CN110222634B - Human body posture recognition method based on convolutional neural network - Google Patents

Human body posture recognition method based on convolutional neural network Download PDF

Info

Publication number
CN110222634B
CN110222634B CN201910481323.XA CN201910481323A CN110222634B CN 110222634 B CN110222634 B CN 110222634B CN 201910481323 A CN201910481323 A CN 201910481323A CN 110222634 B CN110222634 B CN 110222634B
Authority
CN
China
Prior art keywords
neural network
human body
convolutional neural
body posture
layer
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.)
Active
Application number
CN201910481323.XA
Other languages
Chinese (zh)
Other versions
CN110222634A (en
Inventor
李建
张袁
罗颖
张亦昕
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Changzhou Campus of Hohai University
Original Assignee
Changzhou Campus of Hohai University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Changzhou Campus of Hohai University filed Critical Changzhou Campus of Hohai University
Priority to CN201910481323.XA priority Critical patent/CN110222634B/en
Publication of CN110222634A publication Critical patent/CN110222634A/en
Application granted granted Critical
Publication of CN110222634B publication Critical patent/CN110222634B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • 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
    • 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
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/20—Movements or behaviour, e.g. gesture recognition
    • G06V40/23—Recognition of whole body movements, e.g. for sport training

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Molecular Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Psychiatry (AREA)
  • Social Psychology (AREA)
  • Human Computer Interaction (AREA)
  • Multimedia (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a human body posture identification method based on a convolutional neural network, which comprises the steps of firstly, acquiring a human body posture data set, and preprocessing the human body posture data set by cutting a video into image frames; then, a convolutional neural network model is built, and unnecessary input of an excitation function is reduced by introducing sparsity at the input position of the RELU excitation function; then, optimizing a traditional target loss function by combining a sparse term, and training the network by iteratively updating parameters so as to obtain an optimal solution; and finally, according to the network model obtained by training, recognizing the human body posture and outputting the human body posture category. The invention has the beneficial effects that: the method adopted by the invention can accelerate the convergence speed and improve the generalization capability of the network while keeping higher gesture recognition rate.

Description

Human body posture recognition method based on convolutional neural network
Technical Field
The invention belongs to the technical field of computer vision, and particularly relates to a human body posture identification method based on a convolutional neural network.
Background
In recent years, with the development of information technology and the popularization of artificial intelligence technology, human body posture recognition technology has come into wide use. Relevant researchers have attempted to explore valid features and classify using collected body posture data sets. The traditional gesture recognition method mainly comprises two steps: (1) Extracting complex artificial features from an original input image; and (2) training a classifier from the acquired features.
In the conventional gesture recognition process, complex artificial features need to be extracted from the original input image. Although effective at recognition accuracy, the feature sizes extracted from bone keypoints and depth image information tend to be relatively high due to the high complexity of the human body. Most depth images need to be preprocessed, so that the feature extraction is difficult, the recognition efficiency is low, and the convergence time is long. The conventional method is not an optimal method.
However, compared with the traditional local combined modeling method, the artificial neural network has nonlinear modeling and self-adaptive capacity, and can perform posture recognition by training a large number of static images and mining deep-level information of the images; the neural network method is used for representing the local characteristics of the objects in the image, and the classification is more robust. In addition, by introducing sparse regularization, unnecessary increase of excitation function input is reduced, the complexity of the model is reduced, and the generalization capability of the convolutional neural network is improved, so that the convergence speed of the model is improved while the higher recognition rate is ensured.
Disclosure of Invention
Aiming at the defects in the prior art, the invention aims to provide a human body posture recognition method based on a convolutional neural network, and through designing a convolutional neural network model introducing sparse regularization, the high recognition rate is ensured, the convergence rate is improved, and the generalization capability of the model is enhanced.
In order to achieve the purpose, the invention is realized by the following technical scheme:
in the training stage, a data-driven self-learning complex network structure model is constructed, the original human body posture image and the corresponding posture label block are used as the input and the output of the network, and the network is subjected to the learning in a supervised instructor mode. In the verification stage, given an unknown input raw image, pose recognition is performed.
A human body posture identification method based on a convolutional neural network comprises the following steps: s01, acquiring a human body posture video data set, preprocessing the human body posture video data set by cutting a video into image frames, and dividing the image data set cut into the image frames into a training set and a verification set;
s02, constructing a neural network model, introducing sparsity at the input of the RELU excitation function, inputting the convolutional neural network into the image preprocessed in the step S01, and outputting the image as a human body posture category; training the convolutional neural network;
s03, recognizing the human body posture by adopting the neural network model in the S02, and performing model training and performance testing on a public human body posture data set KTH; when an unknown video is input, firstly calling the step S01 to carry out preprocessing, and then carrying out posture recognition by using the neural network model in the step S02 to obtain the human body posture category.
In the above human body posture identification method based on the convolutional neural network, in the step S01, the acquiring of the human body posture video data set specifically includes the following steps:
s11, acquiring a public KTH human posture video data set;
s12, cutting the video into frames and storing the images of each frame;
s13, screening out images of human body behaviors with complete human bodies and corresponding actions of labels from the images, deleting blank images or images without complete human bodies or corresponding actions of labels, and classifying and marking the screened images; the corresponding postures of the tags comprise boxing, hand waving, clapping, jogging, running and jogging; the classification marks are specifically marked according to 6 postures of boxing, waving hands, clapping palms, jogging, running and jogging;
s14, extracting a foreground, namely a moving human body, by using a Gaussian mixture model;
s15, carrying out normalization processing on the image screened in the step S13;
and S16, randomly dividing the image set subjected to the normalization processing in the step S15 into a training set and a verification set according to the proportion of 8.
In the above human body posture recognition method based on the convolutional neural network, in the step S02, the convolutional neural network includes 4 convolutional layers, 4 pooling layers, 2 full-connection layers, and 1 classification layer; the sizes of convolution kernels in the first two convolution layers are set to be 5 multiplied by 5, and the sizes of the last two convolution kernels are set to be 3 multiplied by 3; the sizes of all the pooling layers are set to be 2 multiplied by 2, and the pooling layers adopt maximum pooling; the number of convolution kernels in the first convolution layer is 32, the number of convolution kernels in the second convolution layer is 64, and the number of convolution kernels in the third convolution layer and the fourth convolution layer is 128; the number of neurons in the first fully-connected layer is set to be 1024, and the number of neurons in the second fully-connected layer is set to be 512; in the convolutional layer and the fully-connected layer, the RELU function is used as an activation function.
The pooling layer mainly decomposes the feature information of each layer and divides a feature set. The sample size is large during initialization, the calculation is complex, but the calculation amount is reduced along with continuous updating, and the local features are increased to the global features. For each feature map, which extracts the maximum value, the maximum pooling can extract more efficient features than the average pooling.
The full-connection layer collects the feature information of each layer, finally obtains the global features of the whole neural network, and plays a key role in subsequent sample image classification and identification.
The classification layer is used for final decision making and is the last layer in the network model. It computes the deviation between the network training penalty prediction and the actual label. The soft-max function is used in the classification layer, and is used in the multi-classification process.
In the above human body posture identification method based on the convolutional neural network, the convolutional neural network is a special deep neural network, and comprises a convolutional layer and a pooling layer. The convolutional layer and the preceding layer of neurons are connected through local connection and value sharing, and the number of training parameters is reduced.
Convolutional layers are the core of convolutional neural networks. The low-level image features are iteratively updated step by step to high-level image features. Each of the convolutional layers is composed of neurons; suppose the nth node O of the mth convolutional layerm,nHas an input value of { xn-1,1；xn-1,2；…；xn-1,kThe value of the output unit is as follows:
Figure RE-BDA0002083935200000041
wherein h isw,b(x) Is the excitation value, f (x) is the excitation function, xiIs the ith input value of the node, wiIs the weight of the ith input value of the node, k is the number of the input values of the current node, and b is a bias term;
the modified linear unit RELU is defined as follows:
f(x)＝max(0,x) (2)
the modified linear unit RELU is a standard excitation function of the convolutional layer, provides a non-linear activation capability for the convolutional neural network, and does not interfere with other convolutional layers.
The above human body posture recognition method based on the convolutional neural network obtains the parameters of the convolutional neural network through training of the convolutional neural network, and specifically includes:
if there are p sample sets, they are denoted as { (x)1,y1),(x2,y2),…,( xp,yp) For each sample, each sample set loss function is defined as follows:
Figure RE-BDA0002083935200000051
wherein h isω,b(x) Is a predicted value after network training, namely an excitation value, y is an actual output value, and w is the weight of an input value;
for the whole sample set, the whole sample set loss function is defined as follows:
Figure RE-BDA0002083935200000052
where N is the total number of layers of the convolutional neural network, skThe number of nodes of the k-1 layer, a is a regularization coefficient, i is the ith node, j is the jth characteristic convolution kernel, and k is the kth layer of the convolution neural network;
Figure RE-BDA0002083935200000053
the connection weight between the ith node and the jth characteristic convolution kernel of the kth layer of the convolution neural network is set;
the goal of convolutional neural network model training is to minimize the value of the overall sample set loss function;
the parameter update formula is as follows:
Figure RE-BDA0002083935200000054
Figure RE-BDA0002083935200000055
where L is the objective loss function, μ is the learning rate,
Figure RE-BDA0002083935200000056
indicating that x is derived over y, the parameters are updated using a small batch approach.
In the human body posture identification method based on the convolutional neural network, the learning rate is set to be 0.001, and optimization is performed by using an adamiptimer algorithm.
In the above human body posture recognition method based on the convolutional neural network, in step S02, sparsity is introduced at the input of the RELU excitation function, that is, at the output of the linear filter, specifically:
input h of k-th layer RELU excitation function in the convolutional neural networkkIs expressed as:
Figure RE-BDA0002083935200000061
wherein h iskIs the input of the k-th layer RELU excitation function, S (h)k) As input h to the k-th layer RELU excitation function in convolutional neural networkskThe sparsity of (a);
defining an optimized objective function to determine convolutional neural network parameters, the objective function being defined as follows:
E＝L+λ∑kS(hk) (8)
where E is the objective function of the optimization, i.e., the final result of the present invention; l is an objective function without introducing sparsity, namely an objective loss function; λ is a tuning parameter that controls sparsity.
By introducing sparsity at the RELU input, unnecessary increase of RELU output can be prevented, unnecessary negative input of RELU can be reduced, and the model generalization capability can be improved.
The invention has the beneficial effects that:
(1) The human body posture is recognized by designing a convolution neural network introducing sparsity, so that the method is applied to man-machine interaction, behavior recognition, action classification, abnormal behavior detection, automatic driving and the like, the high recognition rate is ensured, the convergence rate is improved, and the generalization capability of a model is enhanced;
(2) The invention designs a brand new convolution neural network model by utilizing the artificial neural network method mentioned in the background technology and introducing the sparse regularization method and combining the advantages of the artificial neural network method and the sparse regularization method, and classifies and identifies the human body posture by utilizing the model.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a flow chart of the method of the present invention;
fig. 2 is a block diagram of a convolutional neural network model structure proposed by the present invention.
Detailed Description
In order to make the technical means, the creation characteristics, the achievement purposes and the effects of the invention easy to understand, the invention is further explained by combining the specific embodiments.
As shown in fig. 1, the human body posture recognition method provided by the present invention includes the following steps:
s01, acquiring a human body posture video KTH data set, wherein the data set has 6 postures of boxing, waving hands, clapping palms, jogging, running and jogging, performing video image frame cutting pretreatment on the data set, and dividing the image data set into a training set and a verification set;
s11, acquiring a KTH human posture video data set;
s12, cutting the video into frames and storing the images of each frame;
s13, screening out images of human body behaviors with complete human bodies and corresponding postures of labels from the images, deleting blank images or images without complete human bodies or corresponding postures of labels, and classifying and marking the screened images; the corresponding postures of the tags comprise boxing, hand waving, clapping, jogging, running and jogging, and the 6 postures are postures obtained from KTH human posture video data in a centralized manner; the classification mark is specifically marked according to 6 types of postures of boxing, waving hands, clapping palms, jogging, running and jogging;
s14, extracting a foreground, namely a moving human body, by using a Gaussian mixture model;
s15, normalizing the image;
and S16, randomly dividing the image set into a training set and a verification set according to the proportion of 8. And each type of posture training set comprises 120 effective pictures, and the verification set comprises 30 effective pictures.
S02, constructing a neural network model, and introducing sparsity at the input position of the RELU excitation function; the input of the convolution neural network is a preprocessed image, and the output is a human body gesture category; and training the convolutional neural network.
S21, constructing a neural network, wherein a convolutional neural network architecture adopted by the invention is shown in figure 2, and the network architecture comprises 7 layers, 4 convolutional layers (comprising a pooling layer), 2 full-connection layers and 1 classification layer. The first two convolution kernels are set to 5 x 5 in size and the last two are set to 3 x 3 in size. The size of all the pooling layers is set to be 2 x 2, and the pooling layer adopts maximum pooling. The number of convolution kernels in the first convolution layer is 32, the number of convolution kernels in the second convolution layer is 64, and the number of convolution kernels in the third convolution layer and the fourth convolution layer are 128. The number of neurons in the first fully-connected layer was set to 1024, and the number of neurons in the second fully-connected layer was set to 512. In the convolutional layer and the fully-connected layer, the RELU function is used as an activation function. The learning rate is set to 0.001 and optimized using the adampotime algorithm.
Suppose the nth node O of the mth convolutional layerm,nIs inputtedA value of { xn-1,1； xn-1,2；…；xn-1,kThe value of the output unit is as follows:
Figure RE-BDA0002083935200000081
wherein h isw,b(x) Is the excitation value, f (x) is the excitation function, xiIs the ith input value of the node, wiIs the weight of the ith input value of the node, k is the number of the input values of the current node, and b is a bias term;
the modified linear unit RELU is defined as follows:
f(x)＝max(0,x) (2)
the modified linear unit RELU is a standard excitation function of the convolutional layer, provides a non-linear activation capability for the convolutional neural network, and does not interfere with other convolutional layers.
Through training of the convolutional neural network, parameters of the convolutional neural network are obtained, and the parameters are specifically as follows:
if there are p sample sets, they are denoted as { (x)1,y1),(x2,y2),…,(xp,yp) For each sample, each sample set loss function is defined as follows:
Figure RE-BDA0002083935200000091
wherein h isω,b(x) Is a predicted value after network training, namely an excitation value, y is an actual output value, and w is the weight of an input value;
for the whole sample set, the whole sample set loss function is defined as follows:
Figure RE-BDA0002083935200000092
where N is the total number of layers of the convolutional neural network, skIs the number of nodes of the k-1 th layer, a is the regularization coefficient, i is the ith node, j is the jth feature volumeA kernel, k is the k layer of the convolutional neural network;
Figure RE-BDA0002083935200000093
the connection weight between the ith node and the jth characteristic convolution kernel of the kth layer of the convolution neural network is set;
the goal of convolutional neural network model training is to minimize the value of the overall sample set loss function;
the parameter update formula is as follows:
Figure RE-BDA0002083935200000094
Figure RE-BDA0002083935200000095
where L is the objective loss function, μ is the learning rate,
Figure RE-BDA0002083935200000096
indicating that x is derived from y and the parameters are updated using a small batch approach.
Introducing sparsity at the RELU excitation function input, i.e. at the output of the linear filter, for sparsity S (h)k) Layers greater than 0.6 are processed. By introducing sparsity at the RELU input, unnecessary increase of RELU output can be prevented, unnecessary negative input of RELU can be reduced, and the model generalization capability can be improved.
At the RELU excitation function input, i.e. at the output of the linear filter, sparsity is introduced, specifically:
input h of k-th-layer RELU excitation function in convolutional neural networkkIs expressed as:
Figure RE-BDA0002083935200000101
wherein h iskIs the input of the k-th layer RELU excitation function, S (h)k) Is the spirit of convolutionInput h via RELU excitation function at k-th layer in networkkThe sparsity of (a);
defining an optimized objective function to determine convolutional neural network parameters, the objective function being defined as follows:
E＝L+λ∑kS(hk) (8)
where E is the optimized objective function, L is the objective function without introducing sparsity, i.e. the objective loss function, and λ is the tuning parameter controlling sparsity.
And S03, recognizing the human body posture by adopting the neural network model in the S02, and performing model training and performance testing on the public human body posture data set KTH. When a new unknown video is input, preprocessing is firstly carried out through S01, and then the posture is judged through network prediction.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrases "comprising a component of' 8230; \8230;" does not exclude the presence of another like element in a process, method, article, or apparatus that comprises the element.
The above-mentioned serial numbers of the embodiments of the present invention are only for description, and do not represent the advantages and disadvantages of the embodiments.
In the embodiments provided in the present application, it should be understood that the disclosed system and method may be implemented in other ways. For example, the above-described system embodiments are merely illustrative, and for example, the division of the units is only one logical functional division, and other divisions may be realized in practice, for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed.
Those of skill would further appreciate that the various illustrative modules and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both, and that such implementation decisions may be made by those skilled in the art using various means for implementing the functions described herein without departing from the scope of the invention.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
The foregoing shows and describes the general principles and broad features of the present invention and advantages thereof. The industry has described only the principles of the invention and is therefore intended to encompass within its spirit and scope all such changes and modifications as fall within the scope of the invention as claimed. The scope of the invention is defined by the appended claims and equivalents thereof.

Claims (6)

1. A human body posture identification method based on a convolutional neural network is characterized by comprising the following steps:
s01, acquiring a human body posture video data set, preprocessing the human body posture video data set by cutting a video into image frames, and dividing the image data set cut into the image frames into a training set and a verification set;
s02, constructing a neural network model, introducing sparsity at the input of the RELU excitation function, inputting the convolution neural network into the image preprocessed in the step S01, and outputting the image as a human body posture category; training the convolutional neural network;
s03, recognizing the human body posture by adopting the neural network model in the S02, and performing model training and performance testing on a public human body posture data set KTH; when an unknown video is input, firstly calling the step S01 to carry out preprocessing, and then carrying out posture recognition by using the neural network model in the step S02 to obtain the human body posture category;
in said step S02, sparsity is introduced at the RELU excitation function input, i.e. at the output of the linear filter, specifically:
input h of k-th-layer RELU excitation function in the convolutional neural networkkIs expressed as:
Figure FDA0003794458940000011
wherein h iskIs the input of the k-th layer RELU excitation function, S (h)k) As input h to the k-th layer RELU excitation function in convolutional neural networkskSparsity of (d);
defining an optimized objective function to determine convolutional neural network parameters, the objective function being defined as follows:
E＝L+λ∑kS(hk)
where E is the optimized objective function, L is the objective function without introducing sparsity, i.e. the objective loss function, and λ is the tuning parameter controlling sparsity.
2. The method for recognizing the human body posture based on the convolutional neural network as claimed in claim 1, characterized in that: in step S01, the acquiring of the human body posture video data set specifically includes the following steps:
s11, acquiring a public KTH human posture video data set;
s12, cutting the video into frames and storing the images of each frame;
s13, screening out images of human body behaviors with complete human bodies and corresponding postures of labels from the images, deleting blank images or images of human body behaviors without complete human bodies or corresponding postures of labels, and carrying out classification marking on the screened images; the corresponding postures of the tags comprise boxing, hand waving, clapping, jogging, running and jogging; the classification marks are specifically marked according to 6 postures of boxing, waving hands, clapping palms, jogging, running and jogging;
s14, extracting a foreground, namely a moving human body, by using a Gaussian mixture model;
s15, carrying out normalization processing on the image screened in the step S13;
and S16, randomly dividing the image set subjected to the normalization processing in the step S15 into a training set and a verification set according to the proportion of 8.
3. The method for recognizing the human body posture based on the convolutional neural network as claimed in claim 1, wherein: in the step S02, the convolutional neural network includes 4 convolutional layers, 4 pooling layers, 2 fully-connected layers, and 1 classification layer; the sizes of convolution kernels in the first two convolution layers are set to be 5 multiplied by 5, and the sizes of the last two convolution kernels are set to be 3 multiplied by 3; the size of all the pooling layers is set to be 2 multiplied by 2, and the pooling layers adopt maximum pooling; the number of convolution kernels in the first convolution layer is 32, the number of convolution kernels in the second convolution layer is 64, and the number of convolution kernels in the third convolution layer and the fourth convolution layer is 128; the number of neurons in the first fully-connected layer is set to 1024, and the number of neurons in the second fully-connected layer is set to 512; in the convolutional layer and the fully-connected layer, the RELU function is used as an activation function.
4. The method for recognizing the human body posture based on the convolutional neural network as claimed in claim 3, characterized in that: suppose the nth node O of the mth convolutional layerm,nHas an input value of { xn-1,1；xn-1,2；…；xn-1,kAnd then the values of the output units are as follows:
Figure FDA0003794458940000031
wherein h isw，b(x) Is the excitation value, f (x) is the excitation function, xiIs the ith input value of the node, wiIs the weight of the ith input value of the node, k is the number of the input values of the current node, and b is a bias term;
the modified linear unit RELU is defined as follows:
f(x)＝max(0，x)
the modified linear unit RELU is a standard excitation function of the convolutional layer, provides a non-linear activation capability for the convolutional neural network, and does not interfere with other convolutional layers.
5. The human body posture recognition method based on the convolutional neural network as claimed in claim 4, characterized in that: through training of the convolutional neural network, parameters of the convolutional neural network are obtained, and the parameters are specifically as follows:
if there are p sample sets, they are denoted as { (x)1，y1)，(x2，y2)，...，(xp，yp) For each sample, each sample set loss function is defined as follows:
Figure FDA0003794458940000032
wherein h isω，b(x) Is a predicted value after network training, namely the excitation value, y is an actual output value, and w is the weight of an input value;
for the whole sample set, the whole sample set loss function is defined as follows:
Figure FDA0003794458940000033
where N is the total number of layers of the convolutional neural network, skThe number of nodes of the k-1 layer, a is a regularization coefficient, i is the ith node, j is the jth characteristic convolution kernel, and k is the kth layer of the convolution neural network;
Figure FDA0003794458940000034
the connection weight between the ith node and the jth characteristic convolution kernel of the kth layer of the convolution neural network is set;
the goal of convolutional neural network model training is to minimize the value of the overall sample set loss function;
the parameter update formula is as follows:
Figure FDA0003794458940000041
Figure FDA0003794458940000042
where L is the objective loss function, μ is the learning rate,
Figure FDA0003794458940000043
indicating that x is derived over y, the parameters are updated using a small batch approach.
6. The human body posture recognition method based on the convolutional neural network as claimed in claim 5, characterized in that: the learning rate is set to 0.001 and optimized using the adampotime algorithm.
CN201910481323.XA 2019-06-04 2019-06-04 Human body posture recognition method based on convolutional neural network Active CN110222634B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910481323.XA CN110222634B (en) 2019-06-04 2019-06-04 Human body posture recognition method based on convolutional neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910481323.XA CN110222634B (en) 2019-06-04 2019-06-04 Human body posture recognition method based on convolutional neural network

Publications (2)

Publication Number Publication Date
CN110222634A CN110222634A (en) 2019-09-10
CN110222634B true CN110222634B (en) 2022-11-01

Family

ID=67819290

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910481323.XA Active CN110222634B (en) 2019-06-04 2019-06-04 Human body posture recognition method based on convolutional neural network

Country Status (1)

Country Link
CN (1) CN110222634B (en)

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110688933A (en) * 2019-09-23 2020-01-14 中国计量大学 Novel convolutional neural network and weighted assignment human body posture estimation algorithm
CN110688980B (en) * 2019-10-12 2023-04-07 南京工程学院 Human body posture classification method based on computer vision
CN110929242B (en) * 2019-11-20 2020-07-10 上海交通大学 Method and system for carrying out attitude-independent continuous user authentication based on wireless signals
CN111178237A (en) * 2019-12-27 2020-05-19 上海工程技术大学 Pavement state recognition method
CN111753736A (en) * 2020-06-24 2020-10-09 北京软通智慧城市科技有限公司 Human body posture recognition method, device, equipment and medium based on packet convolution
CN111950496B (en) * 2020-08-20 2023-09-15 广东工业大学 Mask person identity recognition method
CN112801176B (en) * 2021-01-25 2022-12-06 西安交通大学 Deep learning identification method for helicopter flight attitude imbalance data
CN113269047B (en) * 2021-04-29 2024-03-22 江苏大学 Three-dimensional human body posture estimation method based on convolutional neural network and spark
CN113850565B (en) * 2021-09-24 2022-06-07 广东诚誉工程咨询监理有限公司 Maturity model-based overall process consultation project management monitoring system and method
CN117649631B (en) * 2024-01-29 2024-04-05 广州宇中网络科技有限公司 Client image processing method and system based on improved convolutional neural network

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106096506A (en) * 2016-05-28 2016-11-09 重庆大学 Based on the SAR target identification method differentiating doubledictionary between subclass class
CN106529442A (en) * 2016-10-26 2017-03-22 清华大学 Pedestrian identification method and apparatus
CN107169117A (en) * 2017-05-25 2017-09-15 西安工业大学 A kind of manual draw human motion search method based on autocoder and DTW
CN109731302A (en) * 2019-01-22 2019-05-10 深圳职业技术学院 Athletic posture recognition methods, device and electronic equipment

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106096506A (en) * 2016-05-28 2016-11-09 重庆大学 Based on the SAR target identification method differentiating doubledictionary between subclass class
CN106529442A (en) * 2016-10-26 2017-03-22 清华大学 Pedestrian identification method and apparatus
CN107169117A (en) * 2017-05-25 2017-09-15 西安工业大学 A kind of manual draw human motion search method based on autocoder and DTW
CN109731302A (en) * 2019-01-22 2019-05-10 深圳职业技术学院 Athletic posture recognition methods, device and electronic equipment

Also Published As

Publication number Publication date
CN110222634A (en) 2019-09-10

Similar Documents

Publication Publication Date Title
CN110222634B (en) Human body posture recognition method based on convolutional neural network
CN111079639B (en) Method, device, equipment and storage medium for constructing garbage image classification model
Thoma Analysis and optimization of convolutional neural network architectures
US20190228268A1 (en) Method and system for cell image segmentation using multi-stage convolutional neural networks
CN107526785A (en) File classification method and device
CN109993100B (en) Method for realizing facial expression recognition based on deep feature clustering
CN110321967B (en) Image classification improvement method based on convolutional neural network
CN110276248B (en) Facial expression recognition method based on sample weight distribution and deep learning
CN109002755B (en) Age estimation model construction method and estimation method based on face image
CN112734775A (en) Image annotation, image semantic segmentation and model training method and device
US20210158166A1 (en) Semi-structured learned threshold pruning for deep neural networks
CN106845510A (en) Chinese tradition visual culture Symbol Recognition based on depth level Fusion Features
CN107223260B (en) Method for dynamically updating classifier complexity
CN111783841A (en) Garbage classification method, system and medium based on transfer learning and model fusion
US20220121949A1 (en) Personalized neural network pruning
CN111832580B (en) SAR target recognition method combining less sample learning and target attribute characteristics
KR20200071865A (en) Image object detection system and method based on reduced dimensional
CN114548256A (en) Small sample rare bird identification method based on comparative learning
CN110163206B (en) License plate recognition method, system, storage medium and device
CN112183464A (en) Video pedestrian identification method based on deep neural network and graph convolution network
CN111694954A (en) Image classification method and device and electronic equipment
CN111310820A (en) Foundation meteorological cloud chart classification method based on cross validation depth CNN feature integration
CN114882278A (en) Tire pattern classification method and device based on attention mechanism and transfer learning
CN114067314B (en) Neural network-based peanut mildew identification method and system
CN114463574A (en) Scene classification method and device for remote sensing image

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
CB03 Change of inventor or designer information

Inventor after: Li Jian

Inventor after: Zhang Yuan

Inventor after: Luo Ying

Inventor after: Zhang Yixin

Inventor before: Zhang Yuan

Inventor before: Li Jian

Inventor before: Luo Ying

Inventor before: Zhang Yixin

CB03 Change of inventor or designer information
GR01 Patent grant
GR01 Patent grant