CN108460407A - A kind of pedestrian's attribute fining recognition methods based on deep learning - Google Patents
A kind of pedestrian's attribute fining recognition methods based on deep learning Download PDFInfo
- Publication number
- CN108460407A CN108460407A CN201810105618.2A CN201810105618A CN108460407A CN 108460407 A CN108460407 A CN 108460407A CN 201810105618 A CN201810105618 A CN 201810105618A CN 108460407 A CN108460407 A CN 108460407A
- Authority
- CN
- China
- Prior art keywords
- attribute
- pedestrian
- layer
- sample
- classification
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of, and pedestrian's attribute based on deep learning refines recognition methods, by improving deep learning model framework, pedestrian's attributive character is arrived in automatic study, then attributive character is input in grader, training pedestrian's attributive classification device independent one by one, can obtain pedestrian sample in this way has the posterior probability of attribute, then by calculating the attribute in pedestrian's training sample and the other proportionate relationship of Attribute class, the other posterior probability of Attribute class is obtained, the attribute classification of pedestrian sample is finally can be obtained by according to Bayesian formula.The present invention can improve accuracy of identification and recognition speed.
Description
Technical field
The present invention relates to mode identification technologies, are refined more particularly to a kind of pedestrian's attribute based on deep learning
Recognition methods.
Background technology
With in recent years, the concept of safe city proposition.Monitor video is distributed in each corner in city, maintains
Urban safety.In case of burst accident, effective information is searched out from the monitoring image of magnanimity, will necessarily expended a large amount of
Manpower and materials.And pedestrian, if can effectively be identified to pedestrian's attribute, can regard as main monitoring objective to monitoring
Frequency retrieval work brings prodigious facility.Pedestrian's Attribute Recognition identifies again in video monitoring, Intelligent Business video, pedestrian, face
The fields such as identification knowledge have a wide range of applications, and receive the concern of more and more researchers.The one of pedestrian's Attribute Recognition
Kind tional identification algorithm is artificial neural network, it is abstracted human brain neuroid from information processing angle, establishes certain
Kind naive model.Training algorithm based on artificial neural network is back-propagation algorithm, it makes network model pass through to a large amount of
The process that training sample is learnt can obtain statistical law, to make prediction to unknown event.Artificial neural network is excellent
Point is there is stronger non-linear mapping capability, self study and adaptive ability, generalization ability and certain fault-tolerant ability, but
It has the following disadvantages, when pedestrian identifies sample training, convergence rate is slow, and its training process is monitor procedure, and to training sample
This mark is cumbersome and time consuming, and video pedestrian identification is related to the calculating and analysis of mass data, some additional environment
The interference of factor, conventional pedestrian's Attribute Recognition algorithm can not extract the preferred feature of image, cause discrimination limited.
Invention content
Technical problem to be solved by the invention is to provide a kind of, and pedestrian's attribute based on deep learning refines identification side
Method can improve accuracy of identification and speed.
The technical solution adopted by the present invention to solve the technical problems is：A kind of pedestrian's attribute based on deep learning is provided
Recognition methods is refined, is included the following steps：
(1) attribute labeling is carried out to pedestrian sample data set, obtains pedestrian's attribute data collection；
(2) build the convolutional neural networks model finely tuned based on AlexNet networks, and to the convolutional neural networks model into
Row training；
(3) to convolutional neural networks mode input pedestrian's test pictures after training, pedestrian's attributive character is extracted；
(4) pedestrian's attributive character of extraction is input in grader, training human body attributive classification device；
(5) attribute x is calculated by pedestrian's attribute data collectioniSample belong to classification yjRatio, i.e., What is indicated is to possess attribute xiAnd belong to classification yjSample size,What is indicated is to possess attribute xi
Sample size, calculate the other quantitative relation of the corresponding Attribute class of each attribute；
(6) by pedestrian's test sample ztIt is input to the convolutional neural networks model that training finishes, and extracts feature；It then will extraction
Feature be input in the human body attributive classification device, obtain pedestrian sample have attribute xiPosterior probability p (xi|zt), pass through pedestrian
Quantitative relation between the corresponding attribute classification of attribute, utilizes Bayesian formula：
Calculating pedestrian sample has classification yjPosterior probability, the classification y calculatedjProbability in the corresponding class of maximum value
It is not exactly the attribute classification of pedestrian sample, wherein the quantity of N pedestrian's attribute,.
42 attributes of pedestrian can be represented in the step 1 to pedestrian's picture mark by picture annotation tool；42
A attribute is divided into gender attribute, hair lengths attribute, upper body clothing type attribute, upper body clothing color attribute, lower garment class
Type attribute, lower garment color attribute, Packet type attribute and footwear type attribute, wherein gender attribute include that man and female, hair are long
Degree attribute includes long hair and bob, and upper body clothing type attribute includes T-shirt, shirt, housing and down jackets, and upper body wears color category clothes
Property include black, white, red, yellow, blue, green, purple, brown, grey, orange and polychrome, lower garment type attribute be trousers, shorts, longuette and
Skirt, lower garment color attribute include black, white, red, yellow, blue, green, purple, brown, grey, orange and polychrome, and Packet type attribute includes single
Shoulder packet, both shoulders packet pull case and wallet, and footwear type includes leather shoes, sport footwear, sandals and boots.
It is finely tuned in the convolutional neural networks model structure of the fine tuning based on AlexNet network models in the step (2)
AlexNet model structures are as follows：First layer is convolutional layer, characteristic plane 96, convolution kernel size 11x11；The second layer is convolution
Layer, characteristic plane 256, convolution kernel size 5x5；Third layer is convolutional layer, characteristic plane 384, convolution kernel size 3x3；The
Four layers are convolutional layers, and characteristic plane is 384, convolution kernel size 3x3；Layer 5 is convolutional layer, and characteristic plane is 256, volume
Product core size 3x3；Layer 6 is full articulamentum, is remained unchanged；Layer 7 is full articulamentum, is remained unchanged；8th layer is to connect entirely
Layer is connect, in this layer according to the number of pedestrian's attribute, changes the node of output.
The convolutional neural networks model is trained in the step (2) and specifically includes following sub-step：
(21) deconvolution operation is carried out to each layer of output of AlexNet convolutional neural networks, then this layer is defeated
Enter the result exported with deconvolution to be compared, obtains error E；
(22) the weights size of the convolution kernel of model is constantly adjusted by gradient descent method, formula is：
W in formula*For newer weights, W is original weights, and η is the learning rate of setting,Indicate local derviation of the error to weights；
(23) training sample is input in model, carries out 20000 repetitive exercises, obtaining one after training changes
Into convolutional neural networks model.
The step (4) includes following sub-step：
(41) it is first each attribute, designs independent grader；
(42) grader selects support vector machines, the kernel function of support vector machines in the training process to select radial base letter
Number, mini-max optimization method determine that the value of radial base parameter σ is 3.35；
(43) pedestrian's attributive character of extraction is input in the grader of the step (42) configuration can be by identification essence
Degree is increased to 94% or more.
Advantageous effect
Due to the adoption of the above technical solution, compared with prior art, the present invention having the following advantages that and actively imitating
Fruit：
The present invention, using the method for deep learning, improves accuracy of identification compared with traditional pedestrian's attribute recognition approach.
Convolutional neural networks can not only extract preferable attributive character, and be unsupervised in training process, reduce pedestrian sample
The cost of attribute labeling.
The present invention can extract high-rise semanteme compared with traditional method for carrying out Attribute Recognition using low-level image feature
Information can also embody better recognition effect under the uncontrollable factors such as illumination, visual angle.
Description of the drawings
Fig. 1 is a kind of flow chart of pedestrian's attribute recognition approach based on deep learning in the embodiment of the present invention.
Specific implementation mode
Present invention will be further explained below with reference to specific examples.It should be understood that these embodiments are merely to illustrate the present invention
Rather than it limits the scope of the invention.In addition, it should also be understood that, after reading the content taught by the present invention, people in the art
Member can make various changes or modifications the present invention, and such equivalent forms equally fall within the application the appended claims and limited
Range.
Embodiments of the present invention are related to a kind of pedestrian's attribute fining recognition methods based on deep learning, are divided into pedestrian
The mark of attribute data collection, the structure of deep learning model and training, the study of pedestrian's attribute, pedestrian's attribute classification mapping relations
It practises and five parts of test sample.Deep learning and the study of pedestrian's attribute are combined, the essence of pedestrian's Attribute Recognition can be improved
Exactness identifies to achieve the purpose that refine pedestrian's attribute.The mark part of pedestrian's attribute data collection, passes through image labeling
Tool is labelled with 42 attributes that can represent pedestrian.The structure of deep learning model and training part, are divided into two stages.It is deep
It spends on the structure of learning model, is finely tuned based on AlexNet convolutional neural networks model, fine tuning is full articulamentum portion
Point, relevant parameter is set, to obtain improved AlexNet network models.Pedestrian's attribute learns, and training sample is input to
In improved AlexNet network models, pedestrian's attributive character is extracted.For one grader of each attribute design.It will extraction
Attributive character be input in each grader, attribute is learnt.Humanized classification mapping relations of being expert at learn part,
Each attribute sample is calculated by training sample first and accounts for the other ratio of Attribute class, to obtain attribute classification mapping relations
Figure.In test sample part, attributive character is input in trained attributive classification device, obtained pedestrian sample belongs to the category
The posterior probability of property is inferred to the posterior probability of pedestrian image classification then according to the posterior probability and attribute classification mapping table,
Choose identification classification of the corresponding classification of maximum probability as sample.
The present invention basic ideas be：The mapping relations between pedestrian sample and pedestrian's attribute are obtained first, then in conjunction with
By the proportionate relationship between the pedestrian's attribute being calculated and attribute classification, so as to release pedestrian sample and pedestrian's attribute
Relationship between classification.
Below by a specific embodiment, the present invention is further explained.
As shown in Figure 1, a kind of pedestrian's attribute based on deep learning of the present invention refines recognition methods, including such as
Under step：
(1) in the step of carrying out attribute labeling to pedestrian sample data set, obtaining pedestrian's attribute data collection, pass through picture mark
Note tool can represent pedestrian's picture mark 42 attributes of pedestrian.These attributes include gender (man, female), and hair lengths are (long
Hair, bob), upper body clothing type (T-shirt, shirt, housing, down jackets), upper body clothing color (it is black, it is in vain, red, it is yellow, it is blue, it is green,
Purple, palm fibre, ash, orange, polychrome), lower garment type (trousers, shorts, longuette, skirt), lower garment color (it is black, it is in vain, red, it is yellow,
Indigo plant, green, purple, palm fibre, ash, orange, polychrome), Packet type (shoulder bag, both shoulders packet pull case, wallet), footwear type (leather shoes, sport footwear,
Sandals, boots).
(2) structure of deep learning model and training.Deep learning model framework is based on the micro- of AlexNet network models
It adjusts improved.Five layers of convolutional layer of AlexNet models and the structural parameters of preceding two layers of full articulamentum remain unchanged, and fine tuning is most
The part of full articulamentum afterwards.According to the attribute class number of mark, the node of output is changed.Training sample is input to structure
In good convolutional neural networks, and by adjusting the power of network using reconstruction error method is successively minimized to Deconvolution Method
Weight.The structure of wherein improved AlexNet networks is as follows：
First layer is convolutional layer, characteristic plane 96, convolution kernel size 11x11；
The second layer is convolutional layer, characteristic plane 256, convolution kernel size 5x5；
Third layer is convolutional layer, characteristic plane 384, convolution kernel size 3x3；
4th layer is convolutional layer, and characteristic plane is 384, convolution kernel size 3x3；
Layer 5 is convolutional layer, and characteristic plane is 256, convolution kernel size 3x3；
Layer 6 is full articulamentum, is remained unchanged；
Layer 7 is full articulamentum, is remained unchanged；
8th layer is full articulamentum, in this layer according to the number of pedestrian's attribute, changes the node of output；
The training step of above-mentioned improved AlexNet network models is as follows：
1) output to each layer carries out deconvolution, and the input of this layer is compared with the result that deconvolution exports, is obtained
To error E
2) the weights size of the convolution kernel of model is constantly adjusted by gradient descent method, formula is：
W in formula*For newer weights, W is original weights, and η is the learning rate of setting,Indicate local derviation of the error to weights.
3) 20000 repetitive exercises are carried out to training sample.After training, a convolutional neural networks mould can be obtained
Type；
(3) pedestrian sample is input in the trained deep learning model of second step, by a series of convolution operation, is obtained
To pedestrian's attributive character.Its specific steps are：
Pedestrian's test sample is input to improved AlexNet network models；
1) pedestrian image feature passes through first layer convolutional layer, by the convolution operation of the convolution kernel of 11X11, exports 96 spies
Sign figure；
2) pedestrian image feature passes through second layer convolutional layer, by the convolution operation of the convolution kernel of 5X5, exports 256 spies
Sign figure；
3) pedestrian image feature passes through third layer convolutional layer, by the convolution operation of the convolution kernel of 3X3, exports 384 spies
Sign figure；
4) pedestrian image feature passes through the 4th layer of convolutional layer, by the convolution operation of the convolution kernel of 3X3, exports 384 spies
Sign figure；
5) pedestrian image feature passes through layer 5 convolutional layer, by the convolution operation of the convolution kernel of 3X3, exports 256 spies
Sign figure；
6) pass through three layers of full articulamentum, export the feature of pedestrian's test image.
(4) it is one grader of each attribute design, it is defeated then by the feature of pedestrian's test image of third step output
Enter into grader training study.Grader is using linear SVM.To possessing attribute xiSample labeling be just
Sample, on the contrary label is.The positive and negative feature that third step is extracted equally is input to each independent linear support
It is trained in vector machine, obtains each attributive classification device.Wherein support vector machines selects radial basis functionWherein, x and x' indicates two samples respectively, is determined using mini-max optimization method radial
Base parameter σ values are 3.35, and each attributive classification device accuracy of identification that is averaged can be allowed to reach 94% or more.
(5) pedestrian's training sample data calculate attribute x by pedestrian's attribute data collectioniSample belong to classification yjRatio
Example, i.e., What is indicated is to possess attribute xiAnd belong to classification yjSample size,What is indicated is to possess
Attribute xiSample size, calculate the other quantitative relation of the corresponding Attribute class of each attribute.
(6) pedestrian's test sample is tested, is as follows：
1) pedestrian's test sample ztIt is input to improved AlexNet networks, the feature extracted；
2) it and then inputs the feature into attributive classification device, obtains with attribute xiPosterior probability p (xi|zt)；
3) the attribute classification quantitative relation drawn by the 5th step can utilize Bayesian formulaCalculating pedestrian sample has classification yjPosterior probability, wherein N pedestrian's attribute
Quantity；
4) the corresponding classification of the calculated maximum probability of previous step is exactly the corresponding attribute classification of pedestrian sample.
The present invention is by improving deep learning model framework, and pedestrian's attributive character is arrived in automatic study, then by attributive character
It is input in grader, training pedestrian's attributive classification device independent one by one, can obtain pedestrian sample in this way has attribute
It is other to obtain Attribute class then by calculating the attribute in pedestrian's training sample and the other proportionate relationship of Attribute class for posterior probability
Posterior probability finally can be obtained by the attribute classification of pedestrian sample according to Bayesian formula.Present invention employs deep learnings
Technology can learn to avoid traditional engineer to preferable feature automatically compared with conventional pedestrian's Attribute Recognition algorithm
The complexity of feature extraction.Due to taking convolutional neural networks to extract feature, the semantic information of higher can be extracted, and by
The influence of the uncontrollable factors such as illumination, visual angle is smaller, therefore Attribute Recognition effect has more robustness, accuracy of identification also higher.
Claims (5)
1. a kind of pedestrian's attribute based on deep learning refines recognition methods, which is characterized in that include the following steps：
(1) attribute labeling is carried out to pedestrian sample data set, obtains pedestrian's attribute data collection；
(2) the convolutional neural networks model finely tuned based on AlexNet networks is built, and the convolutional neural networks model is instructed
Practice；
(3) to convolutional neural networks mode input pedestrian's test pictures after training, pedestrian's attributive character is extracted；
(4) pedestrian's attributive character of extraction is input in grader, training human body attributive classification device；
(5) attribute x is calculated by pedestrian's attribute data collectioniSample belong to classification yjRatio, i.e.,
What is indicated is to possess attribute xiAnd belong to classification yjSample size,What is indicated is to possess attribute xiSample size, calculate
Go out the other quantitative relation of the corresponding Attribute class of each attribute；
(6) by pedestrian's test sample ztIt is input to the convolutional neural networks model that training finishes, and extracts feature；Then by extraction
Feature is input in the human body attributive classification device, and obtaining pedestrian sample has attribute xiPosterior probability p (xi|zt), pass through pedestrian
Quantitative relation between the corresponding attribute classification of attribute, utilizes Bayesian formula：
Calculating pedestrian sample has classification yjPosterior probability, the classification y calculatedjProbability in the corresponding class of maximum value
Be not exactly pedestrian sample attribute classification wherein, the quantity of N pedestrian's attribute.
2. pedestrian's attribute according to claim 1 based on deep learning refines recognition methods, which is characterized in that described
42 attributes of pedestrian can be represented in the step 1 to pedestrian's picture mark by picture annotation tool；42 being divided into property of attribute
Other attribute, hair lengths attribute, upper body clothing type attribute, upper body clothing color attribute, lower garment type attribute, lower part of the body clothing
Color attribute, Packet type attribute and footwear type attribute, wherein gender attribute includes man and female, hair lengths attribute include long hair
And bob, upper body clothing type attribute includes T-shirt, shirt, housing and down jackets, upper body clothing color attribute include it is black, white,
Red, yellow, blue, green, purple, brown, grey, orange and polychrome, lower garment type attribute are trousers, shorts, longuette and skirt, lower garment
Color attribute includes black, white, red, yellow, blue, green, purple, brown, grey, orange and polychrome, Packet type attribute include shoulder bag, both shoulders packet,
Case and wallet are pulled, footwear type includes leather shoes, sport footwear, sandals and boots.
3. pedestrian's attribute according to claim 1 based on deep learning refines recognition methods, which is characterized in that described
The AlexNet model knots finely tuned in the convolutional neural networks model structure of the fine tuning based on AlexNet network models in step (2)
Structure is as follows：First layer is convolutional layer, characteristic plane 96, convolution kernel size 11x11；The second layer is convolutional layer, characteristic plane 256
It is a, convolution kernel size 5x5；Third layer is convolutional layer, characteristic plane 384, convolution kernel size 3x3；4th layer is convolutional layer, special
It is 384 to levy plane, convolution kernel size 3x3；Layer 5 is convolutional layer, and characteristic plane is 256, convolution kernel size 3x3；6th
Layer is full articulamentum, is remained unchanged；Layer 7 is full articulamentum, is remained unchanged；8th layer is full articulamentum, in this layer of basis
The number of pedestrian's attribute changes the node of output.
4. pedestrian's attribute according to claim 1 based on deep learning refines recognition methods, which is characterized in that described
The convolutional neural networks model is trained in step (2) and specifically includes following sub-step：
(21) to each layer of output of AlexNet convolutional neural networks carry out deconvolution operation, then the input of this layer with
The result of deconvolution output is compared, and obtains error E；
(22) the weights size of the convolution kernel of model is constantly adjusted by gradient descent method, formula is：Formula
Middle W*For newer weights, W is original weights, and η is the learning rate of setting,Indicate local derviation of the error to weights；
(23) training sample is input in model, carry out 20000 repetitive exercises, training after obtain one it is improved
Convolutional neural networks model.
5. pedestrian's attribute according to claim 1 based on deep learning refines recognition methods, which is characterized in that described
Step (4) includes following sub-step：
(41) it is first each attribute, designs independent grader；
(42) grader selects support vector machines, the kernel function of support vector machines in the training process to select radial basis function, most
Small largest optimization method determines that the value of radial base parameter σ is 3.35；
(43) pedestrian's attributive character of extraction is input in the grader of the step (42) configuration and accuracy of identification can be carried
Height is to 94% or more.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810105618.2A CN108460407A (en) | 2018-02-02 | 2018-02-02 | A kind of pedestrian's attribute fining recognition methods based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810105618.2A CN108460407A (en) | 2018-02-02 | 2018-02-02 | A kind of pedestrian's attribute fining recognition methods based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108460407A true CN108460407A (en) | 2018-08-28 |
Family
ID=63239368
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810105618.2A Pending CN108460407A (en) | 2018-02-02 | 2018-02-02 | A kind of pedestrian's attribute fining recognition methods based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108460407A (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108805111A (en) * | 2018-09-07 | 2018-11-13 | 杭州善贾科技有限公司 | A kind of detection of passenger flow system and its detection method based on recognition of face |
| CN109598186A (en) * | 2018-10-12 | 2019-04-09 | 高新兴科技集团股份有限公司 | A kind of pedestrian's attribute recognition approach based on multitask deep learning |
| CN109614928A (en) * | 2018-12-07 | 2019-04-12 | 成都大熊猫繁育研究基地 | Panda recognition algorithms based on limited training data |
| CN109711267A (en) * | 2018-12-03 | 2019-05-03 | 浙江大华技术股份有限公司 | A kind of pedestrian identifies again, pedestrian movement's orbit generation method and device |
| CN109815902A (en) * | 2019-01-24 | 2019-05-28 | 北京邮电大学 | A kind of pedestrian attribute region information acquisition method, device and equipment |
| CN109934081A (en) * | 2018-08-29 | 2019-06-25 | 厦门安胜网络科技有限公司 | A kind of pedestrian's attribute recognition approach, device and storage medium based on deep neural network |
| CN109961009A (en) * | 2019-02-15 | 2019-07-02 | 平安科技（深圳）有限公司 | Pedestrian detection method, system, device and storage medium based on deep learning |
| CN110287370A (en) * | 2019-06-26 | 2019-09-27 | 中国人民公安大学 | Suspect's method for tracing, device and storage medium based on field shoe print |
| CN110688888A (en) * | 2019-08-02 | 2020-01-14 | 浙江省北大信息技术高等研究院 | Pedestrian attribute identification method and system based on deep learning |
| CN112016490A (en) * | 2020-08-28 | 2020-12-01 | 中国科学院重庆绿色智能技术研究院 | Pedestrian attribute identification method based on generation countermeasure learning |
| CN112766180A (en) * | 2021-01-22 | 2021-05-07 | 重庆邮电大学 | Pedestrian re-identification method based on feature fusion and multi-core learning |
| CN113221796A (en) * | 2021-05-24 | 2021-08-06 | 厦门市美亚柏科信息股份有限公司 | Vector neuron-based pedestrian attribute identification method and system |
| WO2022227772A1 (en) * | 2021-04-27 | 2022-11-03 | 北京百度网讯科技有限公司 | Method and apparatus for training human body attribute detection model, and electronic device and medium |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104992142A (en) * | 2015-06-03 | 2015-10-21 | 江苏大学 | Pedestrian recognition method based on combination of depth learning and property learning |
| CN105975929A (en) * | 2016-05-04 | 2016-09-28 | 北京大学深圳研究生院 | Fast pedestrian detection method based on aggregated channel features |
| CN106599832A (en) * | 2016-12-09 | 2017-04-26 | 重庆邮电大学 | Method for detecting and recognizing various types of obstacles based on convolution neural network |
| CN106845373A (en) * | 2017-01-04 | 2017-06-13 | 天津大学 | Towards pedestrian's attribute forecast method of monitor video |
| CN106845415A (en) * | 2017-01-23 | 2017-06-13 | 中国石油大学(华东) | A kind of pedestrian based on deep learning becomes more meticulous recognition methods and device |
| CN106845527A (en) * | 2016-12-29 | 2017-06-13 | 南京江南博睿高新技术研究院有限公司 | A kind of vegetable recognition methods |
| CN106951872A (en) * | 2017-03-24 | 2017-07-14 | 江苏大学 | A kind of recognition methods again of the pedestrian based on unsupervised depth model and hierarchy attributes |
-
2018
- 2018-02-02 CN CN201810105618.2A patent/CN108460407A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104992142A (en) * | 2015-06-03 | 2015-10-21 | 江苏大学 | Pedestrian recognition method based on combination of depth learning and property learning |
| CN105975929A (en) * | 2016-05-04 | 2016-09-28 | 北京大学深圳研究生院 | Fast pedestrian detection method based on aggregated channel features |
| CN106599832A (en) * | 2016-12-09 | 2017-04-26 | 重庆邮电大学 | Method for detecting and recognizing various types of obstacles based on convolution neural network |
| CN106845527A (en) * | 2016-12-29 | 2017-06-13 | 南京江南博睿高新技术研究院有限公司 | A kind of vegetable recognition methods |
| CN106845373A (en) * | 2017-01-04 | 2017-06-13 | 天津大学 | Towards pedestrian's attribute forecast method of monitor video |
| CN106845415A (en) * | 2017-01-23 | 2017-06-13 | 中国石油大学(华东) | A kind of pedestrian based on deep learning becomes more meticulous recognition methods and device |
| CN106951872A (en) * | 2017-03-24 | 2017-07-14 | 江苏大学 | A kind of recognition methods again of the pedestrian based on unsupervised depth model and hierarchy attributes |
Non-Patent Citations (5)
| Title |
|---|
| DANGWEI LI 等: "Multi-attribute Learning for Pedestrian Attribute Recognition in Surveillance Scenarios", 《2015 3RD IAPR ASIAN CONFERENCE ON PATTERN RECOGNITION》 * |
| 刘健: "基于深度学习的溯源视频目标检测与识别", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 王枫 等: "基于双卷积神经网络的行人精细化识别", 《中国科技论文》 * |
| 胡诚 等: "基于卷积神经网络的监控场景下行人属性识别", 《图形图像》 * |
| 赵鑫欣: "基于卷积神经网络的铁轨扣件异常状态检测研究", 《中国优秀硕士学位论文全文数据库 工程科技II辑》 * |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109934081A (en) * | 2018-08-29 | 2019-06-25 | 厦门安胜网络科技有限公司 | A kind of pedestrian's attribute recognition approach, device and storage medium based on deep neural network |
| CN108805111A (en) * | 2018-09-07 | 2018-11-13 | 杭州善贾科技有限公司 | A kind of detection of passenger flow system and its detection method based on recognition of face |
| CN109598186A (en) * | 2018-10-12 | 2019-04-09 | 高新兴科技集团股份有限公司 | A kind of pedestrian's attribute recognition approach based on multitask deep learning |
| CN109711267A (en) * | 2018-12-03 | 2019-05-03 | 浙江大华技术股份有限公司 | A kind of pedestrian identifies again, pedestrian movement's orbit generation method and device |
| CN109614928A (en) * | 2018-12-07 | 2019-04-12 | 成都大熊猫繁育研究基地 | Panda recognition algorithms based on limited training data |
| CN109815902A (en) * | 2019-01-24 | 2019-05-28 | 北京邮电大学 | A kind of pedestrian attribute region information acquisition method, device and equipment |
| CN109961009B (en) * | 2019-02-15 | 2023-10-31 | 平安科技（深圳）有限公司 | Pedestrian detection method, system, device and storage medium based on deep learning |
| CN109961009A (en) * | 2019-02-15 | 2019-07-02 | 平安科技（深圳）有限公司 | Pedestrian detection method, system, device and storage medium based on deep learning |
| CN110287370A (en) * | 2019-06-26 | 2019-09-27 | 中国人民公安大学 | Suspect's method for tracing, device and storage medium based on field shoe print |
| CN110688888A (en) * | 2019-08-02 | 2020-01-14 | 浙江省北大信息技术高等研究院 | Pedestrian attribute identification method and system based on deep learning |
| CN112016490B (en) * | 2020-08-28 | 2022-08-02 | 中国科学院重庆绿色智能技术研究院 | Pedestrian attribute identification method based on generation countermeasure learning |
| CN112016490A (en) * | 2020-08-28 | 2020-12-01 | 中国科学院重庆绿色智能技术研究院 | Pedestrian attribute identification method based on generation countermeasure learning |
| CN112766180A (en) * | 2021-01-22 | 2021-05-07 | 重庆邮电大学 | Pedestrian re-identification method based on feature fusion and multi-core learning |
| CN112766180B (en) * | 2021-01-22 | 2022-07-12 | 重庆邮电大学 | Pedestrian re-identification method based on feature fusion and multi-core learning |
| WO2022227772A1 (en) * | 2021-04-27 | 2022-11-03 | 北京百度网讯科技有限公司 | Method and apparatus for training human body attribute detection model, and electronic device and medium |
| CN113221796A (en) * | 2021-05-24 | 2021-08-06 | 厦门市美亚柏科信息股份有限公司 | Vector neuron-based pedestrian attribute identification method and system |
| CN113221796B (en) * | 2021-05-24 | 2022-07-01 | 厦门市美亚柏科信息股份有限公司 | Vector neuron-based pedestrian attribute identification method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108460407A (en) | A kind of pedestrian's attribute fining recognition methods based on deep learning | |
| CN104992142B (en) | A kind of pedestrian recognition method being combined based on deep learning and attribute study | |
| CN106650806B (en) | A kind of cooperating type depth net model methodology for pedestrian detection | |
| CN109919177B (en) | Feature selection method based on hierarchical deep network | |
| CN108510000A (en) | The detection and recognition methods of pedestrian's fine granularity attribute under complex scene | |
| CN108960073A (en) | Cross-module state image steganalysis method towards Biomedical literature | |
| WO2019223302A1 (en) | Dress collocation method and system based on attention knowledge extraction, and storage medium | |
| CN107092894A (en) | A kind of motor behavior recognition methods based on LSTM models | |
| CN109670591A (en) | A kind of training method and image matching method, device of neural network | |
| CN106874296A (en) | A kind of style recognition methods of commodity and device | |
| CN106991386A (en) | A kind of gesture identification method based on depth residual error network | |
| CN108961675A (en) | Fall detection method based on convolutional neural networks | |
| CN108564012B (en) | Pedestrian analysis method based on human body feature distribution | |
| CN109614508A (en) | A kind of image of clothing searching method based on deep learning | |
| CN109558810A (en) | Divided based on position and merges target person recognition methods | |
| CN107330360A (en) | A kind of pedestrian's clothing colour recognition, pedestrian retrieval method and device | |
| CN111597870A (en) | Human body attribute identification method based on attention mechanism and multi-task learning | |
| CN112100908B (en) | Clothing design method for generating countermeasure network based on multi-condition deep convolution | |
| Liu et al. | Affective image classification by jointly using interpretable art features and semantic annotations | |
| CN108229503A (en) | A kind of feature extracting method for clothes photo | |
| CN110263174A (en) | - subject categories the analysis method based on focus | |
| CN110222636A (en) | The pedestrian's attribute recognition approach inhibited based on background | |
| CN108595558A (en) | A kind of image labeling method of data balancing strategy and multiple features fusion | |
| CN101021926A (en) | Personal image design platform and method for designing by the same | |
| Ly et al. | Large‐Scale Coarse‐to‐Fine Object Retrieval Ontology and Deep Local Multitask Learning |
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 |
Application publication date: 20180828 |
|
| RJ01 | Rejection of invention patent application after publication |