CN106778452A - Service robot is based on human testing and the tracking of binocular vision - Google Patents
Service robot is based on human testing and the tracking of binocular vision Download PDFInfo
- Publication number
- CN106778452A CN106778452A CN201510830932.3A CN201510830932A CN106778452A CN 106778452 A CN106778452 A CN 106778452A CN 201510830932 A CN201510830932 A CN 201510830932A CN 106778452 A CN106778452 A CN 106778452A
- Authority
- CN
- China
- Prior art keywords
- tracking
- human
- service robot
- binocular vision
- human body
- 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
- G06F18/285—Selection of pattern recognition techniques, e.g. of classifiers in a multi-classifier system
Abstract
The invention provides human testing and tracking that a kind of service robot is based on binocular vision, including human testing step and human body tracking step；The human testing step, being connected by five-category device carries out human body image detection；The human body tracking step, by being obtained per frame human body information after human testing, performs tracking, confirms target body.The function that service robot intelligently leads, follows is realized by above-mentioned interaction.
Description
【Technical field】
The people of binocular vision is based on the present invention relates to robotics, more particularly to a kind of service robot
Physical examination survey and tracking.
【Background technology】
Human testing and the study hotspot that tracking technique is in computer vision field.The technology is mainly used
In the human-computer interaction function of indoor video monitoring system, auxiliary driving and service robot.At present, this side
The matured product in face is relatively fewer, especially in service robot field.
The scheme conventional with tracking for human testing, the difference according to sensor is divided three classes：1st, based on list
Mesh camera；2nd, based on laser sensor；3rd, based on RGBD sensors.
During using monocular camera, full detail only has image, for common human testing algorithm, easily goes out
The situation of existing error detection, certain region that will be in scene is as being human body.Because scene information is two dimension,
Three-dimensional information will be lost, this kind of error detection is difficult to avoid；Simultaneously for tracking process, due to lacking three-dimensional letter
Breath, it is impossible to pose estimation is carried out to target body, to obtain the movement locus of target.
The detection and tracking of human leg are realized using 2D laser sensors, the benefit of the method is processing frequency
Comparatively fast, required system resources in computation is relatively fewer.But, it also has obvious defect, due to information content mistake
It is few, cause the detection of human leg to be easy to error detection, need by a relatively simple scene just can correctly recognize
People's leg；Simultaneously for tracking process, easily with wrong object if there is nearer people's leg apart.
Using the algorithm of RGBD sensors, the algorithm sharpest edges are the letter that sensor can provide comprehensive and abundant
Breath, including image two-dimensional signal and corresponding depth information.Due to there is depth information, therefore can obtain
The track of human motion is obtained to help robot to be tracked；Simultaneously because informative, can be by depth
Information effectively reduces false recognition rate.
【The content of the invention】
Based on this, it is an object of the invention to provide the human testing that a kind of service robot is based on binocular vision
With tracking.
In order to realize the purpose of the present invention, there is provided a kind of service robot be based on binocular vision human testing with
Tracking, including human testing step and human body tracking step；The human testing step, by Pyatyi
Grader series connection carries out human body image detection；The human body tracking step, by being obtained per frame after human testing
Human body information, performs tracking, confirms target body.
Preferably, the first order grader in the five-category device, detects figure in the first order grader
As the size and human body of detection block are fixed with the distance of camera, and the abscissa positions of detection image detection block and
Human body is associated with the distance of camera.
Preferably, the second level grader in the five-category device, the second level grader passes through
Haar-like features carry out human body image detection with Adaboost graders.
Preferably, the Haar features include edge feature, linear character, central feature and diagonal feature,
And it is combined into feature templates.
Preferably, the feature masterplate includes two kinds of white and black, sets the characteristic value of the feature masterplate
It is white rectangle pixel and the value for subtracting black rectangle pixel sum.
Preferably, the training process of the Adaboost graders is：S11, given training sample (x1,
Y1) ..., (xi, yi) ..., (xn, yn), wherein xi represents i-th sample, and yi=0 is expressed as negative sample
This, yi=1 is expressed as positive sample, and n is training sample sum；The weight of S12, initialization training sample；S13、
First time iteration, trains Weak Classifier, and calculate error rates of weak classifiers；Selected threshold so that error
It is minimum；Update sample weights；S14, T circulation, obtain T Weak Classifier, evaluate each weak typing
The weight of the importance of device is weighted superposition, finally gives strong classifier.
Preferably, the third level grader in the five-category device, the third level grader passes through
Haar-like features are detected that the third level grader is enterprising in gray-scale map with Adaboost graders
Row training, classification.
Preferably, the level V grader in the five-category device, the level V grader passes through Hog
Feature is combined with SVM classifier.
Preferably, the HOG feature extractions are, S21, carry out gray processing to image；S22, image is entered
The standardization of row color space；S23, the gradient for calculating each pixel of image；S24, divide an image into it is small
Unit；S25, the histogram of gradients for counting each unit, form the describer of each unit；S26, will not
Fixed number unit constitutes a block, and the profiler series connection of all units, obtains described piece in described piece
HOG profilers；S27, the HOG profilers series connection by all pieces in image, obtain the figure
The HOG profilers of picture.
Preferably, the human body tracking using color histogram match confirm in currently detected human body whether
There is target body.
Prior art is different from, above-mentioned service robot is based on human testing and the tracking of binocular vision,
Including human testing step and human body tracking step.Wherein, the human testing step, by five-category
Device series connection carries out human body image detection；The human body tracking step, by being obtained per frame human body after human testing
Information, performs tracking, confirms target body.Service robot is set intelligently to lead, follow by above-mentioned interaction
Function.
【Brief description of the drawings】
Fig. 1 is human testing and track side of the service robot based on binocular vision in one embodiment of the invention
The classifier training procedure chart of method.
Fig. 2 is human testing and track side of the service robot based on binocular vision in one embodiment of the invention
The HOG feature extraction figures of method.
【Specific embodiment】
To describe technology contents of the invention, structural feature, the objects and the effects in detail, below in conjunction with
Implementation method simultaneously coordinates accompanying drawing to be explained in detail.It should be appreciated that specific embodiment described herein is only used to solve
The present invention is released, is not used to limit the present invention.
A kind of service robot is based on human testing and the tracking of binocular vision, including human testing step
With human body tracking step.Wherein, the human testing step, being connected by five-category device carries out human figure
As detection；The human body tracking step, by being obtained per frame human body information after human testing, performs tracking,
Confirm target body.
In one particular embodiment of the present invention, it is a kind of based on binocular vision service robot detection with
Track method, including human testing step and human body tracking step.
Wherein, the human testing step, being connected by five-category device carries out human body image detection；It is described
Human body tracking step, by being obtained per frame human body information after human testing, performs tracking, confirms target body
First order grader in the five-category device, can be greatly reduced the number of detection block, lift human body
The time of detection process, and for the correct classification of subsequent classifier provides basis.The first order grader
Size and human body and the distance of camera of middle detection image detection block are fixed, and detection image detection block horizontal seat
Cursor position and human body are associated with the distance of camera.
Based on this, the excursion of relevant parameter is obtained using logistic regression algorithm, the number of detection block is relative
There is the reduction of matter in the method for general sliding window.
It may be preferred that the second level grader in the five-category device, the second level grader passes through
Haar-like features carry out human body image detection with Adaboost graders.
Wherein, Haar features are used to characterize characteristics of human body, and Adaboost graders are used for the human figure for recognizing
Picture.Haar features include edge feature, linear character, central feature and diagonal feature, and are combined into spy
Levy template.
Feature masterplate includes two kinds of white and black, and it is white rectangle pixel to set the characteristic value of feature masterplate
With the value for subtracting black rectangle pixel sum.The Haar characteristic values reflect the grey scale change situation of image.
, only to some simple graphic structures, such as edge, line segment is more sensitive, therefore can only describe for rectangular characteristic
Particular orientation, level, vertical, diagonal structure.
As shown in figure 1, Adaboost enters the strong classifier that row set is obtained by by multiple Weak Classifiers, tool
Body ground the Adaboost graders training process be：
S11, given training sample (x1, y1) ..., (xi, yi) ..., (xn, yn), wherein xi is represented
I-th sample, yi=0 is expressed as negative sample, and yi=1 is expressed as positive sample, and n is training sample sum；
The weight of S12, initialization training sample；
S13, first time iteration, train Weak Classifier, and calculate error rates of weak classifiers；Selected threshold,
So that error is minimum；Update sample weights；
S14, by T times circulate after, obtain T Weak Classifier, evaluate the importance of each Weak Classifier
Weight be weighted superposition, finally give strong classifier.
It may be preferred that the third level grader in the five-category device, the third level grader passes through
Haar-like features are detected with Adaboost graders, wherein, the third level grader is in gray scale
It is trained on figure, is classified.The third level grader is acted on disparity map, color by disparity map relative two dimensional
Color image, possesses and becomes apparent and terse feature, it is easier to carry out the confirmation of human body target.
Fourth stage grader in the five-category device is made using Haar-like features combination supporting vector machine
It is grader.The SVMs (SVM) is a classifying quality preferably grader, by feature space
It is middle to construct a hyperplane for optimum segmentation to realize two-value classification.
In certain embodiments, the level V grader in the five-category device, the level V grader
Combined with SVM classifier by Hog features.
As shown in Fig. 2 the HOG feature extractions are,
S21, gray processing is carried out to image；Will image regard a 3-D view for x, y, z gray scale as；
S22, the standardization that color space is carried out to image；
Specifically, the standardization (normalization) of color space is carried out to input picture using Gamma correction methods；
To adjust the contrast of image, the influence caused by the shade and illumination variation of image local is reduced, while can
To suppress the interference of noise.
S23, the gradient for calculating each pixel of image；The gradient information includes size and Orientation, is taken turns for capturing
Wide information, while the interference that further weakened light shines；
S24, junior unit cells is divided an image into, in the embodiment of the present invention, for example, 6*6 pixels/cell；
S25, the histogram of gradients for counting each unit cell, the number of different gradients form each unit
Describer descriptor；
S26, indefinite several unit cell are constituted into a block block, all units in described piece of block
The profiler series connection of cell, obtains described piece of HOG profilers descriptor；
S27, the HOG profilers series connection by all pieces of block in image, obtain the HOG of described image
Profiler descriptor.
The human body tracking step, by being obtained per frame human body information after human testing, performs tracking, confirms
Target body.Produced tracking target is floated during can effectively avoiding tracking for a long time using which
The problem of shifting.
The human body tracking confirms to whether there is target in currently detected human body using color histogram match
Human body.Specifically, color histogram calculating is carried out using two passages of the H in hsv color space and V,
And build hsv color histogram model.
The embodiment of the present invention, is built in embedded board in concrete application, is applied to indoor service robot
On.To reduce the cost of system, biocular systems are voluntarily built using two network USB cameras, use OpenCV
The stereo calibration of binocular camera is carried out, and the image of 320*240 resolution sizes is used using its SGBM algorithm
Stereo matching is carried out, so as to obtain the RGBD data of scene.
To complete human testing process, using multistage human body grader, human sample is made first.This hair
Bright first order grader has a direct relation with the parameter of camera, the height etc. installed, therefore by binocular camera
After on service robot, human body image and corresponding anaglyph are gathered.
In the embodiment of the present invention, gather positive negative sample each 600 multiple, the mode for manually marking, by people
Body region is drawn as sample, then is read out training.Because feature is in actual implementation process, carry out many
The simplification of the degree of kind, therefore the feature calculation time is very short, so as to ensure the training speed of algorithm quickly, is easy to
The adjustment of model.
Service robot based on the binocular vision detection of the embodiment of the present invention and tracking, can be very good
Realize that the intelligence of robot leads, follows function, so that the intelligent of service robot product is effectively lifted,
Lift the practical value of its product.Function is led by adding service robot, makes robot application in exhibition
Shop, dining room and other needs show and lead the place of function.
It should be noted that in the present invention, such as first and second or the like relational terms are used merely to
One entity or operation are made a distinction with another entity or operation, and is not necessarily required or is implied this
There is any this actual relation or order between a little entities or operation.And, term " including ", " bag
Containing " or any other variant thereof is intended to cover non-exclusive inclusion, so that including a series of key elements
Process, method, article or terminal device not only include those key elements, but also including not arranging clearly
Other key elements for going out, or also include intrinsic for this process, method, article or terminal device
Key element.In the absence of more restrictions, limited by sentence " including ... " or " including ... "
Key element, it is not excluded that also exist in the process including above-mentioned key element, method, article or terminal device another
Outer key element.Additionally, herein, " being more than ", " being less than ", " exceeding " etc. are interpreted as not including this number；
" more than ", " below ", " within " etc. be interpreted as including this number.
Although being described to the various embodiments described above, those skilled in the art once know
Basic creative concept, then can make other change and modification to these embodiments, so the above is above-mentioned only
It is the embodiment of type of the present invention, not thereby limits scope of patent protection of the invention, it is every using the present invention
Equivalent structure or equivalent flow conversion that specification and accompanying drawing content are made, or directly or indirectly it is used in other
Related technical field, is similarly included within scope of patent protection of the invention.
Claims (10)
1. service robot is based on human testing and the tracking of binocular vision, it is characterised in that including people
Body detecting step and human body tracking step；The human testing step, is connected into pedestrian by five-category device
Body image detection；The human body tracking step, by being obtained after human testing per frame human body information, perform with
Track, confirms target body.
2. service robot according to claim 1 is based on human testing and the tracking of binocular vision,
Characterized in that, the first order grader in the five-category device, detects figure in the first order grader
As the size and human body of detection block are fixed with the distance of camera, and the abscissa positions of detection image detection block and
Human body is associated with the distance of camera.
3. service robot according to claim 1 is based on human testing and the tracking of binocular vision,
Characterized in that, the second level grader in the five-category device, the second level grader passes through
Haar-like features carry out human body image detection with Adaboost graders.
4. service robot according to claim 3 is based on human testing and the tracking of binocular vision,
Characterized in that, the Haar features include edge feature, linear character, central feature and diagonal feature,
And it is combined into feature templates.
5. service robot according to claim 4 is based on human testing and the tracking of binocular vision,
Characterized in that, the feature masterplate includes two kinds of white and black, the characteristic value of the feature masterplate is set
It is white rectangle pixel and the value for subtracting black rectangle pixel sum.
6. service robot according to claim 3 is based on human testing and the tracking of binocular vision,
Characterized in that, the training process of the Adaboost graders is：
S11, given training sample (x1, y1) ..., (xi, yi) ..., (xn, yn), wherein xi is represented
I-th sample, yi=0 is expressed as negative sample, and yi=1 is expressed as positive sample, and n is training sample sum；
The weight of S12, initialization training sample；
S13, first time iteration, train Weak Classifier, and calculate error rates of weak classifiers；Selected threshold,
So that error is minimum；Update sample weights；
S14, T circulation, obtain T Weak Classifier, evaluate the weight of the importance of each Weak Classifier
Superposition is weighted, strong classifier is finally given.
7. service robot according to claim 1 is based on human testing and the tracking of binocular vision,
Characterized in that, the third level grader in the five-category device, the third level grader passes through
Haar-like features are detected that the third level grader is enterprising in gray-scale map with Adaboost graders
Row training, classification.
8. service robot according to claim 1 is based on human testing and the tracking of binocular vision,
Characterized in that, the level V grader in the five-category device, the level V grader is by Hog
Feature is combined with SVM classifier.
9. service robot according to claim 8 is based on human testing and the tracking of binocular vision,
Characterized in that, the HOG feature extractions are,
S21, gray processing is carried out to image；
S22, the standardization that color space is carried out to image；
S23, the gradient for calculating each pixel of image；
S24, divide an image into junior unit；
S25, the histogram of gradients for counting each unit, form the describer of each unit；
S26, indefinite several units are constituted into a block, the profiler series connection of all units in described piece,
Obtain described piece of HOG profilers；
S27, the HOG profilers series connection by all pieces in image, obtain the HOG features of described image
Describer.
10. service robot according to claim 1 is based on human testing and the track side of binocular vision
Method, it is characterised in that the human body tracking is confirmed in currently detected human body using color histogram match
With the presence or absence of target body.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510830932.3A CN106778452A (en) | 2015-11-24 | 2015-11-24 | Service robot is based on human testing and the tracking of binocular vision |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510830932.3A CN106778452A (en) | 2015-11-24 | 2015-11-24 | Service robot is based on human testing and the tracking of binocular vision |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106778452A true CN106778452A (en) | 2017-05-31 |
Family
ID=58964083
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510830932.3A Pending CN106778452A (en) | 2015-11-24 | 2015-11-24 | Service robot is based on human testing and the tracking of binocular vision |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106778452A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108737362A (en) * | 2018-03-21 | 2018-11-02 | 北京猎户星空科技有限公司 | Register method, device, equipment and storage medium |
| CN110163033A (en) * | 2018-02-13 | 2019-08-23 | 京东方科技集团股份有限公司 | Positive sample acquisition methods, pedestrian detection model generating method and pedestrian detection method |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101196991A (en) * | 2007-12-14 | 2008-06-11 | 同济大学 | Close passenger traffic counting and passenger walking velocity automatic detection method and system thereof |
| CN101980242A (en) * | 2010-09-30 | 2011-02-23 | 徐勇 | Human face discrimination method and system and public safety system |
| CN102411368A (en) * | 2011-07-22 | 2012-04-11 | 北京大学 | Active vision human face tracking method and tracking system of robot |
| CN102496001A (en) * | 2011-11-15 | 2012-06-13 | 无锡港湾网络科技有限公司 | Method of video monitor object automatic detection and system thereof |
| CN104063689A (en) * | 2014-06-20 | 2014-09-24 | 南京航空航天大学 | Face image identification method based on binocular stereoscopic vision |
| CN104240264A (en) * | 2014-08-29 | 2014-12-24 | 青岛海信电器股份有限公司 | Height detection method and device for moving object |
| CN104821010A (en) * | 2015-05-04 | 2015-08-05 | 清华大学深圳研究生院 | Binocular-vision-based real-time extraction method and system for three-dimensional hand information |
-
2015
- 2015-11-24 CN CN201510830932.3A patent/CN106778452A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101196991A (en) * | 2007-12-14 | 2008-06-11 | 同济大学 | Close passenger traffic counting and passenger walking velocity automatic detection method and system thereof |
| CN101980242A (en) * | 2010-09-30 | 2011-02-23 | 徐勇 | Human face discrimination method and system and public safety system |
| CN102411368A (en) * | 2011-07-22 | 2012-04-11 | 北京大学 | Active vision human face tracking method and tracking system of robot |
| CN102496001A (en) * | 2011-11-15 | 2012-06-13 | 无锡港湾网络科技有限公司 | Method of video monitor object automatic detection and system thereof |
| CN104063689A (en) * | 2014-06-20 | 2014-09-24 | 南京航空航天大学 | Face image identification method based on binocular stereoscopic vision |
| CN104240264A (en) * | 2014-08-29 | 2014-12-24 | 青岛海信电器股份有限公司 | Height detection method and device for moving object |
| CN104821010A (en) * | 2015-05-04 | 2015-08-05 | 清华大学深圳研究生院 | Binocular-vision-based real-time extraction method and system for three-dimensional hand information |
Non-Patent Citations (2)
| Title |
|---|
| 张淑平: "基于视觉的并联机器人位姿检测方法研究", 《中国博士学位论文全文数据库 信息科技辑》 * |
| 耿盛涛: "基于双目视觉的机器人目标检测与跟踪研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110163033A (en) * | 2018-02-13 | 2019-08-23 | 京东方科技集团股份有限公司 | Positive sample acquisition methods, pedestrian detection model generating method and pedestrian detection method |
| US11238296B2 (en) | 2018-02-13 | 2022-02-01 | Boe Technology Group Co., Ltd. | Sample acquisition method, target detection model generation method, target detection method, computing device and computer readable medium |
| CN108737362A (en) * | 2018-03-21 | 2018-11-02 | 北京猎户星空科技有限公司 | Register method, device, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103530599B (en) | The detection method and system of a kind of real human face and picture face | |
| CN101406390B (en) | Method and apparatus for detecting part of human body and human, and method and apparatus for detecting objects | |
| CN106682603B (en) | Real-time driver fatigue early warning system based on multi-source information fusion | |
| CN103824070B (en) | A kind of rapid pedestrian detection method based on computer vision | |
| CN106127137A (en) | A kind of target detection recognizer based on 3D trajectory analysis | |
| CN107301378B (en) | Pedestrian detection method and system based on multi-classifier integration in image | |
| CN106446779B (en) | Personal identification method and device | |
| CN102609724B (en) | Method for prompting ambient environment information by using two cameras | |
| CN103870843B (en) | Head posture estimation method based on multi-feature-point set active shape model (ASM) | |
| CN107305635A (en) | Object identifying method, object recognition equipment and classifier training method | |
| CN106650619A (en) | Human action recognition method | |
| CN103530640B (en) | Unlicensed vehicle checking method based on AdaBoost Yu SVM | |
| CN109255375A (en) | Panoramic picture method for checking object based on deep learning | |
| CN104077577A (en) | Trademark detection method based on convolutional neural network | |
| CN104715238A (en) | Pedestrian detection method based on multi-feature fusion | |
| CN107784291A (en) | target detection tracking method and device based on infrared video | |
| CN110032932B (en) | Human body posture identification method based on video processing and decision tree set threshold | |
| CN108388871B (en) | Vehicle detection method based on vehicle body regression | |
| CN103390164A (en) | Object detection method based on depth image and implementing device thereof | |
| KR20170006355A (en) | Method of motion vector and feature vector based fake face detection and apparatus for the same | |
| CN104881662A (en) | Single-image pedestrian detection method | |
| CN105335725A (en) | Gait identification identity authentication method based on feature fusion | |
| CN106682641A (en) | Pedestrian identification method based on image with FHOG- LBPH feature | |
| CN103218605A (en) | Quick eye locating method based on integral projection and edge detection | |
| CN105096307A (en) | Method for detecting objects in paired stereo images |
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: 20170531 |