CN110659595A - Tumble type and injury part detection method based on feature classification - Google Patents
Tumble type and injury part detection method based on feature classification Download PDFInfo
- Publication number
- CN110659595A CN110659595A CN201910853809.1A CN201910853809A CN110659595A CN 110659595 A CN110659595 A CN 110659595A CN 201910853809 A CN201910853809 A CN 201910853809A CN 110659595 A CN110659595 A CN 110659595A
- Authority
- CN
- China
- Prior art keywords
- falling
- fall
- data
- characteristic
- detection
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
- G06V40/23—Recognition of whole body movements, e.g. for sport training
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The invention belongs to the technical field of electronic information detection, and discloses a fall type and injury part detection method based on feature classification, which is based on data of a user accelerometer and a gyroscope collected by a wearable sensor system; carrying out numerical value normalization processing on the acquired sensor data; acquiring time domain and frequency domain characteristics of the preprocessed data; performing characteristic screening by adopting principal component analysis; establishing a random forest-based fall detection model, and performing fall detection and fall type judgment; the falling injury part is judged and matched according to the falling type, and the falling type can be judged by adopting the method. When the detection is carried out on the falling type, the accuracy rate reaches 91%, and when the detection is carried out on different falling types, the accuracy rate reaches 89%. Through comparison, the detection rate of all types of falling is higher than the detection result of the current falling direction discrimination research, and the effectiveness of the random forest model provided by the invention is verified.
Description
Technical Field
The invention belongs to the technical field of electronic information detection, and particularly relates to a method for detecting falling types and injury parts based on feature classification.
Background
Currently, the closest prior art:
aging of the population is an important issue. According to a recent report from the european union, the proportion of the population over 65 years of age in europe is expected to exceed 20% by 2020 and 30% by 2060. People at high risk of falling are prone to fall and suffer serious injuries. Automatic fall detection can provide real-time alarm of falls and relevant information when falls occur, and is of great importance for quick response in medical assistance. The goal of this work was to propose a novel fall detection system that can capture and analyze motion data and detect falls.
The traditional falling detection is mainly realized by observing the field condition and recalling and judging the falling condition after people are concerned. The most important problem is that the conventional fall detection cannot timely detect the occurrence of a fall. The other type is that the sensor unit is used for acquiring patient motion data, time domain and frequency domain characteristics are extracted through sensor time sequence data, and falling is judged through methods such as threshold value and machine learning.
The main problems in current falls are: 1. because the falling is a random event, the occurrence time is not fixed, when the old is in a single place, the old cannot find the falling in time, and the old can be injured secondarily after falling for a long time. 2. The body of the old is seriously injured by falling, the falling occurrence condition can be judged only through follow-up inquiry, but the falling occurrence condition cannot be reflected in time because the occurrence time is too short when the person concerned falls.
Chinese patent 'CN 108447225A a human body falling detection method and device' provides a human body falling detection method and device, using a three-axis acceleration sensor and a three-axis angular velocity sensor, converting data collected by the acceleration sensor and the angular velocity sensor into a world coordinate system through quaternion for fusion, calculating characteristics of human body acceleration, velocity, displacement and rotation angle, and detecting falling occurrence through a set threshold. According to the technical scheme, when a person falls down, the person is judged to fall down by setting an acceleration threshold, a speed threshold, a displacement threshold, an angular speed threshold and a rotation angle threshold when the person is safe. The method has the advantages that due to the influence of individuals and health conditions, the physical quality is greatly different, and misjudgment is easily caused by a threshold value method; the human body falling and daily actions can be classified, but further research classification is lacked on the further classification of specific falling categories. The subdivision of the falling types can provide a basis for further research, and the falling injury parts, the injury degree and the like are analyzed according to the judgment of the falling types.
In summary, the problems of the prior art are as follows:
(1) the traditional falling detection cannot find the falling in time, and the old can be injured secondarily after falling for a long time.
(2) The specific fall categories cannot be classified, and the scene and the injury condition of the fall cannot be carefully analyzed.
The difficulty of solving the technical problems is as follows:
because the action difference between the falling and the daily action is large, the falling can generate a rapid falling process and a collision process, and the falling process and the collision process are easy to distinguish from the daily action, but the similarity between the falling of various types is high, so that various types of falling are difficult to distinguish by methods such as a threshold value method and the like.
The significance of solving the technical problems is as follows:
the falling scene can be further restored by distinguishing various types of falls, and a more complete basis is provided for fall detection. By judging the falling type, the falling injury part, the falling severity degree and the like can be further analyzed, the falling detection can be embodied, and support can be provided for injury processing after falling.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a method for detecting a falling type and an injury part based on feature classification.
The invention is realized in such a way, and provides a method for detecting falling types and injury parts based on feature classification. The fall type and injury part detection method based on feature classification comprises the following steps:
the method comprises the following steps of firstly, collecting user accelerometer and gyroscope data based on a wearable sensor system;
secondly, carrying out numerical value normalization processing on the acquired sensor data;
thirdly, acquiring time domain and frequency domain characteristics of the preprocessed data;
step four, performing characteristic screening by adopting principal component analysis;
establishing a random forest-based fall detection model, and performing fall detection and fall type judgment;
and step six, judging and matching the falling injury part according to the falling type.
Further, the user accelerometer and gyroscope data collected based on the wearable sensor system in the first step include:
(1) user data is collected using a wearable sensor that includes an Inertial Measurement Unit (IMU) that can measure three-axis acceleration and three-axis angular velocity in daily activities.
(2) The sensor is placed in the waist, particularly in the front middle position of the hip bone.
(3) Action category label: the actions are divided into lateral falling, forward falling, backward falling and daily activities (including six activities such as sitting, standing, lying, jogging, running, going up and down stairs, etc.).
Further, the performing of the numerical normalization processing on the acquired sensor data in the second step includes:
and normalizing the signal value, and mapping the data to a range of 0-1. By using a linear function normalization method, the conversion function is as follows:
where max is the maximum value of the sample data,min is the minimum value of sample data. R is raw sensor signal data, R*Is the result after normalization.
Further, the step three of obtaining time domain and frequency domain characteristics of the preprocessed data includes:
acquiring characteristic data: in order to realize fall detection based on sensor data, 2-second data with a fall data peak as a center is extracted from a preprocessed data signal through a sliding window, and 12 types of characteristic data such as mean, maximum, minimum, standard deviation, variance, median, signal energy, quarter median, three-quarter median, quartile, frequency domain signal skewness, frequency domain signal peak and the like of an accelerometer and a gyroscope are extracted. Because the sensor data includes acceleration and gyroscope data, a total of 24 feature data are obtained.
Further, the feature screening by using principal component analysis in the fourth step comprises:
in the fall specific type identification, the first n characteristics of the characteristics are screened by a principal component analysis method from a plurality of extracted characteristic vectors, wherein the n value selects the first 6 items according to the experimental result.
Finding a feature x(i)Is to find the feature vector matrix W corresponding to the first n eigenvalues of the covariance matrix of the feature set, and then for each feature x(i)Is transformed as follows(i)=WTx(i)The purpose of reducing dimension is achieved, and the process is as follows:
1) centralizing all features:m is the number of features;
2) calculating a covariance matrix of the sample;
3) carrying out eigenvalue decomposition on the covariance matrix;
4) extracting the eigenvector (w) corresponding to the largest n eigenvalues1,w2,...,wn) After all the eigenvectors are normalized, an eigenvector matrix W is formed.
5) For sample setEach feature x in(i)Conversion into a new feature z(i)=WTx(i);
6) Obtaining the characteristic set Z after characteristic screening (Z ═ Z)(1),z(2),…,z(m))。
The fall and daily movement training data with m-dimensional features can be obtained by a front n-dimensional sample set Z ═ Z (Z-dimensional)(1),z(2),…,z(n)) Represents; through experimental result decision, 6 characteristics are selected.
Further, a random forest-based fall detection model is established in the fifth step, and after the data feature extraction is completed, a decision tree and a random forest algorithm are used for setting fall discrimination rules to complete the construction of the fall detection model; and judging the falling types through the random forest, and using the decision tree as a classification tree in the random forest. The fall detection and fall type judgment comprises the following steps:
(1) construction of decision tree models
Firstly, a decision tree is constructed, a characteristic sample set Z is a characteristic set after characteristic screening, 6 characteristics are selected through an experimental result, and Z is expressed as x1,x2,x3,x4,x5,x6}; the fall detection categories have 5 total classes, and the fall detection categories are recorded as a class set D ═ D1,d2,d3,d4In which d is1For non-falling, and daily activities, d2To d4Respectively denoted as forward fall, backward fall, lateral fall; by s representing diNumber of middle samples, and piIndicates that the sample belongs to diThen the information entropy of the sample set can be defined as shown in the formula:
the cross entropy calculation formula is as follows:
wherein p isij=xij/|xjI is used to estimate xjWherein each type of sample belongs to diThe probability of (c).
Suppose feature x is selected1As a test feature, and x1A sample set X may be divided into v subsets { X1,X2,…,XvIs provided with xijIs subset XjClass DiThe number of samples of (1) is x1The information gain calculation of the partition subset obtained as the test feature is shown in the formula:
g(D,xi)=H(D)-H(D|xi)
decision tree C4.5 defines the splitting information as shown in the formula:
the information gain ratio is defined as shown in the formula:
deriving the adopted features xiAs the information gain rate of the test feature, then all the features are tested in sequence. The larger the gain entropy is, the characteristic is adopted as a test characteristic, so that the falling, the daily action and different falling categories can be better distinguished, and a decision tree is constructed;
by randomly selecting a part of fall and daily movement sample characteristics on a sample set as ssubThen at randomly selected ssubAnd selecting the features with the largest gain entropy as the optimal features to divide left and right subtrees of the decision tree from the features of the fall and the daily action samples.
(2) Construction of random forest-based fall detection model
Constructing a random forest based on the classification tree constructed by the decision tree in the last step; voting and judging through u random decision tree test results, and selecting 200 classification trees according to the classification precision and the time efficiency of experimental results to obtain daily actions and classification results of different falling categories;
the random forest algorithm process:
for T1, 2, T:
a) random sampling is carried out on the training set for the t time, v times are collected in total, and a sampling set D containing v samples is obtainedt;
b) Using a sample set DtTraining the tth decision tree model Gt(x) When the nodes of the decision tree model are trained, a part of sample features are selected from all sample features on the nodes, and an optimal feature is selected from the randomly selected part of sample features to divide left and right subtrees of the decision tree.
(3) Classifying according to fall detection models
The specific classification method for classifying daily actions and different falling categories based on the model obtained in the last step comprises the following steps: classifying the input actions through each decision tree to obtain an action classification result; and (4) totally obtaining 200 decision trees in the random forest, totally obtaining 200 action classification results, performing voting judgment, and selecting the result with the highest voting as the classification result of the current input action.
Further, the step six of judging and matching the falling injury part according to the falling category comprises:
judging the possible injured part of the human body according to the falling type, wherein the falling is mainly backward falling, and the hip and hip injury and the hindbrain injury correspond to the backward falling; the trip is mainly forward fall, corresponding forehead injury and knee joint injury; the main reasons for the fainting are lateral falls and the corresponding main reasons are shoulder injuries and elbow injuries.
In summary, the advantages and positive effects of the invention are:
the acceleration data and the gyroscope data are different in numerical value range, and the accuracy can be improved through normalization processing. The invention provides a method for fall detection and injury part judgment based on feature classification. The method comprises the steps of obtaining human body action data through an accelerometer and a gyroscope in a wearable sensor, and extracting time domain and frequency domain features to form a feature vector; the falling injury part is judged by detecting different falling categories. By detecting the falling, the purposes of timely finding the falling of the patient and avoiding secondary damage caused by long-term falling are achieved; meanwhile, the falling type is judged, the possibly injured part is presumed, medical workers are helped to judge the injured part and condition of the patient, the falling condition of the patient is known in time, and the patient is treated in time.
The falling detection method of the invention is adopted in a data set ContentLab[1]Experiments were performed as above and compared to KNN and SVM methods. ContentLab was from Ojetola et al[1]The paper contains 46 experimental data of people, including 644 fall (right fall, left fall, forward fall, backward fall) data and 1196 data of daily activities (standing, advancing, lying down, going up and down stairs, sitting on a bed, sitting on a chair, going back, falling nearly, lying down, etc.) by placing acceleration sensors on chest and thigh positions. The experimental results are shown in table 1, the detection rate of the trained random forest mixed model is tested, the highest accuracy rate reaches 91% when detecting falling or not falling, and the highest accuracy rate reaches 89% when detecting different falling types. Through comparison, the detection rate of various types of falling is higher than that of the detection result of the current falling type discrimination research, and the effectiveness of the model provided by the invention is verified.
Table 1 fall detection results
[1]Ojetola,O.;Gaura,E.;Brusey,J.Data Set for Fall Events and Daily Activities from Inertial Sensors.In Proceedings ofthe 6th ACM MultimediaSystems Conference(MMSys’15),Portland,OR,USA,18–20March 2015;pp.243–248.
Drawings
Fig. 1 is a flowchart of a fall type and injury detection method based on feature classification according to an embodiment of the present invention.
Fig. 2 is a flow chart of fall detection provided by the embodiment of the invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
In order to solve the problems in the background art, the invention provides a method for detecting a falling type and an injury part based on feature classification. The human body activity signal is objectively measured through wearable sensor equipment, time domain and frequency domain features are extracted through data normalization processing, feature screening is carried out through principal component analysis, and finally falling classification detection is carried out through random forests.
The following detailed description of the principles of the invention is provided in connection with the accompanying drawings.
As shown in fig. 1-2, a method for detecting a fall type and a damaged part based on feature classification according to an embodiment of the present invention includes:
s101: based on user accelerometer and gyroscope data collected by the wearable sensor system;
s102: carrying out numerical value normalization processing on the acquired sensor data;
s103: acquiring time domain and frequency domain characteristics of the preprocessed data;
s104: performing characteristic screening by adopting principal component analysis;
s105: establishing a random forest-based fall detection model, and performing fall detection and fall type judgment;
s106: and judging and matching the falling injury part according to the falling category.
Further, the user accelerometer and gyroscope data collected based on the wearable sensor system in the first step include:
(1) user data is collected using a wearable sensor that includes an Inertial Measurement Unit (IMU) that can measure three-axis acceleration and three-axis angular velocity in daily activities.
(2) The sensor is placed in the waist, particularly in the front middle position of the hip bone.
(3) Action category label: the actions are divided into lateral falling, forward falling, backward falling and daily activities (including six activities such as sitting, standing, lying, jogging, running, going up and down stairs, etc.).
Further, the performing of the numerical normalization processing on the acquired sensor data in the second step includes:
and normalizing the signal value, and mapping the data to a range of 0-1. By using a linear function normalization method, the conversion function is as follows:
where max is the maximum value of the sample data and min is the minimum value of the sample data. R is raw sensor signal data, R*Is the result after normalization.
Further, the step three of obtaining time domain and frequency domain characteristics of the preprocessed data includes:
acquiring characteristic data: in order to realize fall detection based on sensor data, 2-second data with a fall data peak as a center is extracted from a preprocessed data signal through a sliding window, and 12 types of characteristic data such as mean, maximum, minimum, standard deviation, variance, median, signal energy, quarter median, three-quarter median, quartile, frequency domain signal skewness, frequency domain signal peak and the like of an accelerometer and a gyroscope are extracted. Because the sensor data includes acceleration and gyroscope data, a total of 24 feature data are obtained.
Further, the feature screening by using principal component analysis in the fourth step comprises:
in the fall specific type identification, the first n characteristics of the characteristics are screened by a principal component analysis method from a plurality of extracted characteristic vectors, wherein the n value selects the first 6 items according to the experimental result.
Sample x is solved(i)The principal component of (1) is the eigenvector matrix W corresponding to the first n eigenvalues of the covariance matrix of the sample setThen for each sample x(i)Is transformed as follows(i)=WTx(i)The purpose of reducing dimension is achieved, and the process is as follows:
2) calculating a covariance matrix of the sample;
3) carrying out eigenvalue decomposition on the covariance matrix;
4) extracting the eigenvector (w) corresponding to the largest n eigenvalues1,w2,...,wn) After all the eigenvectors are normalized, an eigenvector matrix W is formed.
5) For each feature x in the sample set(i)Conversion into a new feature z(i)=WTx(i);
6) Obtaining the screened characteristic set D ═ (z)(1),z(2),…,z(m))。
The fall and daily movement training data with m-dimensional features can be obtained by a front n-dimensional sample set Z ═ Z (Z-dimensional)(1),z(2),…,z(n)) Represents; through experimental result decision, 6 characteristics are selected.
Further, a random forest-based fall detection model is established in the fifth step, and after the data feature extraction is completed, a decision tree and a random forest algorithm are used for setting fall discrimination rules to complete the construction of the fall detection model; and judging the falling types through the random forest, and using the decision tree as a classification tree in the random forest. The fall detection and fall type judgment comprises the following steps:
(1) construction of decision tree models
Firstly, a decision tree is constructed, a characteristic sample set Z is a characteristic set after characteristic screening, 6 characteristics are selected through an experimental result, and Z is expressed as x1,x2,x3,x4,x5,x6}; the fall detection categories have 5 total classes, and the fall detection categories are recorded as a class set D ═ D1,d2,d3,d4In which d is1For non-falling, and daily activities, d2To d4Respectively denoted as forward fall, backward fall, lateral fall; by s representing diNumber of middle samples, and piIndicates that the sample belongs to diThen the information entropy of the sample set can be defined as shown in the formula:
the cross entropy calculation formula is as follows:
wherein p isij=xij/|xjI is used to estimate xjWherein each type of sample belongs to diThe probability of (c).
Suppose feature x is selected1As a test feature, and x1The sample set X may be divided into v subsets { X1,X2,…,XvIs provided with xijIs subset XjClass DiThe number of samples of (1) is x1The information gain calculation of the partition subset obtained as the test feature is shown in the formula:
g(D,xi)=H(D)-h(D|xi)
decision tree C4.5 defines the splitting information as shown in the formula:
the information gain ratio is defined as shown in the formula:
deriving the adopted features xiAs the information gain rate of the test feature, then all the features are tested in sequence. The larger the gain entropy, the more this characteristic is adoptedThe characteristics are used as test characteristics, so that the falling, daily actions and different falling categories can be better distinguished, and a decision tree is constructed.
By randomly selecting a part of fall and daily movement sample characteristics on a sample set as ssubThen at randomly selected ssubAnd selecting the features with the largest gain entropy as the optimal features to divide left and right subtrees of the decision tree from the features of the fall and the daily action samples.
(2) Construction of random forest-based fall detection model
Constructing a random forest based on the classification tree constructed by the decision tree in the last step; and (4) voting and judging through u random decision tree test results, and selecting 200 classification trees according to the classification precision and the time efficiency of the experimental results to obtain daily actions and classification results of different falling categories.
The random forest algorithm process:
for T1, 2, T:
a) random sampling is carried out on the training set for the t time, v times are collected in total, and a sampling set D containing v samples is obtainedt;
b) Using a sample set DtTraining the tth decision tree model Gt(x) When the nodes of the decision tree model are trained, a part of sample features are selected from all sample features on the nodes, and an optimal feature is selected from the randomly selected part of sample features to divide left and right subtrees of the decision tree.
(3) Classifying according to fall detection models
The specific classification method for classifying daily actions and different falling categories based on the model obtained in the last step comprises the following steps: classifying the input actions through each decision tree to obtain an action classification result; and (4) totally obtaining 200 decision trees in the random forest, totally obtaining 200 action classification results, performing voting judgment, and selecting the result with the highest voting as the classification result of the current input action.
Further, the step six of judging and matching the falling injury part according to the falling category comprises:
judging the possible injured part of the human body according to the falling type, wherein the falling is mainly backward falling, and the hip and hip injury and the hindbrain injury correspond to the backward falling; the trip is mainly forward fall, corresponding forehead injury and knee joint injury; the main reasons for the fainting are lateral falls and the corresponding main reasons are shoulder injuries and elbow injuries. The injury site is judged according to the table correspondence, as shown in table 2.
TABLE 2 Fall directions and potential injury sites
According to the old people falling identification method, the human body action data are extracted, the characteristic values of the original data are extracted and combined to form the characteristic vectors, the characteristic vectors are screened, the characteristic extraction is carried out through principal component analysis, and finally the random forest is used for classification, so that falling and injury type identification is realized, and very effective decision reference information is provided for further rescue of medical staff under emergency.
According to the invention, the wearable sensor is used for fall detection and fall type judgment, so that the fall occurrence can be detected in real time, and a judgment of a fall possibly injured part is given according to different types of falls; the data preprocessing considers the influence of different sensor data magnitudes on the accuracy of falling judgment; extracting time domain and frequency domain characteristics from the characteristics; the method is different from common fall detection in that the decision tree and the random forest are used for fall detection and fall type detection, and the method not only detects fall and simultaneously detects fall types, but also is helpful for further helping medical workers to know fall situations and judge the possible injured parts of the fallen workers.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.
Claims (8)
1. A method for detecting a fall type and a damage part based on feature classification is characterized by comprising the following steps:
the method comprises the following steps of firstly, collecting user accelerometer and gyroscope data based on a wearable sensor system;
secondly, carrying out numerical value normalization processing on the acquired sensor data;
thirdly, acquiring time domain and frequency domain characteristics of the preprocessed data;
step four, performing characteristic screening by adopting principal component analysis;
establishing a random forest-based fall detection model, and performing fall detection and fall type judgment;
and step six, judging and matching the falling injury part according to the falling type.
2. A method as claimed in claim 1, wherein the first step of collecting data based on user accelerometer and gyroscope of the wearable sensor system comprises:
(1) a wearable sensor comprising an inertial measurement unit IMU is used for collecting user data, and the IMU can measure three-axis acceleration and three-axis angular velocity in daily actions;
(2) the sensor is placed at the waist, in particular to the middle position of the front side of the hip body;
(3) action category label: the actions are divided into lateral falling, forward falling, backward falling and daily activities.
3. A method for detecting a fall type and a damage part based on feature classification as claimed in claim 1, wherein the second step of performing numerical normalization processing on the collected sensor data comprises:
normalizing the signal value, and mapping the data to 0-1; by using a linear function normalization method, the conversion function is as follows:
where max is the maximum value of the sample dataMin is the minimum value of sample data; r is raw sensor signal data, R*Is the result after normalization.
4. A method as claimed in claim 1, wherein the step three of obtaining time domain and frequency domain features from the preprocessed data comprises:
acquiring characteristic data: in order to realize fall detection based on sensor data, 2-second data taking a fall data peak value as a center is extracted from a preprocessed data signal through a sliding window, and 12 types of characteristic data of an accelerometer and a gyroscope, such as mean, maximum, minimum, standard deviation, variance, median, signal energy, quarter median, three-quarter median, quartile, frequency domain signal skewness and frequency domain signal peak value are extracted; the sensor data includes acceleration and gyroscope data, and 24 items of characteristic data are obtained.
5. A method for detecting a fall type and a damage part based on feature classification as claimed in claim 1, wherein the feature screening using principal component analysis in the fourth step comprises:
in the fall specific type identification, the front n characteristics of the characteristics are screened by a principal component analysis method from a plurality of extracted characteristic vectors, wherein the value of n is the front 6 items selected according to an experimental result;
finding a feature x(i)Is to find the first n eigenvalues of the covariance matrix of the sample set corresponding to the eigenvector matrix W, and then for each eigenvalue x(i)Is transformed as follows(i)=WTx(i)The purpose of reducing dimension is achieved, and the process is as follows:
1) centralizing all features:wherein m is a characteristic number;
2) calculating a covariance matrix of the sample;
3) carrying out eigenvalue decomposition on the covariance matrix;
4) extracting the eigenvector (w) corresponding to the largest n eigenvalues1,w2,...,wn) After all the eigenvectors are standardized, forming an eigenvector matrix W;
5) for each feature x in the sample set(i)Conversion into a new feature z(i)=WTx(i);
6) Obtaining the characteristic set z (z) after characteristic screening(1),z(2),...,z(m));
The fall and daily movement training data with m-dimensional features can be obtained by a front n-dimensional sample set of z ═ z (z-dimensional)(1),z(2),...,z(n)) Represents; through experimental result decision, 6 characteristics are selected.
6. A fall type and injury part detection method based on feature classification as claimed in claim 1, wherein the fall detection model based on random forest is established in the fifth step, and after the extraction of data features is completed, a decision tree and a random forest algorithm are used to set fall discrimination rules, so as to complete the establishment of the fall detection model; the method comprises the following steps of judging falling types through random forests, using decision trees as classification trees in the random forests, and carrying out falling detection and falling type judgment, wherein the falling type judgment comprises the following steps:
(1) construction of decision tree models
Firstly, a decision tree is constructed, a characteristic sample set Z is a characteristic set after characteristic screening, 6 characteristics are selected through an experimental result, and Z is expressed as x1,x2,x3,x4,x5,x6}; the fall detection categories have 5 total classes, and the fall detection categories are recorded as a class set D ═ D1,d2,d3,d4In which d is1For non-falling, and daily activities, d2To d4Respectively denoted as forward fall, backward fall, lateral fall; by s representing diNumber of middle samples, and piIndicates that the sample belongs to diThen the information entropy of the sample set may beTo define as shown in the formula:
the cross entropy calculation formula is as follows:
wherein p isij=xij/|xjI is used to estimate xjWherein each type of sample belongs to diThe probability of (d);
selecting feature x1As a test feature, and x1A sample set X may be divided into v subsets { X1,X2,...,XvIs provided with xijIs subset XjClass DiThe number of samples of (1) is x1The information gain calculation of the partition subset obtained as the test feature is shown in the formula:
g(D,xi)=H(D)-H(D|xi)
decision tree C4.5 defines the splitting information as shown in the formula:
the information gain ratio is calculated as follows:
deriving the adopted features xiThe information gain rate is used as a test characteristic, and then all the characteristics are tested in sequence; the larger the gain entropy is, the characteristic is adopted as a test characteristic, so that the falling, the daily action and different falling categories can be better distinguished, and a decision tree is constructed;
by randomly selecting a part of fall and daily movement sample characteristics on a sample set as ssubThen at randomly selected ssubFall and daySelecting the characteristic with the maximum gain entropy as the optimal characteristic to divide left and right subtrees of the decision tree in the constant-motion sample characteristics;
(2) construction of random forest-based fall detection model
Constructing a random forest based on the classification tree constructed by the decision tree in the last step; voting and judging through u random decision tree test results, and selecting 200 classification trees according to the classification precision and the time efficiency of experimental results to obtain daily actions and classification results of different falling categories;
the random forest algorithm process:
for T1, 2, T:
a) random sampling is carried out on the training set for the t time, v times are collected in total, and a sampling set D containing v samples is obtainedt;
b) Using a sample set DtTraining the tth decision tree model Gt(x) When the nodes of the decision tree model are trained, selecting a part of sample characteristics from all sample characteristics on the nodes, and selecting an optimal characteristic from the randomly selected part of sample characteristics to divide left and right subtrees of the decision tree;
(3) classifying according to fall detection models
The specific classification method for classifying daily actions and different falling categories based on the model obtained in the last step comprises the following steps: classifying the input actions through each decision tree to obtain an action classification result; and (4) totally obtaining 200 decision trees in the random forest, totally obtaining 200 action classification results, performing voting judgment, and selecting the result with the highest voting as the classification result of the current input action.
7. A method as claimed in claim 1, wherein the sixth step of determining the matching of the fall injury part according to the fall category comprises:
judging the possible injured part of the human body according to the falling type, wherein the falling is mainly backward falling, and the hip and hip injury and the hindbrain injury correspond to the backward falling; the trip is mainly forward fall, corresponding forehead injury and knee joint injury; the main reasons for the fainting are lateral falls and the corresponding main reasons are shoulder injuries and elbow injuries.
8. An information data processing terminal applying the fall type and injury part detection method based on feature classification as claimed in any one of claims 1 to 7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910853809.1A CN110659595A (en) | 2019-09-10 | 2019-09-10 | Tumble type and injury part detection method based on feature classification |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910853809.1A CN110659595A (en) | 2019-09-10 | 2019-09-10 | Tumble type and injury part detection method based on feature classification |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110659595A true CN110659595A (en) | 2020-01-07 |
Family
ID=69036906
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910853809.1A Pending CN110659595A (en) | 2019-09-10 | 2019-09-10 | Tumble type and injury part detection method based on feature classification |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110659595A (en) |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111401507A (en) * | 2020-03-12 | 2020-07-10 | 大同公元三九八智慧养老服务有限公司 | Adaptive decision tree fall detection method and system |
CN111657918A (en) * | 2020-06-12 | 2020-09-15 | 电子科技大学 | Falling detection method and system combining electrocardio and inertial sensing data |
CN111681389A (en) * | 2020-06-12 | 2020-09-18 | 电子科技大学 | Old people falling behavior detection method based on blind source separation |
CN111887859A (en) * | 2020-08-05 | 2020-11-06 | 安徽华米智能科技有限公司 | Fall behavior recognition method and device, electronic device and medium |
CN112380946A (en) * | 2020-11-09 | 2021-02-19 | 上海泗科智能科技有限公司 | Fall detection method and device based on end-side AI chip |
CN112405568A (en) * | 2020-10-20 | 2021-02-26 | 同济大学 | Humanoid robot falling prediction method |
CN112699744A (en) * | 2020-12-16 | 2021-04-23 | 南开大学 | Fall posture classification identification method and device and wearable device |
CN112998697A (en) * | 2021-02-22 | 2021-06-22 | 电子科技大学 | Tumble injury degree prediction method and system based on skeleton data and terminal |
CN113095379A (en) * | 2021-03-26 | 2021-07-09 | 厦门中翎易优创科技有限公司 | Human motion state identification method based on wearable six-axis sensing data |
CN113180596A (en) * | 2021-04-07 | 2021-07-30 | 中山大学 | Non-contact sleep analysis method and device and storage medium |
CN113358111A (en) * | 2021-05-28 | 2021-09-07 | 北京金山云网络技术有限公司 | Fall detection method and device and electronic equipment |
CN114720878A (en) * | 2022-03-24 | 2022-07-08 | 长安大学 | Method for detecting state of retired battery |
CN114937341A (en) * | 2022-04-07 | 2022-08-23 | 同济大学 | Wheelchair falling risk monitoring method and system based on multi-dimensional force |
US20240021317A1 (en) * | 2022-07-14 | 2024-01-18 | Brian Matthew Hood | Systems and Methods for Health Analysis |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106355142A (en) * | 2016-08-24 | 2017-01-25 | 深圳先进技术研究院 | A Method and Device for Recognizing Human Falling State |
CN106709471A (en) * | 2017-01-05 | 2017-05-24 | 宇龙计算机通信科技(深圳)有限公司 | Fall detection method and device |
CN106875630A (en) * | 2017-03-13 | 2017-06-20 | 中国科学院计算技术研究所 | A kind of wearable fall detection method and system based on hierarchical classification |
CN107169512A (en) * | 2017-05-03 | 2017-09-15 | 苏州大学 | The construction method of HMM SVM tumble models and the fall detection method based on the model |
CN107273920A (en) * | 2017-05-27 | 2017-10-20 | 西安交通大学 | A kind of non-intrusion type household electrical appliance recognition methods based on random forest |
CN108549900A (en) * | 2018-03-07 | 2018-09-18 | 浙江大学 | Tumble detection method for human body based on mobile device wearing position |
-
2019
- 2019-09-10 CN CN201910853809.1A patent/CN110659595A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106355142A (en) * | 2016-08-24 | 2017-01-25 | 深圳先进技术研究院 | A Method and Device for Recognizing Human Falling State |
CN106709471A (en) * | 2017-01-05 | 2017-05-24 | 宇龙计算机通信科技(深圳)有限公司 | Fall detection method and device |
CN106875630A (en) * | 2017-03-13 | 2017-06-20 | 中国科学院计算技术研究所 | A kind of wearable fall detection method and system based on hierarchical classification |
CN107169512A (en) * | 2017-05-03 | 2017-09-15 | 苏州大学 | The construction method of HMM SVM tumble models and the fall detection method based on the model |
CN107273920A (en) * | 2017-05-27 | 2017-10-20 | 西安交通大学 | A kind of non-intrusion type household electrical appliance recognition methods based on random forest |
CN108549900A (en) * | 2018-03-07 | 2018-09-18 | 浙江大学 | Tumble detection method for human body based on mobile device wearing position |
Non-Patent Citations (5)
Title |
---|
SERKAN BALLI 等: "Human activity recognition from smart watch sensor data using a hybrid of principal component analysis and random forest algorithm", 《MEASUREMENT AND CONTROL》 * |
SHANGMING YANG 等: "A Class of Manifold Regularized Multiplicative Update Algorithms for Image Clustering", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 * |
丁世军 中国矿业大学出版社有限责任公司: "《高级人工智能》", 31 January 2015 * |
刘玉琪: "基于随机森林算法的人体运动模式识别研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
忽丽莎 等: "基于可穿戴设备的跌倒检测算法综述", 《浙江大学学报(工学版)》 * |
Cited By (19)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111401507A (en) * | 2020-03-12 | 2020-07-10 | 大同公元三九八智慧养老服务有限公司 | Adaptive decision tree fall detection method and system |
CN111401507B (en) * | 2020-03-12 | 2021-01-26 | 大同公元三九八智慧养老服务有限公司 | Adaptive decision tree fall detection method and system |
CN111657918A (en) * | 2020-06-12 | 2020-09-15 | 电子科技大学 | Falling detection method and system combining electrocardio and inertial sensing data |
CN111681389A (en) * | 2020-06-12 | 2020-09-18 | 电子科技大学 | Old people falling behavior detection method based on blind source separation |
CN111681389B (en) * | 2020-06-12 | 2021-02-26 | 电子科技大学 | Old people falling behavior detection method based on blind source separation |
CN111887859A (en) * | 2020-08-05 | 2020-11-06 | 安徽华米智能科技有限公司 | Fall behavior recognition method and device, electronic device and medium |
CN112405568A (en) * | 2020-10-20 | 2021-02-26 | 同济大学 | Humanoid robot falling prediction method |
CN112380946A (en) * | 2020-11-09 | 2021-02-19 | 上海泗科智能科技有限公司 | Fall detection method and device based on end-side AI chip |
CN112699744A (en) * | 2020-12-16 | 2021-04-23 | 南开大学 | Fall posture classification identification method and device and wearable device |
CN112998697A (en) * | 2021-02-22 | 2021-06-22 | 电子科技大学 | Tumble injury degree prediction method and system based on skeleton data and terminal |
CN112998697B (en) * | 2021-02-22 | 2022-06-14 | 电子科技大学 | Tumble injury degree prediction method and system based on skeleton data and terminal |
CN113095379A (en) * | 2021-03-26 | 2021-07-09 | 厦门中翎易优创科技有限公司 | Human motion state identification method based on wearable six-axis sensing data |
CN113180596A (en) * | 2021-04-07 | 2021-07-30 | 中山大学 | Non-contact sleep analysis method and device and storage medium |
CN113180596B (en) * | 2021-04-07 | 2024-02-06 | 中山大学 | Non-contact sleep analysis method, device and storage medium |
CN113358111A (en) * | 2021-05-28 | 2021-09-07 | 北京金山云网络技术有限公司 | Fall detection method and device and electronic equipment |
CN114720878A (en) * | 2022-03-24 | 2022-07-08 | 长安大学 | Method for detecting state of retired battery |
CN114937341A (en) * | 2022-04-07 | 2022-08-23 | 同济大学 | Wheelchair falling risk monitoring method and system based on multi-dimensional force |
CN114937341B (en) * | 2022-04-07 | 2023-09-26 | 同济大学 | Wheelchair fall risk monitoring method and system based on multidimensional force |
US20240021317A1 (en) * | 2022-07-14 | 2024-01-18 | Brian Matthew Hood | Systems and Methods for Health Analysis |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110659595A (en) | Tumble type and injury part detection method based on feature classification | |
CN109166275B (en) | Human body falling detection method based on acceleration sensor | |
Khan et al. | Accelerometer signal-based human activity recognition using augmented autoregressive model coefficients and artificial neural nets | |
Jatoba et al. | Context-aware mobile health monitoring: Evaluation of different pattern recognition methods for classification of physical activity | |
Shan et al. | A wearable pre-impact fall detector using feature selection and support vector machine | |
CN104127181B (en) | Electromyographic signal tumble detection method based on WKFDA | |
CN108509897A (en) | A kind of human posture recognition method and system | |
CN105139029B (en) | A kind of Activity recognition method and device of prison prisoner | |
KR102134154B1 (en) | Pattern Recognition System and Mehod of Ultra-Wideband Respiration Data Based on 1-Dimension Convolutional Neural Network | |
CN104473648A (en) | Physiological parameter monitoring-combined human body tumble warning and detecting method | |
Su et al. | A multi-sensor based pre-impact fall detection system with a hierarchical classifier | |
CN116269355B (en) | Safety monitoring system based on figure gesture recognition | |
Ashwini et al. | Skeletal data based activity recognition system | |
Bayat et al. | Classifying human walking patterns using accelerometer data from smartphone | |
Nouredanesh et al. | Detection of compensatory balance responses using wearable electromyography sensors for fall-risk assessment | |
Luštrek et al. | Behavior analysis based on coordinates of body tags | |
CN110547806A (en) | gesture action online recognition method and system based on surface electromyographic signals | |
Hai et al. | PCA-SVM algorithm for classification of skeletal data-based eigen postures | |
CN112863515B (en) | Method for recognizing human tongue and mandible actions by using mobile phone loudspeaker and microphone | |
Khokhlova et al. | Kinematic covariance based abnormal gait detection | |
Kołodziej et al. | Fall Detection Using a Smartphone | |
Burghardt | A brief review of biometric identification | |
Saponaro et al. | Estimating Physical Activity Intensity And Energy Expenditure Using Computer Vision On Videos | |
CN109692004B (en) | Method and device for detecting standing posture of human body | |
Kukharenko et al. | Picking a human fall detection algorithm for wrist-worn electronic device |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200107 |