CN104899575A - Human body assembly dividing method based on face detection and key point positioning - Google Patents
Human body assembly dividing method based on face detection and key point positioning Download PDFInfo
- Publication number
- CN104899575A CN104899575A CN201510347507.9A CN201510347507A CN104899575A CN 104899575 A CN104899575 A CN 104899575A CN 201510347507 A CN201510347507 A CN 201510347507A CN 104899575 A CN104899575 A CN 104899575A
- Authority
- CN
- China
- Prior art keywords
- human body
- point
- key point
- row
- index
- 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
- 241000282414 Homo sapiens Species 0.000 title claims abstract description 214
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000001514 detection method Methods 0.000 title claims abstract description 10
- 230000008569 process Effects 0.000 claims description 5
- 101000911390 Homo sapiens Coagulation factor VIII Proteins 0.000 claims description 3
- 102000057593 human F8 Human genes 0.000 claims description 3
- 230000008520 organization Effects 0.000 claims description 3
- 229940047431 recombinate Drugs 0.000 claims description 3
- 238000003708 edge detection Methods 0.000 abstract description 4
- 238000005259 measurement Methods 0.000 abstract description 2
- 230000000712 assembly Effects 0.000 abstract 2
- 238000000429 assembly Methods 0.000 abstract 2
- 239000000284 extract Substances 0.000 abstract 1
- 230000035800 maturation Effects 0.000 abstract 1
- 238000010586 diagram Methods 0.000 description 4
- 238000011160 research Methods 0.000 description 3
- 230000009471 action Effects 0.000 description 2
- 230000000052 comparative effect Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 230000031702 trunk segmentation Effects 0.000 description 2
- 241000282412 Homo Species 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000037237 body shape Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 229940079593 drug Drugs 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 239000004615 ingredient Substances 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
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
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/12—Acquisition of 3D measurements of objects
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Human Computer Interaction (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a human body assembly dividing method based on face detection and key point positioning; the method mainly aims to an image with no overlap limb for dividing human body assemblies; the method comprises the following steps: inputting a human body image; using face detection technology to position a human body substantially scope; using edge detection to obtain an accurate human body layout; using a cross list storage and retrieval method to complete automatic dividing of the human body assemblies through key point search. The method uses the maturation face detection algorithm to complete face identification, and substantially positions the human body position according to face information and extracts the accurate human body layout, and searches and positions key points on the human body layout according to human body proportion constraint, and finally fast divides layouts of each human body assembly according to the key points. The method is high in instantaneity and accuracy, can be applied to human body three dimensional reconstruction, human body measurement, and behavior identification fields.
Description
Technical field
The invention belongs to computer image processing technology field, be specifically related to a kind of human body component clustering method of locating based on Face datection and key point.
Background technology
In video or picture, the component clustering of human body can serve that human body accurate three-dimensional is rebuild, human body assembly is measured and the research of many association areas such as Activity recognition and practical application.In more than ten years in the past, the research of this problem has attracted this concern of many research.Document 1:Haritaoglu, D.Harwood, and L.S.Davis, " W4:Real-time surveillance of people and their activities; " IEEE Trans.Pattern Anal.Mach.Intell., vol.22, no.8, pp.809 – 830, Aug.2000. proposes a kind of recurrence convex hull algorithm and constructs and find possible body part based on skeleton pattern.Document 2:S.S.Micilotta, E.J.Ong, and R.Bowden, " Detecting and tracking of humans by probabilistic body part assembly; " in Proc.British Machine Vision Conf., Sep.2005, vol.1, pp.429 – 438. uses and severally detects different body parts with the body part detector that Adaboost algorithm is trained in advance.Document 3:S.Weik and C.E.Liedtke, " Hierarchical 3D pose estimation for articu-lated human body models from a sequence of volume data; " in Proc.Int.Workshop Robot Vis., Auckland, New Zealand, Feb.2001, pp.27 – 34. follows the tracks of along the negative minimum curvature of the profile of health, then uses correction iterative closest point algorithms to analyze each body part.Document 4:Y.Song, L.Goncalves, and P.Perona, " Unsupervised learning of human motion, " IEEE Trans.Pattern Anal.Mach.Intell., vol.25, no.7, pp.814 – 827, Jul.2003. proposes the posture of a kind of technique study mankind based on angle, and determines body part from video sequence.Document 5:Jun-Wei Hsieh, Chi-Hung Chuang, and Sin-Yu Chen et al, " Segmentation of Human Body Parts Using Deformable Triangulation, " IEEE Trans.Systems, Man, And Cybernetics, vol.40, no.3, May 2010 proposes a kind of triangle partitioning algorithm of novelty to do the basic algorithm of trunk location.But it is comparatively significantly not enough still to have some on the whole, mainly comprise the pre-service that (1) does not have to consider to reject " similar body shape " fast, (2) may need a large amount of training pictures.
Summary of the invention
Goal of the invention: the object of the invention is the defect for above human body component clustering method and deficiency, propose a kind of New methods in working solving human body component clustering problem.
For realizing described object, the present invention uses comparative maturity Face datection algorithm to carry out Face detection, then carries out component clustering according to the key point information in face location and profile.It is characterized in that after input picture, comprise following treatment step:
Step 1, input the human body image without limbs overlap, use Adaboost learning algorithm in the picture, the method for detecting human face namely based on Haar-like cascade classifier carries out Face datection, according to detect the center of positional information determination face of the face obtained and face radius;
Step 2, detects according to the positional information of face and obtains corresponding human body outline;
Step 3, carries out key point location based on orthogonal list structure detecting on the human body contour outline that obtains;
Step 4, utilizes key point human body contour outline to be divided into the profile of each human body assembly.
Wherein, step 2 comprises: use Canny operator to carry out rim detection, find out the outline set { C of all objects in human body image
i, i ∈ 1,2 ..., m}, and C
i={ P
j(X
j, Y
j), j ∈ 1,2 ..., n}, m, n be greater than 1 natural number, P
j(X
j, Y
j) represent pixel on outline, X
jand Y
jrepresent horizontal ordinate and the ordinate of the pixel on outline respectively, X
o, Y
oand R
orepresent the face center horizontal ordinate, face center ordinate and the face radius that detect and obtain respectively, meet following formula:
Wherein outline set { C
iin the maximum profile of element be human body contour outline.
Step 3 comprises the steps:
Step 3-1, the simple chain structure of storage organization to the pixel on human body outline of orthogonal list is adopted to recombinate, and carry out searching and locating of key point, key point comprises: the upper bound point of face center 1, neck left and right edges key point totally 2, human body left arm and right arm and the lower boundary point totally 4 of human body left arm and right arm, human body waist left and right edges key point totally 2,1, human body crotch mid point;
Step 3-2, set up orthogonal list: a pointwise chain type scanning is carried out to the pixel point range on human body outline, in scanning process, during scanning first point, set up a wardrobe pointer and a row head pointer respectively, all point to this point of scanning, a newly-built total pointer again, point to wardrobe pointer to the right, downward sensing row head pointer, when then scanning other one by one by the chain structure of human body outline, row or column is created as belonged to, just add in orthogonal list by row or column coordinate order, row or column is created if do not belonged to, just a newly-built row or column head pointer, first this pointer is added in row or column head pointer chain, again this analyzing spot of pointed, statistics is row and the some number total often arranged often, and with index identify each point when namely the sequence number being read out is being read out be which,
Step 3-3, determine neck left and right edge key point, algorithm steps is as follows:
Step 3-3-1, obtains cross chain table row chain coordinate Y=Y
o+ R
oline pointer;
Step 3-3-2, in traversal orthogonal list, there is a P in institute
j,if there is some P
jhorizontal ordinate X
jmeet X
o-R
o<X
j<X
o, then P
jfor neck left hand edge key point, if P
jfor in orthogonal list, last is put and X
o<X
j<X
o+ R
o, then P
jfor neck right hand edge key point, if neck left and right edges key point finds entirely, then algorithm stops, otherwise performs step 3-3-3;
Step 3-3-3, line pointer moves up a row, if row chain coordinate Y=Y
o-R
o, algorithm stops;
Step 3-3-4, repeats step 3-3-2 ~ step 3-3-3.
Step 3-4, determine human body crotch mid point key point, algorithm steps is as follows:
Step 3-4-1, obtains cross chain table row chain coordinate Y=(2R
o× 5.5/4) × (1-0.467)/(1-0.844)-(Y
o-R
o) line pointer;
Step 3-4-2, in traversal orthogonal list, there is a P in institute
jif there is some P
jhorizontal ordinate X
jmeet X
a<X
j<X
b, then step 3-4-3, P is performed
aand P
brepresent neck left and right edge key point respectively, X
aand X
bbe respectively the horizontal ordinate of neck left and right edge key point, if not exist and line pointer did not move, perform step 3-4-4, if not exist and line pointer moved, perform step 3-4-5;
Step 3-4-3, definition current line chain coordinate is Y
k, upwards travel through row chain coordinate in orthogonal list line by line and meet Y
k<Y<=Y
kthe row of-hoG, wherein hoG is the vertical height of navel to stern lower edge, hoG=(2R
o× 5.5/4), if there is some P in orthogonal list in × (0.600-0.467)/(1-0.844)
jhorizontal ordinate X
jmeet X
a<X
j<X
brow chain, move on to this journey by line pointer, continue perform step 3-4-2, otherwise perform step 3-4-5;
Step 3-4-4, travels through in orthogonal list downwards line by line and has a P
j, find first to there is some P
jhorizontal ordinate X
jmeet X
a<X
j<X
brow;
Points all in the row that step 3-4-5, traversal step 3-4-4 find, | X
j-(X
a+ X
bthe minimum point of)/2| is human body crotch mid point key point;
Step 3-5, determine human body waist left and right edges key point, algorithm steps is as follows:
Step 3-5-1, obtains row chain coordinate Y=Y in orthogonal list
ithe line pointer of-hoG, P
irepresent human body crotch mid point key point, Y
irepresent some P
iordinate;
Step 3-5-2, if row chain coordinate Y>Y
i, then perform step 3-5-3, otherwise traversal orthogonal list in institute a little, if there is horizontal ordinate X
jmeet X
i-WoS/2<=X
j<X
isome P
j, X
irepresent some P
ihorizontal ordinate, wherein X
jminimum some P
jbe human body waist left hand edge key point, meet X if exist
i<X
j<=X
ithe point P of+WoS/2
j, wherein X
jmaximum some P
jbe human body waist right hand edge key point, WoS is shoulder breadth, if human body waist left and right edge key point does not find entirely, then performs step 3-5-4, if human body waist left and right edge key point finds entirely, then algorithm stops;
Step 3-5-3, (5 is empirical values, and its existence is the lateral extent in order to expand search waist point, and it can get the arbitrary integer being greater than 0 WoS to be updated to WoS+5, value is too small, the number of times of repeated execution of steps 3-5-2 will be too much, and value is excessive, likely the point on arm is defined as the point on waist, so compromise considers that value is 5), perform step 3-5-1, if this step performs more than four times, algorithm stops;
Step 3-5-4, line pointer line down, returns and performs step 3-5-1
Step 3-6, determines the upper bound point of human body left arm and right arm, and the lower boundary point of human body left arm and right arm, and algorithm steps is as follows:
Step 3-6-1, to make in orthogonal list sequence number index a little deduct a value and make (index)
a=0, (index)
arepresent some P
athe sequence number being read out;
Step 3-6-2, obtains row chain coordinate X=X
gcolumn pointer, P
grepresent human body waist left hand edge key point, X
grepresent some P
ghorizontal ordinate;
Step 3-6-3, in traversal orthogonal list, there is a P in institute
j, for finger point P
f, (index)
fequal (index)
aadd the curve distance of neck left hand edge key point to finger point, if there is some P
jmeet (index)
a< (index)
j< (index)
f, then this point is human body left arm upper bound point, performs step 3-6-4, if do not exist, then performs step 3-6-5, (index)
jrepresent some P
jthe sequence number being read out, (index)
frepresent some P
fthe sequence number being read out;
Step 3-6-4, in traversal orthogonal list, institute a little, if there is some P
jmeet (index)
a< (index)
j< (index)
f(index)
j-(index)
a> (index)
g-(index)
j, then this point is human body left arm lower boundary point, and algorithm stops, if do not exist, then performs step 3-6-5, (index)
grepresent some P
gthe sequence number being read out;
Step 3-6-5, column pointer moves to left row, returns and performs step 3-6-2.
Step 4 comprises: the orthogonal list structure utilizing the human body outline pixel established, from first point, off-take point sets up first man body profile one by one, first after scanning neck right hand edge key point by chain structure, start read point write in newly-built human body contour outline by the point of reading one by one, just stop reading in until run into neck left hand edge key point, complete human head profile;
Newly-built second human body contour outline, after scanning human body left arm upper bound point, starts read point being write in second newly-built human body contour outline by the point of reading one by one, just stops reading in, complete human body left arm profile until run into human body left arm lower boundary point;
Newly-built 3rd human body contour outline, after scanning human body waist left hand edge key point, starts to read in the 3rd newly-built human body contour outline by point one by one, just stops reading in, complete left leg profile until run into human body crotch mid point key point;
Newly-built 4th human body contour outline, reads in the 4th newly-built human body contour outline by point one by one from human body crotch mid point key point, just stops reading in, complete right leg profile until run into human body waist right hand edge key point;
Newly-built 5th human body contour outline, after scanning human body right arm lower boundary point, starts to read in the 5th newly-built human body contour outline by point one by one, just stops reading in, complete right arm profile until run into human body right arm upper bound point.
Suppose original human body contour outline read point out time point by point scanning sense of rotation be turn clockwise, the order of newly-built profile is contrary with when being rotated counterclockwise, first point that each profile reads in and last point also just in contrast.
The present invention comprises that Lis Hartel is levied, AdaBoost algorithm and Canny edge detection operator.Lis Hartel is levied (Haar-like features) is a kind of digital picture feature for object identification.They are gained the name because of changing very similar to Haar wavelet transform, are a kind of real-time Face datection operators.AdaBoost, is the abbreviation of English " Adaptive Boosting " (self-adaptation enhancing), is a kind of machine learning method, is proposed by Yoav Freund and Robert Schapire.The self-adaptation of AdaBoost method is: the sample of previous sorter misclassification can be used to train next sorter.AdaBoost method for noise data and abnormal data very sensitive.But in some problems, AdaBoost method, for other learning algorithm of great majority, can not be easy to occur Expired Drugs.May very weak (such as there is very serious mistake rate) in the sorter used in AdaBoost method, as long as but its classifying quality better than at random (such as two class Question Classification error rates are slightly less than 0.5), just can improve the model finally obtained.And error rate is also useful higher than the Weak Classifier of probabilistic classifier, because in the linear combination of the multiple sorters finally obtained, can give negative coefficient to them, equally also can promote classifying quality.Canny edge detection operator is the multistage edge detection algorithm that John F.Canny developed in 1986.
Beneficial effect: adopt method of the present invention, inputs the image without limbs overlap or video, automatically can complete the division of human body assembly.The field such as automatic measurement and action recognition of the three-dimensional reconstruction of human body, human body assembly can be further used for.The method has higher universal, is applicable to various engineering fields.
Accompanying drawing explanation
To do the present invention below in conjunction with the drawings and specific embodiments and further illustrate, above-mentioned and/or otherwise advantage of the present invention will become apparent.
Fig. 1 is process flow diagram of the present invention.
Fig. 2 is the key point schematic diagram of human body
Fig. 3 is the chain structure of wire-frame image vegetarian refreshments along profile.
Fig. 4 a ~ Fig. 4 d is trunk segmentation example 1.
Fig. 5 a ~ Fig. 5 d is trunk segmentation example 2.
Embodiment
A kind of basic point of departure of human body component clustering method based on Face datection and key point location that the present invention proposes is by using comparatively ripe Face datection algorithm, by central point and the radius of recognition of face determination face.Again according to face location and size, accurately locate human body contour outline, then according to the restriction relation that partes corporis humani divides ratio, profile finds key point, realize the division of human body assembly according to key point and profile information.
The present invention uses comparative maturity Face datection algorithm to carry out Face detection, then carries out component clustering according to the key point information in face location and profile.It is characterized in that after input picture, comprise following treatment step:
Step 1, input the human body image without limbs overlap, use Adaboost learning algorithm in the picture, the method for detecting human face namely based on Haar-like cascade classifier carries out Face datection, according to detect the center of positional information determination face of the face obtained and face radius;
Step 2, detects according to the positional information of face and obtains corresponding human body outline;
Step 3, carries out key point location based on orthogonal list structure detecting on the human body contour outline that obtains;
Step 4, utilizes key point human body contour outline to be divided into the profile of each human body assembly.
Wherein, step 2 comprises: use Canny operator to carry out rim detection, find out the outline set { C of all objects in human body image
i, i ∈ 1,2 ..., m}, and C
i={ P
j(X
j, Y
j), j ∈ 1,2 ..., n}, m, n be greater than 1 natural number, P
j(X
j, Y
j) represent pixel on outline, X
jand Y
jrepresent horizontal ordinate and the ordinate of the pixel on outline respectively, X
o, Y
oand R
orepresent the face center horizontal ordinate, face center ordinate and the face radius that detect and obtain respectively, meet following formula:
Wherein outline set { C
iin the maximum profile of element be human body contour outline.
Step 3 comprises the steps:
Step 3-1, the simple chain structure of storage organization to the pixel on human body outline of orthogonal list is adopted to recombinate, and carry out searching and locating of key point, key point comprises: the upper bound point of face center 1, neck left and right edges key point totally 2, human body left arm and right arm and the lower boundary point totally 4 of human body left arm and right arm, human body waist left and right edges key point totally 2,1, human body crotch mid point;
Step 3-2, set up orthogonal list: a pointwise chain type scanning is carried out to the pixel point range on human body outline, in scanning process, during scanning first point, set up a wardrobe pointer and a row head pointer respectively, all point to this point of scanning, a newly-built total pointer again, point to wardrobe pointer to the right, downward sensing row head pointer, when then scanning other one by one by the chain structure of human body outline, row or column is created as belonged to, just add in orthogonal list by row or column coordinate order, row or column is created if do not belonged to, just a newly-built row or column head pointer, first this pointer is added in row or column head pointer chain, again this analyzing spot of pointed, statistics is row and the some number total often arranged often, and with index identify each point when namely the sequence number being read out is being read out be which,
Step 3-3, determine neck left and right edge key point, algorithm steps is as follows:
Step 3-3-1, obtains cross chain table row chain coordinate Y=Y
o+ R
oline pointer;
Step 3-3-2, in traversal orthogonal list, there is a P in institute
j,if there is some P
jhorizontal ordinate meet X
jx
o-R
o<X
j<X
o, then P
jfor neck left hand edge key point, if P
jfor in orthogonal list, last is put and X
o<X
j<X
o+ R
o, then P
jfor neck right hand edge key point, if neck left and right edges key point finds entirely, then algorithm stops, otherwise performs step 3-3-3;
Step 3-3-3, line pointer moves up a row, if row chain coordinate Y=Y
o-R
o, algorithm stops;
Step 3-3-4, repeats step 3-3-2 and step 3-3-3.
Step 3-4, determine human body crotch mid point key point, algorithm steps is as follows:
Step 3-4-1, obtains cross chain table row chain coordinate Y=(2R
o× 5.5/4) × (1-0.467)/(1-0.844)-(Y
o-R
o) line pointer;
Step 3-4-2, in traversal orthogonal list, there is a P in institute
jif there is some P
jhorizontal ordinate X
jmeet X
a<X
j<X
b, then step 3-4-3, P is performed
aand P
brepresent neck left and right edge key point respectively, X
aand X
bbe respectively the horizontal ordinate of neck left and right edge key point, if not exist and line pointer did not move, perform step 3-4-4, if not exist and line pointer moved, perform step 3-4-5;
Step 3-4-3, definition current line chain coordinate is Y
k, upwards travel through row chain coordinate in orthogonal list line by line and meet Y
k<Y<=Y
kthe row of-hoG, wherein hoG is the vertical height of navel to stern lower edge, hoG=(2R
o× 5.5/4), if there is some P in orthogonal list in × (0.600-0.467)/(1-0.844)
jhorizontal ordinate X
jmeet X
a<X
j<X
brow chain, move on to this journey by line pointer, continue perform step 3-4-2, otherwise perform step 3-4-5;
Step 3-4-4, travels through in orthogonal list downwards line by line and has a P
j, find first to there is some P
jhorizontal ordinate X
jmeet X
a<X
j<X
brow;
Points all in the row that step 3-4-5, traversal step 3-4-4 find, | X
j-(X
a+ X
bthe minimum point of)/2| is human body crotch mid point key point;
Step 3-5, determine human body waist left and right edges key point, algorithm steps is as follows:
Step 3-5-1, obtains row chain coordinate Y=Y in orthogonal list
ithe line pointer of-hoG, P
irepresent human body crotch mid point key point, Y
irepresent some P
iordinate;
Step 3-5-2, if row chain coordinate Y>Y
i, then perform step 3-5-3, otherwise traversal orthogonal list in institute a little, if there is horizontal ordinate X
jmeet X
i-WoS/2<=X
j<X
isome P
j, X
irepresent some P
ihorizontal ordinate, wherein X
jminimum some P
jbe human body waist left hand edge key point, meet X if exist
i<X
j<=X
ithe point P of+WoS/2
j, wherein X
jmaximum some P
jbe human body waist right hand edge key point, WoS is shoulder breadth, if human body waist left and right edge key point does not find entirely, then performs step 3-5-4, if waist left and right edge key point finds entirely, then algorithm stops;
Step 3-5-3, (5 is empirical values, and its existence is the lateral extent in order to expand search waist point, and it can get the arbitrary integer being greater than 0 WoS to be updated to WoS+5, value is too small, the number of times of repeated execution of steps 3-5-2 will be too much, and value is excessive, likely the point on arm is defined as the point on waist, so compromise considers that value is 5), perform step 3-5-1, if this step performs more than four times, algorithm stops;
Step 3-5-4, line pointer line down, returns and performs step 3-5-1
Step 3-6, determine human body left and right arms bound point, algorithm steps is as follows:
Step 3-6-1, to make in orthogonal list sequence number index a little deduct a value and make (index)
a=0, (index)
arepresent some P
athe sequence number being read out;
Step 3-6-2, obtains row chain coordinate X=X
gcolumn pointer, P
grepresent human body waist left hand edge key point, X
grepresent some P
ghorizontal ordinate;
Step 3-6-3, in traversal orthogonal list, there is a P in institute
j, for finger point P
f, (index)
fequal (index)
aadd the curve distance of neck left hand edge key point to finger point, if there is some P
jmeet (index)
a< (index)
j< (index)
f, then this point is human body left arm upper bound point, performs step 3-6-4, if do not exist, then performs step 3-6-5, (index)
jrepresent some P
jthe sequence number being read out, (index)
frepresent some P
fthe sequence number being read out;
Step 3-6-4, in traversal orthogonal list, institute a little, if there is some P
jmeet (index)
a< (index)
j< (index)
f(index)
j-(index)
a> (index)
g-(index)
j, then this point is human body left arm lower boundary point, and algorithm stops, if do not exist, then performs step 3-6-5, (index)
grepresent some P
gthe sequence number being read out;
Step 3-6-5, column pointer moves to left row, returns and performs step 3-6-2.
Step 4 comprises: the orthogonal list structure utilizing the human body outline pixel established, from first point, off-take point sets up first man body profile one by one, first after scanning neck right hand edge key point by chain structure, start read point write in newly-built human body contour outline by the point of reading one by one, just stop reading in until run into neck left hand edge key point, complete human head profile;
Newly-built second human body contour outline, after scanning human body left arm upper bound point, starts read point being write in second newly-built human body contour outline by the point of reading one by one, just stops reading in, complete human body left arm profile until run into human body left arm lower boundary point;
Newly-built 3rd human body contour outline, after scanning human body waist left hand edge key point, starts to read in the 3rd newly-built human body contour outline by point one by one, just stops reading in, complete left leg profile until run into human body crotch mid point key point;
Newly-built 4th human body contour outline, reads in the 4th newly-built human body contour outline by point one by one from human body crotch mid point key point, just stops reading in, complete right leg profile until run into human body waist right hand edge key point;
Newly-built 5th human body contour outline, after scanning human body right arm lower boundary point, starts to read in the 5th newly-built human body contour outline by point one by one, just stops reading in, complete right arm profile until run into human body right arm upper bound point.
Embodiment
As shown in Figure 1.In Fig. 1, step 1 is initialization action.Step 2 inputs an achiasmate positive face human body image.
What step 3 adopted is method for detecting human face based on Adaboost learning algorithm, also namely based on the method for detecting human face of Haar-like cascade classifier.
Step 4, according to the face location information detected, finds corresponding human body contour outline.First use Canny operator to carry out rim detection, find out the outline set { C of all objects in image
i(i ∈ 1,2 ..., m}), and C
i={ P
j(X
j, Y
j) (j ∈ 1,2 ..., n}), P
j(X
j, Y
j) pixel on finger wheel exterior feature.Meet
with
(wherein X
o, Y
oand R
obe respectively and face center transverse and longitudinal coordinate and face radius detected) and C
ithe profile that element is maximum is human body contour outline.
Step 5 completes the foundation of human body contour outline orthogonal list.As shown in Figure 3, for wire-frame image vegetarian refreshments is along the chain structure of profile.A pointwise chain type scanning is carried out to human body contour outline point range, in the process, during scanning first point, first set up a wardrobe pointer and a row head pointer respectively, all point to this point, a newly-built total pointer again, point to wardrobe pointer to the right, downward sensing row head pointer, when then scanning other one by one by the chain structure of profile, row (or row) is created as belonged to, just by row (or row) coordinate order adds in orthogonal list, row (or row) is created if do not belonged to, just a newly-built row (or row) head pointer, first this pointer is added and enter a profession in (or row) head pointer chain, again it is pointed to this analyzing spot.Statistics is row and the every some number total that arranges often, and with index identify each point when being read out be which.
Step 6 determines neck left and right edges key point, and its step is as follows:
I () obtains row chain coordinate Y=Y
o+ R
oline pointer;
(ii) P is had in ergodic chain
iif, X
o-R
o<X
i<X
o, then P
ibe neck left hand edge key point, if P
ifor in chain, last is put and X
o<X
i<X
o+ R
o, then P
ifor neck right hand edge key point, if neck left and right edges key point finds entirely, then algorithm stops, otherwise performs (iii);
(iii) pointer moves up a row, if row chain coordinate Y=Y
o-R
o, algorithm stops;
(iv) repeat i-th i step and the i-th ii walk.
Step 7 determines human body crotch central point, and its step is as follows:
I () obtains row chain coordinate Y=(2R
o× 5.5/4) × (1-0.467)/(1-0.844)-(Y
o-R
o) line pointer, be herein by scan line general location to human body in the middle part of a line, to make use of the table 1 experimental knowledge crown be (1-0.467)/(1-0.844) and the crown to vertical height and the crown of stern lower edge to the ratio of the vertical height of shoulder to the column direction distance of shoulder is substantially equal to 5.5/4 times of face diameter of a circle;
| Code name | Title | Ratio |
| 1 | Height | H |
| 2 | Shoulder height | 0.844H |
| 3 | Navel is high | 0.600H |
| 4 | Stern is high | 0.467H |
| 5 | Shoulder-shoulder distance | 0.222H |
Table 1
(ii) P is had in ergodic chain
iif there is point and meet X
a<X
i<X
b, then carry out the i-th ii step, if not exist and line pointer did not move, carry out the i-th v step, if not exist and line pointer moved, carry out v step;
(iii) suppose that current line chain coordinate is Y
j, upwards travel through row chain coordinate line by line and meet Y
j<Y<=Y
jthe row of-hoG, wherein hoG is the vertical height of navel to stern lower edge, and navel obtains hoG=(2R to vertical height and the crown of stern lower edge to the ratio of the vertical height of shoulder for (0.600-0.467)/(1-0.844) based on experience
o× 5.5/4), if there is chain mid point to meet X in × (0.600-0.467)/(1-0.844)
a<X
i<X
brow chain, move on to this journey by line pointer, continue execution i-th i step, otherwise carry out v step;
(iv) travel through downwards line by line, find first to there is point and meet X
a<X
i<X
brow;
V in () traversal row, institute is a little, | X
i-(X
a+ X
bthe minimum point of)/2| is crotch mid point.
Step 8 determines human body waist left and right edges key point, and its step is as follows:
I () obtains row chain coordinate Y=Y
ithe line pointer of-hoG;
(ii) if row chain coordinate Y>Y
i, then perform the i-th ii step, otherwise in ergodic chain, institute a little, meets X if exist
i-WoS/2<=X
i<X
ipoint, wherein X
iminimum is human body waist left hand edge key point, meets X if exist
i<X
i<=X
ithe point of+WoS/2, wherein X
imaximum is human body waist right hand edge key point.If 2 are not found entirely, then perform the i-th v step, if 2 are found entirely, then algorithm stops;
(iii) WoS=WoS+5, performs the i-th step, if this step execution is more than four times, algorithm stops;
(iv) line pointer line down.
Step 9 determines human arm up-and-down boundary point, and its step is as follows:
(i) order index a little deduct a value and make (index)
a=0;
(ii) row chain coordinate X=X is obtained
gcolumn pointer;
(iii) P is had in ergodic chain
iif there is point and meet (index)
a< (index)
i< (index)
j, then this point is human body left arm upper bound point, performs the i-th v step, if do not exist, then performs v step;
(iv) P is had in ergodic chain
iif there is point and meet (index)
a< (index)
i< (index)
j(index)
i-(index)
a> (index)
g-(index)
i, then this point is human body left arm lower boundary point, and algorithm stops, if do not exist, then performs v step;
V () column pointer moves to left row.
Step 10 utilizes key point to divide human body contour outline.As shown in Figure 2, for the key point schematic diagram of human body, wherein HEAD is orthogonal list gauge outfit, A is neck left hand edge key point, B is neck right hand edge key point, C is left arm upper bound point, and D is left arm lower boundary point, and E is right arm upper bound point, F is right arm lower boundary point, G is waist left hand edge, and H is waist right hand edge key point, and I is crotch central point, O is face center, sense of rotation when determining starting point and start read point most, utilizes all point along the chain structure of profile, and from first point, off-take point sets up new segmentation contour one by one.As shown in Figure 3, be profile screening schematic diagram.Suppose original human body contour outline read point out time point by point scanning sense of rotation be rotated counterclockwise, after then scanning neck right hand edge key point by chain structure, start read point one by one to read in newly-built profile, just stop reading in until run into neck left hand edge key point, contouring head completes.Newly-built second profile, after scanning left arm upper bound point, start to read in newly-built profile by point one by one, just stop reading in until run into left arm lower boundary point, left arm profile completes.Newly-built 3rd profile, after scanning waist left hand edge key point, start to read in newly-built profile by point one by one, just stop reading in until run into crotch mid point, left leg profile completes.Newly-built 4th profile, reads in newly-built profile by point one by one from crotch mid point, and just stop reading in until run into waist right hand edge key point, right leg profile completes.Newly-built 5th profile, after scanning right arm lower boundary point, start to read in newly-built profile by point one by one, just stop reading in until run into right arm upper bound point, right arm profile completes.Suppose original human body contour outline read point out time point by point scanning sense of rotation be turn clockwise, the order of newly-built profile is contrary with when being rotated counterclockwise, first point that each profile reads in and last point also just in contrast.
Step 11 utilizes key point to measure human body assembly.According to the positional information of paired key point, can the relative length of computation module, that is to say and use the number of pixel to represent length component.
Step 12 EOP (end of program), completes the result of human body component clustering as shown in Fig. 4 a ~ Fig. 4 d and Fig. 5 a ~ Fig. 5 d, and Fig. 4 a is that Fig. 4 b is the division result of Fig. 4 a containing human body and non-human (profile of human body is considered as chaff interference) original image; Fig. 4 c is that Fig. 4 d is the division result of Fig. 4 c only containing the original image of human body; Fig. 5 a is the original image containing single human body, and Fig. 5 b is the profile division result of Fig. 5 a; Fig. 5 c is the original image containing multiple human body, and Fig. 5 d is the division result of Fig. 5 c.
The invention provides a kind of human body component clustering method of locating based on Face datection and key point; the method and access of this technical scheme of specific implementation is a lot; the above is only the preferred embodiment of the present invention; should be understood that; for those skilled in the art; under the premise without departing from the principles of the invention, can also make some improvements and modifications, these improvements and modifications also should be considered as protection scope of the present invention.The all available prior art of each ingredient not clear and definite in the present embodiment is realized.
Claims (8)
1., based on the human body component clustering method that Face datection and key point are located, it is characterized in that, comprise the following steps:
Step 1, input the human body image without limbs overlap, use Adaboost learning algorithm in the picture, the method for detecting human face namely based on Haar-like cascade classifier carries out Face datection, according to the center of positional information determination face and the radius of face that detect the face obtained;
Step 2, detects according to the positional information of face and obtains corresponding human body outline;
Step 3, carries out key point location based on orthogonal list structure detecting on the human body contour outline that obtains;
Step 4, utilizes key point human body contour outline to be divided into the profile of each human body assembly.
2. a kind of human body component clustering method based on Face datection and key point location as claimed in claim 1, it is characterized in that, step 2 comprises: use Canny operator to carry out rim detection, find out the outline set { C of all objects in human body image
i, i ∈ 1,2 ..., m}, and C
i={ P
j(X
j, Y
j), j ∈ 1,2 ..., n}, m, n be greater than 1 natural number, P
j(X
j, Y
j) represent pixel on outline, X
jand Y
jrepresent horizontal ordinate and the ordinate of the pixel on outline respectively, X
o, Y
oand R
orepresent the face center horizontal ordinate, face center ordinate and the face radius that detect and obtain respectively, meet following formula:
Wherein outline set { C
iin the maximum profile of element be human body contour outline.
3. as claimed in claim 2 a kind of based on Face datection and key point location human body component clustering method, it is characterized in that, step 3 comprises the steps:
Step 3-1, the simple chain structure of storage organization to the pixel on human body outline of orthogonal list is adopted to recombinate, and carry out searching and locating of key point, key point comprises: the upper bound point of face center 1, neck left and right edges key point totally 2, human body left arm and right arm and the lower boundary point totally 4 of human body left arm and right arm, human body waist left and right edges key point totally 2,1, human body crotch mid point;
Step 3-2, set up orthogonal list: a pointwise chain type scanning is carried out to the pixel point range on human body outline, in scanning process, during scanning first point, set up a wardrobe pointer and a row head pointer respectively, all point to this point of scanning, a more newly-built total pointer, point to wardrobe pointer to the right, point to row head pointer downwards; Then other point is scanned one by one by the chain structure of human body outline, point as scanning belongs to and creates row or column, just add in orthogonal list by row or column coordinate order, the row or column created if do not belonged to, just a newly-built row or column head pointer, first adds this pointer in row or column head pointer chain, then this analyzing spot of pointed, statistics is row and the every some number that arranges often, and by the sequence number that the point that index identifies each scanning is being read out;
Step 3-3, determines neck left and right edge key point;
Step 3-4, determines human body crotch mid point key point;
Step 3-5, determines human body waist left and right edge key point;
Step 3-6, determines the upper bound point of human body left arm and right arm, and the lower boundary point of human body left arm and right arm.
4. as claimed in claim 3 a kind of based on Face datection and key point location human body component clustering method, it is characterized in that, step 3-3 comprises the steps:
Step 3-3-1, obtains cross chain table row chain coordinate Y=Y
o+ R
oline pointer;
Step 3-3-2, in traversal orthogonal list, there is a P in institute
jif there is some P
jhorizontal ordinate X
jmeet X
o-R
o<X
j<X
o, then P
jfor neck left hand edge key point, if P
jfor in orthogonal list, last is put and X
o<X
j<X
o+ R
o, then P
jfor neck right hand edge key point, if neck left and right edges key point finds entirely, then algorithm stops, otherwise performs step 3-3-3;
Step 3-3-3, line pointer moves up a row, if row chain coordinate Y=Y
o-R
o, algorithm stops;
Step 3-3-4, repeats step 3-3-2 ~ step 3-3-3.
5. as claimed in claim 4 a kind of based on Face datection and key point location human body component clustering method, it is characterized in that, step 3-4 comprises the steps:
Step 3-4-1, obtains the line pointer meeting the row chain coordinate Y of following formula in orthogonal list:
Y＝(2R
O×5.5/4)×(1-0.467)/(1-0.844)-(Y
O-R
O)；
Step 3-4-2, in traversal orthogonal list, there is a P in institute
jif there is some P
jhorizontal ordinate X
jmeet X
a<X
j<X
b, P
aand P
brepresent neck left and right edge key point respectively, X
aand X
bbe respectively the horizontal ordinate of neck left and right edge key point, then perform step 3-4-3, if not exist and line pointer did not move, perform step 3-4-4, if not exist and line pointer moved, perform step 3-4-5;
Step 3-4-3, definition current line chain coordinate is Y
k, upwards travel through row chain coordinate in orthogonal list line by line and meet Y
k<Y<=Y
kthe row of-hoG, wherein hoG is the vertical height of navel to stern lower edge:
hoG＝(2R
O×5.5/4)×(0.600-0.467)/(1-0.844)，
If there is some P in orthogonal list
jhorizontal ordinate X
jmeet X
a<X
j<X
b, move on to this journey by line pointer, continue to perform step 3-4-2, otherwise perform step 3-4-5;
Step 3-4-4, travels through in orthogonal list downwards line by line and has a P
j, find first to there is some P
jhorizontal ordinate X
jmeet X
a<X
j<X
brow;
Points all in the row that step 3-4-5, traversal step 3-4-4 find, judges | X
j-(X
a+ X
bthe minimum point of)/2| is human body crotch mid point key point.
6. as claimed in claim 5 a kind of based on Face datection and key point location human body component clustering method, it is characterized in that, step 3-5 comprises the steps:
Step 3-5-1, obtains row chain coordinate Y=Y in orthogonal list
ithe line pointer of-hoG, P
irepresent human body crotch mid point key point, Y
irepresent some P
iordinate;
Step 3-5-2, if row chain coordinate Y>Y
i, then perform step 3-5-3, otherwise traversal orthogonal list in institute a little, if there is horizontal ordinate X
jmeet X
i-WoS/2<=X
j<X
isome P
j, X
irepresent some P
ihorizontal ordinate, wherein X
jminimum some P
jbe human body waist left hand edge key point, meet X if exist
i<X
j<=X
ithe point P of+WoS/2
j, wherein X
jmaximum some P
jbe human body waist right hand edge key point, WoS is shoulder breadth, if human body waist left and right edge key point does not find entirely, then performs step 3-5-4, if human body waist left and right edge key point finds entirely, then algorithm stops;
Step 3-5-3, is updated to WoS+5 by WoS, performs step 3-5-1, if this step performs more than four times, algorithm stops;
Step 3-5-4, line pointer line down, returns and performs step 3-5-1.
7. as claimed in claim 6 a kind of based on Face datection and key point location human body component clustering method, it is characterized in that, step 3-6 comprises the steps:
Step 3-6-1, to make in orthogonal list sequence number index a little deduct a value and make (index)
a=0, (index)
arepresent some P
athe sequence number being read out;
Step 3-6-2, obtains row chain coordinate X=X
gcolumn pointer, P
grepresent human body waist left hand edge key point, X
grepresent some P
ghorizontal ordinate;
Step 3-6-3, in traversal orthogonal list, institute a little, for finger point P
f, (index)
fequal (index)
aadd the curve distance of neck left hand edge key point to finger point, if there is some P
jmeet (index)
a< (index)
j< (index)
f, then this point is human body left arm upper bound point, performs step 3-6-4, if do not exist, then performs step 3-6-5, (index)
jrepresent some P
jthe sequence number being read out, (index)
frepresent some P
fthe sequence number being read out;
Step 3-6-4, in traversal orthogonal list, institute a little, if there is some P
jmeet (index)
a< (index)
j< (index)
f(index)
j-(index)
a> (index)
g-(index)
j, then this point is human body left arm lower boundary point, and algorithm stops, if do not exist, then performs step 3-6-5, (index)
grepresent some P
gthe sequence number being read out;
Step 3-6-5, column pointer moves to left row, returns and performs step 3-6-2.
8. as claimed in claim 7 a kind of based on Face datection and key point location human body component clustering method, it is characterized in that, step 4 comprises:
Utilize the orthogonal list structure of the human body outline pixel established, from first point, off-take point sets up first man body profile one by one, first after scanning neck right hand edge key point by chain structure, start read point write in newly-built human body contour outline by the point of reading one by one, just stop reading in until run into neck left hand edge key point, complete human head profile;
Newly-built second human body contour outline, after scanning human body left arm upper bound point, starts read point being write in second newly-built human body contour outline by the point of reading one by one, just stops reading in, complete human body left arm profile until run into human body left arm lower boundary point;
Newly-built 3rd human body contour outline, after scanning human body waist left hand edge key point, starts to read in the 3rd newly-built human body contour outline by point one by one, just stops reading in, complete left leg profile until run into human body crotch mid point key point;
Newly-built 4th human body contour outline, reads in the 4th newly-built human body contour outline by point one by one from human body crotch mid point key point, just stops reading in, complete right leg profile until run into human body waist right hand edge key point;
Newly-built 5th human body contour outline, after scanning human body right arm lower boundary point, starts to read in the 5th newly-built human body contour outline by point one by one, just stops reading in, complete right arm profile until run into human body right arm upper bound point.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510347507.9A CN104899575A (en) | 2015-06-19 | 2015-06-19 | Human body assembly dividing method based on face detection and key point positioning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510347507.9A CN104899575A (en) | 2015-06-19 | 2015-06-19 | Human body assembly dividing method based on face detection and key point positioning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN104899575A true CN104899575A (en) | 2015-09-09 |
Family
ID=54032229
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510347507.9A Pending CN104899575A (en) | 2015-06-19 | 2015-06-19 | Human body assembly dividing method based on face detection and key point positioning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104899575A (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105389801A (en) * | 2015-10-20 | 2016-03-09 | 厦门美图之家科技有限公司 | Figure outline setting method, shooting terminal, figure image blurring method and system |
| CN105608448A (en) * | 2016-02-22 | 2016-05-25 | 海信集团有限公司 | LBP characteristic extraction method based on face key points and LBP characteristic extraction device based on face key points |
| CN106845432A (en) * | 2017-02-07 | 2017-06-13 | 深圳市深网视界科技有限公司 | The method and apparatus that a kind of face is detected jointly with human body |
| CN108229418A (en) * | 2018-01-19 | 2018-06-29 | 北京市商汤科技开发有限公司 | Human body critical point detection method and apparatus, electronic equipment, storage medium and program |
| CN105224935B (en) * | 2015-10-28 | 2018-08-24 | 南京信息工程大学 | A kind of real-time face key independent positioning method based on Android platform |
| CN108550384A (en) * | 2018-03-30 | 2018-09-18 | 百度在线网络技术（北京）有限公司 | Method and apparatus for pushed information |
| CN109191552A (en) * | 2018-08-16 | 2019-01-11 | Oppo广东移动通信有限公司 | Threedimensional model processing method, device, electronic equipment and storage medium |
| CN109325450A (en) * | 2018-09-25 | 2019-02-12 | Oppo广东移动通信有限公司 | Image processing method, device, storage medium and electronic equipment |
| CN109740516A (en) * | 2018-12-29 | 2019-05-10 | 深圳市商汤科技有限公司 | A kind of user identification method, device, electronic equipment and storage medium |
| CN110298785A (en) * | 2019-06-29 | 2019-10-01 | 北京字节跳动网络技术有限公司 | Image beautification method, device and electronic equipment |
| CN110738654A (en) * | 2019-10-18 | 2020-01-31 | 中国科学技术大学 | Key point extraction and bone age prediction method in hip joint image |
| CN110880156A (en) * | 2018-09-05 | 2020-03-13 | 武汉斗鱼网络科技有限公司 | Long-leg special effect implementation method, storage medium, equipment and system |
| CN110895809A (en) * | 2019-10-18 | 2020-03-20 | 中国科学技术大学 | Method for accurately extracting key points in hip joint image |
| CN113031465A (en) * | 2021-03-22 | 2021-06-25 | 北京市商汤科技开发有限公司 | Control method, device and equipment of intelligent equipment and storage medium |
| 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 |
| WO2023165369A1 (en) * | 2022-03-01 | 2023-09-07 | 北京沃东天骏信息技术有限公司 | Image processing method and apparatus |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080037836A1 (en) * | 2006-08-09 | 2008-02-14 | Arcsoft, Inc. | Method for driving virtual facial expressions by automatically detecting facial expressions of a face image |
| CN101499128A (en) * | 2008-01-30 | 2009-08-05 | 中国科学院自动化研究所 | Three-dimensional human face action detecting and tracing method based on video stream |
| CN102360421A (en) * | 2011-10-19 | 2012-02-22 | 苏州大学 | Face identification method and system based on video streaming |
-
2015
- 2015-06-19 CN CN201510347507.9A patent/CN104899575A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080037836A1 (en) * | 2006-08-09 | 2008-02-14 | Arcsoft, Inc. | Method for driving virtual facial expressions by automatically detecting facial expressions of a face image |
| CN101499128A (en) * | 2008-01-30 | 2009-08-05 | 中国科学院自动化研究所 | Three-dimensional human face action detecting and tracing method based on video stream |
| CN102360421A (en) * | 2011-10-19 | 2012-02-22 | 苏州大学 | Face identification method and system based on video streaming |
Non-Patent Citations (1)
| Title |
|---|
| 马 旋等: ""基于人脸检测和关键点识别的快速人体组件划分"", 《计算机应用与软件》 * |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105389801A (en) * | 2015-10-20 | 2016-03-09 | 厦门美图之家科技有限公司 | Figure outline setting method, shooting terminal, figure image blurring method and system |
| CN105389801B (en) * | 2015-10-20 | 2018-09-21 | 厦门美图之家科技有限公司 | Character contour setting method, character image weakening method, system and camera terminal |
| CN105224935B (en) * | 2015-10-28 | 2018-08-24 | 南京信息工程大学 | A kind of real-time face key independent positioning method based on Android platform |
| CN105608448B (en) * | 2016-02-22 | 2019-08-09 | 海信集团有限公司 | A kind of LBP feature extracting method and device based on face's key point |
| CN105608448A (en) * | 2016-02-22 | 2016-05-25 | 海信集团有限公司 | LBP characteristic extraction method based on face key points and LBP characteristic extraction device based on face key points |
| CN106845432A (en) * | 2017-02-07 | 2017-06-13 | 深圳市深网视界科技有限公司 | The method and apparatus that a kind of face is detected jointly with human body |
| CN106845432B (en) * | 2017-02-07 | 2019-09-17 | 深圳市深网视界科技有限公司 | A kind of method and apparatus that face detects jointly with human body |
| CN108229418A (en) * | 2018-01-19 | 2018-06-29 | 北京市商汤科技开发有限公司 | Human body critical point detection method and apparatus, electronic equipment, storage medium and program |
| CN108550384A (en) * | 2018-03-30 | 2018-09-18 | 百度在线网络技术（北京）有限公司 | Method and apparatus for pushed information |
| CN108550384B (en) * | 2018-03-30 | 2022-05-03 | 百度在线网络技术（北京）有限公司 | Method and device for pushing information |
| CN109191552A (en) * | 2018-08-16 | 2019-01-11 | Oppo广东移动通信有限公司 | Threedimensional model processing method, device, electronic equipment and storage medium |
| CN110880156A (en) * | 2018-09-05 | 2020-03-13 | 武汉斗鱼网络科技有限公司 | Long-leg special effect implementation method, storage medium, equipment and system |
| CN109325450A (en) * | 2018-09-25 | 2019-02-12 | Oppo广东移动通信有限公司 | Image processing method, device, storage medium and electronic equipment |
| CN109740516A (en) * | 2018-12-29 | 2019-05-10 | 深圳市商汤科技有限公司 | A kind of user identification method, device, electronic equipment and storage medium |
| CN110298785A (en) * | 2019-06-29 | 2019-10-01 | 北京字节跳动网络技术有限公司 | Image beautification method, device and electronic equipment |
| CN110895809A (en) * | 2019-10-18 | 2020-03-20 | 中国科学技术大学 | Method for accurately extracting key points in hip joint image |
| CN110738654A (en) * | 2019-10-18 | 2020-01-31 | 中国科学技术大学 | Key point extraction and bone age prediction method in hip joint image |
| CN110738654B (en) * | 2019-10-18 | 2022-07-15 | 中国科学技术大学 | Key point extraction and bone age prediction method in hip joint image |
| CN110895809B (en) * | 2019-10-18 | 2022-07-15 | 中国科学技术大学 | Method for accurately extracting key points in hip joint image |
| CN113031465A (en) * | 2021-03-22 | 2021-06-25 | 北京市商汤科技开发有限公司 | Control method, device and equipment of intelligent equipment and storage medium |
| 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 |
| WO2023165369A1 (en) * | 2022-03-01 | 2023-09-07 | 北京沃东天骏信息技术有限公司 | Image processing method and apparatus |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104899575A (en) | Human body assembly dividing method based on face detection and key point positioning | |
| Zhao et al. | A survey on deep learning-based fine-grained object classification and semantic segmentation | |
| Harouni et al. | Online Persian/Arabic script classification without contextual information | |
| CN101763503B (en) | Face recognition method of attitude robust | |
| CN104200240B (en) | A kind of Sketch Searching method based on content-adaptive Hash coding | |
| CN105893968B (en) | The unrelated person's handwriting recognition methods end to end of text based on deep learning | |
| Gorelick et al. | Shape representation and classification using the poisson equation | |
| CN109685013B (en) | Method and device for detecting head key points in human body posture recognition | |
| CN105956560A (en) | Vehicle model identification method based on pooling multi-scale depth convolution characteristics | |
| CN104978549A (en) | Three-dimensional face image feature extraction method and system | |
| Sie Ho Lee et al. | Detecting curved symmetric parts using a deformable disc model | |
| CN107103326A (en) | The collaboration conspicuousness detection method clustered based on super-pixel | |
| Michel et al. | Scale invariant and deformation tolerant partial shape matching | |
| CN105139004A (en) | Face expression identification method based on video sequences | |
| CN103679151A (en) | LBP and Gabor characteristic fused face clustering method | |
| CN103310195A (en) | LLC-feature-based weak-supervision recognition method for vehicle high-resolution remote sensing images | |
| CN101140623A (en) | Video frequency objects recognition method and system based on supporting vectors machine | |
| Aouadi et al. | Word extraction and recognition in arabic. handwritten Text | |
| Pham et al. | Accurate junction detection and characterization in line-drawing images | |
| Joshi et al. | A random forest approach to segmenting and classifying gestures | |
| CN103714554A (en) | Video tracking method based on spread fusion | |
| CN103218825A (en) | Quick detection method of spatio-temporal interest points with invariable scale | |
| Wang et al. | S 3 d: scalable pedestrian detection via score scale surface discrimination | |
| CN106056074A (en) | Single training sample face identification method based on area sparse | |
| CN106408579A (en) | Video based clenched finger tip tracking method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into 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: 20150909 |