CN109101865A - A kind of recognition methods again of the pedestrian based on deep learning - Google Patents
A kind of recognition methods again of the pedestrian based on deep learning Download PDFInfo
- Publication number
- CN109101865A CN109101865A CN201810549705.7A CN201810549705A CN109101865A CN 109101865 A CN109101865 A CN 109101865A CN 201810549705 A CN201810549705 A CN 201810549705A CN 109101865 A CN109101865 A CN 109101865A
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- image
- distance
- feature
- deep learning
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
Abstract
The present invention discloses a kind of recognition methods again of the pedestrian based on deep learning, it include: step 1: pre-training CNN model: including pedestrian's feature extraction and characteristic measure, pedestrian's feature extraction uses the method for blending global characteristics and local feature, characteristic measure is using Euclidean distance as similarity measurement, under the distance constraints of feature vector, establish the loss function based on metric matrix, the loss function of use is to increase constraint function on the basis of traditional Triplet Loss, is optimized to CNN model;Step 2: test data set: will be in the trained CNN model of test data set image input step 1, it obtains characteristics of image and calculates the similarity between target pedestrian image and reference pedestrian image with Euclidean distance, finally pedestrian's weight recognition result will be obtained with reference to pedestrian image according to the big minispread of similarity.This method is suitable for the identification again of pedestrian under complex scene, strong for the portability of scene changes, and algorithmic stability, speed are fast, practical.
Description
Technical field
The invention belongs to Digital Image Processing, computer vision field, more particularly to a kind of row based on deep learning
People's recognition methods again.
Background technique
Extensive application with monitoring camera in each field, traditional artificial monitoring method can not be coped with resulting
Magnanimity monitor video.Pedestrian identifies again refers to progress pedestrian's matching under multiple cameras monitoring, that is, gives a pedestrian target,
The target is found in the video of the video camera different moments shooting of more different locations.Pedestrian's weight identification technology is intelligent video
The core technology of the numerous areas such as analysis, video monitoring, human-computer interaction, oneself is through becoming the research hotspot of computer vision field.
But because illumination, visual angle, posture, blocking influence with factors such as resolution ratio so that pedestrian's weight identification technology exist it is very big
Challenge.Pedestrian identifies the feature for designing effective description pedestrian first usually mainly comprising two steps again, then degree of passing through
It measures learning algorithm and carries out similarity measurement.Recognition methods relies on the feature of engineer to traditional pedestrian again, but due to same
Pedestrian may have very big difference in different images, and different pedestrians may seem very alike, so that these are special by hand
Sign is very difficult to apply in complicated actual environment.Deep learning has been successfully applied to many necks of computer vision at present
Domain, such as Handwritten Digits Recognition, target detection, image classification, recognition of face, identifying field again also in pedestrian has certain grind
Study carefully.
Pedestrian identifies that mainly there are two key components again: firstly, feature extraction, i.e. extraction target pedestrian and candidate pedestrian
Feature.Then, metric learning, the i.e. characteristic similarity of calculating target pedestrian and candidate pedestrian, judge whether candidate pedestrian is to want
The target looked for.Wherein, feature description is the basis of distance metric.
Many is not being considered based on the method for convolutional neural networks (Convolutional Neural Network, CNN)
Learn global characteristics in the case where the space structure of people.This has the shortcomings that several main: 1) inaccurate pedestrian detection frame may
It will affect feature learning;2) incoherent situation may be introduced the feature of study by the human body being blocked;3) global special
Local difference in sign be it is critically important, especially when we have to differentiate between two closely similar people of appearance;4) deformation and
Fuzzy posture causes metric learning highly difficult.
It is different since the visual angle of video camera, scale, illumination, dress ornament are different from attitudes vibration, resolution ratio and presence is blocked
Continuous position and motion information may be lost between camera, use the distance metric of the standards such as Euclidean distance, Pasteur's distance
Weight recognition effect well cannot be obtained to measure the similarity of pedestrian's appearance features.
Summary of the invention
In order to overcome the shortcomings of the prior art described above, a kind of based on deep learning it is an object of the invention to propose
Pedestrian's recognition methods again, this method are a kind of new methods of feature learning, and this method still learns global characteristics, but is learning
Period executes automatic section aligned, without additional supervision or specific Attitude estimation.
In order to achieve the above object, the technical scheme adopted by the invention is that:
A kind of recognition methods again of the pedestrian based on deep learning, which is characterized in that described method includes following steps:
Step 1: pre-training CNN model: pre-training CNN method, including pedestrian's feature extraction and characteristic measure, pedestrian's feature
It extracts using the method for blending global characteristics and local feature, characteristic measure is using Euclidean distance as similarity measurements
Amount establishes the loss function based on metric matrix, the loss function of use is to pass under the distance constraints of feature vector
Increase constraint function on the basis of the Triplet Loss of system, CNN model is optimized;
Step 2: test data set: by the trained CNN model of test data set image input step 1, obtaining image
Feature simultaneously calculates target pedestrian image with Euclidean distance and with reference to the similarity between pedestrian image, will finally refer to pedestrian image
According to the big minispread of similarity, pedestrian's weight recognition result is obtained.
Further, step 1 detailed process includes:
Step 1.1, input picture: input marking image, respectively from different video flowings obtain target pedestrian image and
With reference to pedestrian image;Tag image is scaled to a certain size, and is input to deep learning network, deep learning network it is final
Output is the Feature Mapping of image;
Step 1.2, local shape factor: for local shape factor, first in the horizontal direction to characteristics of image carry out by
Row extracts, and then carries out the convolution operation of 1x1 again, obtained feature represents a horizontal component of human body picture, in local spy
In the study of sign, alignment operation is carried out by calculating shortest path;
Step 1.3, global characteristics extract: extraction is slided in the extraction for global characteristics with global pool on characteristic pattern
Feature, in conjunction with multiple local features in step 1.2, one global characteristics of feature of last image and multiple parts are special
Sign replaces;
Step 1.4, characteristic distance calculates: the Euclidean distance of the feature vector of step 1.2 and 1.3 extractions is calculated separately, it is public
Formula is as follows:
Wherein d is on two images from point x1iTo x2iBetween Euclidean distance;
Step 1.5, loss function: under the distance constraints of feature vector, the loss letter based on metric matrix is established
Number, does loss function using improved Triplet Loss, uses Euclidean distance as similarity measurement.
Further, step 1.2 specifically includes:
F={ f1..., fHAnd G={ g1..., gHBe two images local feature, the calculating of each Distance matrix D
Formula:
Wherein, dI, jIt is between i-th of vertical component of first image and j-th of vertical component of second image
Distance, Distance matrix D is formed based on these distances, wherein (i, j) element is dI, j, define the part between two images
Distance is the total distance of the shortest path of from (1,1) to (H, H) in matrix D;
State transition equation used by shortest path is sought in Dynamic Programming:
Wherein, SI, jIt is S when from (1,1) to (i, j) walks in Distance matrix DH, HIt is final most short between two images
The total distance in path.
Further, step 1.5 specific algorithm such as following formula:
Above formula LquadIndicate improved Triplet Loss function, one shares two, and previous item is traditional Triplet
Loss, latter is for further reducing gap in class.Since the importance of previous item is bigger, α1> α2.Wherein xi、xj
Belong to one kind, xi、xkBelong to another kind of, α1And α2For preset threshold value, siIndicate pedestrian image xi, g (xi, xj) indicate a kind of new
Similarity measure;
For normative model, strategy is set using a kind of adaptive threshold, threshold value is set as similar apart from mean value and foreign peoples
Difference apart from mean value, ω are used to be sized, and sample is as follows to feature vector coefficient relevant to constraint condition:
si=sj, si≠sk
Wherein α is preset threshold value, μnAnd μpIt is the value of two distributions, NnAnd NpIt is two positive and negative logarithms, ω is related right
Number;
Model is trained with loss function, until final result is less than preset value, obtains corresponding pedestrian image
Deep learning model.
Further, step 2 detailed process includes:
Step 2.1, input picture: input test data images obtain target pedestrian from different video flowings respectively
Image and it is multiple refer to pedestrian image;
Step 2.2, it feature extraction: by deep learning network model trained in test image input step 1, extracts
Target pedestrian's characteristics of image is the first characteristics of image, and extracting with reference to pedestrian's characteristics of image is the second characteristics of image;
Step 2.3, it calculates characteristic measure distance: the first characteristics of image and the second characteristics of image phase is calculated using Euclidean distance
Like degree, Euclidean distance formula is as follows:
Wherein d is on two images from point x1iTo x2iBetween Euclidean distance;
Step 2.4, export result: using characteristic distance measurement d as the standard for judging similarity, characteristic measure distance is got over
Small, similarity is higher, will arrange with reference to pedestrian image according to similarity size descending, and obtain pedestrian's weight recognition result.
Compared with prior art, the beneficial effects of the present invention are: the pedestrian proposed by the present invention based on deep learning knows again
Other method, compared with existing algorithm, remarkable advantage is:
(1) in characteristic extraction procedure, depth convolutional neural networks not only extract global characteristics, while also mentioning to each part
Take local message.For any pair of local message in two pictures, the distance between they are calculated, constitute one apart from square
Battle array.Again by Dynamic Programming, a shortest path from matrix upper left corner the to the lower right corner is calculated.One in this shortest path
Side has just corresponded to the matching of a pair of of local feature, it gives a kind of mode of human body alignment, is guaranteeing that body parts are opposite
In the case where sequence, the total distance of this alignment thereof is shortest.When training, the length of shortest path is added into
Loss function, the global feature of assisted learning pedestrian.
(2) traditional Triplet Loss extensive effect on test set is general.The present invention is with the Triplet improved
Loss increases new constraint on the basis of script, and for reducing variance within clusters and increasing inter-class variance, this method can be with
The effective irrelevant variables such as image background, noise that mitigate learn matrix to generate large effect, to avoid over-fitting
Occur, obtained metric matrix generalization ability is strong.
(3) pedestrian provided by the invention again identification of the recognition methods suitable for pedestrian under complex scene again, becomes scene
The portability of change is strong, and algorithmic stability, speed are fast, can effectively solve monitor video and database to be searched is of low quality asks
Topic, it is practical.
Detailed description of the invention
Fig. 1 is that the present invention is based on the flow charts of the pedestrian of deep learning again recognition methods.
Fig. 2 is pre-training CNN model.
Specific embodiment
For the ease of those of ordinary skill in the art understand and implement the present invention, below with reference to embodiment to the present invention make into
The detailed description of one step, it should be understood that implementation example described herein is merely to illustrate and explain the present invention, and is not used to limit
The fixed present invention.
As shown in Figure 1, be the present invention is based on the flow chart of the pedestrian of deep learning again recognition methods, the pedestrian identifies again
Specific step is as follows for method:
Step 1: pre-training CNN model
Pre-training CNN method proposed by the present invention, including pedestrian's feature extraction and characteristic measure.Pedestrian's feature extraction uses
The method that global characteristics and local feature are blended, characteristic measure is using Euclidean distance as similarity measurement.In feature
Under the distance constraints of vector, the loss function based on metric matrix is established.The loss function that the present invention uses is traditional
Increase constraint function on the basis of Triplet Loss, CNN model is optimized, error is made to reach minimum.
1.1 input picture
Input marking image obtains target pedestrian image from different video flowings respectively and refers to pedestrian image.
Tag image is scaled to a certain size, and is input to deep learning network, the final output of deep learning network
For the Feature Mapping of image.Global characteristics are extracted using global pool in Feature Mapping.In the study of local feature, lead to
It crosses calculating shortest path and carries out alignment operation.Finally, the characteristics of image of a figure can use a global characteristics and multiple offices
Portion's feature replaces.
1.2 local feature
For local shape factor, characteristics of image is extracted in the horizontal direction first line by line, then carries out 1x1's again
Convolution operation.The feature obtained in this way represents a horizontal component of human body picture.In the study of local feature, pass through calculating
Shortest path carries out alignment operation.
F={ f1..., fHAnd G={ g1..., gHBe two images local feature, the calculating of each Distance matrix D
Formula:
Wherein, dI, jIt is between i-th of vertical component of first image and j-th of vertical component of second image
Distance, Distance matrix D is formed based on these distances, wherein (i, j) element is dI, j, define the part between two images
Distance is the total distance of the shortest path of from (1,1) to (H, H) in matrix D.
State transition equation used by shortest path is sought in Dynamic Programming,
Wherein SI, jIt is S when from (1,1) to (i, j) walks in Distance matrix DH, HIt is the final shortest path between two images
The total distance of diameter.
1.3 global characteristics
Extraction for global characteristics is slided on characteristic pattern with global pool and extracts feature.In conjunction with more in step 1.2
The feature of a local feature, last image can be replaced with a global characteristics and multiple local features.
1.4 characteristic distance
The Euclidean distance of the feature vector of step 2.1 and 2.2 extractions is calculated separately, formula is as follows:
Wherein d is on two images from point x1iTo x2iBetween Euclidean distance.
1.5 loss function
Under the distance constraints of feature vector, the loss function based on metric matrix is established.The present invention is using improvement
Triplet Loss do loss function, use Euclidean distance as similarity measurement.Traditional Triplet Loss is being tested
Extensive effect is general on collection, and it is still bigger to be primarily due to variance within clusters.The present invention increases new constraint to this, for subtracting
Small variance within clusters and increase inter-class variance, propose a kind of improved Triplet Loss.
Specific algorithm such as following formula,
Above formula LquadIndicate improved Triplet Loss function, one shares two, and previous item is traditional Triplet
Loss, latter is for further reducing gap in class.Since the importance of previous item is bigger, α1> α2.Wherein xi、xj
Belong to one kind, xi、xkBelong to another kind of, α1And α2For preset threshold value, siIndicate pedestrian image xi, g (xi, xj) indicate a kind of new
Similarity measure.
For normative model, present invention employs a kind of adaptive threshold setting strategies, and it is equal that threshold value is set as similar distance
The difference of value and foreign peoples apart from mean value, ω are used to be sized, and sample is as follows to feature vector coefficient relevant to constraint condition,
si=sj, si≠sk
Wherein α is preset threshold value, μnAnd μpIt is the value of two distributions, NnAnd NpIt is two positive and negative logarithms, ω is related right
Number.
Model is trained with loss function, until final result is less than preset value.Obtain corresponding pedestrian image
Deep learning model, detailed process are as shown in Figure 2.
Step 2: test data set
2.1 input picture
Input test data images, respectively from different video flowings obtain target pedestrian image and it is multiple refer to pedestrian
Image.
2.2 feature extraction
By in deep learning network model trained in test image input step 1, target pedestrian characteristics of image is extracted
For the first characteristics of image, extracting with reference to pedestrian's characteristics of image is the second characteristics of image.
2.3 calculate characteristic measure distance
First characteristics of image and the second characteristics of image similarity are calculated using Euclidean distance.Euclidean distance formula is as follows:
Wherein d is on two images from point x1iTo x2iBetween Euclidean distance.
2.4 output results
Using characteristic distance measurement d as the standard for judging similarity, for characteristic measure apart from smaller, similarity is higher.It will ginseng
It examines pedestrian image to arrange according to similarity size descending, obtains pedestrian's weight recognition result.
It should be understood that the part that this specification does not elaborate belongs to the prior art.
It should be understood that the above-mentioned description for preferred embodiment is more detailed, can not therefore be considered to this
The limitation of invention patent protection range, those skilled in the art under the inspiration of the present invention, are not departing from power of the present invention
Benefit requires to make replacement or deformation under protected ambit, fall within the scope of protection of the present invention, this hair
It is bright range is claimed to be determined by the appended claims.
Claims (5)
1. a kind of recognition methods again of the pedestrian based on deep learning, which is characterized in that described method includes following steps:
Step 1: pre-training CNN model: pre-training CNN method, including pedestrian's feature extraction and characteristic measure, pedestrian's feature extraction
Using the method for blending global characteristics and local feature, characteristic measure using Euclidean distance as similarity measurement,
Under the distance constraints of feature vector, the loss function based on metric matrix is established, the loss function of use is traditional
Increase constraint function on the basis of Triplet Loss, CNN model is optimized;
Step 2: test data set: by the trained CNN model of test data set image input step 1, obtaining characteristics of image
And with Euclidean distance calculate target pedestrian image and with reference to the similarity between pedestrian image, finally will with reference to pedestrian image according to
The big minispread of similarity obtains pedestrian's weight recognition result.
2. the recognition methods again of the pedestrian based on deep learning according to claim 1, which is characterized in that step 1 specifically flows
Journey includes:
Step 1.1, input picture: input marking image obtains target pedestrian image and reference from different video flowings respectively
Pedestrian image;Tag image is scaled to a certain size, and is input to deep learning network, the final output of deep learning network
For the Feature Mapping of image;
Step 1.2, local shape factor: for local shape factor, characteristics of image is mentioned line by line in the horizontal direction first
It takes, then carries out the convolution operation of 1x1 again, obtained feature represents a horizontal component of human body picture, in local feature
In study, alignment operation is carried out by calculating shortest path;
Step 1.3, global characteristics extract: the extraction for global characteristics is slided on characteristic pattern with global pool and extracts feature,
In conjunction with multiple local features in step 1.2, feature one global characteristics and multiple local feature generations of last image
It replaces;
Step 1.4, characteristic distance calculates: calculating separately the Euclidean distance of the feature vector of step 1.2 and 1.3 extractions, formula is such as
Under:
Wherein d is on two images from point x1iTo x2iBetween Euclidean distance;
Step 1.5, loss function: under the distance constraints of feature vector, the loss function based on metric matrix is established, is adopted
Loss function is done with improved Triplet Loss, uses Euclidean distance as similarity measurement.
3. the recognition methods again of the pedestrian based on deep learning according to claim 2, which is characterized in that step 1.2 is specific
Include:
F={ f1..., fHAnd G={ g1..., gHBe two images local feature, the calculation formula of each Distance matrix D:
Wherein, dI, jIt is i-th of vertical component of first image and the distance between j-th of vertical component of second image,
Distance matrix D is formed based on these distances, wherein (i, j) element is dI, j, define two images between local distance be
The total distance of the shortest path of from (1,1) to (H, H) in matrix D;
State transition equation used by shortest path is sought in Dynamic Programming:
Wherein, SI, jIt is S when from (1,1) to (i, j) walks in Distance matrix DH, HIt is the final shortest path between two images
Total distance.
4. the recognition methods again of the pedestrian based on deep learning according to claim 2, which is characterized in that step 1.5 is specific
Algorithm such as following formula:
si=sj, sl≠sk, si=sl, si≠sk
Above formula LquadIndicate improved Triplet Loss function, one shares two, and previous item is traditional Triplet
Loss, latter is for further reducing gap in class.Since the importance of previous item is bigger, α1> α2.Wherein xi、xj
Belong to one kind, xi、xkBelong to another kind of, α1And α2For preset threshold value, siIndicate pedestrian image xi, g (xi, xj) indicate a kind of new
Similarity measure;
For normative model, strategy is set using a kind of adaptive threshold, threshold value is set as similar apart from mean value and foreign peoples's distance
The difference of mean value, ω are used to be sized, and sample is as follows to feature vector coefficient relevant to constraint condition:
si=sj, si≠sk
Wherein α is preset threshold value, μnAnd μpIt is the value of two distributions, NnAnd NpIt is two positive and negative logarithms, ω is related logarithm;
Model is trained with loss function, until final result is less than preset value, obtains the depth of corresponding pedestrian image
Learning model.
5. the recognition methods again of the pedestrian based on deep learning according to claim 1, which is characterized in that step 2 specifically flows
Journey includes:
Step 2.1, input picture: input test data images obtain target pedestrian image from different video flowings respectively
Pedestrian image is referred to multiple;
Step 2.2, feature extraction: by deep learning network model trained in test image input step 1, target is extracted
Pedestrian image feature is the first characteristics of image, and extracting with reference to pedestrian's characteristics of image is the second characteristics of image;
Step 2.3, calculate characteristic measure distance: it is similar to the second characteristics of image to calculate the first characteristics of image using Euclidean distance
Degree, Euclidean distance formula are as follows:
Wherein d is on two images from point x1iTo x2iBetween Euclidean distance;
Step 2.4, export result: using characteristic distance measurement d as the standard for judging similarity, characteristic measure is apart from smaller, phase
It is higher like spending, it will be arranged with reference to pedestrian image according to similarity size descending, and obtain pedestrian's weight recognition result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810549705.7A CN109101865A (en) | 2018-05-31 | 2018-05-31 | A kind of recognition methods again of the pedestrian based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810549705.7A CN109101865A (en) | 2018-05-31 | 2018-05-31 | A kind of recognition methods again of the pedestrian based on deep learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109101865A true CN109101865A (en) | 2018-12-28 |
Family
ID=64796584
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810549705.7A Pending CN109101865A (en) | 2018-05-31 | 2018-05-31 | A kind of recognition methods again of the pedestrian based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109101865A (en) |
Cited By (30)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109815908A (en) * | 2019-01-25 | 2019-05-28 | 同济大学 | It is a kind of based on the discrimination method again of the pedestrian that measures between deep learning and overlapping image block |
CN109902590A (en) * | 2019-01-30 | 2019-06-18 | 西安理工大学 | Pedestrian's recognition methods again of depth multiple view characteristic distance study |
CN109993070A (en) * | 2019-03-13 | 2019-07-09 | 华南理工大学 | A kind of pedestrian's recognition methods again based on global distance scale loss function |
CN110110755A (en) * | 2019-04-04 | 2019-08-09 | 长沙千视通智能科技有限公司 | Based on the pedestrian of PTGAN Regional disparity and multiple branches weight recognition detection algorithm and device |
CN110443174A (en) * | 2019-07-26 | 2019-11-12 | 浙江大学 | A kind of pedestrian's recognition methods again based on decoupling self-adaptive identification feature learning |
CN110490060A (en) * | 2019-07-10 | 2019-11-22 | 特斯联(北京)科技有限公司 | A kind of security protection head end video equipment based on machine learning hardware structure |
CN110598033A (en) * | 2019-08-14 | 2019-12-20 | 中国平安财产保险股份有限公司 | Intelligent self-checking vehicle method and device and computer readable storage medium |
CN110619271A (en) * | 2019-08-12 | 2019-12-27 | 浙江浩腾电子科技股份有限公司 | Pedestrian re-identification method based on depth region feature connection |
CN110728216A (en) * | 2019-09-27 | 2020-01-24 | 西北工业大学 | Unsupervised pedestrian re-identification method based on pedestrian attribute adaptive learning |
CN110929558A (en) * | 2019-10-08 | 2020-03-27 | 杭州电子科技大学 | Pedestrian re-identification method based on deep learning |
CN111104867A (en) * | 2019-11-25 | 2020-05-05 | 北京迈格威科技有限公司 | Recognition model training and vehicle heavy recognition method and device based on component segmentation |
CN111177447A (en) * | 2019-12-26 | 2020-05-19 | 南京大学 | Pedestrian image identification method based on depth network model |
CN111291705A (en) * | 2020-02-24 | 2020-06-16 | 北京交通大学 | Cross-multi-target-domain pedestrian re-identification method |
CN111666843A (en) * | 2020-05-25 | 2020-09-15 | 湖北工业大学 | Pedestrian re-identification method based on global feature and local feature splicing |
CN111783526A (en) * | 2020-05-21 | 2020-10-16 | 昆明理工大学 | Cross-domain pedestrian re-identification method using posture invariance and graph structure alignment |
CN111783736A (en) * | 2020-07-23 | 2020-10-16 | 上海高重信息科技有限公司 | Pedestrian re-identification method, device and system based on human body semantic alignment |
CN111985367A (en) * | 2020-08-07 | 2020-11-24 | 湖南大学 | Pedestrian re-recognition feature extraction method based on multi-scale feature fusion |
CN112307847A (en) * | 2019-08-01 | 2021-02-02 | 复旦大学 | Multi-scale attention pedestrian re-recognition deep learning system based on guidance |
CN112651262A (en) * | 2019-10-09 | 2021-04-13 | 四川大学 | Cross-modal pedestrian re-identification method based on self-adaptive pedestrian alignment |
CN113065495A (en) * | 2021-04-13 | 2021-07-02 | 深圳技术大学 | Image similarity calculation method, target object re-identification method and system |
CN113095123A (en) * | 2019-12-23 | 2021-07-09 | 晶睿通讯股份有限公司 | Object recognition method and related monitoring camera equipment |
WO2021196547A1 (en) * | 2020-03-31 | 2021-10-07 | 北京迈格威科技有限公司 | Person re-identification method and apparatus, electronic device and storage medium |
CN113609886A (en) * | 2021-03-11 | 2021-11-05 | 上海汉时信息科技有限公司 | Real-time pedestrian number statistical method and device, camera and server |
CN113780319A (en) * | 2020-09-27 | 2021-12-10 | 北京沃东天骏信息技术有限公司 | Closed loop detection method and device and computer-readable storage medium |
CN113935482A (en) * | 2021-10-13 | 2022-01-14 | 北京百度网讯科技有限公司 | Training method and device for pedestrian re-recognition network |
CN114038197A (en) * | 2021-11-24 | 2022-02-11 | 浙江大华技术股份有限公司 | Scene state determination method and device, storage medium and electronic device |
CN114550236A (en) * | 2022-01-24 | 2022-05-27 | 北京百度网讯科技有限公司 | Image recognition and training method, device, equipment and storage medium of model thereof |
CN114663839A (en) * | 2022-05-12 | 2022-06-24 | 中科智为科技(天津)有限公司 | Method and system for re-identifying blocked pedestrians |
CN115661722A (en) * | 2022-11-16 | 2023-01-31 | 北京航空航天大学 | Pedestrian re-identification method combining attributes and orientation |
CN115862060A (en) * | 2022-11-25 | 2023-03-28 | 天津大学四川创新研究院 | Pig face identification and pig weight identification based pig unique identification method and system |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107273872A (en) * | 2017-07-13 | 2017-10-20 | 北京大学深圳研究生院 | The depth discrimination net model methodology recognized again for pedestrian in image or video |
CN107330918A (en) * | 2017-06-26 | 2017-11-07 | 华中科技大学 | A kind of football video sportsman's tracking based on online multi-instance learning |
CN107844753A (en) * | 2017-10-20 | 2018-03-27 | 珠海习悦信息技术有限公司 | Pedestrian in video image recognition methods, device, storage medium and processor again |
-
2018
- 2018-05-31 CN CN201810549705.7A patent/CN109101865A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107330918A (en) * | 2017-06-26 | 2017-11-07 | 华中科技大学 | A kind of football video sportsman's tracking based on online multi-instance learning |
CN107273872A (en) * | 2017-07-13 | 2017-10-20 | 北京大学深圳研究生院 | The depth discrimination net model methodology recognized again for pedestrian in image or video |
CN107844753A (en) * | 2017-10-20 | 2018-03-27 | 珠海习悦信息技术有限公司 | Pedestrian in video image recognition methods, device, storage medium and processor again |
Non-Patent Citations (3)
Title |
---|
WEIHUA CHEN 等: "Beyond Triplet Loss: A Deep Quadruplet Network for Person Re-identification", 《2017 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)》 * |
XUAN ZHANG 等: "AlignedReID: Surpassing Human-Level Performance in Person Re-Identification", 《ARXIV:1711.08184V2》 * |
熊炜 等: "基于支持向量机的低质量文档图像二值化", 《计算机应用与软件》 * |
Cited By (46)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109815908A (en) * | 2019-01-25 | 2019-05-28 | 同济大学 | It is a kind of based on the discrimination method again of the pedestrian that measures between deep learning and overlapping image block |
CN109902590A (en) * | 2019-01-30 | 2019-06-18 | 西安理工大学 | Pedestrian's recognition methods again of depth multiple view characteristic distance study |
CN109902590B (en) * | 2019-01-30 | 2022-09-16 | 西安理工大学 | Pedestrian re-identification method for deep multi-view characteristic distance learning |
CN109993070B (en) * | 2019-03-13 | 2021-06-08 | 华南理工大学 | Pedestrian re-identification method based on global distance scale loss function |
CN109993070A (en) * | 2019-03-13 | 2019-07-09 | 华南理工大学 | A kind of pedestrian's recognition methods again based on global distance scale loss function |
CN110110755A (en) * | 2019-04-04 | 2019-08-09 | 长沙千视通智能科技有限公司 | Based on the pedestrian of PTGAN Regional disparity and multiple branches weight recognition detection algorithm and device |
CN110110755B (en) * | 2019-04-04 | 2021-02-26 | 长沙千视通智能科技有限公司 | Pedestrian re-identification detection method and device based on PTGAN region difference and multiple branches |
CN110490060A (en) * | 2019-07-10 | 2019-11-22 | 特斯联(北京)科技有限公司 | A kind of security protection head end video equipment based on machine learning hardware structure |
CN110490060B (en) * | 2019-07-10 | 2020-09-11 | 特斯联(北京)科技有限公司 | Security protection front-end video equipment based on machine learning hardware architecture |
CN110443174B (en) * | 2019-07-26 | 2021-08-10 | 浙江大学 | Pedestrian re-identification method based on decoupling self-adaptive discriminant feature learning |
CN110443174A (en) * | 2019-07-26 | 2019-11-12 | 浙江大学 | A kind of pedestrian's recognition methods again based on decoupling self-adaptive identification feature learning |
CN112307847A (en) * | 2019-08-01 | 2021-02-02 | 复旦大学 | Multi-scale attention pedestrian re-recognition deep learning system based on guidance |
CN110619271A (en) * | 2019-08-12 | 2019-12-27 | 浙江浩腾电子科技股份有限公司 | Pedestrian re-identification method based on depth region feature connection |
CN110598033A (en) * | 2019-08-14 | 2019-12-20 | 中国平安财产保险股份有限公司 | Intelligent self-checking vehicle method and device and computer readable storage medium |
CN110728216A (en) * | 2019-09-27 | 2020-01-24 | 西北工业大学 | Unsupervised pedestrian re-identification method based on pedestrian attribute adaptive learning |
CN110929558B (en) * | 2019-10-08 | 2022-10-18 | 杭州电子科技大学 | Pedestrian re-identification method based on deep learning |
CN110929558A (en) * | 2019-10-08 | 2020-03-27 | 杭州电子科技大学 | Pedestrian re-identification method based on deep learning |
CN112651262B (en) * | 2019-10-09 | 2022-10-14 | 四川大学 | Cross-modal pedestrian re-identification method based on self-adaptive pedestrian alignment |
CN112651262A (en) * | 2019-10-09 | 2021-04-13 | 四川大学 | Cross-modal pedestrian re-identification method based on self-adaptive pedestrian alignment |
CN111104867B (en) * | 2019-11-25 | 2023-08-25 | 北京迈格威科技有限公司 | Recognition model training and vehicle re-recognition method and device based on part segmentation |
CN111104867A (en) * | 2019-11-25 | 2020-05-05 | 北京迈格威科技有限公司 | Recognition model training and vehicle heavy recognition method and device based on component segmentation |
CN113095123A (en) * | 2019-12-23 | 2021-07-09 | 晶睿通讯股份有限公司 | Object recognition method and related monitoring camera equipment |
CN113095123B (en) * | 2019-12-23 | 2024-05-03 | 晶睿通讯股份有限公司 | Object recognition method and related monitoring camera equipment thereof |
CN111177447A (en) * | 2019-12-26 | 2020-05-19 | 南京大学 | Pedestrian image identification method based on depth network model |
CN111177447B (en) * | 2019-12-26 | 2021-04-30 | 南京大学 | Pedestrian image identification method based on depth network model |
CN111291705A (en) * | 2020-02-24 | 2020-06-16 | 北京交通大学 | Cross-multi-target-domain pedestrian re-identification method |
CN111291705B (en) * | 2020-02-24 | 2024-04-19 | 北京交通大学 | Pedestrian re-identification method crossing multiple target domains |
WO2021196547A1 (en) * | 2020-03-31 | 2021-10-07 | 北京迈格威科技有限公司 | Person re-identification method and apparatus, electronic device and storage medium |
CN111783526B (en) * | 2020-05-21 | 2022-08-05 | 昆明理工大学 | Cross-domain pedestrian re-identification method using posture invariance and graph structure alignment |
CN111783526A (en) * | 2020-05-21 | 2020-10-16 | 昆明理工大学 | Cross-domain pedestrian re-identification method using posture invariance and graph structure alignment |
CN111666843A (en) * | 2020-05-25 | 2020-09-15 | 湖北工业大学 | Pedestrian re-identification method based on global feature and local feature splicing |
CN111783736A (en) * | 2020-07-23 | 2020-10-16 | 上海高重信息科技有限公司 | Pedestrian re-identification method, device and system based on human body semantic alignment |
CN111985367A (en) * | 2020-08-07 | 2020-11-24 | 湖南大学 | Pedestrian re-recognition feature extraction method based on multi-scale feature fusion |
CN113780319A (en) * | 2020-09-27 | 2021-12-10 | 北京沃东天骏信息技术有限公司 | Closed loop detection method and device and computer-readable storage medium |
CN113609886A (en) * | 2021-03-11 | 2021-11-05 | 上海汉时信息科技有限公司 | Real-time pedestrian number statistical method and device, camera and server |
CN113065495A (en) * | 2021-04-13 | 2021-07-02 | 深圳技术大学 | Image similarity calculation method, target object re-identification method and system |
CN113065495B (en) * | 2021-04-13 | 2023-07-14 | 深圳技术大学 | Image similarity calculation method, target object re-recognition method and system |
CN113935482A (en) * | 2021-10-13 | 2022-01-14 | 北京百度网讯科技有限公司 | Training method and device for pedestrian re-recognition network |
CN114038197A (en) * | 2021-11-24 | 2022-02-11 | 浙江大华技术股份有限公司 | Scene state determination method and device, storage medium and electronic device |
CN114550236B (en) * | 2022-01-24 | 2023-08-15 | 北京百度网讯科技有限公司 | Training method, device, equipment and storage medium for image recognition and model thereof |
CN114550236A (en) * | 2022-01-24 | 2022-05-27 | 北京百度网讯科技有限公司 | Image recognition and training method, device, equipment and storage medium of model thereof |
CN114663839B (en) * | 2022-05-12 | 2022-11-04 | 中科智为科技(天津)有限公司 | Method and system for re-identifying blocked pedestrians |
CN114663839A (en) * | 2022-05-12 | 2022-06-24 | 中科智为科技(天津)有限公司 | Method and system for re-identifying blocked pedestrians |
CN115661722A (en) * | 2022-11-16 | 2023-01-31 | 北京航空航天大学 | Pedestrian re-identification method combining attributes and orientation |
CN115862060A (en) * | 2022-11-25 | 2023-03-28 | 天津大学四川创新研究院 | Pig face identification and pig weight identification based pig unique identification method and system |
CN115862060B (en) * | 2022-11-25 | 2023-09-26 | 天津大学四川创新研究院 | Pig unique identification method and system based on pig face identification and pig re-identification |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109101865A (en) | A kind of recognition methods again of the pedestrian based on deep learning | |
CN109583342B (en) | Human face living body detection method based on transfer learning | |
CN107832672B (en) | Pedestrian re-identification method for designing multi-loss function by utilizing attitude information | |
CN106960202B (en) | Smiling face identification method based on visible light and infrared image fusion | |
CN103942577B (en) | Based on the personal identification method for establishing sample database and composite character certainly in video monitoring | |
Zeng et al. | Silhouette-based gait recognition via deterministic learning | |
CN109558810B (en) | Target person identification method based on part segmentation and fusion | |
CN109949341B (en) | Pedestrian target tracking method based on human skeleton structural features | |
CN108537136A (en) | The pedestrian's recognition methods again generated based on posture normalized image | |
CN108563999A (en) | A kind of piece identity's recognition methods and device towards low quality video image | |
CN108921107A (en) | Pedestrian's recognition methods again based on sequence loss and Siamese network | |
CN109472198A (en) | A kind of video smiling face's recognition methods of attitude robust | |
CN103902992B (en) | Human face recognition method | |
CN105976397B (en) | A kind of method for tracking target | |
CN109190475A (en) | A kind of recognition of face network and pedestrian identify network cooperating training method again | |
CN108537181A (en) | A kind of gait recognition method based on the study of big spacing depth measure | |
CN109101915A (en) | Face and pedestrian and Attribute Recognition network structure design method based on deep learning | |
CN110796101A (en) | Face recognition method and system of embedded platform | |
CN111027481A (en) | Behavior analysis method and device based on human body key point detection | |
CN112464844A (en) | Human behavior and action recognition method based on deep learning and moving target detection | |
CN108280421A (en) | Human bodys' response method based on multiple features Depth Motion figure | |
CN108446642A (en) | A kind of Distributive System of Face Recognition | |
CN109522865A (en) | A kind of characteristic weighing fusion face identification method based on deep neural network | |
CN108830222A (en) | A kind of micro- expression recognition method based on informedness and representative Active Learning | |
CN108334870A (en) | The remote monitoring system of AR device data server states |
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: 20181228 |
|
RJ01 | Rejection of invention patent application after publication |