CN114327045A - Fall detection method and system based on category unbalanced signals - Google Patents
Fall detection method and system based on category unbalanced signals Download PDFInfo
- Publication number
- CN114327045A CN114327045A CN202111448595.3A CN202111448595A CN114327045A CN 114327045 A CN114327045 A CN 114327045A CN 202111448595 A CN202111448595 A CN 202111448595A CN 114327045 A CN114327045 A CN 114327045A
- Authority
- CN
- China
- Prior art keywords
- test data
- action
- data
- motion
- threshold
- 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.)
- Granted
Links
Images
Landscapes
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
本发明涉及一种基于类别不平衡信号的跌倒检测方法及系统，属于行为识别技术领域，解决了现有技术因跌倒检测类别不平衡，很难从大量的日常活动中准确地识别出跌倒事件信号的问题。该方法包括：实时获取智能可穿戴设备采集的使用者的动作测试数据；所述动作测试数据，包括：加速度数据和角速度值；将所述动作测试数据输入最优深度学习模型，进行所述动作测试数据的动作类别识别，获取各动作类别概率值；将所述各动作类别概率值与最佳阈值进行比较，预测所述动作测试数据对应的动作类别；其中，所述最佳阈值用于根据深度学习模型训练时使用的样本数据集的不平衡率使预测结果向发生概率低的动作类别偏移。实现了基于类别不平衡数据的跌倒检测。
The invention relates to a fall detection method and system based on category unbalanced signals, belongs to the technical field of behavior recognition, and solves the problem that it is difficult to accurately identify fall event signals from a large number of daily activities due to unbalanced fall detection categories in the prior art The problem. The method includes: acquiring in real time the motion test data of the user collected by the smart wearable device; the motion test data includes: acceleration data and angular velocity value; inputting the motion test data into an optimal deep learning model to perform the motion The action category identification of the test data, and the probability value of each action category is obtained; the probability value of each action category is compared with the best threshold value, and the action category corresponding to the action test data is predicted; wherein, the best threshold value is used according to The imbalance rate of the sample dataset used in deep learning model training shifts the prediction results to action categories with low probability of occurrence. Implemented fall detection based on class-imbalanced data.
Description
Technical Field
The invention relates to the technical field of behavior recognition, in particular to a falling detection method and system based on category unbalanced signals.
Background
It is well known that falls can have serious consequences such as paralysis, fractures, head injuries, and the like, particularly in the elderly. The old people often suffer from cardiovascular and cerebrovascular diseases and gait disorder, which can cause the patients to fall down, so that the patients are unconscious and faint. Falls have become the biggest threat to life safety of the elderly. In addition, the situation of the mouth aging is increasingly severe in the world, the number of the old people and the proportion of the old people living alone are increased year by year, and the old people living alone are difficult to save themselves once falling down, so that the method has important application significance for the falling detection of the old people.
Currently, fall detection can be classified into the following three categories according to the source of the data: video image-based fall detection, environmental sensor-based fall detection, and wearable device-based fall detection. The fall detection based on the video image is limited by the coverage range of the camera, and the detected person is required not to be shielded by other objects; tumble detection based on an environment sensor also requires the arrangement of equipment in advance, and can only detect in a fixed range; the fall detection device based on the wearable device is small in size and convenient to carry, and data obtained during movement are directly extracted in a feature extraction mode and classified.
However, conventional algorithms for fall detection based on wearable devices include a threshold method and a machine learning method, the threshold method is simple to calculate and easy to deploy, but the threshold method is not very generalized among different users; the machine learning method requires manually setting the characteristics of data in advance, and the selection of the characteristics is a difficult problem. In addition, most of the traditional detection algorithms perform model training in a data set under an ideal condition, the categories of daily activity data and falling event data in the data set under the ideal condition are balanced, in real life, a falling event is a sporadic event, the daily life data acquired by a sensor are far more than the falling event data, and the detection problem of unbalanced categories belongs to, and the traditional detection algorithms are difficult to accurately identify falling event signals from a large amount of daily activities.
Therefore, a fall detection method and system suitable for category imbalance data are lacking in the prior art.
Disclosure of Invention
In view of the above analysis, embodiments of the present invention are directed to a method and a system for fall detection based on a category imbalance signal, so as to solve the problem that it is difficult to accurately identify a fall event signal from a large amount of daily activities due to the imbalance of categories in the conventional fall detection.
In one aspect, an embodiment of the present invention provides a fall detection method based on a category unbalanced signal, including:
acquiring action test data of a user acquired by intelligent wearable equipment in real time; the motion test data comprises: acceleration data and angular velocity values;
inputting the motion test data into an optimal deep learning model, identifying motion types of the motion test data, and acquiring probability values of the motion types;
comparing the probability value of each action category with an optimal threshold value, and predicting the action category corresponding to the action test data;
and the optimal threshold is used for shifting the prediction result to the action type with low occurrence probability according to the unbalance rate of the sample data set used in the deep learning model training.
Further, the optimal deep learning model comprises:
the residual error connecting module is used for extracting the characteristics of the motion test data;
the fully-connected softmax layer is used for classifying the motion test data characteristics and outputting probability values of the motion test data corresponding to the motion categories;
and the threshold moving algorithm layer is used for adjusting the classification threshold, acquiring the optimal threshold, and judging the action type of the action test data according to the relation between the probability value of each action type and the optimal threshold.
Further, the adjusting the classification threshold to obtain the optimal threshold includes:
acquiring a sample data set, dividing data in the sample data set into daily activity sample data and fall sample data, and counting the number of the daily activity sample data and the number of the fall sample data;
calculating the sample unbalance rate according to the ratio of the number of the daily life sample data to the number of the falling sample data;
adjusting a classification threshold according to the sample unbalance rate, and determining the optimal threshold, wherein the optimal threshold is a falling action threshold lambda*。
Further, the optimum threshold λ*Expressed as:
λ*＝k×e-ρ/a+b
wherein λ is*K is the default threshold of the classifier, ρ is the imbalance rate of the sample data set, a and b are constants, nmaxNumber of daily data, n, for sample data setminThe number of fall sample data in the sample data set.
Further, the action categories include fall actions and daily activity actions; judging the action type of the action test data according to the relation between the probability value of each action type and the optimal threshold, wherein the action type comprises the following steps:
when a fall probability value p and an optimal threshold lambda are output by certain action test data*Comparing if p is more than or equal to lambda*If so, the optimal deep learning model predicts that the motion test data is a falling motion;
when the daily activity probability value q of a certain action test data output and the optimal threshold lambda*Comparing if q is less than or equal to 1-lambda*And predicting the motion test data to be the falling motion by the optimal deep learning model.
Further, the residual connecting module includes:
the convolution layer is used for extracting characteristic information in the original data;
the batch normalization layer is used for carrying out normalization processing on the characteristic information and then carrying out nonlinear calculation on the activation function;
and the residual connection computing unit is used for linearly superposing the extracted characteristic information and the original data.
Further, the convolutional layer comprises: a first convolution layer and a second convolution layer;
the batch normalization layer includes: a first batch of normalization layers and a second batch of normalization layers;
the residual error connecting module comprises four first to fourth residual error connecting units which are sequentially connected, and the first to fourth residual error connecting units all comprise the following components in sequence: the first convolution layer, the first batch of normalization layers, the second convolution layer, the second batch of normalization layers and the residual error connection calculation subunit;
the first convolution layer and the second convolution layer in the first residual error connection unit respectively comprise 64 convolution kernels;
the first convolution layer and the second convolution layer in the second residual connecting unit respectively comprise 128 convolution kernels;
the first convolution layer and the second convolution layer in the third residual connecting unit respectively comprise 256 convolution kernels;
the first convolution layer and the second convolution layer in the fourth residual connection unit both comprise 512 convolution kernels;
all convolution kernels in the first to fourth residual connecting units are 1x3 convolution kernels.
Further, in the process of model training of the deep learning model, the error of the model training process is calculated through a cross entropy loss function, and when the error is stable, the deep learning model is the optimal deep learning model.
In another aspect, an embodiment of the present invention provides a fall detection system based on a category imbalance signal, including:
the data acquisition module is used for acquiring action test data of the user, which is acquired by the intelligent wearable equipment, in real time; the motion test data comprises: acceleration data and angular velocity values;
the probability calculation module is used for inputting the motion test data into an optimal deep learning model, performing motion category identification on the motion test data and acquiring probability values of all motion categories;
the prediction module is used for comparing the probability value of each action category with an optimal threshold value and predicting the action category corresponding to the action test data; and the optimal threshold is used for shifting the prediction result to the action type with low occurrence probability according to the unbalance rate of the sample data set used in the deep learning model training.
Further, the optimal deep learning model in the probability calculation module comprises:
the residual error connecting module is used for extracting the characteristics of the motion test data;
the fully-connected softmax layer is used for classifying the motion test data characteristics and outputting probability values of the motion test data corresponding to the motion categories;
and the threshold moving algorithm layer is used for adjusting the classification threshold, acquiring the optimal threshold, and judging the action type of the action test data according to the relation between the probability value of each action type and the optimal threshold.
Compared with the prior art, the invention can realize at least one of the following beneficial effects:
1. according to the invention, the unbalanced category condition of real life activities is optimized by an algorithm value moving method, so that the prediction result of the optimal deep learning model moves to a small number of falling data categories, the deviation of most daily activity categories is reduced, the model can be ensured to detect accidental falling event data with low occurrence probability from a large number of daily activity data, and the detection precision is high;
2. based on a deep learning network model, the deep learning method can automatically identify deep features from mass data and adjust the parameters of the model, compared with the traditional algorithm model, the method can more accurately detect and identify the falling event, the network model is trained and the parameters are adjusted through the category unbalance data collected in the real life, the trained model has good detection capability on the falling event with small probability, falling data signals can be identified from mass daily activity data, and the condition of missed detection is reduced;
3. according to the invention, the real life falling data set is established after certain preprocessing is carried out according to the data acquired by the accelerometer and the gyroscope in the intelligent wearable device, and the established type unbalanced falling detection model is trained and tested on the real life falling data set, so that the intelligent wearable device is more suitable for being applied to real life.
In the invention, the technical schemes can be combined with each other to realize more preferable combination schemes. Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and drawings.
Drawings
The drawings are only for purposes of illustrating particular embodiments and are not to be construed as limiting the invention, wherein like reference numerals are used to designate like parts throughout.
Fig. 1 is a flowchart of a fall detection method based on class imbalance signals according to an embodiment of the present invention;
FIG. 2 is a block diagram of a deep learning model according to an embodiment of the present invention;
FIG. 3 is a diagram illustrating a raw data preprocessing process according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of a residual linking module according to an embodiment of the present invention;
FIG. 5 is an image of a threshold shift function fitted to experimental data according to one embodiment of the present invention;
fig. 6 is a block diagram of a fall detection system based on class imbalance signals according to an embodiment of the invention.
Detailed Description
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate preferred embodiments of the invention and together with the description, serve to explain the principles of the invention and not to limit the scope of the invention.
As shown in fig. 1, an embodiment of the present invention discloses a fall detection method based on a category imbalance signal, which includes:
s10, acquiring action test data of the user acquired by the intelligent wearable equipment in real time; the motion test data comprises: acceleration data and angular velocity values; specifically, acceleration data and angular velocity values of various action categories of a user in daily life are collected according to an accelerometer and a gyroscope of the intelligent wearable device on the user. In particular, the accelerometer and gyroscope are integrated on the smart wearable device, the position of which relative to the wearable device is fixed, optionally, one wearable device comprising one accelerometer and one gyroscope, the wearable device being fixed on the waist of the user.
S20, inputting the motion test data into an optimal deep learning model, identifying motion types of the motion test data, and acquiring probability values of the motion types; in particular, the categories of actions include fall actions and daily activity actions; and inputting the acceleration data and the angular velocity value in the real-time collected data into an optimal deep learning model, and outputting a falling action probability value and a daily activity action probability value corresponding to the real-time collected data by the optimal deep learning model.
As shown in fig. 2, the optimal deep learning model includes:
a residual error connection module 10, configured to perform feature extraction on the motion test data;
the fully-connected softmax layer 20 is used for classifying the motion test data characteristics and outputting probability values of the motion test data corresponding to the motion categories;
and the threshold moving algorithm layer 30 is configured to adjust the classification threshold, obtain an optimal threshold, and determine the action category of the action test data according to a relationship between the probability value of each action category and the optimal threshold.
S30, comparing the probability value of each action type with an optimal threshold value, and predicting the action type corresponding to the action test data; and the optimal threshold is used for shifting the prediction result to the action type with low occurrence probability according to the unbalance rate of the sample data set used in the deep learning model training. Specifically, the falling action probability value and the daily activity action probability value are compared with an optimal threshold value to predict whether a falling action occurs. Specifically, the determining the action category of the action test data according to the relationship between the probability value of each action category and the optimal threshold includes:
when a fall probability value p and an optimal threshold lambda are output by certain action test data*Comparing if p is more than or equal to lambda*If so, the optimal deep learning model predicts that the motion test data is a falling motion;
when the daily activity probability value q of a certain action test data output and the optimal threshold lambda*Comparing if q is less than or equal to 1-lambda*And predicting the motion test data to be the falling motion by the optimal deep learning model.
After a deep learning model is built, the model is trained by using a training set and a verification set, parameters in the model are determined, in a life application scene, data of an accelerometer and a gyroscope are collected by user wearable equipment in real time, as shown in fig. 3, data within 3s before and after a peak value of the data are marked as falling actions, continuous time sequence data are intercepted into fragments by using a sliding window with the width of 1s, and after the real-time collection is preprocessed, the data are input into the trained model to predict whether a falling event occurs.
The method described above focuses on training the optimal deep learning model and determining the optimal threshold, and therefore, this embodiment provides a specific training method for the optimal deep learning model and a specific determining method for the optimal threshold, which are specifically as follows:
specifically, as shown in fig. 2, the deep learning model includes:
a residual error connection module 10, configured to perform feature extraction on the motion test data; specifically, the residual error connection module 10 includes: the convolution layer is used for extracting characteristic information in the original signal; the batch normalization layer is used for carrying out normalization processing on the characteristic information and then carrying out nonlinear calculation on the activation function; and the residual connection computing unit is used for linearly superposing the extracted characteristic information and the original data. Specifically, the residual error connection calculation unit may be introduced to skip one or more layers, and linearly superimpose the input original data and the feature information extracted from the original data through nonlinear change, so that on one hand, the residual error connection module 10 can better fit a classification function to obtain higher classification accuracy, and on the other hand, the residual error connection module 10 solves the problem of optimization training of the network when the number of layers is deepened.
Specifically, in the deep learning model training process, the training process of the residual error connection module is divided into two stages, specifically, the first stage is a stage in which data is propagated from a lower level to a higher level, i.e., a forward propagation stage, and the second stage is a process in which an error is propagated and trained from a higher level to a lower level when a result obtained by current forward propagation does not match an expected result, i.e., a backward propagation stage. Specifically, the training process of the residual error connection module 10 is as follows: 1. initializing a weight value by the network; 2. the input data is propagated forwards through the convolution layer and the like to obtain an output value; 3. calculating the error between the output value of the network and the target value; 4. when the error is larger than 0, the error is transmitted back to the network; 5. and updating the weight according to the obtained error, and then carrying out forward propagation calculation again.
Specifically, the residual connecting module 10 includes four first to fourth residual connecting units connected in sequence, and specifically, as shown in fig. 4, each residual connecting unit includes: a first convolution layer, a first Batch Normalization layer, a second convolution layer, a second Batch Normalization layer and a residual connection calculation subunit; the first convolution layer and the second convolution layer in the first residual connection unit respectively comprise 64 convolution kernels, the first convolution layer and the second convolution layer in the second residual connection unit respectively comprise 128 convolution kernels, the first convolution layer and the second convolution layer in the third residual connection unit respectively comprise 256 convolution kernels, and the first convolution layer and the second convolution layer in the fourth residual connection unit respectively comprise 512 convolution kernels; more specifically, all convolution kernels in the first to fourth residual connecting units are 1 × 3 convolution kernels; the stacking of four residual connecting units and the number of convolution kernels are increased so as to better extract the characteristics of the data deep level.
Specifically, time sequence data of an accelerometer and a gyroscope of a person in daily life are collected according to intelligent wearable equipment on a large number of users, and a sample data set of daily activities and falls of the person is formed; specifically, in the process of establishing the data set, daily life data and fall simulation data of a plurality of users are collected, a sample data set containing a large amount of data is established, and parameters of the algorithm model are adjusted before the sample data set is used, so that the error of model prediction is reduced.
Randomly dividing the sample data set into a training set and a verification set according to the proportion of 3: 1; preprocessing the data in the originally acquired sample data set, specifically, the preprocessing includes data segmentation and labeling: the data of each segment is marked, 0 represents daily activity action data, 1 represents falling event action data, and the original continuous data is cut into segments by using a sliding window.
Specifically, before the residual error connection module 10 is trained, the original parameters of the residual error connection module 10 are initialized, the weight parameters adopt an Xavier initialization mode, and the bias parameters adopt an all-0 initialization mode; and taking the training set as the input of the residual connecting module 10, training the residual connecting module 10 for multiple times, wherein after multiple parameter adjustment and training, the error is reduced along with the number of times of model back propagation calculation, and when the number of calculation times is large enough, the error of the model is rapidly reduced from the initial training and tends to be stable, so that the network model weight which is well represented in the test set is obtained. Optionally, in this embodiment, the function for calculating the error by the residual connecting module 10 uses a cross entropy loss function, and it is found through experiments that the error of the model is stable after 50 times of back propagation calculation, so the training calculation parameter of the model is set to be 50, and the model training is finished after 50 times of back propagation calculation.
In the embodiment of the fully-connected softmax layer, firstly, input data is converted into output with the same number as the number of classes through the weight matrix obtained by the residual error connection module 10, and then, the output probabilities of the classes are converted into probabilities within the range of 0 to 1 through a softmax activation function, and the sum of the probabilities of all the classes is ensured to be 1. And comparing the probability value of each action category corresponding to the action test data output by the fully-connected softmax layer 20 with the optimal threshold value, and predicting the action category corresponding to the action test data.
In the deep learning network, with the increase of the network depth, the features in the sample data set can be better extracted, the model performance is better, but with the increase of the network depth, the problems of gradient disappearance and the like are also brought. The residual connecting module 10 can well solve the problems of gradient disappearance and gradient explosion of the deep learning model.
Specifically, the adjusting the classification threshold to obtain the optimal threshold includes:
s100, acquiring a sample data set, dividing data in the sample data set into daily activity sample data and falling sample data, and counting the number of the daily activity sample data and the number of the falling sample data;
s200, calculating the sample unbalance rate according to the ratio of the number of the daily life sample data to the number of the falling sample data;
s300, adjusting a classification threshold according to the sample unbalance rate, and determining the optimal threshold, wherein the optimal threshold is a falling action threshold lambda*. The optimum threshold value lambda*Expressed as:
λ*＝k×e-ρ/a+b
wherein λ is*K is the default threshold of the classifier, ρ is the imbalance rate of the sample data set, a and b are constants, nmaxNumber of daily data, n, for sample data setminThe number of fall sample data in the sample data set. Optionally, the default threshold k of the classifier is 0.5; a and b are determined through a threshold value moving function curve, and the threshold value moving function curve is obtained by fitting according to falling motion test threshold values corresponding to different unbalance rates in an experimental data set;
specifically, the obtaining of the threshold moving function curve obtained by fitting the experimental data specifically includes:
acquiring an experimental data set, dividing the experimental data set into a training set and a test set, performing feature extraction on the input training set by using a threshold value moving algorithm layer in a deep learning model, and outputting a falling action probability value p 'and a daily activity action probability value q' corresponding to the data, wherein p '+ q' is 1, and the falling action threshold value is set to be lambda ', and the daily activity action threshold value is 1-lambda';
when the probability value p 'of the falling action corresponding to the output data is more than or equal to lambda', classifying the prediction result as the falling action; when the probability value q 'of the daily activity actions is more than or equal to 1-lambda', classifying the prediction result as the daily activity actions; when the falling action threshold value meets the formula (1), the falling action threshold value is a falling action test threshold value corresponding to the unbalance rate of the experimental data set
As shown in fig. 5, a threshold movement function curve is obtained by fitting different imbalance rates and fall action test thresholds, where a in the graph is a fall action test threshold point corresponding to different imbalance rates, and the curve is a threshold movement function curve obtained by fitting according to the fall action test thresholds and the imbalance rates. Corresponding falling action test threshold values under different unbalance rates
Expressed as:
wherein f represents a small numberProbability of occurrence of several types of data (i.e. fall maneuver data) in the training set, fλ′Which represents the probability that the minority of classes of data predicted at the threshold lambda', i.e. fall motion data, is in the test set.
The higher the imbalance rate of the raw data collected, the more the threshold of the classifier shifts to a few classes. The optimal threshold determined by the optimal threshold determination formula can improve the classification precision of the deep learning model on the category unbalanced data, and reduce the missing detection of a few categories.
Compared with the prior art, in the fall detection method based on the category unbalanced signal, the category unbalanced condition of real life activities is optimized by an algorithm value moving method, so that the prediction result of the optimal deep learning model moves to a small number of fall data categories, the deviation of most daily activity categories is reduced, the model can detect accidental fall incident data with low occurrence probability from a large amount of daily activity data, and the detection precision is high; based on a deep learning network model, the deep learning method can automatically identify deep features from mass data and adjust the parameters of the model, compared with the traditional algorithm model, the method can more accurately detect and identify the falling event, the network model is trained and the parameters are adjusted through the category unbalance data collected in the real life, the trained model has good detection capability on the falling event with small probability, falling data signals can be identified from mass daily activity data, and the condition of missed detection is reduced; finally, according to data collected by an accelerometer and a gyroscope in the intelligent wearable device, the real life falling data set is established after certain preprocessing, and the established type unbalanced falling detection model is used for training and testing the real life falling data set, so that the intelligent wearable device is more suitable for being applied to real life.
As shown in fig. 6, an embodiment of the present invention discloses a fall detection system based on a category imbalance signal, including:
the data acquisition module 10 is used for acquiring action test data of a user, which is acquired by the intelligent wearable device, in real time; the motion test data comprises: acceleration data and angular velocity values;
the probability calculation module 20 is configured to input the motion test data into an optimal deep learning model, perform motion category identification on the motion test data, and obtain probability values of the motion categories;
a prediction module 30, configured to compare the probability values of the motion categories with an optimal threshold, and predict motion categories corresponding to the motion test data; and the optimal threshold is used for shifting the prediction result to the action type with low occurrence probability according to the unbalance rate of the sample data set used in the deep learning model training.
Specifically, the optimal deep learning model in the probability calculation module includes:
the residual error connecting module is used for extracting the characteristics of the motion test data;
the fully-connected softmax layer is used for classifying the motion test data characteristics and outputting probability values of the motion test data corresponding to the motion categories;
and the threshold moving algorithm layer is used for adjusting the classification threshold, acquiring the optimal threshold, and judging the action type of the action test data according to the relation between the probability value of each action type and the optimal threshold. Specifically, the threshold moving algorithm layer includes:
the system comprises a sample data acquisition unit, a data processing unit and a data processing unit, wherein the sample data acquisition unit is used for acquiring a sample data set, dividing data in the sample data set into daily activity sample data and falling sample data, and counting the number of the daily activity sample data and the number of the falling sample data;
the unbalance rate obtaining unit is used for calculating the sample unbalance rate according to the ratio of the number of the daily life sample data to the number of the falling sample data;
an optimal threshold confirming unit, configured to adjust a classification threshold according to the sample imbalance rate, and determine the optimal threshold, where the optimal threshold is a fall motion threshold λ*(ii) a The optimum threshold value lambda*Expressed as:
λ*＝k×e-ρ/a+b
wherein λ is*K is the default threshold of the classifier, ρ is the imbalance rate of the sample data set, a and b are constants, nmaxNumber of daily data, n, for sample data setminThe number of fall sample data in the sample data set.
Those skilled in the art will appreciate that all or part of the flow of the method implementing the above embodiments may be implemented by a computer program, which is stored in a computer readable storage medium, to instruct related hardware. The computer readable storage medium is a magnetic disk, an optical disk, a read-only memory or a random access memory.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention.
Claims (10)
1. A method for fall detection based on a category imbalance signal, comprising:
acquiring action test data of a user acquired by intelligent wearable equipment in real time; the motion test data comprises: acceleration data and angular velocity values;
inputting the motion test data into an optimal deep learning model, identifying motion types of the motion test data, and acquiring probability values of the motion types;
comparing the probability value of each action category with an optimal threshold value, and predicting the action category corresponding to the action test data;
and the optimal threshold is used for shifting the prediction result to the action type with low occurrence probability according to the unbalance rate of the sample data set used in the deep learning model training.
2. A fall detection method based on class imbalance signals according to claim 1,
the optimal deep learning model comprises:
the residual error connecting module is used for extracting the characteristics of the motion test data;
the fully-connected softmax layer is used for classifying the motion test data characteristics and outputting probability values of the motion test data corresponding to the motion categories;
and the threshold moving algorithm layer is used for adjusting the classification threshold, acquiring the optimal threshold, and judging the action type of the action test data according to the relation between the probability value of each action type and the optimal threshold.
3. A fall detection method based on class imbalance signals according to claim 2,
the adjusting the classification threshold to obtain the optimal threshold includes:
acquiring a sample data set, dividing data in the sample data set into daily activity sample data and fall sample data, and counting the number of the daily activity sample data and the number of the fall sample data;
calculating the sample unbalance rate according to the ratio of the number of the daily life sample data to the number of the falling sample data;
adjusting a classification threshold according to the sample unbalance rate, and determining the optimal threshold, wherein the optimal threshold is a falling action threshold lambda*。
4. A fall detection method based on class imbalance signals according to claim 3,
the optimum threshold value lambda*Expressed as:
λ*＝k×e-ρ/a+b
wherein λ is*K is the default threshold of the classifier, ρ is the imbalance rate of the sample data set, a and b are constants, nmaxNumber of daily data, n, for sample data setminThe number of fall sample data in the sample data set.
5. A fall detection method based on class imbalance signals according to claim 4,
the action categories include fall actions and daily activity actions; judging the action type of the action test data according to the relation between the probability value of each action type and the optimal threshold, wherein the action type comprises the following steps:
when a fall probability value p and an optimal threshold lambda are output by certain action test data*Comparing if p is more than or equal to lambda*If so, the optimal deep learning model predicts that the motion test data is a falling motion;
when the daily activity probability value q of a certain action test data output and the optimal threshold lambda*Comparing if q is less than or equal to 1-lambda*And predicting the motion test data to be the falling motion by the optimal deep learning model.
6. A fall detection method based on class imbalance signals according to claim 2,
the residual connecting module comprises:
the convolution layer is used for extracting characteristic information in the original data;
the batch normalization layer is used for carrying out normalization processing on the characteristic information and then carrying out nonlinear calculation on the activation function;
and the residual connection computing unit is used for linearly superposing the extracted characteristic information and the original data.
7. A fall detection method based on class imbalance signals according to claim 6,
the convolutional layer includes: a first convolution layer and a second convolution layer;
the batch normalization layer includes: a first batch of normalization layers and a second batch of normalization layers;
the residual error connecting module comprises four first to fourth residual error connecting units which are sequentially connected, and the first to fourth residual error connecting units all comprise the following components in sequence: the first convolution layer, the first batch of normalization layers, the second convolution layer, the second batch of normalization layers and the residual error connection calculation subunit;
the first convolution layer and the second convolution layer in the first residual error connection unit respectively comprise 64 convolution kernels;
the first convolution layer and the second convolution layer in the second residual connecting unit respectively comprise 128 convolution kernels;
the first convolution layer and the second convolution layer in the third residual connecting unit respectively comprise 256 convolution kernels;
the first convolution layer and the second convolution layer in the fourth residual connection unit both comprise 512 convolution kernels;
all convolution kernels in the first to fourth residual connecting units are 1x3 convolution kernels.
8. A fall detection method based on class imbalance signals according to claim 3,
in the process of carrying out model training on the deep learning model, calculating the error of the model training process through a cross entropy loss function, and when the error is stable, the deep learning model is the optimal deep learning model.
9. A fall detection system based on a class imbalance signal, comprising:
the data acquisition module is used for acquiring action test data of the user, which is acquired by the intelligent wearable equipment, in real time; the motion test data comprises: acceleration data and angular velocity values;
the probability calculation module is used for inputting the motion test data into an optimal deep learning model, performing motion category identification on the motion test data and acquiring probability values of all motion categories;
the prediction module is used for comparing the probability value of each action category with an optimal threshold value and predicting the action category corresponding to the action test data; and the optimal threshold is used for shifting the prediction result to the action type with low occurrence probability according to the unbalance rate of the sample data set used in the deep learning model training.
10. The category imbalance signal based fall detection system of claim 9,
the optimal deep learning model in the probability calculation module comprises:
the residual error connecting module is used for extracting the characteristics of the motion test data;
the fully-connected softmax layer is used for classifying the motion test data characteristics and outputting probability values of the motion test data corresponding to the motion categories;
and the threshold moving algorithm layer is used for adjusting the classification threshold, acquiring the optimal threshold, and judging the action type of the action test data according to the relation between the probability value of each action type and the optimal threshold.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111448595.3A CN114327045B (en) | 2021-11-30 | 2021-11-30 | Fall detection method and system based on class imbalance signal |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111448595.3A CN114327045B (en) | 2021-11-30 | 2021-11-30 | Fall detection method and system based on class imbalance signal |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114327045A true CN114327045A (en) | 2022-04-12 |
| CN114327045B CN114327045B (en) | 2024-07-26 |
Family
ID=81048062
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111448595.3A Active CN114327045B (en) | 2021-11-30 | 2021-11-30 | Fall detection method and system based on class imbalance signal |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114327045B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114983447A (en) * | 2022-08-01 | 2022-09-02 | 广东海洋大学 | A wearable device for human action recognition, analysis and storage based on AI technology |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107578061A (en) * | 2017-08-16 | 2018-01-12 | 哈尔滨工业大学深圳研究生院 | Based on the imbalanced data classification issue method for minimizing loss study |
| US20190102674A1 (en) * | 2017-09-29 | 2019-04-04 | Here Global B.V. | Method, apparatus, and system for selecting training observations for machine learning models |
| CN111488850A (en) * | 2020-04-17 | 2020-08-04 | 电子科技大学 | Neural network-based old people falling detection method |
| CN112016597A (en) * | 2020-08-12 | 2020-12-01 | 河海大学常州校区 | Depth sampling method based on Bayesian imbalance measurement in machine learning |
| CN112595906A (en) * | 2020-10-09 | 2021-04-02 | 国网浙江省电力有限公司湖州供电公司 | Method for judging abnormal operation of transformer area |
| CN113220960A (en) * | 2021-05-12 | 2021-08-06 | 西安理工大学 | Unbalanced time series data classification method based on autonomous learning |
| US11182691B1 (en) * | 2014-08-14 | 2021-11-23 | Amazon Technologies, Inc. | Category-based sampling of machine learning data |
-
2021
- 2021-11-30 CN CN202111448595.3A patent/CN114327045B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11182691B1 (en) * | 2014-08-14 | 2021-11-23 | Amazon Technologies, Inc. | Category-based sampling of machine learning data |
| CN107578061A (en) * | 2017-08-16 | 2018-01-12 | 哈尔滨工业大学深圳研究生院 | Based on the imbalanced data classification issue method for minimizing loss study |
| US20190102674A1 (en) * | 2017-09-29 | 2019-04-04 | Here Global B.V. | Method, apparatus, and system for selecting training observations for machine learning models |
| CN111488850A (en) * | 2020-04-17 | 2020-08-04 | 电子科技大学 | Neural network-based old people falling detection method |
| CN112016597A (en) * | 2020-08-12 | 2020-12-01 | 河海大学常州校区 | Depth sampling method based on Bayesian imbalance measurement in machine learning |
| CN112595906A (en) * | 2020-10-09 | 2021-04-02 | 国网浙江省电力有限公司湖州供电公司 | Method for judging abnormal operation of transformer area |
| CN113220960A (en) * | 2021-05-12 | 2021-08-06 | 西安理工大学 | Unbalanced time series data classification method based on autonomous learning |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114983447A (en) * | 2022-08-01 | 2022-09-02 | 广东海洋大学 | A wearable device for human action recognition, analysis and storage based on AI technology |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114327045B (en) | 2024-07-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Umadevi et al. | A survey on data mining classification algorithms | |
| US11416717B2 (en) | Classification model building apparatus and classification model building method thereof | |
| EP2523149A2 (en) | A method and system for association and decision fusion of multimodal inputs | |
| CN104504366A (en) | System and method for smiling face recognition based on optical flow features | |
| Li et al. | Adaptive deep feature fusion for continuous authentication with data augmentation | |
| Su et al. | HDL: Hierarchical deep learning model based human activity recognition using smartphone sensors | |
| CN106156702A (en) | Identity identifying method and equipment | |
| CN113343123B (en) | A Generative Adversarial Multi-Relational Graph Network Training Method and Detection Method | |
| JP2008262331A (en) | Object tracking device and object tracking method | |
| CN111767932B (en) | Action determination method and device, computer equipment and computer readable storage medium | |
| JP2019016268A (en) | Image processing apparatus, image processing method and image processing program | |
| CN106874874A (en) | Motion state identification method and device | |
| CN113361554A (en) | Biological feature recognition multi-modal fusion method and device, storage medium and equipment | |
| CN111898479B (en) | A mask wearing recognition method and device based on a fully convolutional single-step target detection algorithm | |
| CN116958584A (en) | Key point detection method, regression model training method and device and electronic equipment | |
| Shang et al. | Improving training and inference of face recognition models via random temperature scaling | |
| Yang et al. | A Face Detection Method Based on Skin Color Model and Improved AdaBoost Algorithm. | |
| Qin et al. | Finger-vein quality assessment based on deep features from grayscale and binary images | |
| CN114064958B (en) | Object detection method, device, computer readable storage medium and processor | |
| CN114327045A (en) | Fall detection method and system based on category unbalanced signals | |
| Ge et al. | CRUFT: Context recognition under uncertainty using fusion and temporal learning | |
| CN119295937B (en) | Multi-scale twin hyperspectral target detection method and system | |
| CN117617888B (en) | System and method for predicting myopic diopter | |
| WO2022049704A1 (en) | Information processing system, information processing method, and computer program | |
| Kumari et al. | Effect of reduced dimensionality on deep learning for human activity recognition |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |