CN111860278B - Human behavior recognition algorithm based on deep learning - Google Patents
Human behavior recognition algorithm based on deep learning Download PDFInfo
- Publication number
- CN111860278B CN111860278B CN202010676134.0A CN202010676134A CN111860278B CN 111860278 B CN111860278 B CN 111860278B CN 202010676134 A CN202010676134 A CN 202010676134A CN 111860278 B CN111860278 B CN 111860278B
- Authority
- CN
- China
- Prior art keywords
- sample
- training
- predicted
- loss
- data set
- 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
Links
- 230000006399 behavior Effects 0.000 title claims abstract description 27
- 238000013135 deep learning Methods 0.000 title claims abstract description 15
- 238000012549 training Methods 0.000 claims abstract description 24
- 230000006870 function Effects 0.000 claims abstract description 12
- 238000012360 testing method Methods 0.000 claims abstract description 10
- 238000007781 pre-processing Methods 0.000 claims abstract description 7
- 238000009826 distribution Methods 0.000 claims description 10
- 238000000034 method Methods 0.000 claims description 9
- 238000013461 design Methods 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 3
- 230000001419 dependent effect Effects 0.000 abstract description 2
- 230000035945 sensitivity Effects 0.000 abstract 1
- 238000013527 convolutional neural network Methods 0.000 description 5
- 238000011176 pooling Methods 0.000 description 4
- 239000013598 vector Substances 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- NVNSXBXKNMWKEJ-UHFFFAOYSA-N 5-[[5-(2-nitrophenyl)furan-2-yl]methylidene]-1,3-diphenyl-2-sulfanylidene-1,3-diazinane-4,6-dione Chemical compound [O-][N+](=O)C1=CC=CC=C1C(O1)=CC=C1C=C1C(=O)N(C=2C=CC=CC=2)C(=S)N(C=2C=CC=CC=2)C1=O NVNSXBXKNMWKEJ-UHFFFAOYSA-N 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
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
- 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
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary 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 provides a human behavior recognition algorithm based on deep learning, which comprises (1) preprocessing an input video segment; (2) constructing a network model RD3D; (3) Defining a loss function, accuracy and operation of an optimizer; (4) training the network model comprising the sub-steps of: (41) initializing parameters; (42) a learning rate of 0.0001, a batch size of 16; (43) Calculating a loss by the forward propagation value of the RD3D model and the real tag, and updating the weight parameter by backward propagation; (44) ending the training after 100 epochs have been trained; (5) test results. The invention pursues the accuracy of the recognition algorithm from the characteristic angle, solves the problem that the existing algorithm is seriously dependent on the data set, reduces the sensitivity to the type of the data set, and can be applied to any behavior recognition data set.
Description
Technical Field
The invention relates to the technical field of computer vision, in particular to a human behavior recognition algorithm based on deep learning.
Background
In recent years, with the rise of related technologies such as deep learning, deep neural networks have made breakthrough progress in various fields such as computer vision. Deep learning is capable of learning its commonality features from training data due to its end-to-end training characteristics and fitting out a network suitable for the current task. Meanwhile, mass data acquisition in the modern society becomes very easy, and convenience is provided for deep learning application to the fields of video understanding, recognition and the like.
Whereas in the conventional method, the local features (such as HOG, HOF and MBH) are mainly extracted, a great a priori knowledge is required. Although apparent and motion information is considered, such information is limited to a single frame, where the contextual appearance of the frame and motion information is ignored, which results in less accurate identification of human behavior. How to design an algorithm for behavior recognition is particularly important.
Therefore, it has been a trend to apply deep learning and human behavior recognition. The behavior recognition method based on deep learning mainly comprises the following steps: the double-flow convolutional neural network, the 3D convolutional neural network, the convolutional neural network and the cyclic neural combination. The invention is based on a 3D convolution network, and improves the recognition precision.
Patent CN 110163133A, "a human body behavior recognition method based on a depth residual error network", discloses a human body behavior recognition method based on a depth residual error network, wherein human body joint data and depth image data are simultaneously input into ResNet for recognition, but the recognition accuracy is improved, but the human body joint data and the depth image are required to be input, so that the end-to-end learning cannot be realized, and the data are lacking in daily life. Patent CN 107862275A, human body behavior recognition model, construction method and human body behavior recognition method, discloses a method for extracting human body behavior feature vectors by adopting a 3D convolutional neural network, inputting the extracted feature vectors into a coulomb force field, and clustering all feature vectors by relative movement under the action of attraction force generated by the same class and repulsion force generated by different classes to finish human body behavior recognition. The RGB diagram and the optical flow diagram are input into the network for learning, and the end-to-end learning is not performed, the whole network has seven layers, wherein only three layers perform feature extraction, and the accuracy is low although the calculated amount is small;
The recognition accuracy is improved from the point of conforming to the data set, the accuracy of behavior recognition can not be improved by means of RGB images, and patent CN 109002808A 'a human behavior recognition method and system' discloses a human behavior recognition method and system, a multi-task deep learning method is used for training a 3D convolutional neural network, continuous video frames of various human behavior attributes and background videos are used as input, and recognition tasks are completed after training. More, how to make a data set in the application of multitask learning is taught, so that a behavior video and a background video are distinguished, feature extraction is completed only by means of a seven-layer common 3D convolution network, and classification is achieved. Identification of human behavior is still accomplished from the perspective of the data set.
Disclosure of Invention
Aiming at the technical problems, the invention provides a human behavior recognition algorithm based on deep learning, which comprises the following steps:
(1) Preprocessing an input video segment;
(2) Constructing a network model RD3D;
(3) Defining a loss function and an operation of an optimizer;
(4) Training the network model comprises the sub-steps of:
(41) Initializing parameters;
(42) Learning rate is 0.0001, and batch size is 16;
(43) Calculating a loss according to a loss function by the forward propagation value of the RD3D model and the real tag, and updating weight parameters by backward propagation of the loss;
(44) Finishing training after training 100 epochs;
(5) And (5) testing results.
Furthermore, in the preprocessing stage of the step (1), in order to comprehensively consider the global motion information of the video, a secondary sampling algorithm is provided and adopted to collect n frames of key video frames, so that the recognition accuracy is improved, and the specific contents are as follows:
a: each video segment is subjected to image frame acquisition (alpha=3) according to an acquisition rate alpha, and an image data set A corresponding to each video is obtained;
b: uniformly acquiring n frames (n=16) from the image data set A by adopting a subsampling algorithm, and scaling the n frames to k×k (k=224) to form a data set B as key frames of video clips;
c: the acquired data set B was processed as 7: the 3 proportion is divided into a training set and a testing set for training and testing, wherein each sample in the training set is a four-element (candidate, positive, negative, label) sample to be predicted, other samples in the same category with the sample to be predicted, other samples in different categories with the sample to be predicted, and category labels of the sample to be predicted.
Furthermore, in step (2), in order to improve the recognition accuracy, a novel network model RD3D (Residual Dense 3D) is proposed and designed by combining the feature multiplexing idea and the shortcut idea. The RD3D model designs 134 layers, i.e., 1+4×4+6×6×3+2×4+1,6 stages.
Further, step (3) proposes and designs a new type of loss function:
F＝H(P,Q)+Lre+Ltr
Wherein:
Cross entropy H (P, Q) = -P (x) log (Q (x)) measures the similarity of the predicted and true distributions, the smaller the loss, the more accurate the classification. Wherein P is the real sample distribution, and Q is the predicted sample distribution;
l2 regularization loss To prevent overfitting, where λ is the penalty factor (λ=0.009), n is the number of weights W;
ternary loss Where ||f (x i)-f(xp i)||2 2 is the Euclidean distance of x i and x p i, ||f (x i)-f(xn i)||2 2 is the Euclidean distance of x i and x n i, |f (x) is the characteristics of sample x extracted by RD3D, bs is batchsize, x i is the sample currently predicted, x p i is the sample of the same class as currently calculated sample x i, x n i is the sample not of the same class as currently calculated sample x i, and β is the distance threshold of x i and x p i、xi and x n i (β=0.2).
The invention overcomes the problem that the existing algorithm is seriously dependent on the data set while pursuing the accuracy of the identification algorithm, realizes the network structure design from the aspect of the characteristics extracted from the human behavior, is insensitive to the type of the data set, and can be applied to any data set.
Drawings
FIG. 1 is a RD3D model of the present invention;
FIG. 2 is a ConvBlock of the present invention;
FIG. 3 is a IDBlock of the present invention;
Fig. 4 is a flow chart of the present invention.
Detailed Description
The specific technical scheme of the invention is described by combining the embodiments.
As shown in fig. 4, a human behavior recognition algorithm based on deep learning includes the following steps:
(1) Preprocessing an input video segment (UCF 101 dataset is taken as an example in this embodiment);
(2) Constructing a network model RD3D;
(3) Defining a loss function, accuracy and operation of an optimizer;
(4) Training the network model comprises the sub-steps of:
(41) Initializing parameters;
(42) Learning rate is 0.0001, and batch size is 16;
(43) Calculating a loss according to a loss function by the forward propagation value of the RD3D model and the real tag, and updating weight parameters by backward propagation of the loss;
(44) Finishing training after training 100 epochs;
(5) And (5) testing results.
Specific:
(1) In the preprocessing stage, in order to comprehensively consider the video global motion information, a secondary sampling algorithm is proposed and adopted to collect n frames of key frames, so that the recognition accuracy is improved, and the specific contents are as follows:
a: each video segment is subjected to image frame acquisition (alpha=3) according to an acquisition rate alpha, and an image data set A corresponding to each video is obtained;
b: uniformly acquiring n frames (n=16) from the image data set A by adopting a subsampling algorithm, and scaling the n frames to k×k (k=224) to form a data set B as key frames of video clips;
c: the acquired data set B was processed as 7: the 3 proportion is divided into a training set and a testing set for training and testing, wherein each sample in the training set is a four-element (candidate, positive, negative, label) sample to be predicted, other samples in the same category with the sample to be predicted, other samples in different categories with the sample to be predicted, and category labels of the sample to be predicted.
(2) In order to improve the recognition accuracy, a novel network model RD3D (Residual Dense 3D) is provided and designed by combining the characteristic multiplexing idea and the shortcut idea, the structure of the novel network model RD3D is shown in figure 1, 127 layers, 6 stages are designed by the RD3D model, and the content is as follows:
a: stage1 consists of Conv3d, BN, relu, maxPool, where Conv3d has 64 filters, convolution kernel 3 x 3, stride= [1, 2], and padding is SAME; the pooling window in MaxPool is 1×3×3, stride= [1, 2]. Stage1 has an input dimension of [16, 16, 224,3] and an output dimension of [16, 16, 56, 56, 64];
b: stage2 consists of Conv Block4, three ID Block4, maxpool, where Conv Block4 is connected together by the addition of channels by a 4-layer 3D convolution group and a shortcut with two-layer convolution, as shown in fig. 2, wherein the number of filters of the 4-layer 3D convolution sets is 64, 64, 128, 128, the convolution kernels are all 3 x 3; in the convolution set, the input of the next layer is the output of all the previous layers in the block. The number of filters in shortcut is 128, the convolution kernels are 1×1,3×3, respectively. ID Block4 is a combination of a 4-layer 3D convolution and the input addition of the Block, as shown in figure 3, wherein the number of filters of the 4-layer 3D convolution sets is 64, 64, 128, 128, the convolution kernels are all 3 x 3; in the convolution set, the input of the next layer is the output of all the previous layers in the block. Pooling window 2 x 2, stride= [2, 2] in Maxpool. Stage2 has an input dimension of [16,16,56,56,64] and an output dimension of [16,8,28,28,128];
c: the composition of stage3 and stage4 is identical to stage2, the only difference is that the number of layers and the number of filters in each block are different, in stage3, the number of layers of ConvBlock and IDBlock is 6, the number of filters in each layer is 128, 128, 256, 256, 512, the number of filters of shortcut in ConvBlock is 512, the input dimension of stage3 is [16,8,28,28,128], and the output dimension is [16,4,14,14,512]; in stage4, the number of layers ConvBlock and IDBlock is 6, the number of filters in each layer is 256, 256, 512, 512, 1024, and the number of filters in shortcut in convblock is 1024.Stage4 has an input dimension of [16,4,14,14,512] and an output dimension of [16,2,7,7,1024];
d: stage5 and stage2 differ in composition in that stage5 does not have MaxPool. In ConvBlock and IDBlock, the number of layers is 6, the number of filters of each layer is 512, 512, 1024, 1024, 2048, the number of filters of shortcut in ConvBlock is 2048, the input dimension of stage5 is [16,2,7,7,1024], and the output dimension is [16,2,7,7,2048];
e: stage6 is composed of AvgPool, flatten, FC, softmax, as shown in fig. 1, in which AvgPool is global average pooling, the pooling window is 2×7×7, and flat is that output reshape of the upper layer is [16,2048], FC is the full connection layer, output dimension is ucf of category number 101, softmax is the classification layer. Stage6 has an input dimension of [16,2,7,7,2048] and an output dimension of [16,101]
(3) The loss function is designed. In order to expand the distinction of different types of samples and improve the recognition precision, the invention adds ternary loss on the basis of the traditional loss function to obtain a new loss function:
F=h (P, Q) +l re+Ltr wherein,
A: cross entropy H (P, Q) = -P (x) log (Q (x)) measures the similarity of the predicted and true distributions, the smaller the loss, the more accurate the classification. Wherein P is the real sample distribution, and Q is the predicted sample distribution;
b: l2 regularization loss To prevent overfitting, where λ is the penalty factor (λ=0.009), n is the number of weights W;
c: ternary loss:
Where ||f (x i)-f(xp i)||2 2 is the Euclidean distance of x i and x p i, ||f (x i)-f(xn i)||22 is the Euclidean distance of x i and x n i, |f (x) is the characteristics of sample x extracted by RD3D, bs is batchsize, x i is the sample currently predicted, x p i is the sample of the same class as currently calculated sample x i, x n i is the sample not of the same class as currently calculated sample x i, and β is the distance threshold of x i and x p i、xi and x n i (β=0.2).
Claims (3)
1. The human behavior recognition algorithm based on deep learning is characterized by comprising the following steps of:
(1) Preprocessing an input video segment;
in the preprocessing stage of the step (1), a secondary sampling algorithm is provided and adopted to collect n frames of key frames, and the method specifically comprises the following steps:
a: acquiring image frames of each video clip according to the acquisition rate alpha to obtain an image data set A corresponding to each video;
b: uniformly acquiring n frames from the image data set A by adopting a subsampling algorithm, taking the n frames as key frames of video clips, and scaling the key frames to k x k to form a data set B;
c: the acquired data set B was processed as 7: the 3 proportion is divided into a training set and a testing set for training and testing, wherein each sample in the training set is a quadruple, and is a sample to be predicted, other samples of the same class of the sample to be predicted, other samples of different classes of the sample to be predicted, and class labels of the sample to be predicted;
(2) Constructing a network model RD3D;
(3) Defining a loss function and an operation of an optimizer;
(4) Training a network model; comprises the following substeps:
(41) Initializing parameters;
(42) Learning rate is 0.0001, and batch size is 16;
(43) Calculating a loss according to a loss function by the forward propagation value of the RD3D model and the real tag, and updating weight parameters by backward propagation of the loss;
(44) Finishing training after training 100 epochs;
(5) And (5) testing results.
2. The human behavior recognition algorithm based on deep learning of claim 1, wherein the RD3D model of step (2) designs 134 layers, i.e., 1+4×4+6×6×3+2×4+1,6 stages.
3. The human behavior recognition algorithm based on deep learning of claim 1, wherein step (3) designs a loss function:
F＝H(P,Q)+Lre+Ltr
wherein, cross entropy H (P, Q) = -P (x) log (Q (x)), the similarity of the predicted distribution and the real distribution is measured, the smaller the loss is, the more accurate the classification is; wherein P is the real sample distribution, and Q is the predicted sample distribution;
l2 regularization loss To prevent overfitting, where λ is a penalty factor and n is the number of weights W;
Ternary loss:
Where ||f (x i)-f(xp i)||2 2 is the Euclidean distance of x i and x p i, ||f (x i)-f(xn i)||2 2 is the Euclidean distance of x i and x n i, |f (x) is the RD3D extracted feature of sample x, bs is batchsize, x i is the current predicted sample, x p i is the same class of sample as the current calculated sample x i, x n i is a sample not the same class as the current calculated sample x i, and β is the distance threshold of x i and x p i、xi and x n i.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010676134.0A CN111860278B (en) | 2020-07-14 | 2020-07-14 | Human behavior recognition algorithm based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010676134.0A CN111860278B (en) | 2020-07-14 | 2020-07-14 | Human behavior recognition algorithm based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111860278A CN111860278A (en) | 2020-10-30 |
| CN111860278B true CN111860278B (en) | 2024-05-14 |
Family
ID=72984833
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010676134.0A Active CN111860278B (en) | 2020-07-14 | 2020-07-14 | Human behavior recognition algorithm based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111860278B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112861752B (en) * | 2021-02-23 | 2022-06-14 | 东北农业大学 | DCGAN and RDN-based crop disease identification method and system |
| CN113361417B (en) * | 2021-06-09 | 2023-10-31 | 陕西理工大学 | Human behavior recognition method based on variable time sequence |
| CN114627424A (en) * | 2022-03-25 | 2022-06-14 | 合肥工业大学 | Gait recognition method and system based on visual angle transformation |
| CN114897146B (en) * | 2022-05-18 | 2023-11-03 | 北京百度网讯科技有限公司 | Model generation method and device and electronic equipment |
Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2017133009A1 (en) * | 2016-02-04 | 2017-08-10 | 广州新节奏智能科技有限公司 | Method for positioning human joint using depth image of convolutional neural network |
| WO2017203262A2 (en) * | 2016-05-25 | 2017-11-30 | Metail Limited | Method and system for predicting garment attributes using deep learning |
| CN108009525A (en) * | 2017-12-25 | 2018-05-08 | 北京航空航天大学 | A kind of specific objective recognition methods over the ground of the unmanned plane based on convolutional neural networks |
| CN108830185A (en) * | 2018-05-28 | 2018-11-16 | 四川瞳知科技有限公司 | Activity recognition and localization method based on multitask combination learning |
| CN109993076A (en) * | 2019-03-18 | 2019-07-09 | 华南理工大学 | A kind of white mouse behavior classification method based on deep learning |
| WO2019169942A1 (en) * | 2018-03-09 | 2019-09-12 | 华南理工大学 | Anti-angle and occlusion interference fast face recognition method |
| CN110348381A (en) * | 2019-07-11 | 2019-10-18 | 电子科技大学 | Video behavior identification method based on deep learning |
| WO2019232894A1 (en) * | 2018-06-05 | 2019-12-12 | 中国石油大学(华东) | Complex scene-based human body key point detection system and method |
| CN110598598A (en) * | 2019-08-30 | 2019-12-20 | 西安理工大学 | Double-current convolution neural network human behavior identification method based on finite sample set |
| CN110619352A (en) * | 2019-08-22 | 2019-12-27 | 杭州电子科技大学 | Typical infrared target classification method based on deep convolutional neural network |
| CN110826462A (en) * | 2019-10-31 | 2020-02-21 | 上海海事大学 | Human body behavior identification method of non-local double-current convolutional neural network model |
| CN110956085A (en) * | 2019-10-22 | 2020-04-03 | 中山大学 | Human behavior recognition method based on deep learning |
| WO2020073951A1 (en) * | 2018-10-10 | 2020-04-16 | 腾讯科技（深圳）有限公司 | Method and apparatus for training image recognition model, network device, and storage medium |
| CN111027487A (en) * | 2019-12-11 | 2020-04-17 | 山东大学 | Behavior recognition system, method, medium, and apparatus based on multi-convolution kernel residual network |
-
2020
- 2020-07-14 CN CN202010676134.0A patent/CN111860278B/en active Active
Patent Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2017133009A1 (en) * | 2016-02-04 | 2017-08-10 | 广州新节奏智能科技有限公司 | Method for positioning human joint using depth image of convolutional neural network |
| WO2017203262A2 (en) * | 2016-05-25 | 2017-11-30 | Metail Limited | Method and system for predicting garment attributes using deep learning |
| CN108009525A (en) * | 2017-12-25 | 2018-05-08 | 北京航空航天大学 | A kind of specific objective recognition methods over the ground of the unmanned plane based on convolutional neural networks |
| WO2019169942A1 (en) * | 2018-03-09 | 2019-09-12 | 华南理工大学 | Anti-angle and occlusion interference fast face recognition method |
| CN108830185A (en) * | 2018-05-28 | 2018-11-16 | 四川瞳知科技有限公司 | Activity recognition and localization method based on multitask combination learning |
| WO2019232894A1 (en) * | 2018-06-05 | 2019-12-12 | 中国石油大学(华东) | Complex scene-based human body key point detection system and method |
| WO2020073951A1 (en) * | 2018-10-10 | 2020-04-16 | 腾讯科技（深圳）有限公司 | Method and apparatus for training image recognition model, network device, and storage medium |
| CN109993076A (en) * | 2019-03-18 | 2019-07-09 | 华南理工大学 | A kind of white mouse behavior classification method based on deep learning |
| CN110348381A (en) * | 2019-07-11 | 2019-10-18 | 电子科技大学 | Video behavior identification method based on deep learning |
| CN110619352A (en) * | 2019-08-22 | 2019-12-27 | 杭州电子科技大学 | Typical infrared target classification method based on deep convolutional neural network |
| CN110598598A (en) * | 2019-08-30 | 2019-12-20 | 西安理工大学 | Double-current convolution neural network human behavior identification method based on finite sample set |
| CN110956085A (en) * | 2019-10-22 | 2020-04-03 | 中山大学 | Human behavior recognition method based on deep learning |
| CN110826462A (en) * | 2019-10-31 | 2020-02-21 | 上海海事大学 | Human body behavior identification method of non-local double-current convolutional neural network model |
| CN111027487A (en) * | 2019-12-11 | 2020-04-17 | 山东大学 | Behavior recognition system, method, medium, and apparatus based on multi-convolution kernel residual network |
Non-Patent Citations (3)
| Title |
|---|
| 基于深度卷积神经网络与中心损失的人脸识别;张延安;王宏玉;徐方;;科学技术与工程;20171218(35);第97-102页 * |
| 基于深度学习的人体行为识别研究;赵新秋;杨冬冬;贺海龙;段思雨;;高技术通讯;20200515(05);第41-49页 * |
| 基于通道注意力机制的视频人体行为识别;解怀奇;乐红兵;;电子技术与软件工程;20200215(04);第146-148页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111860278A (en) | 2020-10-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111860278B (en) | Human behavior recognition algorithm based on deep learning | |
| CN109472194B (en) | Motor imagery electroencephalogram signal feature identification method based on CBLSTM algorithm model | |
| CN109146921B (en) | Pedestrian target tracking method based on deep learning | |
| CN110399850B (en) | Continuous sign language recognition method based on deep neural network | |
| CN110969087B (en) | Gait recognition method and system | |
| CN110555387B (en) | Behavior identification method based on space-time volume of local joint point track in skeleton sequence | |
| CN110781829A (en) | Light-weight deep learning intelligent business hall face recognition method | |
| CN109101108B (en) | Method and system for optimizing human-computer interaction interface of intelligent cabin based on three decisions | |
| CN111461025B (en) | Signal identification method for self-evolving zero-sample learning | |
| CN114038037A (en) | Expression label correction and identification method based on separable residual attention network | |
| CN110795990A (en) | Gesture recognition method for underwater equipment | |
| CN108288048B (en) | Facial emotion recognition feature selection method based on improved brainstorming optimization algorithm | |
| CN110728694A (en) | Long-term visual target tracking method based on continuous learning | |
| CN111259735B (en) | Single-person attitude estimation method based on multi-stage prediction feature enhanced convolutional neural network | |
| CN115294658B (en) | Personalized gesture recognition system and gesture recognition method for multiple application scenes | |
| CN114360005B (en) | Micro-expression classification method based on AU region and multi-level transducer fusion module | |
| CN111753683A (en) | Human body posture identification method based on multi-expert convolutional neural network | |
| CN116363712B (en) | Palmprint palm vein recognition method based on modal informativity evaluation strategy | |
| CN110751027A (en) | Pedestrian re-identification method based on deep multi-instance learning | |
| EP2115737B1 (en) | Method and system to improve automated emotional recognition | |
| CN115170868A (en) | Clustering-based small sample image classification two-stage meta-learning method | |
| CN115797827A (en) | ViT human body behavior identification method based on double-current network architecture | |
| CN116884067B (en) | Micro-expression recognition method based on improved implicit semantic data enhancement | |
| CN106250818B (en) | A kind of total order keeps the face age estimation method of projection | |
| CN117333908A (en) | Cross-modal pedestrian re-recognition method based on attitude feature alignment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |