CN107122726A - A kind of multi-pose pedestrian detection method - Google Patents
A kind of multi-pose pedestrian detection method Download PDFInfo
- Publication number
- CN107122726A CN107122726A CN201710258662.2A CN201710258662A CN107122726A CN 107122726 A CN107122726 A CN 107122726A CN 201710258662 A CN201710258662 A CN 201710258662A CN 107122726 A CN107122726 A CN 107122726A
- Authority
- CN
- China
- Prior art keywords
- dpm
- sample
- pedestrian
- mrow
- mtd
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
- G06V40/23—Recognition of whole body movements, e.g. for sport training
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
Abstract
The invention discloses a kind of multi-pose pedestrian detection method, comprise the following steps:Picture is sampled and sample set is created, DPM feature extractions are carried out to the sample in sample set and obtain DPM characteristic vectors, the DPM features input LSVM graders that sample extraction is obtained carry out sample training, and the fraction for calculating testing image by pedestrian detector carries out the particular location that pedestrian detection draws pedestrian on image.Multi-pose pedestrian detection method proposed by the present invention, the characteristics of can adapt to pedestrian's non-rigid shape deformations is detected for the pedestrian of different scenes, different postures.
Description
Technical field
The present invention relates to a kind of detection method, especially a kind of multi-pose pedestrian detection method.
Background technology
Pedestrian detection may be defined as:Judge whether input picture includes pedestrian, if so, providing positional information.It is car
The first step in the applications such as auxiliary drives, intelligent video monitoring and human body behavioural analysis, be also employed in recent years Aerial Images,
In the emerging fields such as victim's rescue.Due to the complexity of background, pedestrian often will not individually occur in the scene, therefore can not
Having of avoiding block and angle and posture the deformation of non-rigid that comes of different band etc. so that pedestrian detection, which turns into, to be calculated
The Research Challenges of machine vision and focus
It is entitled《Multi-pose pedestrian detection based on posteriority HOG features》Paper.The paper is referred to after one kind is based on
The pedestrian detection method of HOG features is tested, common information --- the Gradient Features energy of whole pedestrian samples is first counted, to individual sample
This HOG features are weighted processing, obtain characterizing the posteriority HOG features at pedestrian edge, then utilize S-Isomap features drop
Dimension method and K-means clustering methods do subclass division to the pedestrian of different postures and visual angle, and integrated each subclass grader
Method.But program model is inflexible, and pedestrian has nonrigid feature, it is difficult to adapt to its motion;Lie down, squat for pedestrian
Under, the posture such as bend over, the corresponding posture training not being directed to, it is impossible to detect such posture;Moreover, being hidden in pedestrian's lower part of the body
The situation of gear, it is difficult to be detected.
The content of the invention
To overcome existing technological deficiency, the present invention proposes a kind of multi-pose pedestrian detection method, and this method is for colourful
State pedestrian's is non-rigid, proposes using pedestrian's head and shoulder as training positive sample, and is divided into two class positive samples according to posture type, trains
Two deformable member models are obtained to be detected for different postures.In detection process, carried out using hyperspin feature figure
The method of detection, has adapted to the pedestrian detection of multi-angle multi-pose.The present invention is achieved using following technical scheme, including with
Lower step:
1) sample set is created:The sample set includes positive sample collection and negative sample collection;
2) DPM feature extractions are carried out to sample set:DPM feature extractions are carried out to the sample in the sample set and obtain DPM
Characteristic vector;
3) sample training:The first kind positive sample and negative sample are extracted obtained DPM features and input LSVM graders
In obtain the first pedestrian detector, the input LSVM classification of obtained DPM features is extracted to the Equations of The Second Kind positive sample and negative sample
Second pedestrian detector is obtained in device;
4) pedestrian detection:Point of testing image is calculated by first pedestrian detector and second pedestrian detector
Number is so as to carry out pedestrian detection.
Further, step 1) in, the positive sample that the positive sample is concentrated is divided into first kind positive sample and the positive sample of Equations of The Second Kind
This, first kind positive sample mainly includes the posture picture standing, walk, being seated, and Equations of The Second Kind positive sample is the posture of people's recumbency
Picture;The negative sample that the negative sample is concentrated is sampled from the background picture of reality scene.
Further, step 2) in, can shape to the DPM that the first kind positive sample and negative sample extract DPM characteristic vectors
Become the components number of partial model into 4, part dimension is 6*6 pixels.
Further, the positive sample concentrates Equations of The Second Kind positive sample also to include carrying out Equations of The Second Kind positive sample appropriate angle
Postrotational sample.
Further, step 2) in, can shape to the DPM that the Equations of The Second Kind positive sample and negative sample extract DPM characteristic vectors
Become the components number of partial model into 5, part dimension is 6*6 pixels.
Further, step 2) in, the HOG obtained after being extracted to the positive sample and negative sample in the sample set is special
It is 36 dimension DPM characteristic vectors to levy vector.
Further, step 3) in, the DPM obtained after the dimensionality reduction is characterized as 13 dimensions.
Further, step 5) in, the pedestrian detection method step includes as follows:
41) image to be detected is inputted, dextrorotation is turn 90 degrees, 180 degree, 270 degree, and calculating obtains 4 DPM features respectively
Figure:F1、F2、F3、F4;
42) by step 51) in all DPM characteristic patterns that obtain examined respectively with first pedestrian detector and the second pedestrian
Survey device convolution and obtain 8 pedestrian detection score charts;
43) merged for two detection score charts that same DPM characteristic patterns convolution is obtained:In new score chart
In, each position takes the larger fractional value of correspondence position in two detection score charts, obtains S1, S2, S3, S4 points of new score chart
Dui Ying not F1, F2, F3, F4;
44) by 90 degree of S2, S3, S4 successively rotate counterclockwise, 180 degree, 270 degree, obtain S2 ', S3, S4 ';
45) according to S1, S2 ', S3 ', S4 ' fractional marks go out pedestrian where position, and merge testing result.
Further, step 4) in, the fractional expression is β Φ (x), and wherein β is the first pedestrian detector or the
Two pedestrian detectors, they are all characteristic vector filters, and Φ (x) is image and the position specified and yardstick, x be feature to
Amount, calculating obtains that fraction is bigger, represents that the possibility of pedestrian in detection window is bigger.
Further, for step 2) specific method of the dimensionality reduction is as follows:36 dimension DPM characteristic vectors are regarded as a 4*
9 matrix, makes V={ u1,...,u9}∪{v1,...,v9, wherein uiAnd viAll it is 36 dimensional vectors, its 4*9 expression matrix shape
Formula meets following condition:
Then with 36 dimension DPM features and each ukAnd vkCarry out 4 normalized values of certain row of dot product, i.e. calculating matrix expression
And obtain DPM features to each ukProjection, certain row of calculating matrix expression 9 normalized values and obtain DPM
Feature is to each vkProjection so that obtain one 13 dimension characteristic vector.
Compared with the prior art, the invention provides a kind of multi-pose pedestrian detection method, this method by pedestrian head and shoulder
Wing is divided into multiple parts as training sample, trained multiple submodels, combines the matching degree of master cast and submodel, with
And using the spatial relation of master cast and submodel, adapted to the non-rigid of many attitude pedestrian in motion process.We
Posture is divided into two classes and is trained by method, and characteristic pattern is rotated in detection process, has adapted to various postures.With pedestrian
Head and shoulder be used as training sample, it is to avoid the lower part of the body is blocked undetectable situation.
Brief description of the drawings
Fig. 1 is flow chart of the method for the present invention;
Fig. 2 is pedestrian detection flow chart of the invention.
Embodiment
To make the object, technical solutions and advantages of the present invention clearer, below in conjunction with the accompanying drawings to embodiment of the present invention
It is described in more detail.
1) sample set is created:The sample set includes positive sample collection and negative sample collection, the positive sample that the positive sample is concentrated
It is divided into first kind positive sample and Equations of The Second Kind positive sample, first kind positive sample mainly includes standing, walking, the posture picture being seated,
Equations of The Second Kind positive sample is the picture of the posture of people's recumbency.The negative sample that the negative sample is concentrated is the background picture from reality scene
Sampled.Appropriate angle rotation is carried out to Equations of The Second Kind positive sample, Equations of The Second Kind positive sample is added, is still being denoted as Equations of The Second Kind just
Sample.
2) sample is entered in DPM feature extractions, DPM features is carried out to the positive sample in the sample set and negative sample, carried
The DPM characteristic vectors obtained after taking are 36 dimension DPM characteristic vectors.
Then dimensionality reduction is carried out to DPM characteristic vectors:The specific method of dimensionality reduction is as follows:
36 dimension DPM characteristic vectors tie up direction histogram from 4 different normalization 9, so 36 dimension DPM characteristic vectors can
To regard 4*9 matrix as.If V={ u1,...,u9}∪{v1,...,v9, wherein uiAnd viAll it is 36 dimensional vectors, its 4*9
Expression matrix form meet following condition:
13 dimensional features are defined, element therein is 36 dimension DPM features and each ukAnd vkDot product.Calculate counterparty
To 4 normalized values and (sums that i.e. certain of expression matrix is arranged) obtain DPM features to each ukProjection, calculate correspondence
9 direction value of method for normalizing and (i.e. the sum of certain row of expression matrix) obtain DPM features to each vkProjection.
3) sample training:Training step is as follows:
DPM characteristic vectors are extracted to first kind positive sample and negative sample and are inputted the first pedestrian inspection is obtained in LSVM graders
Device is surveyed, the wherein components number of DPM deformable members pixel model is 4, and part dimension is 6*6 pixels.
DPM characteristic vectors are extracted to Equations of The Second Kind positive sample and negative sample and are inputted the second pedestrian inspection is obtained in LSVM graders
Device is surveyed, the wherein components number of DPM deformable members pixel model is 5, and part dimension is 6*6 pixels.
4) pedestrian detection:Because all detectors are all characteristic vector filters, fraction β Φ (x), wherein β are calculated
It is the first pedestrian detector or the second pedestrian detector, they are all characteristic vector filters, Φ (x) is image and the position specified
Put and yardstick, x is characteristic vector.Calculating obtains that fraction is bigger, represents that the possibility of pedestrian in detection window is bigger.Pedestrian detection
Basic step it is as follows:
41) image to be detected is inputted, dextrorotation is turn 90 degrees, 180 degree, 270 degree, and calculating obtains 4 DPM features respectively
Figure:F1、F2、F3、F4;
42) by step 41) in obtain all DPM characteristic patterns respectively with the first pedestrian detector and the second pedestrian detector
Convolution obtains 8 pedestrian detection score charts;
43) merged for two detection score charts that same DPM characteristic patterns convolution is obtained:In new score chart
In, each position takes the larger fractional value of correspondence position in two detection score charts, obtains S1, S2, S3, S4 points of new score chart
Dui Ying not F1, F2, F3, F4;
44) by 90 degree of S2, S3, S4 successively rotate counterclockwise, 180 degree, 270 degree, obtain S2 ', S3 ', S4 ';
45) according to S1, S2 ', S3 ', S4 ' fractional marks go out pedestrian where position, and merge testing result, thus
Go out the particular location of pedestrian on image.
Multi-pose pedestrian detection method proposed by the present invention, the characteristics of can adapt to pedestrian's non-rigid shape deformations, for difference
The pedestrian of posture difference angle is detected, it is adaptable to detected under scene, the pedestrian of different postures and camera angle.
Claims (10)
1. a kind of multi-pose pedestrian detection method, it is characterised in that comprise the following steps:
1) sample set is created:The sample set includes positive sample collection and negative sample collection;
2) DPM feature extractions are carried out to sample set:DPM feature extractions are carried out to the sample in the sample set and obtain DPM features
Vector;
3) sample training:The first kind positive sample and negative sample are extracted and obtained in obtained DPM features input LSVM graders
To first pedestrian detector, obtained DPM features are extracted to the Equations of The Second Kind positive sample and negative sample and input LSVM graders
In obtain second pedestrian detector;
4) pedestrian detection:By first pedestrian detector and second pedestrian detector calculate the fraction of testing image from
And carry out pedestrian detection.
2. multi-pose pedestrian detection method according to claim 1, it is characterised in that step 1) in, the positive sample collection
In positive sample be divided into first kind positive sample and Equations of The Second Kind positive sample, first kind positive sample includes the posture standing, walk, being seated
Picture, Equations of The Second Kind positive sample is the posture picture of people's recumbency;Background of the negative sample that the negative sample is concentrated from reality scene
Piece is sampled.
3. multi-pose pedestrian detection method according to claim 2, it is characterised in that step 2) in, to the first kind
The components number that positive sample and negative sample extract the DPM deformable member models of DPM characteristic vectors is 4, and part dimension is 6*6
Pixel.
4. multi-pose pedestrian detection method according to claim 2, it is characterised in that the positive sample is concentrating Equations of The Second Kind just
Sample also includes carrying out Equations of The Second Kind positive sample in the appropriate postrotational sample of angle.
5. multi-pose pedestrian detection method according to claim 4, it is characterised in that step 2) in, to the Equations of The Second Kind
The components number that positive sample and negative sample extract the DPM deformable member models of DPM characteristic vectors is 5, and part dimension is 6*6
Pixel.
6. the multi-pose pedestrian detection method according to claim 1 or 4, it is characterised in that step 2) in, to the sample
The DPM characteristic vectors that the positive sample and negative sample of concentration are obtained after being extracted are 36 dimension DPM characteristic vectors.
7. multi-pose pedestrian detection method according to claim 6, it is characterised in that step 2) in, after the dimensionality reduction
The DPM characteristic vectors arrived are 13 dimensions.
8. the multi-pose pedestrian detection method according to claim 1 or 2 or 4, it is characterised in that step 4) in, the row
People's detection method step includes as follows:
41) image to be detected is inputted, dextrorotation is turn 90 degrees, 180 degree, 270 degree, and calculating obtains 4 DPM characteristic patterns respectively:
F1、F2、F3、F4;
42) by step 41 obtain all DPM characteristic patterns respectively with first pedestrian detector and the second pedestrian detector
Convolution obtains 8 pedestrian detection score charts;
43) merged for two detection score charts that same DPM characteristic patterns convolution is obtained:In new score chart, often
Individual position takes the larger fractional value of correspondence position in two detection score charts, obtains new score chart S1, S2, S3, S4 right respectively
Answer F1, F2, F3, F4;
44) by 90 degree of S2, S3, S4 successively rotate counterclockwise, 180 degree, 270 degree, obtain S2 ', S3 ', S4 ';
55) according to S1, S2 ', S3 ', S4 ' fractional marks go out pedestrian where position, and merge testing result.
9. multi-pose pedestrian detection method according to claim 8, it is characterised in that step 4) in, the score graph reaches
Formula is β Φ (x), and wherein β is the first pedestrian detector or the second pedestrian detector, and they are all characteristic vector filter, Φ
(x) it is image and the position specified and yardstick, x is characteristic vector, calculating obtains that fraction is bigger, represents pedestrian in detection window
Possibility is bigger.
10. multi-pose pedestrian detection method according to claim 6, it is characterised in that for step 2) dimensionality reduction
Specific method is as follows:
36 dimension DPM characteristic vectors are regarded as 4*9 matrix, make V={ u1,...,u9}∪{v1,...,v9, wherein uiAnd vi
All it is 36 dimensional vectors, its 4*9 expression matrix form meets following condition:
<mrow>
<msub>
<mi>u</mi>
<mi>k</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mn>1</mn>
</mtd>
<mtd>
<mrow>
<mi>i</mi>
<mi>f</mi>
<mi> </mi>
<mi>j</mi>
<mo>=</mo>
<mi>k</mi>
<mo>;</mo>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mrow>
<mi>o</mi>
<mi>t</mi>
<mi>h</mi>
<mi>e</mi>
<mi>r</mi>
<mi>w</mi>
<mi>i</mi>
<mi>s</mi>
<mi>e</mi>
<mo>;</mo>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
</mrow>
<mrow>
<msub>
<mi>v</mi>
<mi>k</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mo>{</mo>
<mrow>
<mtable>
<mtr>
<mtd>
<mn>1</mn>
</mtd>
<mtd>
<mrow>
<mi>i</mi>
<mi>f</mi>
<mi> </mi>
<mi>j</mi>
<mo>=</mo>
<mi>k</mi>
<mo>;</mo>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mrow>
<mi>o</mi>
<mi>t</mi>
<mi>h</mi>
<mi>e</mi>
<mi>r</mi>
<mi>w</mi>
<mi>i</mi>
<mi>s</mi>
<mi>e</mi>
<mo>;</mo>
</mrow>
</mtd>
</mtr>
</mtable>
<mo>;</mo>
</mrow>
<mo>;</mo>
</mrow>
Then with 36 dimension DPM features and each ukAnd vkCarry out dot product, i.e. 4 normalized values of certain row of calculating matrix expression
With obtain DPM features to each ukProjection, certain row of calculating matrix expression it is 9 normalized values and special to obtain DPM
Levy to each vkProjection so that obtain one 13 dimension DPM characteristic vectors.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710258662.2A CN107122726A (en) | 2017-04-19 | 2017-04-19 | A kind of multi-pose pedestrian detection method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710258662.2A CN107122726A (en) | 2017-04-19 | 2017-04-19 | A kind of multi-pose pedestrian detection method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107122726A true CN107122726A (en) | 2017-09-01 |
Family
ID=59725827
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710258662.2A Pending CN107122726A (en) | 2017-04-19 | 2017-04-19 | A kind of multi-pose pedestrian detection method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107122726A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109117719A (en) * | 2018-07-02 | 2019-01-01 | 东南大学 | Driving gesture recognition method based on local deformable partial model fusion feature |
CN109697390A (en) * | 2017-10-23 | 2019-04-30 | 北京京东尚科信息技术有限公司 | Pedestrian detection method, device, medium and electronic equipment |
CN109886086A (en) * | 2019-01-04 | 2019-06-14 | 南京邮电大学 | Pedestrian detection method based on HOG feature and Linear SVM cascade classifier |
CN110084118A (en) * | 2019-03-25 | 2019-08-02 | 哈尔滨工业大学(深圳) | Method for building up, pedestrian detection method and the device of pedestrian detection tranining database |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102842045A (en) * | 2012-08-03 | 2012-12-26 | 华侨大学 | Pedestrian detection method based on combined features |
CN103136524A (en) * | 2011-11-24 | 2013-06-05 | 北京三星通信技术研究有限公司 | Object detecting system and method capable of restraining detection result redundancy |
CN104318578A (en) * | 2014-11-12 | 2015-01-28 | 苏州科达科技股份有限公司 | Video image analyzing method and system |
CN104484680A (en) * | 2014-09-26 | 2015-04-01 | 徐晓晖 | Multi-model multi-threshold combined pedestrian detection method |
CN106407943A (en) * | 2016-09-28 | 2017-02-15 | 天津工业大学 | Pyramid layer positioning based quick DPM pedestrian detection method |
-
2017
- 2017-04-19 CN CN201710258662.2A patent/CN107122726A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103136524A (en) * | 2011-11-24 | 2013-06-05 | 北京三星通信技术研究有限公司 | Object detecting system and method capable of restraining detection result redundancy |
CN102842045A (en) * | 2012-08-03 | 2012-12-26 | 华侨大学 | Pedestrian detection method based on combined features |
CN104484680A (en) * | 2014-09-26 | 2015-04-01 | 徐晓晖 | Multi-model multi-threshold combined pedestrian detection method |
CN104318578A (en) * | 2014-11-12 | 2015-01-28 | 苏州科达科技股份有限公司 | Video image analyzing method and system |
CN106407943A (en) * | 2016-09-28 | 2017-02-15 | 天津工业大学 | Pyramid layer positioning based quick DPM pedestrian detection method |
Non-Patent Citations (4)
Title |
---|
JANFLUSSER,TOMASSUK,BARBARAZITOVA著: "《模式识别中的矩和矩不变量》", 31 December 2014 * |
PEDRO F. FELZENSZWALB 等: "Object Detection with Discriminatively Trained Part-Based Models", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
PEDRO FELZENSZWALB 等: "A Discriminatively Trained, Multiscale, Deformable Part Model", 《CVPR》 * |
熊聪 等: "基于DPM模型的行人检测技术的研究", 《电子设计工程》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109697390A (en) * | 2017-10-23 | 2019-04-30 | 北京京东尚科信息技术有限公司 | Pedestrian detection method, device, medium and electronic equipment |
CN109697390B (en) * | 2017-10-23 | 2020-12-22 | 北京京东尚科信息技术有限公司 | Pedestrian detection method, device, medium, and electronic apparatus |
CN109117719A (en) * | 2018-07-02 | 2019-01-01 | 东南大学 | Driving gesture recognition method based on local deformable partial model fusion feature |
CN109117719B (en) * | 2018-07-02 | 2020-04-14 | 东南大学 | Driving posture recognition method based on local deformable component model fusion characteristics |
CN109886086A (en) * | 2019-01-04 | 2019-06-14 | 南京邮电大学 | Pedestrian detection method based on HOG feature and Linear SVM cascade classifier |
CN109886086B (en) * | 2019-01-04 | 2020-12-04 | 南京邮电大学 | Pedestrian detection method based on HOG (histogram of oriented gradient) features and linear SVM (support vector machine) cascade classifier |
CN110084118A (en) * | 2019-03-25 | 2019-08-02 | 哈尔滨工业大学(深圳) | Method for building up, pedestrian detection method and the device of pedestrian detection tranining database |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2021047232A1 (en) | Interaction behavior recognition method, apparatus, computer device, and storage medium | |
CN103942577B (en) | Based on the personal identification method for establishing sample database and composite character certainly in video monitoring | |
JP6664163B2 (en) | Image identification method, image identification device, and program | |
Sidla et al. | Pedestrian detection and tracking for counting applications in crowded situations | |
CN108520226B (en) | Pedestrian re-identification method based on body decomposition and significance detection | |
US9639748B2 (en) | Method for detecting persons using 1D depths and 2D texture | |
CN106446930A (en) | Deep convolutional neural network-based robot working scene identification method | |
WO2019080203A1 (en) | Gesture recognition method and system for robot, and robot | |
CN107122726A (en) | A kind of multi-pose pedestrian detection method | |
US11380010B2 (en) | Image processing device, image processing method, and image processing program | |
CN104794737B (en) | A kind of depth information Auxiliary Particle Filter tracking | |
Jammalamadaka et al. | Has my algorithm succeeded? an evaluator for human pose estimators | |
CN108288047A (en) | A kind of pedestrian/vehicle checking method | |
CN107766864B (en) | Method and device for extracting features and method and device for object recognition | |
CN106295564A (en) | The action identification method that a kind of neighborhood Gaussian structures and video features merge | |
CN106934380A (en) | A kind of indoor pedestrian detection and tracking based on HOG and MeanShift algorithms | |
CN105938551A (en) | Video data-based face specific region extraction method | |
CN103500345A (en) | Method for learning person re-identification based on distance measure | |
CN106529441B (en) | Depth motion figure Human bodys' response method based on smeared out boundary fragment | |
Vadlapati et al. | Facial recognition using the OpenCV Libraries of Python for the pictures of human faces wearing face masks during the COVID-19 pandemic | |
CN105488541A (en) | Natural feature point identification method based on machine learning in augmented reality system | |
Lee et al. | Head and body orientation estimation using convolutional random projection forests | |
CN111310720A (en) | Pedestrian re-identification method and system based on graph metric learning | |
CN109977862A (en) | A kind of recognition methods of parking stall limiter | |
CN111639562A (en) | Intelligent positioning method for palm region of interest |
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: 20170901 |