CN114897025A - Human body posture recognition model establishing method and human body posture recognition method - Google Patents
Human body posture recognition model establishing method and human body posture recognition method Download PDFInfo
- Publication number
- CN114897025A CN114897025A CN202210565324.4A CN202210565324A CN114897025A CN 114897025 A CN114897025 A CN 114897025A CN 202210565324 A CN202210565324 A CN 202210565324A CN 114897025 A CN114897025 A CN 114897025A
- Authority
- CN
- China
- Prior art keywords
- human body
- posture recognition
- body posture
- model
- human
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- 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
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/285—Selection of pattern recognition techniques, e.g. of classifiers in a multi-classifier system
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
Abstract
The invention relates to a human body posture recognition model establishing method and a human body posture recognition method. The human body posture recognition model establishing method comprises the following steps: acquiring human body motion signals under different postures, wherein the source of the human body motion signals comprises intelligent wearable equipment worn on a human body; preprocessing the human body motion signal to obtain a model input signal; training a preset model based on the model input signal to obtain a human body posture recognition model, wherein the human body posture recognition model is used for outputting a posture classification result, the preset model comprises a bidirectional long-short term memory network and a support vector machine classifier which are sequentially connected, and the model input signal is used for training the bidirectional long-short term memory network. The technical scheme of the invention is beneficial to improving the accuracy and efficiency of human body posture recognition.
Description
Technical Field
The invention relates to the technical field of computer application, in particular to a human body posture recognition model establishing method and a human body posture recognition method.
Background
Human gesture recognition is widely used in health monitoring, athletic competition analysis, and human-computer interaction scenarios. The traditional human posture recognition algorithm is divided into two types, one is a statistical machine learning method, for example, the original signals are classified through models such as threshold analysis, a support vector machine or a perception vector machine, and the problems of insufficient feature capability extraction and low recognition accuracy rate exist; the other is that the shallow machine learning algorithm is combined with the signal processing technology, for example, signal features are extracted through wavelet transformation, empirical mode decomposition or fast fourier transformation and then input into a machine learning model for classification, and although the classification effect is still good, the problems of complex extraction process, high filtering delay and incomplete feature extraction exist.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a human body posture recognition model establishment method, a human body posture recognition device and a storage medium.
In a first aspect, the present invention provides a method for establishing a human body posture recognition model, which comprises the following steps:
acquiring human body motion signals under different postures, wherein the source of the human body motion signals comprises intelligent wearable equipment worn on a human body;
preprocessing the human body motion signal to obtain a model input signal;
training a preset model based on the model input signal to obtain a human body posture recognition model, wherein the human body posture recognition model is used for outputting a posture classification result, the preset model comprises a bidirectional long-short term memory network and a support vector machine classifier which are sequentially connected, and the model input signal is used for training the bidirectional long-short term memory network.
In a second aspect, the invention provides a human body posture recognition model establishing device, which comprises a memory and a processor; the memory for storing a computer program; the processor is configured to implement the human body posture identification model building method as described above when the computer program is executed.
In a third aspect, the present invention provides a human body posture identifying method, including the following steps:
acquiring a human body motion signal;
and inputting the human body motion signal into the human body posture recognition model established according to the human body posture recognition model establishing method to obtain a posture classification result.
In a fourth aspect, the present invention provides a human body posture identifying apparatus, comprising a memory and a processor; the memory for storing a computer program; the processor is configured to implement the human body posture recognition method as described above when executing the computer program.
In a fifth aspect, the present invention provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements a human body posture recognition model building method as described above or a human body posture recognition method as described above.
In the invention, the human motion signals can be acquired through intelligent wearable equipment such as an intelligent belt, so that the method can be used for various scenes needing real-time posture recognition, and can be used for training a preset model after the human motion signals are preprocessed, wherein the preset model comprises a bidirectional long-short term memory network and a support vector machine classifier, bidirectional deep feature extraction can be carried out through the bidirectional long-short term memory network, for example, fine features of bidirectional time sequence before and after time can be extracted, the classification performance of a small sample data set can be ensured through the support vector machine classifier, and the recognition precision and efficiency of the model on a posture recognition task are finally obviously improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, a brief description will be given below to the drawings required for the description of the embodiments or the prior art, and it is obvious that the drawings in the following description are some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without inventive exercise.
FIG. 1 is a schematic flow chart of a human body posture recognition model building method according to an embodiment of the present invention;
fig. 2 is a schematic view of an intelligent wearable device according to an embodiment of the present invention;
FIG. 3 is a schematic structural diagram of a human body posture recognition model according to an embodiment of the present invention;
fig. 4 is a flowchart illustrating a human body posture recognition method according to an embodiment of the present invention.
Detailed Description
The principles and features of this invention are described below in conjunction with the following drawings, which are set forth by way of illustration only and are not intended to limit the scope of the invention.
As shown in fig. 1, a method for establishing a human body posture recognition model according to an embodiment of the present invention includes the following steps:
the method comprises the steps of obtaining human motion signals under different postures, wherein the sources of the human motion signals comprise intelligent wearable equipment worn on a human body.
Specifically, intelligence wearing equipment can be the intelligence waistband to guarantee the travelling comfort of wearing and the uniformity of man-machine motion, be favorable to accurate data collection. The main sensing equipment carried by the device comprises an accelerometer and a gyroscope, wherein the accelerometer can be used for collecting acceleration data, and the gyroscope can be used for collecting angular velocity data and angle data to serve as human motion signals. In addition, the human body behavior postures usually comprise 6 types of walking, running, going upstairs, going downstairs, standing and sitting, human body motion signals corresponding to each posture are different, the human body motion signals in specific postures can be collected to form corresponding relations, and the corresponding behavior postures can also be determined through random human body motion signals.
And preprocessing the human body motion signal to obtain a model input signal.
Specifically, after human motion signals in different postures are acquired through, for example, an intelligent belt, information mining or noise reduction processing such as filtering can be performed according to data characteristics. The preprocessed signals are more favorable for establishing an efficient and accurate classification model.
Training a preset model based on the model input signal to obtain a human body posture recognition model, wherein the human body posture recognition model is used for outputting a posture classification result, the preset model comprises a bidirectional long-short term memory network and a support vector machine classifier which are sequentially connected, and the model input signal is used for training the bidirectional long-short term memory network.
Specifically, a bidirectional long-short term memory network (BLSTM) may be trained using the motion signals as samples and the respective corresponding poses as corresponding labels. The BLSTM obtained by training is used as a signal feature extractor, and the BLSTM and a Support Vector Machine (SVM) classifier jointly form a human body posture recognition model.
In this embodiment, the human motion signal can be collected through intelligent wearing equipment such as an intelligent waistband, so that the human motion signal can be used in various scenes needing real-time posture recognition, after the human motion signal is preprocessed, the human motion signal can be used for training a preset model, wherein the preset model comprises a bidirectional long-short term memory network and a support vector machine classifier, bidirectional deep feature extraction can be performed through the bidirectional long-short term memory network, for example, fine features of bidirectional time sequences before and after time can be extracted, the classification performance of a small sample data set can be guaranteed through the support vector machine classifier, and finally, the recognition accuracy and the efficiency of the model on a posture recognition task are remarkably improved.
Optionally, the acquiring human motion signals in different postures includes:
the method comprises the steps of obtaining the human motion signals under walking, running, going upstairs, going downstairs, standing and sitting postures, wherein the human motion signals comprise three-dimensional acceleration data, three-dimensional angular velocity data and three-dimensional angular data collected by the intelligent wearable device.
Specifically, as shown in fig. 2, after the user wears the smart belt, the vertical direction of the view is taken as the X-axis direction, the horizontal direction of the view is taken as the Y-axis direction, and the direction perpendicular to the view is taken as the Z-axis direction. At this time, the intelligent waistband can acquire acceleration data, angular velocity data and angle data of three dimensions of an X axis, a Y axis and a Z axis, and the intelligent waistband comprises 9 motion signals, namely X-axis direction acceleration data, Y-axis direction acceleration data, Z-axis direction acceleration data, X-axis direction angular velocity data, Y-axis direction angular velocity data, Z-axis direction angular velocity data, X-axis direction angle (roll angle) data, Y-axis direction angle (pitch angle) data and Z-axis direction angle (yaw angle) data. When data used for model training is collected, a user can respectively move in walking, running, going upstairs, going downstairs, standing and sitting postures, and human motion signals corresponding to different postures are obtained.
Optionally, the preprocessing the human motion signal to obtain a model input signal includes:
and obtaining three-dimensional gravity acceleration data and three-dimensional linear acceleration data according to the three-dimensional acceleration data.
Specifically, the acquired acceleration data is a superimposed amount of the gravitational acceleration and the linear acceleration. In consideration of the high frequency characteristic of the linear acceleration of the body and the low frequency characteristic of the gravitational acceleration during the exercise, the raw acceleration data is separated by first-order low-pass filtering, and can be calculated as follows.
g i (n)＝αg i (n-1)+(1-α)A i (n) i＝x,y,z；
a i (n)＝A i (n)-g i (n) i＝x,y,z；
Wherein n is time, A i (n) is the sum of the accelerations at time n, g i (n) is a gravitational acceleration component at time n, a i (n) is a linear acceleration at time n, α is a parameter relating to a sampling period and a time constant, and α may be set to 0.8, where X, Y, and Z represent components in the X, Y, and Z-axis directions, respectively.
Obtaining the model input signal based on the three-dimensional acceleration data, the three-dimensional angular velocity data, the three-dimensional angular data, the three-dimensional gravitational acceleration data, and the three-dimensional linear acceleration data.
Specifically, in the upstairs posture, the included angle between the gravity direction and the Z-axis direction is greater than 90 degrees, so that the component of the acceleration data in the Z-axis direction is a negative value; under the posture of going downstairs, the included angle between the gravity direction and the Z-axis direction is smaller than 90 degrees, so the component of the acceleration data in the Z-axis direction is a positive value. The acceleration data is decomposed into the gravity acceleration data and the linear acceleration data, external parameters are not required to be introduced, but the relevant characteristics under different motion postures can be more accurately reflected through the latter two data, so that the latter two data are also introduced into the model input signals, and the precision of the training model is further improved. Therefore, this corresponds to expanding the motion signal from 9 to 15.
Optionally, the preprocessing the human motion signal to obtain a model input signal further includes:
and carrying out moving average filtering on the human motion signal.
Specifically, to ensure the smoothness of the signal variation, the original signal needs to be filtered and denoised first. Compared with filtering modes such as median filtering, Kalman filtering and the like, the moving average filtering method has the advantages of small calculated amount and difficulty in distortion of processed signals, and is more suitable for processing motion sensor sequence signals. Therefore, the moving average filtering is used for carrying out noise reduction processing on the human motion signal, and the filtering process can be calculated according to the following formula.
Wherein x is i Is the original signal value recorded at the ith sample point, y i Is the moving average of the sample point, k represents the window size of the mean filtering, y i The new sequence is the signal sequence after noise reduction.
And performing moving average filtering processing on all signals and setting the size of an optimal filtering window to finish signal denoising processing. The noise signal can be effectively eliminated after the moving average filtering processing, and simultaneously, the main motion information in the original signal can be well reserved.
And carrying out data standardization processing on the human motion signals.
Specifically, because the sensing device is embedded inside the intelligent belt, the sensing device may slightly displace along with the shaking of the belt during the violent exercise, and the acquired data may have slight deviation. Meanwhile, considering that the amplitude differences of the acceleration, the angular velocity and the angle are large, the 15 items of signal data are processed by Min-Max standardization, and can be calculated by the following formula.
Wherein x is i Represents the respective signal values, min represents the minimum value of the sampled sample data, and max represents the maximum value of the sampled sample data.
In addition, the data overlapping sampling can be carried out on the human motion signals.
Optionally, the bidirectional long-short term memory network includes a first full connection layer, a forward LSTM layer, a backward LSTM layer, and a second full connection layer, two outputs of the first full connection layer are respectively connected to the forward LSTM layer and the backward LSTM layer, outputs of the forward LSTM layer and the backward LSTM layer are connected to the second full connection layer after being spliced, and an output of the second full connection layer is used for being connected to the support vector machine classifier.
Specifically, as shown in fig. 3, a bidirectional long-short term memory network (BLSTM) includes a first fully-connected layer, a forward LSTM layer, a backward LSTM layer, and a second fully-connected layer, wherein the forward LSTM layer and the backward LSTM layer respectively include two LSTM units.
The bidirectional long-short term memory network is used as a motion signal feature extractor. In the model training stage, the above 15 motion signals can be used as samples, and the corresponding postures can be used as corresponding labels. BLSTM can perform bidirectional deep feature extraction, in which fine features of bidirectional temporal sequences of motion behaviors around time are mined separately through forward and backward LSTM layers.
After a human body posture recognition model is obtained through training, or in a classification recognition stage, after a human body motion signal is subjected to BLSTM after training, the output of a forward LSTM layer and a reverse LSTM layer at the last moment is spliced, extracted characteristic information is input into an SVM classifier through a full connection layer, a multi-classification task is realized through a soft interval and a kernel function in the SVM classifier, and finally a posture classification result is output to finish posture recognition. The nonlinear SVM classifier replaces a Softmax layer in a neural network model to serve as a behavior attitude classifier, and the accuracy of behavior attitude identification can be improved by utilizing the good classification performance of the SVM in low-dimensional features.
Optionally, the training a preset model based on the model input signal includes:
and combining the output of the forward LSTM layer and the output of the backward LSTM layer at the last moment with a Softmax classifier, updating the parameters of each layer of neural network in the forward LSTM layer and the backward LSTM layer through a back propagation algorithm along time, and finishing the training of the preset model by utilizing a minimum loss function.
Specifically, in the model training process, the BLSTM and the Softmax intermediate layer are combined, the probability value of 6 types of behavior postures is output, corresponding indexes are output through maximum likelihood estimation to serve as posture classification results, parameters of each internal layer of neural network are updated through a time back propagation algorithm (BPTT), the training of the model is completed through a minimum loss function, and the model parameters are stored.
The calculation of the Softmax function is shown below.
Wherein i is a behavior gesture number, y i Is the probability of the output.
The BLSTM realizes training through weight updating of each internal network layer, and sets a cross entropy function as a loss function aiming at a multi-classification task of gesture recognition, and the calculation is shown as the following formula.
Wherein the content of the first and second substances,
for each training sample's label, y i Is the output value, or output probability, of the classification result.
In the actual training process, the hyper-parameters of the model need to be set to finish the model training and tuning. Specifically, the optimal hyperparameters may be determined by a grid search method, including dropout coefficient setting and neuron number selection. Due to a complex structure, the neural network often has an overfitting phenomenon in the training process, and except for setting regularization in a loss function, a drop layer can be added to set neuron discarding probability mitigation. Through comparison, when a dropout layer is added to each middle layer and the discarding probability is set to be 0.2, the identification accuracy of the model is stable and the overfitting is avoided. In addition, the neuron number has a significant influence on the feature extraction capability of the model, not only directly influences the training precision of the model, but also influences the training duration of the model, and the neuron number of each LSTM layer is set to 64 in consideration of the accuracy, the overfitting condition and the time consumption of operation caused by the model parameter quantity. In order to obtain the best training effect, all the hyper-parameters are subjected to traversal combination in a grid search mode, and the best hyper-parameter combination is determined according to the training effect. The human posture recognition model hyper-parameter settings are shown in table 1.
TABLE 1
The design and training of the human posture recognition model can be completed through model structure adjustment and super-parameter tuning. In order to verify the training effect of the model, the human motion signals collected by the intelligent waistband and the UCI-HAR data set can be respectively used for verification. The UCI-HAR data set is a public data set of California university, and comprises sensor signal data of 30 volunteers in 6 types of behavioral postures of walking, going upstairs, going downstairs, sitting, standing and lying. The original signals can be input into the human body posture recognition model for comparison and verification after corresponding preprocessing.
Since the recognition of the human body posture belongs to the classification task, Precision, Recall, F1-score (F1 value), and accuracy can be used as evaluation indexes, and the calculation is obtained by 4 kinds of sample statistics, which are:
(1) true Positive (TP): the classification result is a positive sample positive case.
(2) True Negative (True Negative, TN): the classification result is a negative sample negative case.
(3) False Positive (FP): the classification result is a positive sample negative case.
(4) False Negative (FN): the classification result is a negative sample positive case.
Calculating the sample statistic of each posture to obtain each evaluation index, which is shown as the following formula:
for human body posture classification, the traditional machine learning method comprises MLP, random forest, decision tree, KNN, SVM and the like. In order to more intuitively compare the effects of the various methods, the methods and the method of the invention are compared by using the same UCI-HAR data set, and finally the average accuracy, the average precision and the average recall rate are obtained, and the corresponding results are shown in Table 2.
TABLE 2
The human posture recognition model has the best performance in each evaluation index, the average accuracy rate, the average recall rate and the average accuracy rate all exceed 95%, and the performance in the UCI-HAR data set is comprehensively superior to that of the traditional machine learning method. The model deeply excavates the fine characteristic difference of each behavior motion signal in time sequence through a bidirectional cyclic neural network, and the traditional machine learning method relies on a single characteristic extraction mode to capture only the local characteristic information of the sensor signal, so that the classification performance level is general. Meanwhile, the human body posture recognition model can directly recognize the human body posture through the motion time sequence signal segment after training is finished, end-to-end input and output are achieved, and the human body posture recognition model is comprehensively superior to a machine learning method in this respect.
Further, the human body posture recognition model of the invention is compared with the posture recognition model designed by researchers at home and abroad on the basis of the neural network on the UCI-HAR data set at present, and the comparison result is shown in Table 3.
TABLE 3
It can be seen that the recognition accuracy of the human body posture recognition model is the highest, and reaches 95.76%, and the classification effect of the human body posture recognition model is superior to the effect shown in corresponding documents of researchers in table 3. The human body posture recognition model excavates the time sequence signal characteristics of the previous moment, the current moment and the next moment in the characteristic learning process, and the CNN and the LSTM do not consider the signal characteristics of the next moment. In addition, during classification and judgment, the method can fully combine the excellent performance of BLSTM in time sequence signal feature extraction and the classification performance advantage of an SVM classifier in a small sample data set, so that the identification accuracy of the model on an attitude identification task is remarkably improved.
In another embodiment of the present invention, a human body posture recognition model building apparatus includes a memory and a processor; the memory for storing a computer program; the processor is configured to implement the human body posture identification model building method as described above when the computer program is executed.
It should be noted that the device may be a computer device such as a server or a mobile terminal.
In another embodiment of the present invention, a computer-readable storage medium has stored thereon a computer program which, when executed by a processor, implements the human body posture recognition model building method as described above.
As shown in fig. 4, a human body posture recognition method according to an embodiment of the present invention includes the following steps:
and acquiring a human body motion signal.
Specifically, the human motion signal of the user can be acquired through an intelligent wearable device such as an intelligent belt. Optionally, the human body operation signal is input into the human body posture recognition model after being correspondingly preprocessed.
And inputting the human body motion signal into the human body posture recognition model established according to the human body posture recognition model establishing method to obtain a posture classification result.
Specifically, after the human motion signal is input into the human posture recognition model, the BLSTM is used as a signal feature extractor, the SVM is used as a classifier, and a posture classification result can be output.
In another embodiment of the present invention, a human body posture recognition apparatus includes a memory and a processor; the memory for storing a computer program; the processor is configured to implement the human body posture recognition method as described above when executing the computer program.
It should be noted that the device may be a computer device such as a server or a mobile terminal.
In another embodiment of the present invention, a computer-readable storage medium has stored thereon a computer program which, when executed by a processor, implements the human body posture recognition method as described above.
The reader should understand that in the description of this specification, reference to the description of the terms "one embodiment," "some embodiments," "an example," "a specific example" or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.
Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention, and that variations, modifications, substitutions and alterations can be made to the above embodiments by those of ordinary skill in the art within the scope of the present invention.
Claims (10)
1. A human body posture recognition model building method is characterized by comprising the following steps:
acquiring human body motion signals under different postures, wherein the source of the human body motion signals comprises intelligent wearable equipment worn on a human body;
preprocessing the human body motion signal to obtain a model input signal;
training a preset model based on the model input signal to obtain a human body posture recognition model, wherein the human body posture recognition model is used for outputting a posture classification result, the preset model comprises a bidirectional long-short term memory network and a support vector machine classifier which are sequentially connected, and the model input signal is used for training the bidirectional long-short term memory network.
2. The human body posture identification model building method according to claim 1, wherein the obtaining human body motion signals under different postures comprises:
the method comprises the steps of obtaining the human motion signals under walking, running, going upstairs, going downstairs, standing and sitting postures, wherein the human motion signals comprise three-dimensional acceleration data, three-dimensional angular velocity data and three-dimensional angular data collected by the intelligent wearable device.
3. The human body posture recognition model building method according to claim 2, wherein the preprocessing the human body motion signal to obtain a model input signal comprises:
obtaining three-dimensional gravity acceleration data and three-dimensional linear acceleration data according to the three-dimensional acceleration data;
obtaining the model input signal based on the three-dimensional acceleration data, the three-dimensional angular velocity data, the three-dimensional angular data, the three-dimensional gravitational acceleration data, and the three-dimensional linear acceleration data.
4. The human body posture identification model building method of claim 3, wherein the preprocessing the human body motion signal to obtain a model input signal further comprises:
carrying out moving average filtering on the human motion signal;
and carrying out data standardization processing on the human motion signals.
5. The human body posture identification model building method according to any one of claims 1 to 4, wherein the bidirectional long-short term memory network comprises a first full connection layer, a forward LSTM layer, a backward LSTM layer and a second full connection layer, two outputs of the first full connection layer are respectively connected to the forward LSTM layer and the backward LSTM layer, outputs of the forward LSTM layer and the backward LSTM layer are connected to the second full connection layer after being spliced, and an output of the second full connection layer is used for being connected to the SVM classifier.
6. The human body posture recognition model building method of claim 5, wherein the training of the preset model based on the model input signal comprises:
and combining the output of the forward LSTM layer and the output of the backward LSTM layer at the last moment with a Softmax classifier, updating the parameters of each layer of neural network in the forward LSTM layer and the backward LSTM layer through a back propagation algorithm along time, and finishing the training of the preset model by utilizing a minimum loss function.
7. A human body posture recognition model establishing device is characterized by comprising a memory and a processor;
the memory for storing a computer program;
the processor, when executing the computer program, is configured to implement the human body posture recognition model building method according to any one of claims 1 to 6.
8. A human body posture recognition method is characterized by comprising the following steps:
acquiring a human body motion signal;
inputting the human motion signal into a human posture recognition model established according to the human posture recognition model establishing method of any one of claims 1 to 6 to obtain a posture classification result.
9. The human body posture recognition device is characterized by comprising a memory and a processor;
the memory for storing a computer program;
the processor, when executing the computer program, for implementing the human gesture recognition method of claim 8.
10. A computer-readable storage medium, characterized in that the storage medium has stored thereon a computer program which, when being executed by a processor, carries out the human posture recognition model building method of any one of claims 1 to 6 or the human posture recognition method of claim 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210565324.4A CN114897025A (en) | 2022-05-23 | 2022-05-23 | Human body posture recognition model establishing method and human body posture recognition method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210565324.4A CN114897025A (en) | 2022-05-23 | 2022-05-23 | Human body posture recognition model establishing method and human body posture recognition method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114897025A true CN114897025A (en) | 2022-08-12 |
Family
ID=82723028
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210565324.4A Pending CN114897025A (en) | 2022-05-23 | 2022-05-23 | Human body posture recognition model establishing method and human body posture recognition method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114897025A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116108391A (en) * | 2023-04-12 | 2023-05-12 | 江西珉轩智能科技有限公司 | Human body posture classification and recognition system based on unsupervised learning |
| CN117218728A (en) * | 2023-11-09 | 2023-12-12 | 深圳市微克科技有限公司 | Body posture recognition method, system and medium of intelligent wearable device |
| CN117216644A (en) * | 2023-11-09 | 2023-12-12 | 北京世纪慈海科技有限公司 | Human body posture recognition method and device based on electric digital data processing |
| CN116226702B (en) * | 2022-09-09 | 2024-04-26 | 武汉中数医疗科技有限公司 | Thyroid sampling data identification method based on bioelectrical impedance |
-
2022
- 2022-05-23 CN CN202210565324.4A patent/CN114897025A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116226702B (en) * | 2022-09-09 | 2024-04-26 | 武汉中数医疗科技有限公司 | Thyroid sampling data identification method based on bioelectrical impedance |
| CN116108391A (en) * | 2023-04-12 | 2023-05-12 | 江西珉轩智能科技有限公司 | Human body posture classification and recognition system based on unsupervised learning |
| CN117218728A (en) * | 2023-11-09 | 2023-12-12 | 深圳市微克科技有限公司 | Body posture recognition method, system and medium of intelligent wearable device |
| CN117216644A (en) * | 2023-11-09 | 2023-12-12 | 北京世纪慈海科技有限公司 | Human body posture recognition method and device based on electric digital data processing |
| CN117216644B (en) * | 2023-11-09 | 2024-02-02 | 北京世纪慈海科技有限公司 | Human body posture recognition method and device based on electric digital data processing |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114897025A (en) | Human body posture recognition model establishing method and human body posture recognition method | |
| CN109670548B (en) | Multi-size input HAR algorithm based on improved LSTM-CNN | |
| Xi et al. | Deep dilated convolution on multimodality time series for human activity recognition | |
| Wang et al. | Event-stream representation for human gaits identification using deep neural networks | |
| WO2018040757A1 (en) | Wearable device and method of using same to monitor motion state | |
| US8000500B2 (en) | System and method for analyzing of human motion based on silhouettes of real time video stream | |
| Uddin et al. | A guided random forest based feature selection approach for activity recognition | |
| Uddin et al. | Human activity recognition from wearable sensors using extremely randomized trees | |
| CN108245172B (en) | Human body posture recognition method free of position constraint | |
| CN104484890B (en) | Video target tracking method based on compound sparse model | |
| CN105184325A (en) | Human body action recognition method and mobile intelligent terminal | |
| CN112464738B (en) | Improved naive Bayes algorithm user behavior identification method based on mobile phone sensor | |
| CN111291865B (en) | Gait recognition method based on convolutional neural network and isolated forest | |
| CN116226691A (en) | Intelligent finger ring data processing method for gesture sensing | |
| CN113763424B (en) | Real-time intelligent target detection method and system based on embedded platform | |
| CN111723662B (en) | Human body posture recognition method based on convolutional neural network | |
| Abdu-Aguye et al. | VersaTL: Versatile Transfer Learning for IMU-based Activity Recognition using Convolutional Neural Networks. | |
| CN107688828B (en) | Bus crowding degree estimation method based on mobile phone sensor | |
| CN106562771B (en) | A kind of pet sleep recognition methods of Embedded platform | |
| Li et al. | Human motion representation and motion pattern recognition based on complex fuzzy theory | |
| CN108257148B (en) | Target suggestion window generation method of specific object and application of target suggestion window generation method in target tracking | |
| CN114601455A (en) | Motion recognition method based on two-stage neural network | |
| CN114821765A (en) | Human behavior recognition method based on fusion attention mechanism | |
| CN114327045A (en) | Fall detection method and system based on category unbalanced signals | |
| CN113057628A (en) | Inertial sensor based motion capture 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 |