CN109919141A - A kind of recognition methods again of the pedestrian based on skeleton pose - Google Patents
A kind of recognition methods again of the pedestrian based on skeleton pose Download PDFInfo
- Publication number
- CN109919141A CN109919141A CN201910281066.5A CN201910281066A CN109919141A CN 109919141 A CN109919141 A CN 109919141A CN 201910281066 A CN201910281066 A CN 201910281066A CN 109919141 A CN109919141 A CN 109919141A
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- template
- image
- target
- skeleton
- 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
Landscapes
- Image Analysis (AREA)
Abstract
A kind of recognition methods again of the pedestrian based on skeleton pose constructs target pedestrian template comprising steps of obtaining pedestrian image；Acquired image frames detect all pedestrian images being successively partitioned into image under various postures by skeleton；All pedestrian images and framework information being partitioned into are carried out according to target line human skeleton template size normalised；Local characteristic region segmentation is successively carried out to all pedestrian images according to pedestrian's framework information, Slant Rectify is carried out to the local characteristic region of all pedestrians, is obtained and the consistent local feature image set of target pedestrian template posture；Target pedestrian's Local Feature Fusion identification model is established, COMPREHENSIVE CALCULATING goes out the similarity of all pedestrians of target pedestrian template and real-time detection, realizes accurately identifying for target pedestrian.The present invention be able to achieve bend under the more people's scenes of complex environment, the pedestrian under the influence of the abnormal posture in part such as limbs inclination effectively identifies, promote pedestrian's recognition accuracy.
Description
Technical field
The invention belongs to pedestrian's identification technology field, specifically a kind of pedestrian based on skeleton pose side of identification again
Method.
Background technique
Pedestrian identifies that (Person re-identification) is again in the video image by different cameras shooting again
Identify target pedestrian.Unlike pedestrian tracking, pedestrian identify again can be realized under complex environment it is long-term cross-border
Target is tracked, therefore very big effect can be played in monitoring field.For example the pedestrian of monitor video tracks again, when police does
Computer can lock suspect automatically when case, without artificial time-consuming and laborious observation identification.
It is usually matched using single features currently, the pedestrian according to pedestrian's characteristic matching identifies again, effect is often not
Ideal, main reason is that single feature can only express a part of attribute of target, it tends to be difficult to meet magnanimity monitor video
Target under data accurately identifies and continuously tracks demand.In addition, traditional can not with the recognition methods that whole body is an entirety
The local feature of prominent pedestrian, encountering the case where blocking will lead to target important feature information and loses so as to cause under discrimination
Drop, and common regional area extracts and similarity mode can cause to match there is a situation where regional area can not be aligned
Existing ambiguity thus greatly reduces discrimination.
Summary of the invention
In order to solve the above technical problems, the present invention provides a kind of recognition methods again of the pedestrian based on skeleton pose.
In order to solve the above-mentioned technical problem, the present invention takes following technical scheme:
A kind of recognition methods again of the pedestrian based on skeleton pose, comprising the following steps:
The image including target pedestrian frontal upright posture is obtained by monitor video or camera, passes through skeleton detection point
Target pedestrian image and target pedestrian's local characteristic region image are cut out, so that target pedestrian's template is constructed, target pedestrian's mould
Plate includes target pedestrian image template, target line human skeleton template and target pedestrian's local feature image set template；
The picture frame in monitor video or camera is acquired, pedestrian's framework information is extracted, is successively divided by skeleton detection
All pedestrian images in image under various postures out, posture include normally, bend over, limbs inclination；
All pedestrian images and framework information being partitioned into are carried out according to target line human skeleton template size normalised；
Local characteristic region segmentation is successively carried out to all pedestrian images according to pedestrian's framework information, to the office of all pedestrians
Portion's characteristic area carries out Slant Rectify, obtains and the consistent local feature image set of target pedestrian template posture；
Target pedestrian local feature image set template is matched with the local feature image set of each pedestrian, and is established
Target pedestrian's Local Feature Fusion identification model, it is similar to all pedestrians' of real-time detection that COMPREHENSIVE CALCULATING goes out target pedestrian template
Degree realizes accurately identifying for target pedestrian.
The target line human skeleton template acquisition methods be using depth convolutional neural networks to the image of acquisition at
Reason, obtains 18 framework characteristic key points, is labeled as Pi={ (xi,yi) | i=0,1 ... 17 }, 18 framework characteristic key points
Including at eye, right eye, left ear, auris dextra, mouth, chest neck, left shoulder, left elbow, left hand, right shoulder, right elbow, the right hand, left hip, left knee, a left side
Foot, right hip, right knee and right crus of diaphragm.
The size normalization procedure of the framework information are as follows: respectively according to target line human skeleton's template B0Row is detected
Human skeleton B1Body joint point coordinate information, calculate B0And B1Lower part of the body leg skeleton and the space of upper body trunk skeleton are long in skeleton
Degree and, for characterizing pedestrian's skeleton height；
To B0Skeleton height length and B1Skeleton height length carry out ratio calculation and obtain K, and will have been detected according to K
Pedestrian image zooms to target pedestrian's image template size, obtain completing it is size normalised after pedestrian image
It is described to use the skeleton dividing method based on framework information, to target pedestrian image template and carry out dimensional standard
The pedestrian image of change carries out regional area segmentation, obtains several targets pedestrian topography's template and pedestrian's regional area figure
Picture, local area image are expressed as Ri, wherein i=0,1 ... m.
The local characteristic region to pedestrian carries out Slant Rectify, specifically:
A vertical reference axis is established in target pedestrian topography's template and the pedestrian's local area image detected,
For image Y direction, target pedestrian topography's template and vertical reference axis angulation are denoted as θ0, pedestrian's regional area
Image and vertical reference axis angulation are denoted as θ1, θ0With θ1Difference be denoted asThen by pedestrian's regional area
The angle of image rotation Δ θ simultaneously carries out background and cuts out, the body local characteristic area after being corrected.
It is described that target pedestrian local feature image set template is matched with the local feature image set of each pedestrian, tool
Body are as follows:
Characteristic matching is carried out to target pedestrian image template and pedestrian's local area image, it is similar to obtain local features
Spend Sij, wherein i=1 ... m indicates some regional area, and it is similar that j=1 ... n indicates that different feature matching methods obtains
Degree is as a result, establish pedestrian's local feature similarity fusion matrix S:
Several local feature similarities are fused to comprehensive characteristics similarity, fusion calculation formula is as follows:
F=(S λ)T·ω
In above formula, different feature weights is distributed for different feature matching methods, λ is characterized weight matrix λ,
Different regional area weights is distributed for each regional area, ω is regional area weight matrix,
Then F=(S λ)Tω expansion are as follows:
It realizes that target pedestrian template carries out similar differentiation to all pedestrian images one by one according to comprehensive characteristics similarity F, selects
The highest pedestrian of similarity in pedestrian image is selected, identification is completed.
It is described that characteristic matching is carried out to target pedestrian image template and pedestrian's local area image, obtain local features
Similarity specifically:
The color characteristic and LBP textural characteristics in target pedestrian local feature image set template and pedestrian image are extracted, point
The feature vector x of target pedestrian's local area image template Xing Cheng not characterized0With the feature vector x of pedestrian's local area image1；
Then, the convolution feature vector y of characterization target pedestrian local area image template is obtained by depth convolutional neural networks0With row
The convolution feature vector y of people's local area image1, then calculate separately feature vector x0With x1, y0With y1Between COS distance, distance
It is smaller to illustrate that feature vector similarity is higher, this feature vector similarity higher i.e. current pedestrian's local area image and target
Pedestrian's local area image template is more similar.
The invention has the benefit that
Using human skeleton detection method, pedestrian's fine granularity local characteristic region can be precisely extracted, and according to skeleton pose
Pedestrian's characteristic area correction to be detected has been carried out, so that pedestrian's local characteristic region to be detected is consistent with target template, has been mentioned
Risen complex environment bend over, target signature template and pedestrian's characteristic matching to be detected under the influence of the abnormal posture in part such as limbs inclination
Accuracy, while constructing pedestrian's local feature similarity matrix and phase is carried out to target pedestrian and the feature of pedestrian each in monitoring image
Like degree COMPREHENSIVE CALCULATING, to realize accurately identifying for target pedestrian.Pedestrian under the more people's scenes of complex environment is able to achieve effectively to identify,
Pedestrian's recognition accuracy again is promoted, there is certain practical value.
Detailed description of the invention
Fig. 1 algorithm flow chart；
Fig. 2 pedestrian's recognition principle block diagram again；
Fig. 3 skeleton segmentation figure；
Fig. 4 target skeleton regional area correction chart；
Fig. 5 regional area color, texture feature extraction Matching Model；
Fig. 6 local features Matching Model；
Fig. 7 skeleton pattern；
The chest Fig. 8 regional area framework information；
Fig. 9 pixel rotates schematic diagram.
Specific embodiment
To further understand the features of the present invention, technological means and specific purposes achieved, function, below with reference to
Present invention is further described in detail with specific embodiment for attached drawing.
As shown in attached drawing 1-9, present invention discloses a kind of recognition methods again of the pedestrian based on skeleton pose, including following step
It is rapid:
S1 obtains the image including target pedestrian frontal upright posture by camera, is partitioned into mesh by skeleton detection
Pedestrian image and target pedestrian's local characteristic region image are marked, to construct target pedestrian's template, target pedestrian's template includes
Target pedestrian image template, target line human skeleton template and target pedestrian's local feature image set template.It is artificial in video
Target pedestrian is calibrated, while choosing targeted attitude in a certain frame is the image of frontal upright as target pedestrian's template M0。
S2 acquires the picture frame in monitor video or camera, extracts pedestrian's framework information, is successively divided by skeleton detection
Cut out all pedestrian images in image under various postures, posture include it is normal, bend over, the limb actions such as limbs inclination.Under
Pedestrian image described in the text indicates to be the image detected.In the present embodiment, using depth convolutional neural networks to mesh
Mark pedestrian's template M0Pedestrian's template M is detected1Skeleton joint point positioned, and according to skeleton joint pixel confidence to pass
Node is attached.And visual modeling is carried out to pedestrian's skeleton using OpenPose, skeleton pattern is as shown in fig. 7, this 18
Framework characteristic key point include eye, right eye, left ear, auris dextra, mouth, at chest neck, left shoulder, left elbow, left hand, right shoulder, right elbow, the right side
Hand, left hip, left knee, left foot, right hip, right knee and right crus of diaphragm.
S3 carries out dimensional standard to all pedestrian images and framework information being partitioned into according to target line human skeleton template
Change.Respectively according to target line human skeleton's template B0Pedestrian's skeleton B is detected1Body joint point coordinate information, calculate B0And B1Bone
In frame the space length of lower part of the body leg skeleton and upper body trunk skeleton and, for characterizing pedestrian's skeleton height；To B0Skeleton body
High length and B1Skeleton height length carry out ratio calculation and obtain K, and target line is zoomed to for pedestrian image has been detected according to K
People's image template size, obtain completing it is size normalised after pedestrian image.
S4 successively carries out local characteristic region segmentation to all pedestrian images according to pedestrian's framework information, to all pedestrians
Local characteristic region carry out Slant Rectify, obtain with the consistent local feature image set of target pedestrian template posture.
S5 matches target pedestrian local feature image set template with the local feature image set of each pedestrian, and
Target pedestrian's Local Feature Fusion identification model is established, COMPREHENSIVE CALCULATING goes out all pedestrians' of target pedestrian template and real-time detection
Similarity realizes accurately identifying for target pedestrian.
In addition, successively carrying out local characteristic region segmentation, detailed process to pedestrian image are as follows: include according in framework information
Body joint point coordinate skeleton is split, it is then that pedestrian image is corresponding with pedestrian's framework local region, obtain pedestrian office
Portion area image Ri, wherein i=0,1 ... m.Pedestrian image is divided into 10 regions, pedestrian's local area image is expressed as Rij
(x, y, w, h), wherein i=0,1 ... 9, successively represent head, chest, left large arm, left forearm, right large arm, right forearm, left thigh,
Left leg, right large arm, right thigh, j=0,1, respectively indicate target pedestrian image and pedestrian image.The center of gravity of local area image
It is expressed as G (x, y), the width of local area image and high respectively w, h.For example, the regional area template in chest is R2j(x,y,w,
H), as shown in FIG. 7 and 8, can be obtained by P1 (x1, y1), P5 (x5, y5), P8 (x8, y8), P11 (x11, y11) coordinate information,
Framework information specific formula for calculation is as follows:
Wherein:
After obtaining pedestrian's local area image, Slant Rectify, detailed process are also carried out are as follows:
As shown in figure 4, in local template RijIn establish a vertical reference axis L L, in target pedestrian topography template
Ri0θ is denoted as with L angulationi0, target template regional area Ri1θ is denoted as with L angulationi1。θi0With θi1Difference note
For Δ θi
The rotation of image is exactly to rotate to each pixel.Since the coordinate origin of image is in the upper left corner of image,
We are turned by coordinate, using the center of image as coordinate origin.Assuming that the width of original image is w, a height of h, (x0,y0) it is former sit
In mark a bit, convert coordinate after point be (x1,y1), then it can be obtained:
Pixel rotates schematic diagram as shown in figure 9, by point (x0,y0) rotate to point (x1,y1), then be easy to get to:
x1=r × cos (b-a)
y1=r × sin (b-a)
Postrotational pixel coordinate is obtained in coordinate system after conversion, as long as these coordinates are reconverted into former coordinate system
?.(b-a) is exactly Δ θ in above-mentioned formulai, then pedestrian topography rotation Δ θ will have been detectediAngle people after carry on the back again
Correction purpose is cut out and then reached to scape.
A vertical reference axis is established in target pedestrian topography's template and the pedestrian's local area image detected,
For image Y direction, target pedestrian topography's template and vertical reference axis angulation are denoted as θ0, pedestrian's regional area
Image and vertical reference axis angulation are denoted as θ1, θ0With θ1Difference be denoted asThen by pedestrian's regional area
The angle of image rotation Δ θ simultaneously carries out background and cuts out, the body local characteristic area after being corrected.
It is described that target pedestrian local feature image set template is matched with the local feature image set of each pedestrian, tool
Body are as follows:
Characteristic matching is carried out to target pedestrian image template and pedestrian's local area image, it is similar to obtain local features
Spend Sij, wherein i=1 ... m indicates some regional area, and it is similar that j=1 ... n indicates that different feature matching methods obtains
Degree is as a result, it also can also be certainly other that feature matching method, which includes color characteristic matching process, textural characteristics matching process,
Feature matching method will not enumerate herein, establish pedestrian's local feature similarity fusion matrix S:
Several local feature similarities are fused to comprehensive characteristics similarity, fusion calculation formula is as follows:
F=(S λ)T·ω
In above formula, different feature weights is distributed for different feature matching methods, λ is characterized weight matrix λ,
Different regional area weights is distributed for each regional area, ω is regional area weight matrix,
Then F=(S λ)Tω expansion are as follows:
It realizes that target pedestrian template carries out similar differentiation to all pedestrian images one by one according to comprehensive characteristics similarity F, selects
The highest pedestrian of similarity in pedestrian image is selected, identification is completed.
It is described that characteristic matching is carried out to target pedestrian image template and pedestrian's local area image, obtain local features
Similarity specifically:
The color characteristic and LBP textural characteristics in target pedestrian local feature image set template and pedestrian image are extracted, point
The feature vector x of target pedestrian's local area image template Xing Cheng not characterized0With the feature vector x of pedestrian's local area image1；
Then, the convolution feature vector y of characterization target pedestrian local area image template is obtained by depth convolutional neural networks0With row
The convolution feature vector y of people's local area image1, then calculate separately feature vector x0With x1, y0With y1Between COS distance, distance
It is smaller to illustrate that feature vector similarity is higher, this feature vector similarity higher i.e. current pedestrian's local area image and target
Pedestrian's local area image template is more similar.
xjBy the Y using color image, first three rank color moment of U, V, three attributes of totally nine components and textural characteristics
Composition.Wherein color moment feature expression are as follows:
Wherein, tri- Color Channel components of i Y, U, V, pi,jIndicate gray scale in i-th of Color Channel component of color image
For the probability that the pixel of j occurs, N indicates the number of pixel in image.
Three attributes of textural characteristics are as follows: energy SE, contrast SConWith entropy SqExpression formula are as follows:
Wherein, p (i, j | d, θ) is indicated on the direction θ, is separated by certain pixel distance d, gray value is respectively the picture of i and j
Frequency of the member to appearance.
Artificial feature vector xjWith convolution feature vector yjIt may be expressed as:
x0(x01,x02,…x0m),
y0(y01,y02,…y0n),
x1(x11,x12,…x1m),
y1(y11,y12,…y1n),
Wherein, xjIt is the n dimension artificial feature vector being made of color and textural characteristics, wherein n=9, (x01,x02,…x012)
=(μY,σY,sY…,SE,Scon,SQ)。
COS distance formula are as follows:
Wherein i=0 ... 9 indicates some regional area, Si0It represents target pedestrian's local area image template and has detected row
COS distance between the artificial feature vector of people's local area image, Si1It represents target pedestrian's local area image template and has examined
Survey the COS distance between the convolution feature vector of pedestrian's local area image.
According to Fig. 6 local features Matching Model, pedestrian's local feature similarity matrix S is constructed:
Due to using multi-mode characteristic matching, so distributing weight according to the significance level of two kinds of features, a power is obtained
Weight matrix λ:
Similarity of a certain regional area in Fusion Features is acquired according to λ.Then, based on regional area fusion
With model, according to weight matrix ω shared by 10 regional areas:
Weighting, which is asked, merges the similarity that similarity obtains final whole body.Similarity fusion calculation formula are as follows:
F=(S λ)Tω,
Finally, realizing that target pedestrian template has detected pedestrian image and carry out one by one with all according to comprehensive characteristics similarity F
Similar differentiation, the highest pedestrian of similarity in preferred image, to realize that target pedestrian identifies again.
It should be noted that these are only the preferred embodiment of the present invention, it is not intended to restrict the invention, although ginseng
According to embodiment, invention is explained in detail, for those skilled in the art, still can be to aforementioned reality
Technical solution documented by example is applied to modify or equivalent replacement of some of the technical features, but it is all in this hair
Within bright spirit and principle, any modification, equivalent replacement, improvement and so on should be included in protection scope of the present invention
Within.
Claims (7)
1. a kind of recognition methods again of the pedestrian based on skeleton pose, comprising the following steps:
The image including target pedestrian frontal upright posture is obtained by monitor video or camera, is partitioned by skeleton detection
Target pedestrian image and target pedestrian's local characteristic region image, so that target pedestrian's template is constructed, target pedestrian's template packet
Include target pedestrian image template, target line human skeleton template and target pedestrian's local feature image set template；
The picture frame in monitor video or camera is acquired, pedestrian's framework information is extracted, figure is successively partitioned by skeleton detection
All pedestrian images as under various postures, posture include normally, bend over, limbs inclination；
All pedestrian images and framework information being partitioned into are carried out according to target line human skeleton template size normalised；
Local characteristic region segmentation is successively carried out to all pedestrian images according to pedestrian's framework information, it is special to the part of all pedestrians
It levies region and carries out Slant Rectify, obtain and the consistent local feature image set of target pedestrian template posture；
Target pedestrian local feature image set template is matched with the local feature image set of each pedestrian, and establishes target
Pedestrian's Local Feature Fusion identification model, COMPREHENSIVE CALCULATING go out the similarity of all pedestrians of target pedestrian template and real-time detection,
Realize accurately identifying for target pedestrian.
2. the recognition methods again of the pedestrian based on skeleton pose according to claim 1, which is characterized in that the target pedestrian
Skeleton template acquisition methods are to be handled using image of the depth convolutional neural networks to acquisition, obtain 18 framework characteristics and close
Key point is labeled as Pi={ (xi,yi) | i=0,1 ... 17 }, 18 framework characteristic key points include eye, right eye, left ear, auris dextra,
At mouth, chest neck, left shoulder, left elbow, left hand, right shoulder, right elbow, the right hand, left hip, left knee, left foot, right hip, right knee and right crus of diaphragm.
3. the recognition methods again of the pedestrian based on skeleton pose according to claim 2, which is characterized in that the skeleton letter
The size normalization procedure of breath are as follows: respectively according to target line human skeleton's template B0Pedestrian's skeleton B is detected1Body joint point coordinate
Information calculates B0And B1In skeleton the space length of lower part of the body leg skeleton and upper body trunk skeleton and, for characterizing pedestrian's skeleton
Height；
To B0Skeleton height length and B1Skeleton height length carry out ratio calculation and obtain K, and pedestrian will have been detected according to K
Image scaling to target pedestrian's image template size, obtain completing it is size normalised after pedestrian image.
4. the recognition methods again of the pedestrian based on skeleton pose according to claim 3, which is characterized in that described use is based on
The skeleton dividing method of framework information to target pedestrian image template and has carried out size normalised pedestrian image progress part
Region segmentation, obtains several targets pedestrian topography's template and pedestrian's local area image, local area image are expressed as
Ri, wherein i=0,1 ... m.
5. the recognition methods again of the pedestrian based on skeleton pose according to claim 4, which is characterized in that described to pedestrian's
Local characteristic region carries out Slant Rectify, specifically:
A vertical reference axis is established in target pedestrian topography's template and the pedestrian's local area image detected, for figure
As Y direction, target pedestrian topography's template and vertical reference axis angulation are denoted as θ0, pedestrian's local area image
θ is denoted as with vertical reference axis angulation1, θ0With θ1Difference be denoted asThen by pedestrian's local area image
It rotates the angle of Δ θ and carries out background and cut out, the body local characteristic area after being corrected.
6. the recognition methods again of the pedestrian based on skeleton pose according to claim 5, which is characterized in that described to target line
People's local feature image set template is matched with the local feature image set of each pedestrian, specifically:
Characteristic matching is carried out to target pedestrian image template and pedestrian's local area image, obtains local features similarity
Sij, wherein i=1 ... m indicates that some regional area, j=1 ... n indicate the similarity result that different feature matching methods obtains,
Feature matching method includes color characteristic matching process, textural characteristics matching process, establishes the fusion of pedestrian's local feature similarity
Matrix S:
Several local feature similarities are fused to comprehensive characteristics similarity, fusion calculation formula is as follows:
F=(S λ)T·ω
In above formula, different feature weights is distributed for different feature matching methods, λ is characterized weight matrix λ,
Different regional area weights is distributed for each regional area, ω is regional area weight matrix,
Then F=(S λ)Tω expansion are as follows:
Realize that target pedestrian template carries out similar differentiation, selection row to all pedestrian images one by one according to comprehensive characteristics similarity F
The highest pedestrian of similarity in people's image completes identification.
7. the recognition methods again of the pedestrian based on skeleton pose according to claim 6, which is characterized in that described to target line
People's image template and pedestrian's local area image carry out characteristic matching, obtain local features similarity specifically:
The color characteristic and LBP textural characteristics in target pedestrian local feature image set template and pedestrian image are extracted, respectively shape
At the feature vector x of characterization target pedestrian local area image template0With the feature vector x of pedestrian's local area image1；Then,
The convolution feature vector y of characterization target pedestrian local area image template is obtained by depth convolutional neural networks0With pedestrian office
The convolution feature vector y of portion's area image1, then calculate separately feature vector x0With x1, y0With y1Between COS distance, apart from smaller
Illustrate that feature vector similarity is higher, this feature vector similarity higher i.e. current pedestrian's local area image and target pedestrian
Local area image template is more similar.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910281066.5A CN109919141A (en) | 2019-04-09 | 2019-04-09 | A kind of recognition methods again of the pedestrian based on skeleton pose |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910281066.5A CN109919141A (en) | 2019-04-09 | 2019-04-09 | A kind of recognition methods again of the pedestrian based on skeleton pose |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109919141A true CN109919141A (en) | 2019-06-21 |
Family
ID=66969151
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910281066.5A Pending CN109919141A (en) | 2019-04-09 | 2019-04-09 | A kind of recognition methods again of the pedestrian based on skeleton pose |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109919141A (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110334675A (en) * | 2019-07-11 | 2019-10-15 | 山东大学 | A kind of pedestrian's recognition methods again based on skeleton key point segmentation and column convolution |
| CN110490171A (en) * | 2019-08-26 | 2019-11-22 | 睿云联(厦门)网络通讯技术有限公司 | A kind of danger gesture recognition method, device, computer equipment and storage medium |
| CN110796079A (en) * | 2019-10-29 | 2020-02-14 | 深圳龙岗智能视听研究院 | Multi-camera visitor identification method and system based on face depth features and human body local depth features |
| CN111126275A (en) * | 2019-12-24 | 2020-05-08 | 广东省智能制造研究所 | Pedestrian re-identification method and device based on multi-granularity feature fusion |
| CN111144229A (en) * | 2019-12-05 | 2020-05-12 | 中国科学院深圳先进技术研究院 | Posture detection system and method |
| CN111353474A (en) * | 2020-03-30 | 2020-06-30 | 安徽建筑大学 | Pedestrian re-identification method based on human posture invariant feature |
| CN111582208A (en) * | 2020-05-13 | 2020-08-25 | 北京字节跳动网络技术有限公司 | Method and device for generating organism posture key point information |
| CN111915898A (en) * | 2020-07-24 | 2020-11-10 | 杭州金通科技集团股份有限公司 | Parking monitoring AI electronic post house |
| WO2021036304A1 (en) * | 2019-08-29 | 2021-03-04 | 深圳市商汤科技有限公司 | Image retrieval method and device |
| CN112668531A (en) * | 2021-01-05 | 2021-04-16 | 重庆大学 | Motion posture correction method based on motion recognition |
| CN113137983A (en) * | 2021-04-30 | 2021-07-20 | 深圳市恒星物联科技有限公司 | Self-learning manhole cover posture monitoring method and monitoring system |
| CN113177442A (en) * | 2021-04-12 | 2021-07-27 | 广东省科学院智能制造研究所 | Human behavior detection method and device based on edge calculation |
| CN114863553A (en) * | 2022-04-08 | 2022-08-05 | 北京踏行天际科技发展有限公司 | Device and method for automatic human body induction recognition of fitness mirror |
| CN115063483A (en) * | 2022-06-14 | 2022-09-16 | 广东天太机器人有限公司 | Template posture correction method and system based on 2d image recognition |
| CN115909394A (en) * | 2022-10-25 | 2023-04-04 | 珠海视熙科技有限公司 | Sitting posture identification method and device, intelligent desk lamp and computer storage medium |
| JP7396364B2 (en) | 2019-10-31 | 2023-12-12 | 日本電気株式会社 | Image processing device, image processing method, and image processing program |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101482928A (en) * | 2009-02-24 | 2009-07-15 | 清华大学 | Behavior description method and apparatus |
| CN105224937A (en) * | 2015-11-13 | 2016-01-06 | 武汉大学 | Based on the semantic color pedestrian of the fine granularity heavily recognition methods of human part position constraint |
| CN105354548A (en) * | 2015-10-30 | 2016-02-24 | 武汉大学 | Surveillance video pedestrian re-recognition method based on ImageNet retrieval |
| CN105518744A (en) * | 2015-06-29 | 2016-04-20 | 北京旷视科技有限公司 | Pedestrian re-identification method and equipment |
| CN105637850A (en) * | 2013-08-02 | 2016-06-01 | 脸谱公司 | Systems and methods for transforming an image |
| CN106874884A (en) * | 2017-03-03 | 2017-06-20 | 中国民航大学 | Human body recognition methods again based on position segmentation |
| CN107832672A (en) * | 2017-10-12 | 2018-03-23 | 北京航空航天大学 | A kind of pedestrian's recognition methods again that more loss functions are designed using attitude information |
| CN108334849A (en) * | 2018-01-31 | 2018-07-27 | 中山大学 | A kind of recognition methods again of the pedestrian based on Riemann manifold |
| CN108351690A (en) * | 2015-11-02 | 2018-07-31 | 索尼互动娱乐股份有限公司 | Information processing unit, information processing system and information processing method |
| CN108549843A (en) * | 2018-03-22 | 2018-09-18 | 南京邮电大学 | A kind of VAT invoice recognition methods based on image procossing |
| US20180315200A1 (en) * | 2017-04-28 | 2018-11-01 | Cherry Labs, Inc. | Monitoring system |
| CN109299646A (en) * | 2018-07-24 | 2019-02-01 | 北京旷视科技有限公司 | Crowd's accident detection method, apparatus, system and storage medium |
| CN109409250A (en) * | 2018-10-08 | 2019-03-01 | 高新兴科技集团股份有限公司 | A kind of across the video camera pedestrian of no overlap ken recognition methods again based on deep learning |
-
2019
- 2019-04-09 CN CN201910281066.5A patent/CN109919141A/en active Pending
Patent Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101482928A (en) * | 2009-02-24 | 2009-07-15 | 清华大学 | Behavior description method and apparatus |
| CN105637850A (en) * | 2013-08-02 | 2016-06-01 | 脸谱公司 | Systems and methods for transforming an image |
| CN105518744A (en) * | 2015-06-29 | 2016-04-20 | 北京旷视科技有限公司 | Pedestrian re-identification method and equipment |
| CN105354548A (en) * | 2015-10-30 | 2016-02-24 | 武汉大学 | Surveillance video pedestrian re-recognition method based on ImageNet retrieval |
| CN108351690A (en) * | 2015-11-02 | 2018-07-31 | 索尼互动娱乐股份有限公司 | Information processing unit, information processing system and information processing method |
| CN105224937A (en) * | 2015-11-13 | 2016-01-06 | 武汉大学 | Based on the semantic color pedestrian of the fine granularity heavily recognition methods of human part position constraint |
| CN106874884A (en) * | 2017-03-03 | 2017-06-20 | 中国民航大学 | Human body recognition methods again based on position segmentation |
| US20180315200A1 (en) * | 2017-04-28 | 2018-11-01 | Cherry Labs, Inc. | Monitoring system |
| CN107832672A (en) * | 2017-10-12 | 2018-03-23 | 北京航空航天大学 | A kind of pedestrian's recognition methods again that more loss functions are designed using attitude information |
| CN108334849A (en) * | 2018-01-31 | 2018-07-27 | 中山大学 | A kind of recognition methods again of the pedestrian based on Riemann manifold |
| CN108549843A (en) * | 2018-03-22 | 2018-09-18 | 南京邮电大学 | A kind of VAT invoice recognition methods based on image procossing |
| CN109299646A (en) * | 2018-07-24 | 2019-02-01 | 北京旷视科技有限公司 | Crowd's accident detection method, apparatus, system and storage medium |
| CN109409250A (en) * | 2018-10-08 | 2019-03-01 | 高新兴科技集团股份有限公司 | A kind of across the video camera pedestrian of no overlap ken recognition methods again based on deep learning |
Non-Patent Citations (6)
| Title |
|---|
| CHI SU等: "Pose-Driven Deep Convolutional Model for Person Re-identification", 《2017 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV)》 * |
| DONG YI等: "Deep Metric Learning for Person Re-identification", 《2014 22ND INTERNATIONAL CONFERENCE ON PATTERN RECOGNITION》 * |
| HAIYU ZHAO等: "Spindle Net: Person Re-identification with Human Body Region Guided Feature Decomposition and Fusion", 《2017 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
| LONGHUI WEI等: "GLAD: Global–Local-Alignment Descriptor for Scalable Person Re-Identification", 《IEEE TRANSACTIONS ON MULTIMEDIA》 * |
| SHAGHAYEGH GHARGHABI等: "People Re-identification Using 3D Descriptor with Skeleton Information", 《2015 INTERNATIONAL CONFERENCE ON INFORMATICS, ELECTRONICS & VISION (ICIEV)》 * |
| 段朦: "基于深度学习的行人再识别技术研究与实现", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110334675B (en) * | 2019-07-11 | 2022-12-27 | 山东大学 | Pedestrian re-identification method based on human skeleton key point segmentation and column convolution |
| CN110334675A (en) * | 2019-07-11 | 2019-10-15 | 山东大学 | A kind of pedestrian's recognition methods again based on skeleton key point segmentation and column convolution |
| CN110490171A (en) * | 2019-08-26 | 2019-11-22 | 睿云联(厦门)网络通讯技术有限公司 | A kind of danger gesture recognition method, device, computer equipment and storage medium |
| WO2021036304A1 (en) * | 2019-08-29 | 2021-03-04 | 深圳市商汤科技有限公司 | Image retrieval method and device |
| CN110796079A (en) * | 2019-10-29 | 2020-02-14 | 深圳龙岗智能视听研究院 | Multi-camera visitor identification method and system based on face depth features and human body local depth features |
| JP7396364B2 (en) | 2019-10-31 | 2023-12-12 | 日本電気株式会社 | Image processing device, image processing method, and image processing program |
| CN111144229A (en) * | 2019-12-05 | 2020-05-12 | 中国科学院深圳先进技术研究院 | Posture detection system and method |
| CN111126275B (en) * | 2019-12-24 | 2023-05-05 | 广东省智能制造研究所 | Pedestrian re-identification method and device based on multi-granularity feature fusion |
| CN111126275A (en) * | 2019-12-24 | 2020-05-08 | 广东省智能制造研究所 | Pedestrian re-identification method and device based on multi-granularity feature fusion |
| CN111353474A (en) * | 2020-03-30 | 2020-06-30 | 安徽建筑大学 | Pedestrian re-identification method based on human posture invariant feature |
| CN111353474B (en) * | 2020-03-30 | 2023-12-19 | 安徽建筑大学 | Pedestrian re-recognition method based on human body posture invariant features |
| CN111582208A (en) * | 2020-05-13 | 2020-08-25 | 北京字节跳动网络技术有限公司 | Method and device for generating organism posture key point information |
| CN111915898A (en) * | 2020-07-24 | 2020-11-10 | 杭州金通科技集团股份有限公司 | Parking monitoring AI electronic post house |
| CN112668531A (en) * | 2021-01-05 | 2021-04-16 | 重庆大学 | Motion posture correction method based on motion recognition |
| CN113177442A (en) * | 2021-04-12 | 2021-07-27 | 广东省科学院智能制造研究所 | Human behavior detection method and device based on edge calculation |
| CN113177442B (en) * | 2021-04-12 | 2024-01-30 | 广东省科学院智能制造研究所 | Human behavior detection method and device based on edge calculation |
| CN113137983A (en) * | 2021-04-30 | 2021-07-20 | 深圳市恒星物联科技有限公司 | Self-learning manhole cover posture monitoring method and monitoring system |
| CN113137983B (en) * | 2021-04-30 | 2023-08-22 | 深圳市恒星物联科技有限公司 | Self-learning well lid posture monitoring method and monitoring system |
| CN114863553A (en) * | 2022-04-08 | 2022-08-05 | 北京踏行天际科技发展有限公司 | Device and method for automatic human body induction recognition of fitness mirror |
| CN115063483A (en) * | 2022-06-14 | 2022-09-16 | 广东天太机器人有限公司 | Template posture correction method and system based on 2d image recognition |
| CN115909394A (en) * | 2022-10-25 | 2023-04-04 | 珠海视熙科技有限公司 | Sitting posture identification method and device, intelligent desk lamp and computer storage medium |
| CN115909394B (en) * | 2022-10-25 | 2024-04-05 | 珠海视熙科技有限公司 | Sitting posture identification method and device, intelligent table lamp and computer storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109919141A (en) | A kind of recognition methods again of the pedestrian based on skeleton pose | |
| CN106056050B (en) | Multi-view gait recognition method based on self-adaptive three-dimensional human motion statistical model | |
| CN107392964B (en) | The indoor SLAM method combined based on indoor characteristic point and structure lines | |
| Rogez et al. | Mocap-guided data augmentation for 3d pose estimation in the wild | |
| CN107423730B (en) | Human gait behavior active detection and recognition system and method based on semantic folding | |
| CN103155003B (en) | Posture estimation device and posture estimation method | |
| CN111881887A (en) | Multi-camera-based motion attitude monitoring and guiding method and device | |
| Uddin et al. | Human activity recognition using body joint‐angle features and hidden Markov model | |
| CN110544301A (en) | Three-dimensional human body action reconstruction system, method and action training system | |
| JP7015152B2 (en) | Processing equipment, methods and programs related to key point data | |
| CN106600626B (en) | Three-dimensional human motion capture method and system | |
| CN104794722A (en) | Dressed human body three-dimensional bare body model calculation method through single Kinect | |
| CN113077519B (en) | Multi-phase external parameter automatic calibration method based on human skeleton extraction | |
| CN104200200B (en) | Fusion depth information and half-tone information realize the system and method for Gait Recognition | |
| CN110555408B (en) | Single-camera real-time three-dimensional human body posture detection method based on self-adaptive mapping relation | |
| CN110544302A (en) | Human body action reconstruction system and method based on multi-view vision and action training system | |
| CN111862299A (en) | Human body three-dimensional model construction method and device, robot and storage medium | |
| JP2012123667A (en) | Attitude estimation device and attitude estimation method | |
| CN106815855A (en) | Based on the human body motion tracking method that production and discriminate combine | |
| CN105022999A (en) | Man code company real-time acquisition system | |
| Uddin et al. | Human Activity Recognition via 3-D joint angle features and Hidden Markov models | |
| CN117671738B (en) | Human body posture recognition system based on artificial intelligence | |
| CN111881888A (en) | Intelligent table control method and device based on attitude identification | |
| CN105488491A (en) | Human body sleep posture detection method based on pyramid matching histogram intersection kernel | |
| CN110477921B (en) | Height measurement method based on skeleton broken line Ridge regression |
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: 20190621 |