CN115410267A - Statistical algorithm based on interaction action analysis data of human skeleton and muscle - Google Patents
Statistical algorithm based on interaction action analysis data of human skeleton and muscle Download PDFInfo
- Publication number
- CN115410267A CN115410267A CN202210841160.3A CN202210841160A CN115410267A CN 115410267 A CN115410267 A CN 115410267A CN 202210841160 A CN202210841160 A CN 202210841160A CN 115410267 A CN115410267 A CN 115410267A
- Authority
- CN
- China
- Prior art keywords
- human body
- human
- action
- data
- motion
- 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.)
- Withdrawn
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to the technical field of human body motion analysis data statistics, and discloses a motion analysis data statistical algorithm based on the mutual matching of human bones and muscles, which comprises the following steps: s1, extracting human body action features; s2, constructing a human behavior action model; s3, recognizing local actions of the human body, and constructing a neural network model; and S4, analyzing and counting human body action data. The statistical algorithm is based on the interaction action analysis data of human skeleton and muscle, and is extracted through human action features; constructing a human behavior action model; recognizing local actions of a human body, and constructing a neural network model; the human body movement data analysis and statistics, the four steps are used for carrying out feature extraction on the movement of the human body during movement and carrying out statistical analysis on the data, the whole steps are simple, the efficiency is high, meanwhile, the calculation method is simple and easy, the posture of the human body during movement can be effectively analyzed and calculated, and a basis is provided for the optimal posture of the human body during movement.
Description
Technical Field
The invention relates to the technical field of human body motion analysis data statistics, in particular to a motion analysis data statistical algorithm based on the mutual matching of human bones and muscles.
Background
The human body realizes division of upper and lower limb activities in the long evolution process, the motion of the human body is in various postures due to the activity of the body, the shape and posture of the external expression of the limb motion is called as the motion form of motion, the basic motion form of the human body can be mainly summarized into the motion forms of pushing, pulling, whipping, buffering, pedaling and stretching, swinging, twisting, opposite motion and the like, and the limb motion can be formed through mutual matching of human skeletons and muscles.
Along with the increasing importance of people on physical health in recent years, the development of sports activities is also more and more emphasized by the country, the national fitness is promoted to be the national strategy, running is the most common fitness activity in recent years, so that the number of marathon projects in the country in 2014 is only 51, the number of marathon projects in 2014 is increased to be nearly 1500 in 2018, the annual growth rate exceeds 700%, however, on average, the number of injured people in all running populations accounts for 40% -70% every year, and researches show that 55% of running people are injured in one city marathon game, and the other research shows a remarkable result; in order to participate in the marathon competition every year, 90% of runners are injured during training, the root cause of the injury is running posture irregularity, and therefore, the running posture analysis is performed on the running population through computer vision, the data are analyzed and counted, and meanwhile, the guiding opinions are provided for the running population, so that the method has great significance.
Disclosure of Invention
Technical problem to be solved
Aiming at the defects of the prior art, the invention provides a statistical algorithm for analyzing data based on the mutual matching action of human bones and muscles, which has the advantages of analyzing the action of a human body and the like and solves the problems of difficult analysis of the existing human body limb action, low efficiency and high cost.
(II) technical scheme
In order to realize the purpose of analyzing the human body action, the invention provides the following technical scheme: a statistical algorithm based on the interaction motion analysis data of human bones and muscles comprises the following steps:
s1, extracting human body action features;
s2, constructing a human behavior action model;
s3, recognizing local actions of the human body, and constructing a neural network model;
and S4, analyzing and counting human body action data.
Preferably, the human body action feature extraction realizes the construction of the multi-scale local space-time domain features through the following steps:
a1, local characteristics and detection description of space-time interest points, and d is drawn up i Representing space-time interest points of the local features and the human body action video V;
a2, constructing S scale domain features by using local features, and replacing human body space-time motion shapes in the video V with sets
Wherein f represents a domain feature set factor;
a3, constructing respective codebooks for the domain characteristics in all scales, and then respectively coding to obtain a coding coefficient c;
a4, utilizing an absolute value average pooling technology
Conversion into pooling vector H i ；
A5, finally cascading pooling vectors under all sizes to obtain H V 。
Wherein, the pool vector under the human motion video V is represented according to H V The support vector machine of the kernel is used as a motion classifier to lift the motion characteristics of the human body, and the extraction process is as follows:
in the formula, H m And H n For two kinds of feature vectors, H m (i)H n (j) Are respectively H m And H n I and j feature components.
Preferably, the method for constructing the human body behavior action model adopts a kalman filtering method to evaluate coordinate information data of the human body joints, based on a previous state sequence of a target, then performs optimal evaluation on a subsequent state by using a meta-deviation, and predicts by using a recursion mode, wherein a specific algorithm formula is as follows:
x k ＝Fg k-1 +Bu k +W k
wherein x is k ∈R n Is a human body action state variable, g k-1 Is the previous human body motion state vector, F is the transfer matrix, B represents the control matrix, u k Represents a control input, W k Is process noise.
Calculating by using a Kalman filtering principle, in a forecasting stage, generating a current time state from a previous time state as a forecasting value, after entering an analysis stage, carrying out state reanalysis on observation data by using a minimum variance estimation method, and continuously advancing the whole process along with continuous state forecasting and continuous input of new observation data to realize the construction of a human body behavior action model, wherein the human body local action can be identified by using the model.
Preferably, the step of constructing the neural network model recognition classification includes:
b1, utilizing windowing to process the action signal transmitted from the sensor;
b2, calculating the following human body motion vectors in the time window: including centroid frequency, ringing count, rise time, wavelet coefficient characteristics, duration, window frequency domain characteristic peak and energy;
b3, constructing an initial neural network model;
b4, taking the characteristic parameters as input neurons, and inputting the input neurons into a neural network for training:
and B5, carrying out quantitative recognition on the classification of the test set by using the trained neural network, and calculating the accuracy of motion recognition.
The mathematical relation expression of the neural network is
Wherein, y j Representing hidden layer output state vector, x i Representing the input state vector, w hij Representing the value of the connection of the input layer to the hidden layer, b j Threshold representing hidden layer neurons, f y Representing an excitation function, using the constructed spiritAnd quantitatively identifying the local behaviors of the human body through a network.
Preferably, the human body motion data is analyzed and counted, and the human body motion is classified based on the extracted data, and the classification accuracy is calculated as follows:
wherein v is i i denotes the classification accuracy, Y i Representing the number of classes, and U representing the total number of human motion classes.
Preferably, the human body local action recognition further comprises quantitative recognition, 16 feature vectors are used for input of the neural network, 16 network input neurons are provided, 8 hidden character neurons are provided, 6 action states are network output, 6 output neurons are provided, sample data are randomly labeled, and then the sample data are divided into two parts to operate, wherein one part is a training set, and the other parts are a testing set.
Preferably, the analysis of the human body motion data on the difference of the extracted number of the human body motions is an index for determining the effect of the recognition method, the recognition effect is judged according to the extracted number of the data, the extracted data number is large, and the recognition effect is good to a certain extent.
(III) advantageous effects
Compared with the prior art, the invention provides a statistical algorithm based on the interaction analysis data of human skeleton and muscle, which has the following beneficial effects:
the statistical algorithm is based on the interaction action analysis data of human skeleton and muscle, and is extracted through human action features; constructing a human behavior action model; identifying local actions of a human body, and constructing a neural network model; the method comprises four steps of analyzing and counting human body movement data, extracting characteristics of movement of a human body during movement, and performing statistical analysis on the data, has simple integral steps and high efficiency, can effectively analyze and calculate postures of the human body during movement, and provides a basis for the optimal postures of the human body during movement.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.
The embodiment provides a statistical algorithm for analysis data based on the mutual matching action of human bones and muscles, which is characterized by comprising the following steps of:
s1, extracting human body action features;
s2, constructing a human behavior action model;
s3, recognizing local actions of the human body, and constructing a neural network model;
and S4, analyzing and counting human body action data.
The human body action feature extraction realizes the construction of multi-scale local space-time domain features through the following steps:
a1, local characteristics and detection description of space-time interest points, and d is drawn up i Representing space-time interest points of the local features and the human motion video V;
a2, constructing S scale domain features by using local features, and replacing human body space-time motion shapes in the video V with sets
Wherein f represents a domain feature set factor;
a3, constructing respective codebooks for the domain features in all scales, and then respectively coding to obtain a coding coefficient c;
a4, utilizing an absolute value average pooling technology
Converted into pooling vector H i ；
A5, finally cascading pooling vectors under all sizes to obtain H V 。
Wherein, the pool vector under the human motion video V is represented according to H V The kernel support vector machine is used as a motion classifier to lift the motion characteristics of the human body, and the extraction process is as follows:
in the formula, H m And H n For two kinds of feature vectors, H m (i)H n (j) Are each H m And H n I and j feature components.
Meanwhile, a human body behavior action model is constructed, the coordinate information data of the human body joints are evaluated by adopting a Kalman filtering method, the state sequence before the target is taken as the basis, then the optimal evaluation of the element deviation is carried out on the later state, the prediction is carried out by utilizing a recursion mode, and the specific algorithm formula is as follows:
x k ＝Fg k-1 +Bu k +W k
wherein x is k ∈R n Is a human body motion state variable, g k-1 Is the previous human body motion state vector, F is the transfer matrix, B represents the control matrix, u k Represents a control input, W k Is process noise.
Calculating by using a Kalman filtering principle, in a forecasting stage, generating a current time state from a previous time state as a forecasting value, after entering an analysis stage, carrying out state reanalysis on observation data by using a minimum variance estimation method, and continuously advancing the whole process along with continuous state forecasting and continuous input of new observation data to realize the construction of a human body behavior action model, wherein the human body local action can be identified by using the model.
In addition, the human body motion data is analyzed and counted, the human body motion is classified on the basis of the extracted data, and the classification precision is calculated as follows:
wherein v is i i denotes the classification accuracy, Y i Representing the number of classes, and U representing the total number of human motion classes.
It should be noted that, the human body local motion recognition also includes quantitative recognition, 16 feature vectors are used for the input of the neural network, 16 network input neurons are provided, 8 hidden character neurons are provided, 6 action states are network output, 6 output neurons are provided, sample data are randomly labeled, and then the sample data are divided into two parts to operate, wherein one part is a training set, and the other parts are a testing set.
It can be understood that the difference of human body motion data analysis on the extracted number of human body motions is an index for determining the effect of the recognition method, the recognition effect is judged according to the extracted number of the data, the extracted data number is large, and the recognition effect is good to a certain extent.
The local action of the human body is a non-stationary signal, and can change along with the change of time when feature statistics is carried out, while the traditional feature point extraction is almost calculated through the aspect of time-frequency domain, then the time domain ten-thousandth square data features of action signal data, namely, the mean value, the maximum amplitude and the standard deviation are extracted for identification or are considered through the frequency domain, and the power spectrum energy, the frequency absolute value and the like of the action signal data are analyzed.
The invention has the beneficial effects that:
extracting through human body action characteristics; constructing a human behavior action model; recognizing local actions of a human body, and constructing a neural network model; the human body movement data analysis and statistics, the four steps are used for carrying out feature extraction on the movement of the human body during movement and carrying out statistical analysis on the data, the whole steps are simple, the efficiency is high, the posture of the human body during movement can be effectively analyzed and calculated, and a basis is provided for the optimal posture of the human body during movement.
Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.
Claims (7)
1. A statistical algorithm based on the interaction between human skeleton and muscle is characterized by comprising the following steps:
s1, extracting human body action features;
s2, constructing a human behavior action model;
s3, recognizing local actions of the human body, and constructing a neural network model;
and S4, analyzing and counting the human body action data.
2. The statistical algorithm for analysis data based on the interaction of human skeleton and muscle actions as claimed in claim 1, wherein the human action feature extraction is implemented by constructing multi-scale local spatiotemporal domain features through the following steps:
a1, local characteristics and detection description of space-time interest points, and d is drawn up i Representing space-time interest points of the local features and the human body action video V;
a2, constructing S scale domain features by using local features, and replacing human body space-time motion shapes in the video V with sets
Wherein f represents a domain feature set factor;
a3, constructing respective codebooks for the domain characteristics in all scales, and then respectively coding to obtain a coding coefficient c;
a4, utilizing an absolute value average pooling technology
Conversion into pooling vector H i ；
A5, finally cascading pooling vectors under all sizes to obtain H V 。
Wherein, the pool vector of the human motion video V is represented according to H V Support vector machine of kernel as action classifierThe human body action characteristics are lifted, and the extraction process is as follows:
in the formula, H m And H n For two kinds of feature vectors, H m (i)H n (j) Are respectively H m And H n I and j feature components.
3. The statistical algorithm for analysis data based on the interaction between human bones and muscles according to claim 1, wherein the constructed human behavior action model adopts a Kalman filtering method to evaluate coordinate information data of human joints, and based on a state sequence before a target, the subsequent state is optimally evaluated for element deviation, and the prediction is performed in a recursion mode, and the specific algorithm formula is as follows:
x k ＝Fg k-1 +Bu k +W k
wherein x is k ∈R n Is a human body motion state variable, g k-1 Is the previous human body motion state vector, F is the transfer matrix, B represents the control matrix, u k Represents a control input, W k Is process noise.
The method comprises the steps of utilizing a Kalman filtering principle to calculate, utilizing a forecasting stage to generate a current time state from a previous time state as a forecasting value, utilizing a minimum variance estimation method to analyze the state of observation data after entering an analysis stage, and continuously advancing the whole process along with the continuous progress of state forecasting and the continuous input of new observation data to realize the construction of a human body behavior action model, wherein the human body local action can be identified by utilizing the model.
4. The statistical algorithm for analyzing data based on the interaction between human bones and muscles as claimed in claim 1, wherein the step of constructing neural network model recognition classification comprises:
b1, utilizing windowing to process the action signal transmitted from the sensor;
b2, calculating the following human body motion vectors in the time window: including centroid frequency, ringing count, rise time, wavelet coefficient characteristics, duration, window frequency domain characteristic peak and energy;
b3, constructing an initial neural network model;
b4, taking the characteristic parameters as input neurons, and inputting the input neurons into a neural network for training:
and B5, quantitatively identifying the classification of the test set by using the trained neural network, and calculating the accuracy of motion identification.
The mathematical relation expression of the neural network is
Wherein, y j Representing hidden layer output state vector, x i Representing the input state vector, w hij Representing the value of the connection of the input layer to the hidden layer, b j Threshold representing hidden layer neurons, f y Representing the excitation function, and quantitatively identifying the local behavior of the human body by using the constructed neural network.
5. The statistical algorithm for human body bone and muscle interaction based motion analysis data as claimed in claim 1, wherein the human body motion data analysis and statistics classify the human body motion based on the extracted data, and the classification accuracy is calculated as follows:
wherein v is i i denotes the classification accuracy, Y i Representing the number of classes, U representing the total number of human action classes.
6. The statistical algorithm for analyzing data based on the interaction between human skeleton and muscle according to claim 1, wherein the recognition of human local motion further comprises quantitative recognition, 16 feature vectors are used for the input of neural network, 16 network input neurons are provided, 8 hidden character neurons are provided, 6 action states are provided for network output, 6 output neurons are provided, the sample data are randomly labeled, and then the operation is divided into two parts, one part is training set, and the other part is testing set.
7. The statistical algorithm for human body bone and muscle interaction based motion analysis data as claimed in claim 1, wherein the difference of the human body motion data analysis to the extracted number of human body motions is an index for determining the recognition method effect, the recognition effect is determined according to the extracted number of the data, the extracted data number is large, and the recognition effect is good to a certain extent.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210841160.3A CN115410267A (en) | 2022-07-18 | 2022-07-18 | Statistical algorithm based on interaction action analysis data of human skeleton and muscle |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210841160.3A CN115410267A (en) | 2022-07-18 | 2022-07-18 | Statistical algorithm based on interaction action analysis data of human skeleton and muscle |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115410267A true CN115410267A (en) | 2022-11-29 |
Family
ID=84156798
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210841160.3A Withdrawn CN115410267A (en) | 2022-07-18 | 2022-07-18 | Statistical algorithm based on interaction action analysis data of human skeleton and muscle |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115410267A (en) |
Cited By (1)
| 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 |
-
2022
- 2022-07-18 CN CN202210841160.3A patent/CN115410267A/en not_active Withdrawn
Cited By (2)
| 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 |
| CN116108391B (en) * | 2023-04-12 | 2023-06-30 | 江西珉轩智能科技有限公司 | Human body posture classification and recognition system based on unsupervised learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Quaid et al. | Wearable sensors based human behavioral pattern recognition using statistical features and reweighted genetic algorithm | |
| CN108133188A (en) | A kind of Activity recognition method based on motion history image and convolutional neural networks | |
| CN103942568B (en) | A kind of sorting technique based on unsupervised feature selection | |
| CN110163127A (en) | A kind of video object Activity recognition method from thick to thin | |
| Ghazali et al. | Common sport activity recognition using inertial sensor | |
| CN103886341A (en) | Gait behavior recognition method based on feature combination | |
| CN104598880A (en) | Behavior identification method based on fuzzy support vector machine | |
| CN108446733A (en) | A kind of human body behavior monitoring and intelligent identification Method based on multi-sensor data | |
| CN108717548B (en) | Behavior recognition model updating method and system for dynamic increase of sensors | |
| CN110490109A (en) | A kind of online human body recovery action identification method based on monocular vision | |
| CN109979161A (en) | A kind of tumble detection method for human body based on convolution loop neural network | |
| US20230149774A1 (en) | Handle Motion Counting Method and Terminal | |
| CN111640483A (en) | Fitness scheme recommendation method based on AKC model | |
| CN112396001A (en) | Rope skipping number statistical method based on human body posture estimation and TPA (tissue placement model) attention mechanism | |
| CN115410267A (en) | Statistical algorithm based on interaction action analysis data of human skeleton and muscle | |
| Fan et al. | Hybrid lightweight Deep-learning model for Sensor-fusion basketball Shooting-posture recognition | |
| CN110008847B (en) | Swimming stroke identification method based on convolutional neural network | |
| CN114384999A (en) | User irrelevant myoelectricity gesture recognition system based on self-adaptive learning | |
| CN116747495A (en) | Action counting method and device, terminal equipment and readable storage medium | |
| Ahmed et al. | Improving prediction of plant disease using k-efficient clustering and classification algorithms | |
| Jia et al. | Deep learning with convolutional neural networks for sleep arousal detection | |
| CN114118167B (en) | Action sequence segmentation method aiming at behavior recognition and based on self-supervision less sample learning | |
| Zhou et al. | A Swarm Intelligence Assisted IoT-Based Activity Recognition System for Basketball Rookies | |
| CN111274908B (en) | Human body action recognition method | |
| CN114613015A (en) | Body-building action image identification method, device, equipment and storage medium |
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 | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20221129 |