;
step three, judging whether a falling behavior occurs according to the three-dimensional point cloud data, if so, entering the step four, and if not, returning to the step one;
and step four, triggering an alarm when the physiological characteristic abnormity is detected in the bed area or the falling behavior is detected in the non-bed area.
3. The geriatric health status detection method based on the millimeter wave radar according to claim 2, wherein the third step is specifically:
step 3.1, a training set made of the data generated in the data conversion unit is used for training a support vector machine model;
and 3.2, inputting the three-dimensional point cloud data of the object into the trained support vector machine model for falling judgment.
4. The method for detecting the health state of the elderly based on the millimeter wave radar according to claim 3, wherein the step 3.1 specifically comprises:
step 3.1.1, smoothing the three-dimensional point cloud data;
step 3.1.2, extracting the characteristics of the three-dimensional point cloud data to obtain the absolute height, height mean, height variance and height-to-width ratio data of a user;
3.1.3, recording the three-dimensional point cloud corresponding to the maximum reflection intensity in the three-dimensional point cloud data as a thoracic cavity position, and determining a body inclination angle according to the thoracic cavity position and the absolute height, height mean, height variance and aspect ratio data of the user;
and 3.1.4, obtaining basic data for training the support vector machine model based on the data obtained in the step 3.1.2 and the step 3.1.3.
5. The geriatric health status detection method based on the millimeter wave radar according to claim 3, wherein the step 3.2 is specifically as follows:
step 3.2.1, detecting the number of people, starting falling detection judgment when one person is detected, and entering step 3.2.2, otherwise, entering step one;
step 3.2.2, acquiring three-dimensional point cloud data of the object;
step 3.2.3, extracting the characteristics of the three-dimensional point cloud data;
and 3.2.4, using the trained fall detection model to carry out fall prediction.
6. The method for detecting the health state of the elderly based on the millimeter wave radar according to claim 3, wherein the step 3.1.2 is specifically as follows:
1) calculating the absolute height of the user:
based on the three-dimensional point cloud data of the users in the ith frame, acquiring the highest point z in the ith frame max [i]As the highest point in the z-axis direction in the three-dimensional point cloud picture of the ith frame user, the absolute height of the user at the ith frame is represented as height [ i [ ] i]＝z max [i]；
2) The height average is calculated in a weighted manner:
wherein, i represents the ith frame, N represents how many previous frames are used for calculating the weighted average value, N is less than 100, mean [ i ] represents the height average value of the ith frame, mean [0] is height [0], and height [ i ] represents the absolute height of the ith frame;
3) calculating the height variance:
delta[i]＝mean[i]-mean[(i-10)％100]
4) calculating the aspect ratio:
calculating the aspect ratio of the user in the lateral direction in the ith frame as
Wherein x is max [i]Represents the maximum value in the x-axis direction, x, of the point cloud in the i-th frame min [i]Representing the minimum value of the x-axis direction in the point cloud in the ith frame;
the aspect ratio of the user in the longitudinal direction in the ith frame is
Wherein, y max [i]Maximum value in y-axis direction in point cloud representing user in ith frame, y min [i]Indicating the user in the ith frameThe minimum value in the y-axis direction in the point cloud of (1);
the step 3.1.3 is specifically as follows:
the point cloud with the maximum emission power intensity is regarded as the characteristic point of the thorax part, and the velocity (v) of the point cloud x ，v y ，v z )
Velocity (v) x ，v y ，v z ) Can be respectively calculated as:
wherein x is max [t1]The x-axis coordinate value, x, corresponding to the point cloud with the maximum reflection intensity on the x-axis of the t1 th frame max [t2]The x-axis coordinate value, y, corresponding to the point cloud with the maximum reflection intensity on the x-axis of the t2 th frame max [t1]The coordinate value of the y axis corresponding to the point cloud with the maximum reflection intensity on the y axis of the t1 th frame max [t2]The coordinate value of the y axis corresponding to the point cloud with the maximum reflection intensity on the y axis of the t2 th frame, z max [t1]Is a z-axis coordinate value z corresponding to the point cloud with the maximum reflection intensity on the z-axis of the tl frame max [t2]A z-axis coordinate value corresponding to the point cloud with the maximum reflection intensity on the z-axis of the t2 th frame; the body tilt angle θ is calculated by:
7. the method for detecting the health state of the elderly based on the millimeter wave radar according to claim 3, wherein the step 3.1.5 is specifically as follows:
converting the sample text into a preset format by using a support vector machine:
<0> absolute height: eigenvalue, height mean: eigenvalue, height variance: eigenvalues, lateral aspect ratio: eigenvalues, longitudinal aspect ratio: eigenvalues, body dip: a characteristic value;
<1> absolute height: eigenvalue, height mean: eigenvalue, height variance: eigenvalues, lateral aspect ratio: eigenvalues, longitudinal aspect ratio: eigenvalues, body dip: a characteristic value;
and normalizing the characteristic value to a [0, 1] range.
8. The elderly health status detection method based on millimeter wave radar according to any of claims 3 to 7, wherein the support vector machine model is:
selecting a kernel function and determining parameters of the kernel function, and selecting a radial basis kernel function to solve the problem that the linearity of nonlinear data in a low-dimensional characteristic space is inseparable, wherein the formula of the kernel function is as follows:
wherein K (x, z) is a kernel function; z is the kernel function center; sigma is a width parameter; x is an indicated quantity;
selecting a punishment parameter C and a kernel function parameter g required by the construction of the model, and determining a final parameter pair C and g by a method of combining cross validation and network parameter optimization;
and inputting the basic data of the training sample into a support vector machine for training to obtain a classification model according to the kernel function with the determined parameters, and optimizing the classification model by using a verification sample to obtain an optimal classification model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210443414.6A CN115089135A (en) | 2022-04-25 | 2022-04-25 | Millimeter wave radar-based elderly health state detection method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210443414.6A CN115089135A (en) | 2022-04-25 | 2022-04-25 | Millimeter wave radar-based elderly health state detection method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115089135A true CN115089135A (en) | 2022-09-23 |
Family
ID=83287731
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210443414.6A Pending CN115089135A (en) | 2022-04-25 | 2022-04-25 | Millimeter wave radar-based elderly health state detection method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115089135A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115359624A (en) * | 2022-10-24 | 2022-11-18 | 中科芯集成电路有限公司 | Millimeter wave radar fall detection algorithm based on state quantity analysis and sign detection |
| CN116687394A (en) * | 2023-08-04 | 2023-09-05 | 亿慧云智能科技（深圳）股份有限公司 | Tumble detection method, device, equipment and storage medium based on millimeter wave radar |
| CN117173793A (en) * | 2023-11-02 | 2023-12-05 | 浙江大华技术股份有限公司 | Target gesture detection method, computer device and storage medium |
| TWI836783B (en) * | 2022-12-12 | 2024-03-21 | 國立臺北科技大學 | Intelligent monitoring method and intelligent monitoring system suitable for individuals living alone |
-
2022
- 2022-04-25 CN CN202210443414.6A patent/CN115089135A/en active Pending
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115359624A (en) * | 2022-10-24 | 2022-11-18 | 中科芯集成电路有限公司 | Millimeter wave radar fall detection algorithm based on state quantity analysis and sign detection |
| TWI836783B (en) * | 2022-12-12 | 2024-03-21 | 國立臺北科技大學 | Intelligent monitoring method and intelligent monitoring system suitable for individuals living alone |
| CN116687394A (en) * | 2023-08-04 | 2023-09-05 | 亿慧云智能科技（深圳）股份有限公司 | Tumble detection method, device, equipment and storage medium based on millimeter wave radar |
| CN116687394B (en) * | 2023-08-04 | 2023-10-20 | 亿慧云智能科技（深圳）股份有限公司 | Tumble detection method, device, equipment and storage medium based on millimeter wave radar |
| CN117173793A (en) * | 2023-11-02 | 2023-12-05 | 浙江大华技术股份有限公司 | Target gesture detection method, computer device and storage medium |
| CN117173793B (en) * | 2023-11-02 | 2024-03-01 | 浙江大华技术股份有限公司 | Target gesture detection method, computer device and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115089135A (en) | Millimeter wave radar-based elderly health state detection method and system | |
| US11819344B2 (en) | Systems for automatic assessment of fall risk | |
| US20240099607A1 (en) | System, sensor and method for monitoring health related aspects of a patient | |
| Shin et al. | Detection of abnormal living patterns for elderly living alone using support vector data description | |
| CN110944577B (en) | Method and system for detecting blood oxygen saturation | |
| US9597016B2 (en) | Activity analysis, fall detection and risk assessment systems and methods | |
| CN113397520B (en) | Information detection method and device for indoor object, storage medium and processor | |
| CN112784662A (en) | Video-based fall risk evaluation system | |
| CN113925475B (en) | Non-contact human health monitoring system and monitoring method | |
| US20210057101A1 (en) | In-home remote monitoring systems and methods for predicting health status decline | |
| US20160128638A1 (en) | System and method for detecting and quantifying deviations from physiological signals normality | |
| KR20210154675A (en) | System of providing complex living support solution by monitoring and analyzing biosignal based on artificial intelligence and operating method thereof | |
| WO2009156936A2 (en) | System and method for determining a personal health related risk | |
| JP2021528135A (en) | Determining the reliability of vital signs of monitored persons | |
| JP2019155071A (en) | Event prediction system, sensor signal processing system, event prediction method, and program | |
| Minamisawa et al. | Dementia scale score classification based on daily activities using multiple sensors | |
| JP2023521416A (en) | Contactless sensor-driven devices, systems, and methods that enable environmental health monitoring and predictive assessment | |
| JP2019097829A (en) | Abnormality notification device and program | |
| Jin et al. | Improving diagnostic accuracy using multiparameter patient monitoring based on data fusion in the cloud | |
| US20240172951A1 (en) | Method and Device for detecting Information of Object In Room, Storage Medium and Processor | |
| JP7327397B2 (en) | Computer-implemented programs, information processing systems, and computer-implemented methods | |
| JP7342863B2 (en) | Computer-executed programs, information processing systems, and computer-executed methods | |
| US11457875B2 (en) | Event prediction system, sensor signal processing system, event prediction method, and non-transitory storage medium | |
| Krabbe et al. | Detection of activities of daily living with decision trees through a technical assistance system | |
| US20230282352A1 (en) | Heart condition detection sensor device and system for providing complex life support solution using same |
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 |