CN110222708A - A kind of fall detection method and system based on Integrated Decision tree - Google Patents
A kind of fall detection method and system based on Integrated Decision tree Download PDFInfo
- Publication number
- CN110222708A CN110222708A CN201910354991.6A CN201910354991A CN110222708A CN 110222708 A CN110222708 A CN 110222708A CN 201910354991 A CN201910354991 A CN 201910354991A CN 110222708 A CN110222708 A CN 110222708A
- Authority
- CN
- China
- Prior art keywords
- decision tree
- tree
- fall detection
- integrated
- error
- 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
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/103—Detecting, measuring or recording devices for testing the shape, pattern, colour, size or movement of the body or parts thereof, for diagnostic purposes
- A61B5/11—Measuring movement of the entire body or parts thereof, e.g. head or hand tremor, mobility of a limb
- A61B5/1116—Determining posture transitions
- A61B5/1117—Fall detection
-
- 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
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- 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
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Pathology (AREA)
- Medical Informatics (AREA)
- Public Health (AREA)
- Veterinary Medicine (AREA)
- Animal Behavior & Ethology (AREA)
- Surgery (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Heart & Thoracic Surgery (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Dentistry (AREA)
- Physiology (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention proposes a kind of fall detection method and system based on Integrated Decision tree, it include: for labeled as the first acceleration information fallen, take the wherein maximum point of quadratic sum, take data segment a little or so as original set according to preset time segment limit, for being labeled as the second acceleration information of non-tumble, take the data segment of preset time segment limit that original set is added every time by sliding window, finally collected, finally concentrates the multidimensional characteristic of each data segment as training characteristics by Fast Fourier Transform (FFT) extraction;More decision trees are created, using training characteristics as input feature vector, decision tree is inputted respectively with each decision tree of repetitive exercise, the residual error for being fitted a upper decision tree is gone by each decision tree, gather the decision tree of training completion as integrated study model;The third acceleration information to fall detection is obtained, the data segment for extracting third acceleration information using the method for sliding window will test data and be input to integrated study model, obtain fall detection result as detection data.
Description
Technical field
The present invention relates to the present invention relates to the fields such as general fit calculation, health supervision, and in particular to based on Integrated Decision tree
Fall detection method and model.
Background technique
Recently as the growth at old man's age, muscular strength ability, balanced capacity and cognitive ability sharp fall, therefore hold
Easily lead to the generation of tumble.It falls and seriously endangers the physical and mental health of old man, result even in disabled and dead generation.Tumble band
The physiology and psychological problems come, can seriously affect its daily behavior, keep its life limited, quality of life substantially reduces.Therefore right
It falls and carries out the attention that effective detection increasingly obtains people.
The prior art mainly utilizes acceleration information for the identification of fall detection, then uses SVM (Support
Vector Machine), KNN (K Nearest Neighbor), the machine learning methods such as convolutional neural networks, threshold method, with
And the method that the two combines is detected.The method of machine learning is using fall detection as a classification problem, the label of classification
To fall and non-tumble, using collected data as training data, training data training pattern is then utilized, then will
Trained model be applied to new data detected, such as patent CN201710024316 to training data extract feature with
Dimensionality reduction and normalization are carried out afterwards, then use SVM algorithm training pattern, then prediction result is exported by fall detection module;Patent
Whether CN201711397818 carries out Kalman filtering to data and is fallen by KNN algorithm analysis detection;Patent
CN201710022067 is trained by carrying out convolutional neural networks after converting to data, obtains the network model of behavior, then
Fall detection is carried out to new data using the model;Pattern match is carried out to convolutional neural networks model, whether judges user
It falls.The threshold value that threshold method mainly passes through feature is compared with the threshold value manually set, if threshold value is more than manually to set
Surely the threshold value given, then it is determined that falling.Patent CN201610601474 carries out fall detection using acceleration rate threshold.
There are also some patents to use multiple sensors data, these patents needs are obtained specific using specific sensor
Data.Such as patent CN201710329754 acquires data using acceleration transducer, heart rate sensor and temperature sensor,
Then tumble behavior is analyzed.Patent CN201610601474 judges whether user falls by atmospheric pressure value and air pressure change rate.Specially
Sharp CN201710755632 carries out fall detection using three-axis sensor and EGC sensor data.Patent CN201710063265
Fall detection is carried out using acceleration and gyro data.
That there are still detection accuracy in practical applications is low for fall detection technology based on wearable device, false alarm rate is high etc.
Challenge.Most of traditional fall detection technology is all based on multiple sensors (accelerometer, gyroscope etc.), falls in analysis
Behavior signal characteristic after, the method using multiple threshold values or machine learning is differentiated.But there are following for these methods
Disadvantage: the sensor that 1. methods having use is more, and specific sensor device is needed to be acquired data.2. using single
The method detection accuracy of the carry out fall detection of one sensing data (such as acceleration) is low, and false alarm rate is high.3. due to falling
The scarcity and daily behavior diversity of behavior cause its data extremely unbalanced, and the data of this lack of balance are used to train
Identification model will cause the offset of model, lead to that false alarm is excessively high, discrimination decline.Therefore need one kind can be with Shandong
Stick is higher, the model that can be learnt in a small amount of tumble sample, and obtains higher detection accuracy and lower wrong report
Rate.
Summary of the invention
The present invention provides a kind of fall detection method based on Integrated Decision tree and system to establish include integrated study model,
Using the robustness of integrated study model, model is enabled to obtain higher detection accuracy and lower rate of false alarm
In view of the deficiencies of the prior art, the present invention proposes a kind of fall detection method based on Integrated Decision tree, wherein wrapping
It includes:
Step 1, the first acceleration information for being labeled as falling, take quadratic sum in first acceleration information maximum
Point takes the data segment of the point or so as original set, for accelerating labeled as the second of non-tumble according to preset time segment limit
Degree evidence takes the data segment of the preset time segment limit that the original set is added every time, is finally collected, by quick by sliding window
Fourier transformation extracts this and finally concentrates the multidimensional characteristic of each data segment as training characteristics;
More step 2, creation decision trees input the decision tree respectively using the training characteristics as input feature vector with iteration
Each decision tree of training, the residual error for being fitted a upper decision tree is gone by each decision tree, is gathered the decision tree that training is completed and is made
For integrated study model;
The third acceleration information of step 3, acquisition to fall detection extracts the third using the method for sliding window and accelerates degree
According to data segment as detection data, which is input to the integrated study model, obtains fall detection result.
The fall detection method based on Integrated Decision tree, wherein the decision tree is regression tree, using gini index
Select optimal characteristics.
The fall detection method based on Integrated Decision tree, wherein in the step 2 each decision tree of repetitive exercise time
Number is the number of the more decision trees.
The fall detection method based on Integrated Decision tree, wherein step 2 include:
Decision tree is specially kth -1 tree on this, the residual error of kth wheel be the kth -1 tree the sum of predicted value with
Error between true value, the fit procedure are specially so that the error between the predicted value and residual error of kth tree is minimum.
The fall detection method based on Integrated Decision tree, wherein obtaining the error, guidance collection using objective function
It is updated at learning model to the smallest direction of error, objective function Obj (Θ) specifically:
Wherein K represents the number of total decision tree, fkThe several marking of kth is represented, i represents i-th of sample, and n is instruction
Practice the number of samples of collection, yiIndicate the true tag of sample,Indicate the prediction label of the sample, l indicates yi, andMistake
Difference.
The invention also provides a kind of fall detection system based on Integrated Decision tree, including:
Module 1, the first acceleration information for being labeled as falling, take quadratic sum in first acceleration information maximum
Point takes the data segment of the point or so as original set, for accelerating labeled as the second of non-tumble according to preset time segment limit
Degree evidence takes the data segment of the preset time segment limit that the original set is added every time, is finally collected, by quick by sliding window
Fourier transformation extracts this and finally concentrates the multidimensional characteristic of each data segment as training characteristics;
More module 2, creation decision trees input the decision tree respectively using the training characteristics as input feature vector with iteration
Each decision tree of training, the residual error for being fitted a upper decision tree is gone by each decision tree, is gathered the decision tree that training is completed and is made
For integrated study model;
The third acceleration information of module 3, acquisition to fall detection extracts the third using the system of sliding window and accelerates degree
According to data segment as detection data, which is input to the integrated study model, obtains fall detection result.
The fall detection system based on Integrated Decision tree, wherein the decision tree is regression tree, using gini index
Select optimal characteristics.
The fall detection system based on Integrated Decision tree, wherein in the module 2 each decision tree of repetitive exercise time
Number is the number of the more decision trees.
The fall detection system based on Integrated Decision tree, wherein module 2 include:
Decision tree is specially kth -1 tree on this, the residual error of kth wheel be the kth -1 tree the sum of predicted value with
Error between true value, the fit procedure are specially so that the error between the predicted value and residual error of kth tree is minimum.
The fall detection system based on Integrated Decision tree, wherein obtaining the error, guidance collection using objective function
It is updated at learning model to the smallest direction of error, objective function Obj (Θ) specifically:
Wherein K represents the number of total decision tree, fkThe several marking of kth is represented, i represents i-th of sample, and n is instruction
Practice the number of samples of collection, yiIndicate the true tag of sample,Indicate the prediction label of the sample, l indicates yi, andMistake
Difference.
As it can be seen from the above scheme the present invention has the advantages that
The present invention proposes the new algorithm of one kind for user's tumble behavioral value and carries out to 3-axis acceleration data
Pretreated method.Compare traditional fall detection method recall rate, rate of false alarm and in terms of have it is biggish
It is promoted.
Detailed description of the invention
Fig. 1 is algorithm flow chart of the invention;
Fig. 2 is CART tree graph of the invention:
Fig. 3 is algorithm training flow chart of the invention:
Fig. 4 is the comparative result figure of algorithm experimental result of the invention on MobiAct:
Fig. 5 is the comparative result figure of algorithm experimental result of the invention on MMSys:
Fig. 6 is the comparative result figure of algorithm experimental result of the invention on SisFall.
Specific embodiment
To allow features described above and effect of the invention that can illustrate more clearly understandable, special embodiment below, and cooperate
Bright book attached drawing is described in detail below.
Algorithm flow chart of the invention as shown in Figure 1, include 3 parts altogether: data preprocessing module, training module,
Prediction module.Each functions of modules is as follows:
Data preprocessing module: the module mainly pre-processes the data of 3-axis acceleration, mainly uses sliding window
The method of mouth extracts time series data in the feature of time domain and frequency domain.
Training module: the module is mainly used for the training to model, is the core of entire algorithm.Entire model is based on
Integrated study model can not be predicted effectively since a decision tree is too simple, therefore more decision trees are integrated into
Together, the residual error for being fitted upper one tree is gone using every one tree, what each decision tree was fitted is all that all decision trees in front are pre-
Error between the sum of measured value and true value, thus their fit object be it is different, the pre- of model can be greatly improved
Survey ability, last prediction result are the sum of the scores of every decision tree.Using the thought of integrated study, more decision trees are collected
The detection accuracy of model can be greatly improved after being combined together.
Prediction module: this module, which is mainly responsible for, to be treated prediction data and is detected.When there is new data to reach server,
It is sent into the module after data preprocessing module to be detected, which mainly utilizes trained in training module
Model is detected, and exports prediction result.
Compared with the prior art, method proposed by the present invention is a kind of method of integrated study, enables the detection of model
Power is stronger, after data prediction, recall rate, rate of false alarm and in terms of have biggish promotion.
Specifically, the data that the present invention is perceived using wearable device establish the tumble behavior of model identification user.
Specifically introduce data preprocessing module, training module, the method is as follows:
Data preprocessing module: this part is mainly handled 3-axis acceleration data, utilizes the method for sliding window, window
It is dimensioned to the half of data sampling frequency;In the training stage, need data carrying out cutting, cutting method: for
The data of tumble take the maximum point of the quadratic sum of 3-axis acceleration, the data of 1s are respectively taken in the point or so, for the number of non-tumble
According to the method for then using sliding window takes the data of 1s every time;Then the methods of Fast Fourier Transform (FFT) is utilized to the data segmented
The features such as frequency domain, time domain, energy, quantile are extracted, this Partial Feature 794 is tieed up totally.
Training module: this part is the training process of the Integrated Algorithm based on decision tree, the base of entire integrated study model
Plinth is regression tree (also referred to as CART, one kind of decision tree, see Fig. 2).Algorithm training flow chart of the invention is shown in Fig. 3.Decision
Setting this algorithm has many good characteristics: training time complexity is lower, and faster, model is interpretable for the process ratio of prediction
It is strong etc..
Training process includes input: data set and corresponding label;
Step (1) establishes the start node of decision tree, which only indicates to start the foundation of decision tree;
Step (2) is if all samples in data set T belong to the same label C, using N as leaf node, and
And it is labeled as C;
Step (3) is if the attribute set that can be used for dividing is sky, using N as leaf node, and by data set T
Label of the most label of middle number of samples as the node;
The attribute set that step (4) traversal can be used for dividing selects the value so that the smallest attribute using Gini coefficient
As split point, node N is changed to the split point;
Step (5) is according to split point partition tree T1, T2 subset, when the attribute set that can be used for dividing is empty
It waits, then end loop;
T1 subset is repeated into step (2)-(5);T2 subset is repeated into step (2)-(5).It wherein can be used for the attribute classified
Refer to the attribute for not being selected as split point in (4) step.
(1) CART classification tree prediction classification discrete data selects optimal characteristics using gini index, while determining the spy
The optimal two-value cut-off of sign.When classification measurement, the classification for including in totality is more mixed and disorderly, and gini index is bigger.Assorting process
In, it is assumed that sample set D has K class, and different classes of probability is p in sample setk, then the gini index definition of probability distribution
Are as follows:
If data set D is split on a certain value a according to feature A, D1 is obtained, after D2 two parts, then in spy
The Gini coefficient for levying set D under A is as follows.Wherein Gini coefficient Gini (D) indicates the uncertainty of set D, Gini coefficient
Gini (D, A) indicates the uncertainty of set D after A=a segmentation.Gini index is bigger, and the uncertainty of sample set is bigger.
For attribute A, (attribute has 794 in the present embodiment, and the limited the application of length only enumerates most important attribute and exists
In the last table 2 of text, wherein first fft_coefficient calculates 404 attributes in total), calculate separately any category
Data set is divided into the Gain_Gini after two parts by property value, is chosen minimum value therein, is obtained as attribute A optimal
Two offshoot programs.
Then for training set S, optimal two offshoot program of all properties is calculated, minimum value therein is chosen, as sample
Collect optimal two offshoot program of S.
But simultaneously, single decision tree has some bad places again, such as be easy over-fitting etc..
(2) it is based on the integrated study model of decision tree (CART tree):
Integrated study model are as follows:
Wherein K represents the number of total decision tree, fk(xi) that represent is kth number xiLabel (label is i.e. by xiIt is defeated
Enter to after kth tree, the tree is to xiLabel result), by all K several couples of xiMarking to add up be exactly integrated study mould
Type, each f are a functions inside function space F, and F has corresponded to the set of all CART trees, the target in this algorithm
Function is used to calculate the error between prediction result and legitimate reading, and guidance model is updated towards the smallest direction of error, the mesh
Scalar functions consist of two parts, loss function and regular terms:
Wherein i represents i-th of sample, and n is the number of samples of training set, yiIndicate the true tag of sample,Indicating should
The prediction label of sample, l indicate yi, andError.
The first part of objective function is training error, and second part is the sum of the complexity of each tree.
For the complexity of each tree, tree is split into structure division q and leaf weight portion w.Structure function q is input
It is mapped to above the call number of leaf and goes, and w is given the corresponding leaf score of each call number;The complexity of each tree
Is defined as:
Wherein T is the number of leaf node, and w is the score of leaf node.
The detectability of experimental verification algorithm on public data collection.Use SisFall, MobiAct, MMsys data set
It is verified, wherein SisFall data set comes from Colombia's Universidad de Antioquia, and the sensor used includes acceleration
Meter, gyroscope (in the present embodiment, only having used the data of acceleration), the data set include 19 kinds of daily behaviors, are fallen in 15
Backward is number of users are as follows: 23 users;MobiAct data set comes from special education in Greece gram and educates technical college, the sensor used
Including accelerometer, gyroscope (in the present embodiment, only having used the data of acceleration), which includes 9 kinds of daily rows
For, behavior of falling in 4, number of users are as follows: 57 users;MMsys data set comes from Coventry, United Kingdom university, the sensor packet used
Accelerometer, gyroscope (in the algorithm of the present embodiment, only having used the data of acceleration) are included, which includes 11 kinds of days
Chang Hangwei, behavior of falling in 4, number of users are as follows: 23 users;
Using inventive method of the invention, the accurate rate (precision) and recall rate (recall) of Activity recognition result,
Specific (specificity) and f1 value are as shown in table 1 below, and the comparing result on three data sets shows Fig. 4, Fig. 5,
Fig. 6, method of the invention can have best performance on three data sets.
Table 1
Table 2
The following are system embodiment corresponding with above method embodiment, present embodiment can be mutual with above embodiment
Cooperation is implemented.The relevant technical details mentioned in above embodiment are still effective in the present embodiment, in order to reduce repetition,
Which is not described herein again.Correspondingly, the relevant technical details mentioned in present embodiment are also applicable in above embodiment.
The invention also provides a kind of fall detection system based on Integrated Decision tree, including:
Module 1, the first acceleration information for being labeled as falling, take quadratic sum in first acceleration information maximum
Point takes the data segment of the point or so as original set, for accelerating labeled as the second of non-tumble according to preset time segment limit
Degree evidence takes the data segment of the preset time segment limit that the original set is added every time, is finally collected, by quick by sliding window
Fourier transformation extracts this and finally concentrates the multidimensional characteristic of each data segment as training characteristics;
More module 2, creation decision trees input the decision tree respectively using the training characteristics as input feature vector with iteration
Each decision tree of training, the residual error for being fitted a upper decision tree is gone by each decision tree, is gathered the decision tree that training is completed and is made
For integrated study model;
The third acceleration information of module 3, acquisition to fall detection extracts the third using the system of sliding window and accelerates degree
According to data segment as detection data, which is input to the integrated study model, obtains fall detection result.
The fall detection system based on Integrated Decision tree, wherein the decision tree is regression tree, using gini index
Select optimal characteristics.
The fall detection system based on Integrated Decision tree, wherein in the module 2 each decision tree of repetitive exercise time
Number is the number of the more decision trees.
The fall detection system based on Integrated Decision tree, wherein module 2 include:
Decision tree is specially kth -1 tree on this, the residual error of kth wheel be the kth -1 tree the sum of predicted value with
Error between true value, the fit procedure are specially so that the error between the predicted value and residual error of kth tree is minimum.
The fall detection system based on Integrated Decision tree, wherein obtaining the error, guidance collection using objective function
It is updated at learning model to the smallest direction of error, objective function Obj (Θ) specifically:
Wherein K represents the number of total decision tree, fkThe several marking of kth is represented, i represents i-th of sample, and n is instruction
Practice the number of samples of collection, yiIndicate the true tag of sample,Indicate the prediction label of the sample, l indicates yi, andMistake
Difference.
Claims (10)
1. a kind of fall detection method based on Integrated Decision tree characterized by comprising
Step 1, the first acceleration information for being labeled as falling, take the maximum point of quadratic sum in first acceleration information,
Take the data segment of the point or so as original set according to preset time segment limit, for accelerating degree labeled as the second of non-tumble
According to taking the data segment of the preset time segment limit that the original set is added every time by sliding window, finally collected, by quick Fu
Leaf transformation extracts this and finally concentrates the multidimensional characteristic of each data segment as training characteristics;
More step 2, creation decision trees input the decision tree respectively using the training characteristics as input feature vector with repetitive exercise
Each decision tree goes the residual error for being fitted a upper decision tree by each decision tree, gathers the decision tree that training is completed and is used as collection
At learning model;
The third acceleration information of step 3, acquisition to fall detection, extracts the third acceleration information using the method for sliding window
Data segment is input to the integrated study model as detection data, by the detection data, obtains fall detection result.
2. the fall detection method as described in claim 1 based on Integrated Decision tree, which is characterized in that the decision tree is to return
Tree selects optimal characteristics using gini index.
3. the fall detection method as described in claim 1 based on Integrated Decision tree, which is characterized in that iteration in the step 2
The number of each decision tree of training is the number of the more decision trees.
4. the fall detection method as described in claim 1 based on Integrated Decision tree, which is characterized in that step 2 includes:
Decision tree is specially kth -1 tree on this, the residual error of kth wheel be the kth -1 tree the sum of predicted value with it is true
Error between value, the fit procedure are specially so that the error between the predicted value and residual error of kth tree is minimum.
5. the fall detection method as claimed in claim 4 based on Integrated Decision tree, which is characterized in that obtained using objective function
To the error, integrated study model is instructed to update to the smallest direction of error, objective function Obj (Θ) specifically:
Wherein K represents the number of total decision tree, fkThe several marking of kth is represented, i represents i-th of sample, and n is training set
Number of samples, yiIndicate the true tag of sample,Indicate the prediction label of the sample, l indicates yi, andError.
6. a kind of fall detection system based on Integrated Decision tree characterized by comprising
Module 1, the first acceleration information for being labeled as falling, take the maximum point of quadratic sum in first acceleration information,
Take the data segment of the point or so as original set according to preset time segment limit, for accelerating degree labeled as the second of non-tumble
According to taking the data segment of the preset time segment limit that the original set is added every time by sliding window, finally collected, by quick Fu
Leaf transformation extracts this and finally concentrates the multidimensional characteristic of each data segment as training characteristics;
More module 2, creation decision trees input the decision tree respectively using the training characteristics as input feature vector with repetitive exercise
Each decision tree goes the residual error for being fitted a upper decision tree by each decision tree, gathers the decision tree that training is completed and is used as collection
At learning model;
The third acceleration information of module 3, acquisition to fall detection, extracts the third acceleration information using the system of sliding window
Data segment is input to the integrated study model as detection data, by the detection data, obtains fall detection result.
7. the fall detection system as claimed in claim 6 based on Integrated Decision tree, which is characterized in that the decision tree is to return
Tree selects optimal characteristics using gini index.
8. the fall detection system as claimed in claim 6 based on Integrated Decision tree, which is characterized in that iteration in the module 2
The number of each decision tree of training is the number of the more decision trees.
9. the fall detection system as claimed in claim 6 based on Integrated Decision tree, which is characterized in that module 2 includes:
Decision tree is specially kth -1 tree on this, the residual error of kth wheel be the kth -1 tree the sum of predicted value with it is true
Error between value, the fit procedure are specially so that the error between the predicted value and residual error of kth tree is minimum.
10. the fall detection system as claimed in claim 9 based on Integrated Decision tree, which is characterized in that use objective function
The error is obtained, instructs integrated study model to update to the smallest direction of error, objective function Obj (Θ) specifically:
Wherein K represents the number of total decision tree, fkThe several marking of kth is represented, i represents i-th of sample, and n is training set
Number of samples, yiIndicate the true tag of sample,Indicate the prediction label of the sample, l indicates yi, andError.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910354991.6A CN110222708A (en) | 2019-04-29 | 2019-04-29 | A kind of fall detection method and system based on Integrated Decision tree |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910354991.6A CN110222708A (en) | 2019-04-29 | 2019-04-29 | A kind of fall detection method and system based on Integrated Decision tree |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110222708A true CN110222708A (en) | 2019-09-10 |
Family
ID=67820407
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910354991.6A Pending CN110222708A (en) | 2019-04-29 | 2019-04-29 | A kind of fall detection method and system based on Integrated Decision tree |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110222708A (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111177502A (en) * | 2019-12-14 | 2020-05-19 | 贵州电网有限责任公司 | Decision tree algorithm-based power communication network redundant data identification method and system |
CN111309852A (en) * | 2020-03-16 | 2020-06-19 | 青岛百洋智能科技股份有限公司 | Method, system, apparatus and storage medium for generating a visual decision tree (set) model |
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 |
CN111464510A (en) * | 2020-03-18 | 2020-07-28 | 华南理工大学 | Network real-time intrusion detection method based on rapid gradient lifting tree model |
CN111832769A (en) * | 2019-09-24 | 2020-10-27 | 北京嘀嘀无限科技发展有限公司 | Method and device for ordering vehicle-entering points and information |
CN113658002A (en) * | 2021-08-17 | 2021-11-16 | 中国平安财产保险股份有限公司 | Decision tree-based transaction result generation method and device, electronic equipment and medium |
CN115054237A (en) * | 2022-08-16 | 2022-09-16 | 武汉理工大学 | Real-time falling prediction and power-assisted recovery method based on decision tree model |
CN115423148A (en) * | 2022-07-29 | 2022-12-02 | 江苏大学 | Agricultural machinery operation performance prediction method and device based on kriging method and decision tree |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105023022A (en) * | 2015-07-09 | 2015-11-04 | 深圳天珑无线科技有限公司 | Tumble detection method and system |
CN108549900A (en) * | 2018-03-07 | 2018-09-18 | 浙江大学 | Tumble detection method for human body based on mobile device wearing position |
WO2018168604A1 (en) * | 2017-03-15 | 2018-09-20 | Nec Corporation | Method, system, storage medium and computer system for determining fall response of subject |
-
2019
- 2019-04-29 CN CN201910354991.6A patent/CN110222708A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105023022A (en) * | 2015-07-09 | 2015-11-04 | 深圳天珑无线科技有限公司 | Tumble detection method and system |
WO2018168604A1 (en) * | 2017-03-15 | 2018-09-20 | Nec Corporation | Method, system, storage medium and computer system for determining fall response of subject |
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 (1)
Title |
---|
冯杨民、丁晓光、杨海彦: "残差修正支持向量机应用于高程异常拟合", 《研究生科技论坛》 * |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111832769A (en) * | 2019-09-24 | 2020-10-27 | 北京嘀嘀无限科技发展有限公司 | Method and device for ordering vehicle-entering points and information |
CN111177502A (en) * | 2019-12-14 | 2020-05-19 | 贵州电网有限责任公司 | Decision tree algorithm-based power communication network redundant data identification method and system |
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 |
CN111309852B (en) * | 2020-03-16 | 2021-09-03 | 青岛百洋智能科技股份有限公司 | Method, system, device and storage medium for generating visual decision tree set model |
CN111309852A (en) * | 2020-03-16 | 2020-06-19 | 青岛百洋智能科技股份有限公司 | Method, system, apparatus and storage medium for generating a visual decision tree (set) model |
CN111464510A (en) * | 2020-03-18 | 2020-07-28 | 华南理工大学 | Network real-time intrusion detection method based on rapid gradient lifting tree model |
CN113658002A (en) * | 2021-08-17 | 2021-11-16 | 中国平安财产保险股份有限公司 | Decision tree-based transaction result generation method and device, electronic equipment and medium |
CN113658002B (en) * | 2021-08-17 | 2023-07-28 | 中国平安财产保险股份有限公司 | Transaction result generation method and device based on decision tree, electronic equipment and medium |
CN115423148A (en) * | 2022-07-29 | 2022-12-02 | 江苏大学 | Agricultural machinery operation performance prediction method and device based on kriging method and decision tree |
CN115423148B (en) * | 2022-07-29 | 2023-10-31 | 江苏大学 | Agricultural machinery operation performance prediction method and device based on Ke Li jin method and decision tree |
CN115054237A (en) * | 2022-08-16 | 2022-09-16 | 武汉理工大学 | Real-time falling prediction and power-assisted recovery method based on decision tree model |
CN115054237B (en) * | 2022-08-16 | 2022-11-29 | 武汉理工大学 | Real-time falling prediction and power-assisted recovery method based on decision tree model |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110222708A (en) | A kind of fall detection method and system based on Integrated Decision tree | |
CN106471516B (en) | Malware judging device, Malware decision-making system, Malware determination method and program | |
CN100581451C (en) | Cardiac diagnostic system and method | |
CN108777873A (en) | The wireless sensor network abnormal deviation data examination method of forest is isolated based on weighted blend | |
CN114564982B (en) | Automatic identification method for radar signal modulation type | |
Gwo et al. | Plant identification through images: Using feature extraction of key points on leaf contours1 | |
CN108717548B (en) | Behavior recognition model updating method and system for dynamic increase of sensors | |
CN107582077A (en) | A kind of human body state of mind analysis method that behavior is touched based on mobile phone | |
CN108309304A (en) | A method of generating freezing of gait intelligent monitor system | |
CN106649276A (en) | Identification method and device for core product word in title | |
CN110866547B (en) | Automatic classification system and method for traditional Chinese medicine decoction pieces based on multiple features and random forests | |
CN110897634A (en) | Electrocardiosignal generation method based on generation countermeasure network | |
CN110458094A (en) | Device class method based on fingerprint similarity | |
CN109145685A (en) | Fruits and vegetables EO-1 hyperion quality detecting method based on integrated study | |
CN109409216A (en) | Speed adaptive indoor human body detection method based on subcarrier dynamic select | |
CN106649264A (en) | Text information-based Chinese fruit variety information extracting method and device | |
CN108062570A (en) | A kind of pattern recognition system for screening lung cancer | |
CN117133057A (en) | Physical exercise counting and illegal action distinguishing method based on human body gesture recognition | |
CN108078563A (en) | A kind of EEG signal analysis method of integrated classifier | |
CN104021295B (en) | Cluster feature fusion method and device for moving identification | |
CN111709440B (en) | Feature selection method based on FSA-choket fuzzy integral | |
Ashish et al. | Survey on detection and classification of plant leaf disease in agriculture environment | |
Sari et al. | Cholesterol level detection through eye image using fractal and decision tree | |
CN110096941A (en) | A kind of Gait Recognition system based on siamese network | |
CN113933334A (en) | Feature selection and machine learning algorithm-based acacia honey authenticity identification method |
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: 20190910 |