CN109344688A - The automatic identifying method of people in a kind of monitor video based on convolutional neural networks - Google Patents
The automatic identifying method of people in a kind of monitor video based on convolutional neural networks Download PDFInfo
- Publication number
- CN109344688A CN109344688A CN201810890872.8A CN201810890872A CN109344688A CN 109344688 A CN109344688 A CN 109344688A CN 201810890872 A CN201810890872 A CN 201810890872A CN 109344688 A CN109344688 A CN 109344688A
- Authority
- CN
- China
- Prior art keywords
- people
- image
- neural networks
- convolutional neural
- model
- 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
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses the automatic identifying methods of people in monitor video based on convolutional neural networks a kind of.The collected video information of the recognition methods carries out processing training, video is changed into a frame frame image, the invention proposes a kind of convolutional neural networks, for the information in image there is people, convolutional neural networks through the invention extract the profile information of people, database is formed for the information extraction characteristic value of the people extracted, and it is tagged to information in database, then classification is compared with the information in server database to tested characteristics of image, calculating similarity is carried out to classification results.Its recognition methods is successively: Image Acquisition, image segmentation, image characteristics extraction, constructs database, and images match realizes an identification process.The advantages that present invention has flexible design, and powerful, system can reduce, and is able to satisfy the following demand to various what's news, can be applied to the various occasions of network security monitoring.
Description
Technical field
The present invention relates to the recognition methods of people in monitor video a kind of, specifically relate to one kind and are based on rolling up in deep learning
The comparison of the features such as the profile of people, shape (including front, side and the back side) and identification side in the video monitoring of product neural network
Method belongs to technical field of network security.
Background technique
With the enhancing that the security protection of people is realized, important component of the video monitoring as safety and protection system, is one
The effective means for planting people's normal life order of fighting crime, safeguard, possesses vast market prospect.Traditional video monitoring system
What system obtained is entire monitored picture, including people and background image.Monitoring management person is needed to spend time and manpower that analysis is gone to adopt
Specific information in the video monitoring image collected.Need to save the time of image recognition using necessary technological means, with
And improve the accuracy of image recognition.Therefore, how to automatically identify the people in video is a very important job, is found out
People wherein included for scene analysis and realizes that people's networking is of great significance.As network security and Internet of Things development need
It wants, the present invention realizes network monitoring using camera, while can be carried out processing to the video that IP Camera monitors, and is divided into
Image one by one analyzes image, finds out people wherein included.Identify that the common method used has for people:
Method, template matching method, subspace method, Hidden Markov Models, neural network method, elastic graph based on geometrical characteristic
Matching process and Flexible Model about Ecology method etc..
The existing safe identification technology of public security is substantially the identification to face, such as application number " 201510471210.3 " one
Real-time face recognition methods and system based on deep learning are planted, is based under application number " 201510063023.1 " monitoring scene deep
It spends patents, these patents such as the face identification system of study and uses deep learning all to carry out recognition of face.But in practice,
It is relatively difficult that face can be collected by video monitoring, need people specially and for it, to sum up, influence factor also has age increases,
The factors such as shelter influence, all can make the degree of difficulty of human face detection and recognition greatly improve.And in many videos, it is easy to
The video-frequency band comprising people is captured, video-frequency band can be decomposed into a frame frame image, it, can be according to the information for the people that image includes
The information such as the shape of people or profile are extracted.Relative to recognition of face just for face part, and what the present invention was directed to
The detection identification of entire people, the identification process of people is carried out according to the profile of entire people, gait or movement.
Summary of the invention
To overcome deficiency existing for above-mentioned existing inventive technique, the main purpose of the present invention is to provide one kind to be based on convolution
The recognition methods of people, the present invention propose convolutional Neural net new in a kind of deep learning in the IP Camera monitoring of neural network
Network realizes the segmentation and identification of people in image, and the method for being built into database.
In order to achieve the above objectives, it adopts the following technical scheme that
The recognition methods of people, sets about final realization in IP Camera monitoring based on convolutional neural networks of the invention
Meter content is broadly divided into four parts: first part is the semantic segmentation that people is carried out based on convolutional neural networks, is mentioned using the present invention
Convolutional neural networks model out carries out image, semantic segmentation, and the portrait in image is split；Second part is based on people
Deep learning feature is extracted, this feature can be expressed as someone feature vector, and the present invention is special with VGG model extraction image
Sign；Part III constructs database, records using the corresponding feature vector of the people of certain figure as one, is saved in data
Library；Part IV is when identifying to people, to extract the deep learning feature of the people, and with the various people's that save in database
Deep learning feature is matched, and pairing identification is carried out, and similar people is shown automatically.
Wherein, the first part carries out the semantic segmentation of people based on convolutional neural networks, and camera is collected
Video, convert image one by one for video first.In order to effectively extract the people in image, the present invention is used
Preparatory trained convolutional neural networks model is split the image one by one in the video of camera shooting, uses
Convolutional network of the invention obtains segmentation result, and the background pixel point value in image in addition to human is then all set to 0,
Only retain the value of the pixel of the part of people.
Wherein, the second part is based on people and extracts deep learning feature, using improved convolutional neural networks VGG mould
Type extracts the deep learning feature of image.As described above, the image for the people that convolutional neural networks are split, as second
The input that partial depth learning characteristic extracts.It inputs an image into VGG model, using the model as a feature extractor.
In the full articulamentum of the model, the characteristics of image that dimension is 4096 is extracted, the information conversion of people in piece image is special in a row
Levy vector.
Wherein, the building of the property data base of the Part III people.The people that the present invention obtains second part
A corresponding feature vector saves in the database as a record data, and tagged in corresponding data
Classification facilitates later comparison to match, and database is constructed using private database system software, and database is stored in service-specific
Inside device, to ensure data safety and secrecy, and the server and network monitoring are networked.
Wherein, the identification matching of the Part IV people.Arbitrarily tested video is chosen, first converts video to
Image one by one.Semantic segmentation is carried out by convolutional neural networks model again, extracts the people in image.Then pass through convolution
Neural network VGG model extracts the deep learning feature of image, obtains the feature vector of people in each frame image.Finally
Classification is compared to the information in the property data base constructed in the feature vector got and the Part III, to point
Class result carries out calculating similarity, and the present invention is calculated using Euclidean distance and realized.The result value calculated for Euclidean distance
Range about illustrates that distance is closer close to 1, that is, more matches between 0 to 1, on the contrary then diversity factor is higher.It is big to similarity
It all exports in matched people's information of some threshold value.
Beneficial effects of the present invention:
Present invention generally provides a kind of convolutional neural networks and the recognition methods of the people based on the convolutional neural networks, difference
In traditional recognition of face.Present invention introduces depth learning technology, the convolutional neural networks semantic segmentation that designs through the invention
Model, convolutional neural networks and improvement VGG model combine, respectively to the image of people is split one by one in video,
Extract deep learning feature.And it is compared with the information in database.The present invention can realize network monitoring using camera,
Processing can be carried out to the video that IP Camera monitors simultaneously, be divided into image one by one, image is analyzed.It looks for
People wherein included out.In the world not there are two completely the same people, each specific people includes an independence
Feature vector, the characteristic vector data library of people can be constructed from the profile of people, shape (including front, side and the back side) etc..
It lays the foundation for the further development of network shooting safety.
Detailed description of the invention
Fig. 1 is convolutional neural networks proposed by the present invention.
Fig. 2 is application method implementation flow chart of the present invention；
Specific embodiment
The present invention will be further explained below with reference to the attached drawings.
Fig. 1 is a kind of convolutional neural networks model for the segmentation for realizing people proposed by the present invention.
The model includes 3 pond layers and convolutional layer, through the feature for the structure fusion interlayer that jumps, on characteristic spectrum
It is merged, obtains final confidence map, i.e. segmentation output.In terms of the pretreatment of this mode input image, pre- instruction is utilized
The residual error network ResNet-18 model perfected, and remove rear 7 layer network in addition to output layer, feature is carried out to original image
It extracts, obtains the characteristic pattern of one (dimension channel × height height × width width), it is then again that pre-training is good
The 7th layer of the inverse in addition to output layer of ResNet-18 (ResNet18 [- 7]), reciprocal 6th layer (ResNet18 [- 6]), reciprocal the
It is used as subsequent pond layer, that is, tri- pond layers of pool3, pool4 and pool5 to respectively correspond for 5 layers (ResNet18 [- 5])
ResNet18 [- 7], ResNet18 [- 6] and ResNet18 [- 5] layer finally obtain 3 intermediate features maps of different sizes.
Convolution sum sampling operation then is carried out to different intermediate maps, obtains the characteristic spectrum of final different layers,
Implementation process includes: that mode input end is an original image, to input original picture size as 3 × 320 × 480
For, pool3, pool4, pool5 are 3 pond layers, carry out one using to the feature 21 × 10 × 15 behind the pond pool3 first
Secondary 32 times of up-samplings, obtain 21 × 320 × 480 confidence map；Secondly also the feature behind the pond pool4 is passed through one time 1 × 1
Characteristic pattern 21 × 20 × 30 that convolution obtains carries out 16 times of up-samplings, obtains 21 × 320 × 480 confidence map；Most
The characteristic pattern 21 × 20 × 30 also obtained afterwards to the feature behind the pond pool5 by 1 × 1 convolution carries out 8 times upper
Sampling, obtains 21 × 320 × 480 confidence map；Finally three characteristic spectrums are merged, obtain final characteristic spectrum
Output obtains the segmentation result of corresponding region, realizes the segmentation of the people based on convolutional neural networks.
Fig. 2 is the realization design scheme of allomeric function of the present invention.Present invention specific implementation is as follows:
(1) the external camera of application acquires video, video is divided into video data one by one, and accomplish fluently mark
Label are compared for later period identification.
(2) preliminary treatment is carried out to collected video image, it is converted into jpg lattice by different video file formats
Formula or bmp format.Image is screened, the image file of the information in image containing someone is remained.
(3) image collection being handled, the convolutional neural networks of the invention realized using Fig. 1 are divided portrait,
Portrait is split from image.
(4) map operation.Then the background pixel point value in image in addition to human is all set to 0, only retains people
Part pixel value.Return to segmented image.
(5) image for there was only colour portrait is returned.
(6) with deep learning model VGG model is improved, to the picture extraction feature handled well.The model has 15 volumes
Lamination, each convolutional layer configure the activation primitive of ReLu.The mode of down-sampling is carried out using maximum down-sampling in a model
Pondization operation.Preceding 13 layers of network structure using VGG model of the model, this is to preferably adjust image in model
Spatial coherence.The present invention only uses convolutional layer without the use of full articulamentum in the model.This is because full convolution operation is mentioning
Ability with shared convolution when taking image, reduces the redundancy of information.The VGG network model has multiple convolutional layers and down-sampling
The output of layer, convolution node layer indicates are as follows: WithThat indicate is l layers and l-
1 layer of corresponding characteristic pattern,Indicate convolution kernel of m-th of the characteristic pattern from l layers to current n-th of characteristic pattern, f (x)=1/
The linear activation primitive of ReLu that [1+exp (- x)] is each layer,Indicate biasing.The output of down-sampling node layer can indicate
Are as follows:Wherein s × s indicates the size of down-sampling,For weight.By the VGG mould
After type extraction feature, the eigenmatrix of a 1*4096 is obtained for every picture.
(7) test set in the feature vector and Database Systems of collected people is calculated, by calculating Europe
Formula distance calculates image similarity.
(8) label of picture most like in database is returned out for the test image of selection.Or by the spy of extraction
Sign vectormatrix is stored in Database Systems, stamps tag along sort to the corresponding picture of this feature vector.
(9) it finally shows comparison result and exports.
The series of detailed descriptions listed above only for feasible embodiment of the invention specifically
Protection scope bright, that they are not intended to limit the invention, it is all without departing from equivalent implementations made by technical spirit of the present invention
Or change should all be included in the protection scope of the present invention.
Claims (8)
1. the automatic identifying method of people in a kind of monitor video based on convolutional neural networks, which is characterized in that including four portions
Point:
First part is the semantic segmentation that people is carried out based on convolutional neural networks, and the portrait in image is split；
Second part is to extract deep learning feature based on people, and the deep learning feature can be expressed as someone feature vector；
Part III is building database, records using the corresponding feature vector of the people of certain figure as one, is saved in number
According to library；
Part IV is to identify to people, extracts the deep learning feature of the people, and with the various people's that save in database
Deep learning feature carries out match cognization, and similar people is shown automatically.
2. the automatic identifying method of people in a kind of monitor video based on convolutional neural networks according to claim 1,
It is characterized in that, the first part is the semantic segmentation that people in image is carried out using convolutional neural networks model, is specifically included: right
In the collected video of camera, image one by one is converted by video first；Then using preparatory trained convolution
Image one by one in the video that neural network model shoots camera is split, then by image in addition to human
Background pixel point value be all set to 0, only retain people part pixel value.
3. the automatic identifying method of people in a kind of monitor video based on convolutional neural networks according to claim 2,
It is characterized in that, the construction of the convolutional neural networks model includes the following: that the model includes 3 pond layers and convolutional layer, benefit
With the good residual error network ResNet-18 model of pre-training, feature extraction is carried out to original image, obtains a dimension channel
× height height × width width characteristic pattern, by respectively obtaining intermediate features figure of different sizes after 3 pond layers
Spectrum obtains three characteristic spectrums and merges, obtains final after carrying out convolution sum sampling operation to intermediate characteristic spectrum respectively
Characteristic spectrum output.
4. the automatic identifying method of people in a kind of monitor video based on convolutional neural networks according to claim 1,
It being characterized in that, the second part is the deep learning feature that image is extracted using improved convolutional neural networks VGG model,
It specifically includes: special as second part deep learning to the image for the people that first part is split using convolutional neural networks
The input extracted is levied, is input an image into improved convolutional neural networks VGG model, using the model as a feature extraction
Device extracts the characteristics of image that dimension is 4096 in the full articulamentum of the model, in a row by the information conversion of people in piece image
Feature vector.
5. the automatic identifying method of people in a kind of monitor video based on convolutional neural networks according to claim 1,
It is characterized in that, every all tagged classification of record in the Part III, the database is stored in private server
Face, and the server and network monitoring are networked.
6. the automatic identifying method of people in a kind of monitor video based on convolutional neural networks according to claim 1,
It is characterized in that, the specific implementation of the Part IV includes: to convert image one by one for tested video first；Lead to again
It crosses convolution neural network model and carries out semantic segmentation, extract the people in image；Then it by convolutional neural networks VGG model, mentions
The deep learning feature for taking image obtains the feature vector of people in each frame image；Finally to the feature vector and institute got
Classification is compared in the information recorded in the property data base constructed in the Part III stated, calculate to classification results similar
Degree, the result value range of similar calculating about illustrates that distance is closer close to 1 between 0 to 1, that is, more matches, on the contrary
Then diversity factor is higher；The matched people's information for being greater than threshold value to similarity all exports.
7. the automatic identifying method of people in a kind of monitor video based on convolutional neural networks according to claim 6,
It is characterized in that, the calculating of the similarity is calculated using Euclidean distance to be realized.
8. the automatic identification of people in a kind of monitor video based on convolutional neural networks according to claim 1-7
Method, which is characterized in that the automatic identifying method is realized by following specific steps:
(1) the external camera of application acquires video, video is divided into video data one by one, and accomplish fluently label, uses
It identifies and compares in the later period；
(2) to collected video image carry out preliminary treatment, by its by different video file formats be converted into jpg format or
Bmp format；Image is screened；
(3) to treated, image collection is handled, and using the convolutional neural networks, portrait is divided, will from image
Portrait is split；
(4) the background pixel point value in image in addition to human is all set to 0 by map operation, only retains the part of people
Pixel value, return segmented image；
(5) image only comprising colour portrait is returned；
(6) with deep learning model VGG model is improved, to the picture extraction feature handled well, the VGG model includes 15
Convolutional layer, each convolutional layer configure the activation primitive of ReLu；The mode of down-sampling is carried out using maximum down-sampling in model
Pondization operation；Preceding 13 layers of network structure using VGG model of the model；Only used in the model convolutional layer without the use of
Full articulamentum；The VGG model has several convolutional layers and down-sampling layer, and wherein the output of convolution node layer indicates are as follows:The output of down-sampling node layer indicates are as follows:
After the VGG model extraction feature, the eigenmatrix of a 1*4096 is obtained for every picture；
(7) test set in the feature vector and Database Systems of collected people is calculated, by calculate it is European away from
From calculating image similarity；
(8) for the label of picture most like in the test image returned data library of selection, or by the feature vector of extraction
Battle array is stored in Database Systems, stamps tag along sort to the corresponding picture of this feature vector；
(9) it shows comparison result and exports.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810890872.8A CN109344688A (en) | 2018-08-07 | 2018-08-07 | The automatic identifying method of people in a kind of monitor video based on convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810890872.8A CN109344688A (en) | 2018-08-07 | 2018-08-07 | The automatic identifying method of people in a kind of monitor video based on convolutional neural networks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109344688A true CN109344688A (en) | 2019-02-15 |
Family
ID=65296802
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810890872.8A Pending CN109344688A (en) | 2018-08-07 | 2018-08-07 | The automatic identifying method of people in a kind of monitor video based on convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109344688A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109978918A (en) * | 2019-03-21 | 2019-07-05 | 腾讯科技（深圳）有限公司 | A kind of trajectory track method, apparatus and storage medium |
| CN110044964A (en) * | 2019-04-25 | 2019-07-23 | 湖南科技大学 | Architectural coating layer debonding defect recognition methods based on unmanned aerial vehicle thermal imaging video |
| CN110222587A (en) * | 2019-05-13 | 2019-09-10 | 杭州电子科技大学 | A kind of commodity attribute detection recognition methods again based on characteristic pattern |
| CN110402840A (en) * | 2019-07-25 | 2019-11-05 | 深圳市阿龙电子有限公司 | Live pig monitoring terminal and live pig monitoring system based on image recognition |
| CN110688884A (en) * | 2019-02-28 | 2020-01-14 | 成都通甲优博科技有限责任公司 | Passenger flow statistical method and device |
| CN110702042A (en) * | 2019-10-15 | 2020-01-17 | 河海大学常州校区 | Thickness estimation method under signal aliasing condition in ultrasonic pulse echo thickness measurement |
| CN111734885A (en) * | 2020-06-30 | 2020-10-02 | 全球能源互联网研究院有限公司 | Converter valve on-line monitoring and evaluating method and system |
| CN111845647A (en) * | 2019-04-28 | 2020-10-30 | 上海汽车集团股份有限公司 | Automobile camera cleaning system and method |
| WO2020220951A1 (en) * | 2019-04-29 | 2020-11-05 | 杭州海康威视数字技术股份有限公司 | Video recording data storage method and device |
| CN113177478A (en) * | 2021-04-29 | 2021-07-27 | 西华大学 | Short video semantic annotation method based on transfer learning |
| CN113657169A (en) * | 2021-07-19 | 2021-11-16 | 浙江大华技术股份有限公司 | Gait recognition method, device, system and computer readable storage medium |
| CN117315591A (en) * | 2023-11-13 | 2023-12-29 | 安徽光谷智能科技股份有限公司 | Intelligent campus safety monitoring prediction management system |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105631427A (en) * | 2015-12-29 | 2016-06-01 | 北京旷视科技有限公司 | Suspicious personnel detection method and system |
| CN105631413A (en) * | 2015-12-23 | 2016-06-01 | 中通服公众信息产业股份有限公司 | Cross-scene pedestrian searching method based on depth learning |
| US20170243058A1 (en) * | 2014-10-28 | 2017-08-24 | Watrix Technology | Gait recognition method based on deep learning |
| CN107169974A (en) * | 2017-05-26 | 2017-09-15 | 中国科学技术大学 | It is a kind of based on the image partition method for supervising full convolutional neural networks more |
| CN107273836A (en) * | 2017-06-07 | 2017-10-20 | 深圳市深网视界科技有限公司 | A kind of pedestrian detection recognition methods, device, model and medium |
| US9940539B2 (en) * | 2015-05-08 | 2018-04-10 | Samsung Electronics Co., Ltd. | Object recognition apparatus and method |
-
2018
- 2018-08-07 CN CN201810890872.8A patent/CN109344688A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170243058A1 (en) * | 2014-10-28 | 2017-08-24 | Watrix Technology | Gait recognition method based on deep learning |
| US9940539B2 (en) * | 2015-05-08 | 2018-04-10 | Samsung Electronics Co., Ltd. | Object recognition apparatus and method |
| CN105631413A (en) * | 2015-12-23 | 2016-06-01 | 中通服公众信息产业股份有限公司 | Cross-scene pedestrian searching method based on depth learning |
| CN105631427A (en) * | 2015-12-29 | 2016-06-01 | 北京旷视科技有限公司 | Suspicious personnel detection method and system |
| CN107169974A (en) * | 2017-05-26 | 2017-09-15 | 中国科学技术大学 | It is a kind of based on the image partition method for supervising full convolutional neural networks more |
| CN107273836A (en) * | 2017-06-07 | 2017-10-20 | 深圳市深网视界科技有限公司 | A kind of pedestrian detection recognition methods, device, model and medium |
Non-Patent Citations (3)
| Title |
|---|
| HE,KAIMING: "Identity Mappings in Deep Residual Networks", 《 LECTURE NOTES IN COMPUTER SCIENCE 》 * |
| 公安部第三研究所: "《多摄像机协同关注目标检测跟踪技术》", 30 June 2017, 东南大学出版社 * |
| 刘键: "基于卷积神经网络的行人检测方法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110688884A (en) * | 2019-02-28 | 2020-01-14 | 成都通甲优博科技有限责任公司 | Passenger flow statistical method and device |
| CN109978918A (en) * | 2019-03-21 | 2019-07-05 | 腾讯科技（深圳）有限公司 | A kind of trajectory track method, apparatus and storage medium |
| CN110044964A (en) * | 2019-04-25 | 2019-07-23 | 湖南科技大学 | Architectural coating layer debonding defect recognition methods based on unmanned aerial vehicle thermal imaging video |
| CN111845647A (en) * | 2019-04-28 | 2020-10-30 | 上海汽车集团股份有限公司 | Automobile camera cleaning system and method |
| WO2020220951A1 (en) * | 2019-04-29 | 2020-11-05 | 杭州海康威视数字技术股份有限公司 | Video recording data storage method and device |
| CN110222587A (en) * | 2019-05-13 | 2019-09-10 | 杭州电子科技大学 | A kind of commodity attribute detection recognition methods again based on characteristic pattern |
| CN110402840A (en) * | 2019-07-25 | 2019-11-05 | 深圳市阿龙电子有限公司 | Live pig monitoring terminal and live pig monitoring system based on image recognition |
| CN110402840B (en) * | 2019-07-25 | 2021-12-17 | 深圳市阿龙电子有限公司 | Live pig monitoring terminal and live pig monitoring system based on image recognition |
| CN110702042A (en) * | 2019-10-15 | 2020-01-17 | 河海大学常州校区 | Thickness estimation method under signal aliasing condition in ultrasonic pulse echo thickness measurement |
| CN110702042B (en) * | 2019-10-15 | 2021-07-02 | 河海大学常州校区 | Thickness estimation method under signal aliasing condition in ultrasonic pulse echo thickness measurement |
| CN111734885A (en) * | 2020-06-30 | 2020-10-02 | 全球能源互联网研究院有限公司 | Converter valve on-line monitoring and evaluating method and system |
| CN111734885B (en) * | 2020-06-30 | 2021-11-30 | 全球能源互联网研究院有限公司 | Converter valve on-line monitoring and evaluating method and system |
| CN113177478A (en) * | 2021-04-29 | 2021-07-27 | 西华大学 | Short video semantic annotation method based on transfer learning |
| CN113657169A (en) * | 2021-07-19 | 2021-11-16 | 浙江大华技术股份有限公司 | Gait recognition method, device, system and computer readable storage medium |
| CN117315591A (en) * | 2023-11-13 | 2023-12-29 | 安徽光谷智能科技股份有限公司 | Intelligent campus safety monitoring prediction management system |
| CN117315591B (en) * | 2023-11-13 | 2024-03-22 | 安徽光谷智能科技股份有限公司 | Intelligent campus safety monitoring prediction management system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109344688A (en) | The automatic identifying method of people in a kind of monitor video based on convolutional neural networks | |
| CN109829443B (en) | Video behavior identification method based on image enhancement and 3D convolution neural network | |
| Bazzani et al. | Multiple-shot person re-identification by hpe signature | |
| CN104050471B (en) | Natural scene character detection method and system | |
| CN106529467B (en) | Group behavior recognition methods based on multi-feature fusion | |
| CN104573111B (en) | Pedestrian's data structured in a kind of monitor video stores and preindexing method | |
| CN106096568A (en) | A kind of pedestrian's recognition methods again based on CNN and convolution LSTM network | |
| CN109522853A (en) | Face datection and searching method towards monitor video | |
| Avgerinakis et al. | Recognition of activities of daily living for smart home environments | |
| CN112836646B (en) | Video pedestrian re-identification method based on channel attention mechanism and application | |
| CN106355154B (en) | Method for detecting frequent passing of people in surveillance video | |
| CN106407966B (en) | A kind of face identification method applied to attendance | |
| CN108416780A (en) | A kind of object detection and matching process based on twin-area-of-interest pond model | |
| CN109190456A (en) | Pedestrian detection method is overlooked based on the multiple features fusion of converging channels feature and gray level co-occurrence matrixes | |
| CN111539445B (en) | Object classification method and system for semi-supervised feature fusion | |
| CN112801037A (en) | Face tampering detection method based on continuous inter-frame difference | |
| Diyasa et al. | Multi-face Recognition for the Detection of Prisoners in Jail using a Modified Cascade Classifier and CNN | |
| CN117095471B (en) | Face counterfeiting tracing method based on multi-scale characteristics | |
| CN111582195B (en) | Construction method of Chinese lip language monosyllabic recognition classifier | |
| CN117710888A (en) | Method and system for re-identifying blocked pedestrians | |
| Wang et al. | Human detection based on a sequence of thermal images using deep learning | |
| CN112488165A (en) | Infrared pedestrian identification method and system based on deep learning model | |
| CN115393788B (en) | Multi-scale monitoring pedestrian re-identification method based on global information attention enhancement | |
| CN115690658B (en) | Priori knowledge-fused semi-supervised video abnormal behavior detection method | |
| CN111008601A (en) | Fighting detection method based on video |
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: 20190215 |