CN110008861A - A kind of recognition methods again of the pedestrian based on global and local feature learning - Google Patents
A kind of recognition methods again of the pedestrian based on global and local feature learning Download PDFInfo
- Publication number
- CN110008861A CN110008861A CN201910219450.2A CN201910219450A CN110008861A CN 110008861 A CN110008861 A CN 110008861A CN 201910219450 A CN201910219450 A CN 201910219450A CN 110008861 A CN110008861 A CN 110008861A
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- feature
- training
- indicate
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 29
- 238000012549 training Methods 0.000 claims abstract description 29
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 21
- 238000012360 testing method Methods 0.000 claims abstract description 11
- 230000006870 function Effects 0.000 claims description 29
- 238000013507 mapping Methods 0.000 claims description 15
- 238000010606 normalization Methods 0.000 claims description 6
- 239000013598 vector Substances 0.000 claims description 5
- 238000011176 pooling Methods 0.000 claims description 4
- 230000004927 fusion Effects 0.000 claims description 3
- 238000011524 similarity measure Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 2
- 230000000644 propagated effect Effects 0.000 claims description 2
- 239000012141 concentrate Substances 0.000 claims 1
- 238000005516 engineering process Methods 0.000 description 8
- 238000013135 deep learning Methods 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000007630 basic procedure Methods 0.000 description 1
- 230000037237 body shape Effects 0.000 description 1
- 238000011840 criminal investigation Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000001815 facial effect Effects 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of recognition methods again of the pedestrian based on global and local feature learning, comprising the following steps: S1, obtains training dataset, training data, which is carried out pretreatment and data, to be enhanced;S2, building depth convolutional neural networks;Step S3, the training data training network is utilized;S4, test data set is obtained, and it is pre-processed, the feature of each test set image is then extracted using trained network;The similarity score of S5, the feature for calculating each Query data and the feature in Gallery data set;S6, it sorts to all similarity scores, the Gallery pedestrian image of highest scoring can consider that with corresponding query pedestrian be the same pedestrian, and then obtain the result of images to be recognized.Network proposed by the present invention not only simply, but also does not need additional pedestrian information and can obtain accuracy rate more higher than other classical ways.
Description
Technical field
The present invention relates to pedestrian identification technology fields again, and in particular to a kind of row based on global and local feature learning
People's recognition methods again.
Background technique
With the gradually development of social economy and science and technology, intelligent monitoring technology is had been to be concerned by more and more people.School, doctor
Institute, railway station, the biggish public place of flows of the people such as airport is assembled with a large amount of camera, by the video data of these magnanimity
It is researched and analysed, this is of great significance in public safety, the fields such as criminal investigation.
Identification technology refers to and had already appeared a pedestrian in some camera pedestrian again, when the pedestrian takes the photograph at another
As head when occurring again, it would be desirable to identify him.Identification technology is different from recognition of face to pedestrian again.Face is known
The facial image background used in not is relatively simple, and face is easy to discrimination than more visible.And pedestrian identify again in pedestrian figure
Picture resolution ratio is lower, and face information is fuzzy, and background is complex, it is difficult to correct matching;Secondly, the shooting between different cameras
There are great differences for angle, and pedestrian is likely to change in the posture occurred every time either figure and features feature.Based on these
Feature, so that our analyses to image, the extraction of pedestrian's feature are all extremely difficult.Current pedestrian identifies the technology in field again
Be roughly divided into two classes: the first kind is to study the character representation method of pedestrian's object, extracts the diagnostic characteristics pair for having more robustness
Pedestrian is indicated;Second class uses learning distance metric method, by learning the distance metric function for having judgement index, makes
The image distance of the same person is obtained less than the distance between different pedestrian images.Recent years with deep learning development, more
Method focus on character representation this aspect of pedestrian, there are three types of the technologies of mainstream: the first kind is global characteristics, global characteristics
It is concerned with global information, such as the gender of pedestrian, body shape, clothes color etc..But global characteristics tend to lose
The mistake in information and pedestrian detection in details.Second class is local feature, and many methods are directly by entire pedestrian's picture
It is divided into the part of several fixations, they is then inputted into training in neural network, but this mode has ignored the posture of pedestrian
Change and block the influence to the picture of cutting.Third class is to combine global characteristics and local message, and this mode directly merges
The global and local information feature descriptor last as pedestrian, the disadvantage is that often having bigger calculating cost and additional
Memory space.From this, above-mentioned three classes method, all cannot sufficiently excavate overall situation and partial situation's feature of pedestrian.
Summary of the invention
The purpose of the present invention is to propose to a kind of recognition methods again of the pedestrian based on global and local feature learning, existing to solve
Some deep learning methods are unable to fully the technical issues of excavating the global and local feature of pedestrian.
The purpose of the present invention is realized at least through following technical solution.
Pedestrian based on global and local feature learning recognition methods again, comprising the following steps:
Step S1, training dataset is obtained, training data, which is carried out pretreatment and data, to be enhanced;
Step S2, depth convolutional neural networks are constructed;
Step S3, the training data training depth convolutional neural networks handled well are utilized;
Step S4, test data set is obtained, and it is pre-processed, then extracts each survey using trained network
Examination collection
The feature of image;
Step S5, the feature and the feature in Candidate Set (Gallery) data set for calculating each query set (Query) data
Similarity score;
Step S6, it sorts to all similarity score, the Gallery pedestrian image of highest scoring can consider and therewith
Corresponding Query pedestrian is the same pedestrian, and then obtains the result of images to be recognized.
Further, the test data set includes Query data set and Gallery data set.
Further, the pretreatment of the step S1 be the RGB image size of each pedestrian is adjusted to 256*144, and
And to its mean normalization;Data enhancement method includes that picture size size is cut to 256*128 and level by random cropping
Overturning etc..
Further, the step S2 depth convolutional neural networks building the following steps are included:
Step S21, the all-network layer (convolution before the last layer convolutional layer (Conv5 layers) of Resnet50 is intercepted
Layer Conv5) and the good parameter of pre-training on data set ImageNet is used to initialize it;The parameter include weight to
Measure θ1,θ2,…,θm,…θnAnd biasing;
Step S22, in practice, it is contemplated that the vertical direction of pedestrian image can intuitively be divided into different parts, example
Such as head, chest, leg, there are also feet etc..After Conv5 layers, part pond (Local is carried out to the output X of Conv5
Average Pooling), it is that the output is cut into k part (Part), then respectively to this k part pond, Qi Chihua
Receptive field be (H/k) * W, wherein H, W and k are the length of the output of Conv5 and the quantity of wide and cutting part respectively, each
Each element representation of Part are as follows:
Here, Xc,i,jIndicate that each element of convolutional layer Conv5 output, i, j are illustrated respectively in the rope in long and wide direction
Draw, c indicates that c ties up channel.Δ=H/k.
Step S23, (Mapping) study is mapped to each Part that cutting obtains, the result after mapping are as follows:
Wherein σ1And σ2It is line rectification function (ReLU) and Sigmoid function respectively.It is convolution kernel ginseng
Number.
Each of step S24, in view of the information of proximate region in pedestrian image is similar, so will be obtained by mapping study
Part first replicates (Repeat) once, then gets up in height (H) dimension splicing (Cat);
Step S25, the tensor being stitched together (Tensor) is multiplied point by point with X, realizes the selection of local feature, selection
As a result it indicates are as follows:
Wherein, XC, i, jIndicate each element of convolutional layer Conv5 output, Sc,i,jIndicate mapping study obtain as a result,
Indicate the operation being multiplied point by point.
Step S26, the fusion of global characteristics is carried out to the result that step S25 is obtained, i.e., global average pond (Global
AveragePooling it) operates.
Further, the step S3 training the following steps are included:
Step S31, by training data, tissue line is good as needed, network described in input step S2.
Step S32, the loss function of the depth convolutional neural networks is set:
Wherein, λ1、λ2、λ3、λ4It is the coefficient of corresponding loss function with u, is respectively set to 0.1,0.1,0.1,0.1 and
0.6, p1、p2、p3、p4Respectively indicate each section respectively corresponded in the k part with G, G correspond to the cutting before it is whole
Body;WithRespectively indicate the loss function of corresponding part;
Step S33, by the parameter of the loss function and depth convolutional neural networks, propagated forward penalty values are obtained;
The parameter includes weight vectors θ1,θ2,…,θm,…θn。
Step S34, backpropagation obtains training error.
Further, this method all uses Softmax classification function to local feature and global characteristics;For Softmax
Loss function first has to calculate image pattern (x(z)) belong to the probability of each classification.Assuming that all samples are divided into n class, it is right
Input sample x(z)(z indicates z-th of sample), belongs to the probability value of classification m are as follows:
Wherein, θ1,θ2,…,θm,…θnIt is the parameter of depth convolutional neural networks, by SmFormula obtain Softmax loss
Function:
Wherein, y is the vector of a 1*n, ymIndicate to be 1 when the corresponding position of the sample is true classification.
Further, test data set its pretreatment mode in the step S4 is that image size is adjusted to 256*
144, and to its mean normalization.
Further, the similarity measure of the step S5 uses Euclidean distance, the Euclidean distance formula are as follows:
Wherein xuAnd xvRespectively indicate v-th of pedestrian in the feature and Gallery of u-th of pedestrian in Query data set
Feature
Compared with prior art, the present invention having the advantages that the present invention cannot for existing depth learning technology
The problem of sufficiently excavating the global and local feature of pedestrian proposes a kind of new network structure.The structure can be automatically real
The selection of current situation portion and global information learns, and simultaneously to local feature and global characteristics using Softmax loss function come into
Row constraint obtains a feature descriptor with very robust, so that improve pedestrian identifies matched accuracy again.
Detailed description of the invention
Fig. 1 is the basic procedure that pedestrian identifies again;
Fig. 2 is that the present invention is based on the depth network structures of the global and local feature selecting of pedestrian.
Specific embodiment
In order to be more clear technical solution of the present invention and advantage, come With reference to embodiment and referring to attached drawing
The present invention is described in more detail.
A kind of recognition methods again of the pedestrian based on global and local feature learning as shown in Figure 1, comprising the following steps:
Step S1, training dataset is obtained, training data, which is carried out pretreatment and data, to be enhanced
The present invention uses three disclosed pedestrians identification database again: Market-1501, DukeMTMC-reID and
CUHK03.Since the picture size in raw data set is different, it is not able to satisfy the input needs of neural network, so every
RGB picture Resize is at 256*144 size, and to its mean normalization.Then in order to promote deep learning e-learning
Robustness prevents over-fitting, carries out data enhancing to obtained data, mode includes that (size is 256* to random cropping
128), the modes such as flip horizontal.
Step S2, depth convolutional neural networks are constructed
As shown in Fig. 2, the step S2 the following steps are included:
Step S21, the last layer convolutional layer Conv5 for intercepting Resnet-50 (depth residual error network -50) (includes Conv5
Layer) before all-network layer as basic network (Base Network), and use pre-training on ImageNet is good
Parameter it is initialized, the network structure of addition is initialized using Gauss.
Step S22, after Conv5 layers, to the output X of Conv5 carry out local pond (Local Average Pooling,
LAP), the receptive field in pond is (H/k) * W, and wherein H, W and k are the length of the output of Conv5 and the part of width and cutting respectively
Quantity,If it is respectively 8 and 4 that the size for inputting picture, which is 256*128, H and W,.In practice, consider
Vertical direction to pedestrian can intuitively be divided into different parts, such as head, and chest, leg, there are also feet etc..So can be with
It is 1,2,3,4 etc. by k value, the value of the present embodiment k is 4.
Each element in each part can indicate are as follows:
Here, Xc,i,jIndicate that each element of convolutional layer Conv5 output, i, j are illustrated respectively in the rope in long and wide direction
Draw, c indicates that c ties up channel.Δ=H/k,
Step S23, mapping study is carried out to each part that cutting obtains.Each mapping is by convolutional layer, ReLU letter
Number, convolutional layer, sigmoid function cascaded, and the parameter of this part are without shared.
Result after mapping are as follows:
Wherein σ1And σ2It is ReLU function and Sigmoid function respectively,WithBeing is convolution nuclear parameter,
Step S24, in view of the information of proximate region in pedestrian image is similar, so will be by mapping (Mapping) study
Obtained each Part first replicates (Repeat) once, then gets up in height (H) dimension splicing (Cat);
Step S25, the Tensor being stitched together is multiplied point by point with X, realizes the selection of local feature.Selection
(selection) result are as follows:
Wherein, Xc,i,jIndicate each element of convolutional layer Conv5 output, Sc,i,jIndicate mapping study obtain as a result,
Indicate the operation being multiplied point by point,
Step S26, the fusion of global characteristics is carried out to the result that step S25 is obtained, i.e., global average pond (Global
AveragePooling, GAP) operation.
Step S3, the training data training depth convolutional neural networks handled well are utilized
Wherein, the step S3 is comprised the steps of:
Step S31, by training data, tissue line is good as needed, network described in input step S2;
The present invention individually puts each different classes of pedestrian image together when implementing.
Step S32, the loss function of the depth convolutional neural networks is set.This method is special to local feature and the overall situation
Sign all employs Softmax classification function, calculates sample (x(z)) belonging to the probability of each classification, calculating process is as follows:
Assuming that all samples are divided into n class, to input sample x(z)(z indicates z-th of sample), belongs to the probability of classification m
Value are as follows:
Wherein, θ1,θ2,…,θm,…θnIt is the parameter of depth convolutional neural networks, by SmFormula obtain Softmax loss
Function:
Wherein, y is the vector of a 1*n, ymIndicate to be 1 when the corresponding position of the sample is true classification.
As shown in Fig. 2, loss function used in the present embodiment is 5 Classification Loss functions (Loss), respectively to part
Feature and global characteristics constraint.The final loss function of depth convolutional neural networks are as follows:
Wherein, λ1、λ2、λ3、λ4It is the coefficient of corresponding loss function with u, is respectively set to 0.1,0.1,0.1,0.1 and
0.6, p1、p2、p3、p4Each section in the k part is respectively corresponded with G, G corresponds to the entirety before the cutting;WithRespectively indicate the loss function of corresponding part;
Step S33, before being obtained by the loss function and depth convolutional neural networks parameter (including weight and biasing)
To propagation loss value;
The deep learning optimization algorithm that the present invention uses is stochastic gradient descent method, trains 60 iteration (Epoch) altogether,
Wherein the learning rate of preceding 40 epoch is set as 0.1, and the learning rate of 20 Epoch is set as 0.01 later.Batch size
(Batchsize) 32 are set as.
Step S34, backpropagation obtains training error
Utilize L described in step S32lossAs benchmark, back transfer training error.
Step S4, test data set (comprising Query data set and Gallery data set) is obtained, and it is located in advance
Reason, so
The feature of all images in test set is extracted using trained network afterwards;
Wherein, the test data set in the step S4 needs to reset size, is uniformly cut to 256*144, and
To its mean normalization.
Step S5, the similarity for calculating the feature and the feature in the Gallery data set of each Query data obtains
Point;The feature refers to the feature in step S4.
Wherein, the similarity measure in the step S5 uses Euclidean distance:
Wherein xuAnd xvRespectively indicate v-th of pedestrian in the feature and Gallery of u-th of pedestrian in Query data set
Feature
Step S6, it sorts to all similarity score, the Gallery pedestrian image of highest scoring can consider and therewith
Corresponding Query pedestrian is the same pedestrian, and then obtains the result of images to be recognized.
Rank-1 index and mAP index have been used to the evaluation criterion of result in the present invention.
The foregoing is merely illustrative embodiments of the invention, the protection scope being not intended to limit the invention is all at this
Within the spirit and principle of invention, any modification, equivalent substitution, improvement and etc. done should be included in protection model of the invention
Within enclosing.
Claims (8)
1. a kind of recognition methods again of the pedestrian based on global and local feature learning, which is characterized in that the method includes following
Step:
Step S1, training dataset is obtained, training data, which is carried out pretreatment and data, to be enhanced;
Step S2, depth convolutional neural networks are constructed;
Step S3, the training data training depth convolutional neural networks handled well are utilized;
Step S4, test data set is obtained, and it is pre-processed, is then mentioned using trained depth convolutional neural networks
Test data is taken to concentrate the feature of all images;
Step S5, the phase of the feature and the feature in Candidate Set (Gallery) data set of each query set (Query) data is calculated
Like degree score;The feature refers to the feature in step S4;
Step S6, sort to all similarity score, the Gallery pedestrian image of highest scoring then think with it is corresponding
Query pedestrian is the same pedestrian, and then obtains the result of images to be recognized.
2. pedestrian according to claim 1 recognition methods again, which is characterized in that the test data set includes Query number
According to collection and Gallery data set.
3. pedestrian according to claim 1 recognition methods again, which is characterized in that the pretreatment of the step S1 is each
The RGB image size of pedestrian is adjusted to 256*144, and to its mean normalization;Data enhancement method include random cropping i.e.
Picture size size is cut to 256*128 and flip horizontal mode.
4. pedestrian according to claim 1 recognition methods again, which is characterized in that the step S2 depth convolutional neural networks
Building the following steps are included:
Step S21, the all-network layer before the last layer convolutional layer Conv5 of Resnet50, including convolutional layer are intercepted
Conv5, and the good parameter of pre-training on ImageNet data set is used to initialize it;The parameter includes weight vectors
θ1, θ2..., θm... θn;
Step S22, local pond (Local Average Pooling) is carried out to the output X of Conv5, is by the outputting cutting
It is divided into k part (Part), then respectively to this k part pond, the receptive field in pond is (H/k) * W, wherein H, W and k points
It is not the length of the output of Conv5 and the quantity of wide and cutting part, each element representation of each Part are as follows:
Here, XC, i, jIndicate that each element of convolutional layer Conv5 output, i, j are illustrated respectively in the index in long and wide direction, c
Indicate that c ties up channel, Δ=H/k;
Step S23, (Mapping) study is mapped to each Part that cutting obtains, the result after mapping are as follows:
Wherein, VC, kIndicate mapping study obtain as a result, σ1And σ2It is line rectification function (ReLU) and Sigmoid letter respectively
Number,WithIt is convolution nuclear parameter;
Step S24, in view of the information of proximate region in pedestrian image is similar, so each Part that will be obtained by mapping study
First duplication (Repeat) once, is then got up in height (H) dimension splicing (Cat);
Step S25, the tensor being stitched together (Tensor) is multiplied point by point with X, realizes the selection of local feature, the result of selection
It indicates are as follows:
Wherein, XC, i, jIndicate each element of convolutional layer Conv5 output, SC, i, jIndicate mapping study obtain as a result,It indicates
The operation being multiplied point by point;
Step S26, the fusion of global characteristics is carried out to the result that step S25 is obtained, i.e., global average pond (Global
Average Pooling) operation.
5. according to the method described in claim 4, it is characterized in that, the training of the step S3 the following steps are included:
Step S31, by depth convolutional neural networks described in training data input step S2;
Step S32, the loss function of the depth convolutional neural networks is set:
Wherein, λ1、λ2、λ3、λ4It is the coefficient of corresponding loss function with u, is respectively set to 0.1,0.1,0.1,0.1 and 0.6, p1、
p2、p3、p4Each section in the k part is respectively corresponded, G corresponds to the entirety before the cutting;WithRespectively indicate the loss function of corresponding part;
Step S33, by the parameter of the loss function and depth convolutional neural networks, propagated forward penalty values are obtained;
Step S34, backpropagation obtains training error.
6. the method according to claim 1, wherein being all made of Softmax points to local feature and global characteristics
Class function calculates image pattern x(z)Belong to the probability of each classification, calculating process is as follows:
Assuming that all samples are divided into n class, to input sample x(z), z indicate z-th of sample, belong to the probability value of classification m are as follows:
Wherein, θ1, θ2..., θm... θnIt is the parameter of depth convolutional neural networks, by SmFormula obtain Softmax loss letter
Number:
Wherein, y is the vector of a l*n, ymIndicate to be 1 when the corresponding position of the sample is true classification.
7. the method according to claim 1, wherein pretreatment described in step S4 is to be adjusted to image size
256*144, and to its mean normalization.
8. being somebody's turn to do the method according to claim 1, wherein the similarity measure of the step S5 uses Euclidean distance
Euclidean distance formula are as follows:
Wherein xuAnd xvRespectively indicate the spy of v-th of pedestrian in the feature and Gallery of u-th of pedestrian in Query data set
Sign.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910219450.2A CN110008861A (en) | 2019-03-21 | 2019-03-21 | A kind of recognition methods again of the pedestrian based on global and local feature learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910219450.2A CN110008861A (en) | 2019-03-21 | 2019-03-21 | A kind of recognition methods again of the pedestrian based on global and local feature learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110008861A true CN110008861A (en) | 2019-07-12 |
Family
ID=67167747
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910219450.2A Pending CN110008861A (en) | 2019-03-21 | 2019-03-21 | A kind of recognition methods again of the pedestrian based on global and local feature learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110008861A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110688976A (en) * | 2019-10-09 | 2020-01-14 | 创新奇智(北京)科技有限公司 | Store comparison method based on image identification |
CN110781817A (en) * | 2019-10-25 | 2020-02-11 | 南京大学 | Pedestrian re-identification method for solving component misalignment |
CN111275712A (en) * | 2020-01-15 | 2020-06-12 | 浙江工业大学 | Residual semantic network training method oriented to large-scale image data |
CN112149517A (en) * | 2020-08-31 | 2020-12-29 | 三盟科技股份有限公司 | Face attendance checking method and system, computer equipment and storage medium |
CN112200093A (en) * | 2020-10-13 | 2021-01-08 | 北京邮电大学 | Pedestrian re-identification method based on uncertainty estimation |
CN113269070A (en) * | 2021-05-18 | 2021-08-17 | 重庆邮电大学 | Pedestrian re-identification method fusing global and local features, memory and processor |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107506703A (en) * | 2017-08-09 | 2017-12-22 | 中国科学院大学 | A kind of pedestrian's recognition methods again for learning and reordering based on unsupervised Local Metric |
CN108062756A (en) * | 2018-01-29 | 2018-05-22 | 重庆理工大学 | Image, semantic dividing method based on the full convolutional network of depth and condition random field |
CN108875696A (en) * | 2018-07-05 | 2018-11-23 | 五邑大学 | The Off-line Handwritten Chinese Recognition method of convolutional neural networks is separated based on depth |
CN108960140A (en) * | 2018-07-04 | 2018-12-07 | 国家新闻出版广电总局广播科学研究院 | The pedestrian's recognition methods again extracted and merged based on multi-region feature |
CN109034044A (en) * | 2018-06-14 | 2018-12-18 | 天津师范大学 | A kind of pedestrian's recognition methods again based on fusion convolutional neural networks |
CN109271926A (en) * | 2018-09-14 | 2019-01-25 | 西安电子科技大学 | Intelligent Radiation source discrimination based on GRU depth convolutional network |
-
2019
- 2019-03-21 CN CN201910219450.2A patent/CN110008861A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107506703A (en) * | 2017-08-09 | 2017-12-22 | 中国科学院大学 | A kind of pedestrian's recognition methods again for learning and reordering based on unsupervised Local Metric |
CN108062756A (en) * | 2018-01-29 | 2018-05-22 | 重庆理工大学 | Image, semantic dividing method based on the full convolutional network of depth and condition random field |
CN109034044A (en) * | 2018-06-14 | 2018-12-18 | 天津师范大学 | A kind of pedestrian's recognition methods again based on fusion convolutional neural networks |
CN108960140A (en) * | 2018-07-04 | 2018-12-07 | 国家新闻出版广电总局广播科学研究院 | The pedestrian's recognition methods again extracted and merged based on multi-region feature |
CN108875696A (en) * | 2018-07-05 | 2018-11-23 | 五邑大学 | The Off-line Handwritten Chinese Recognition method of convolutional neural networks is separated based on depth |
CN109271926A (en) * | 2018-09-14 | 2019-01-25 | 西安电子科技大学 | Intelligent Radiation source discrimination based on GRU depth convolutional network |
Non-Patent Citations (2)
Title |
---|
李姣等: "多置信度重排序的行人再识别算法", 《模式识别与人工智能》 * |
王鹏 等: "Local-Global Extraction Unit for Person Re-identification", 《 INTERNATIONAL CONFERENCE ON BRAIN INSPIRED COGNITIVE SYSTEM》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110688976A (en) * | 2019-10-09 | 2020-01-14 | 创新奇智(北京)科技有限公司 | Store comparison method based on image identification |
CN110781817A (en) * | 2019-10-25 | 2020-02-11 | 南京大学 | Pedestrian re-identification method for solving component misalignment |
CN111275712A (en) * | 2020-01-15 | 2020-06-12 | 浙江工业大学 | Residual semantic network training method oriented to large-scale image data |
CN112149517A (en) * | 2020-08-31 | 2020-12-29 | 三盟科技股份有限公司 | Face attendance checking method and system, computer equipment and storage medium |
CN112200093A (en) * | 2020-10-13 | 2021-01-08 | 北京邮电大学 | Pedestrian re-identification method based on uncertainty estimation |
CN113269070A (en) * | 2021-05-18 | 2021-08-17 | 重庆邮电大学 | Pedestrian re-identification method fusing global and local features, memory and processor |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110008861A (en) | A kind of recognition methods again of the pedestrian based on global and local feature learning | |
CN105512680B (en) | A kind of more view SAR image target recognition methods based on deep neural network | |
CN108460403A (en) | The object detection method and system of multi-scale feature fusion in a kind of image | |
CN111666843B (en) | Pedestrian re-recognition method based on global feature and local feature splicing | |
CN104166841B (en) | The quick detection recognition methods of pedestrian or vehicle is specified in a kind of video surveillance network | |
CN108537136A (en) | The pedestrian's recognition methods again generated based on posture normalized image | |
CN109934176A (en) | Pedestrian's identifying system, recognition methods and computer readable storage medium | |
CN109101865A (en) | A kind of recognition methods again of the pedestrian based on deep learning | |
CN108710868A (en) | A kind of human body critical point detection system and method based under complex scene | |
CN107330396A (en) | A kind of pedestrian's recognition methods again based on many attributes and many strategy fusion study | |
CN110070010A (en) | A kind of face character correlating method identified again based on pedestrian | |
CN109447115A (en) | Zero sample classification method of fine granularity based on multilayer semanteme supervised attention model | |
CN108229444A (en) | A kind of pedestrian's recognition methods again based on whole and local depth characteristic fusion | |
CN107463920A (en) | A kind of face identification method for eliminating partial occlusion thing and influenceing | |
CN108960184A (en) | A kind of recognition methods again of the pedestrian based on heterogeneous components deep neural network | |
CN106529499A (en) | Fourier descriptor and gait energy image fusion feature-based gait identification method | |
CN110008913A (en) | The pedestrian's recognition methods again merged based on Attitude estimation with viewpoint mechanism | |
Liu et al. | A contrario comparison of local descriptors for change detection in very high spatial resolution satellite images of urban areas | |
CN107133569A (en) | The many granularity mask methods of monitor video based on extensive Multi-label learning | |
CN109190475A (en) | A kind of recognition of face network and pedestrian identify network cooperating training method again | |
CN108416295A (en) | A kind of recognition methods again of the pedestrian based on locally embedding depth characteristic | |
CN109741240A (en) | A kind of more flat image joining methods based on hierarchical clustering | |
CN108564012A (en) | A kind of pedestrian's analytic method based on characteristics of human body's distribution | |
CN109784130A (en) | Pedestrian recognition methods and its device and equipment again | |
CN104298974A (en) | Human body behavior recognition method based on depth video sequence |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190712 |