CN105320917B - A kind of pedestrian detection and tracking based on head-shoulder contour and BP neural network - Google Patents
A kind of pedestrian detection and tracking based on head-shoulder contour and BP neural network Download PDFInfo
- Publication number
- CN105320917B CN105320917B CN201410302238.XA CN201410302238A CN105320917B CN 105320917 B CN105320917 B CN 105320917B CN 201410302238 A CN201410302238 A CN 201410302238A CN 105320917 B CN105320917 B CN 105320917B
- Authority
- CN
- China
- Prior art keywords
- head
- shoulder
- human
- point
- target
- 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.)
- Expired - Fee Related
Links
Landscapes
- Image Analysis (AREA)
Abstract
The present invention proposes a kind of pedestrian detection and tracking based on head-shoulder contour and BP neural network.First, the movement human target in video sequence is extracted using adaptive mixed Gaussian context update algorithm, and the levels of precision of background estimating is improved by changing the Studying factors of mixed Gauss model;Secondly, it uses Canny operators to go out the initial profile of original object for template extraction, and average drifting Mean shift algorithms is combined to carry out profile cluster to obtain more complete human body contour outline;Again, it in conjunction with head and shoulder the ratio of width to height of human body, establishes head-shoulder contour model and extracts head-shoulder contour feature vector, input BP neural network, cluster out multiple human head and shoulder models, carry out human bioequivalence;Finally, using particle filter to the pedestrian target that identifies into line trace.It judges by accident and misjudges caused by the present invention overcomes imperfect due to identification target, improve the accuracy rate of pedestrian target identification, while reducing calculation amount.
Description
Technical field
The invention belongs to moving object detection and tracking technical fields, and in particular to one kind is based on head-shoulder contour and BP nerves
The pedestrian detection and tracking of network.
Background technology
The detection of human body target, recognition and tracking are one of the research hotspot problems in Computer Vision Recognition field,
Order of accuarcy affects being smoothed out for the follow-up works such as target following, Activity recognition and analysis.
Human bioequivalence judges whether moving target is human body target by the information characteristics of acquisition.N.Dalal et al. will be whole
A human body as an identification model, by calculate the model HOG (Histograms of Oriented Gradients,
Histograms of oriented gradients) feature, and SVM (Support-Vector Machines, support vector machines) graders are combined to realize
Human bioequivalence;Kuno people etc. analyzes target shape using projection histogram, distinguishes people and inhuman;Nicolaou etc. utilizes standard
Square and artificial neural network ANN (Artificial Neural Networks) identify human body target.These methods it is common
Feature is, is all complete for the target of human testing and identification.However, since human body is non-rigid object, during exercise
Posture it is uncertain, and complete human body is easy to be blocked by external object, these can all influence above method human bioequivalence
Accuracy.
Invention content
The present invention proposes that a kind of pedestrian detection and tracking based on head-shoulder contour and BP neural network, this method are not necessarily to
According to complete human body as identification target, it is only necessary to be trained identification human body, algorithm meter to the head-shoulder contour model of foundation
Calculation amount is smaller.
The pedestrian detection based on head-shoulder contour and BP neural network that in order to solve the above technical problem, the present invention provides a kind of
And tracking, include the following steps:
Step 1:The moving target in video sequence is examined using adaptive mixed Gaussian context update algorithm
It surveys, obtains the residual plot of moving target;
Step 2:Canny operator edge detections are carried out to moving target residual plot, extract the rough profile of moving target;
Rough profile is clustered using average drifting Mean shift algorithms, in conjunction with moving target residual plot, reservation belongs to human body
Class, and the human body parts clustered out are added in rough profile, obtain objective contour bianry image;
Step 3:According to human head and shoulder wide high proportion and objective contour bianry image, human head and shoulder skeleton pattern is established;
The feature vector for extracting human head and shoulder skeleton pattern, using the feature vector of human head and shoulder skeleton pattern as the defeated of BP neural network
Enter value, establish the correspondence of human head and shoulder Outline Feature Vector and moving target, clusters out multiple human head and shoulder skeleton patterns,
Human bioequivalence is carried out, movement human target is obtained and moves non-human target;
Step 4:The movement human target that step 3 is identified using particle filter algorithm carry out real time kinematics target with
Track.
Compared with prior art, the present invention its remarkable advantage is, (1) is based on ADAPTIVE MIXED Gaussian Background more new algorithm
Moving target recognition algorithm can adapt to background at any time slowly varying, there is preferable detection to the scene of minor change
Effect;(2) edge detection method that Canny operators are combined with average drifting Mean shift algorithms is used to extract moving target
Profile, without that, as identification target, only the head-shoulder contour of detection target need to be used as human bioequivalence mould according to complete human body
Type reduces the calculation amount of algorithm;(3) particle filter is used to be detected the tracking of human body target, it is contemplated that target
Complexity, it is preferable for the tracking effect of the human body target of the self-movement in the camera supervised video sequence of separate unit.
Description of the drawings
Fig. 1 is the pedestrian detection and tracking flow chart the present invention is based on head-shoulder contour and BP neural network.
Fig. 2 is human head and shoulder proportionate relationship schematic diagram in invention.
Fig. 3 is the human head and shoulder skeleton pattern schematic diagram established in invention.
Fig. 4 is that the present invention tests the 300th frame original image in used video sequence.
Fig. 5 is to carry out moving target inspection using adaptive mixed Gaussian context update algorithm to Fig. 2 in present invention experiment
The residual plot obtained after survey.
Fig. 6 is the flow chart of moving target contours extract in invention experiment.
Fig. 7 is human bioequivalence design sketch in invention experiment.
Fig. 8 is invention experiment movement human target following result figure.
Specific implementation mode
As shown in Figure 1, the present invention is based on head-shoulder contour and the pedestrian detection and tracking of BP neural network, including it is following
Step:
Step 1:The moving target in video sequence is examined using adaptive mixed Gaussian context update algorithm
It surveys, obtains the residual plot of moving target.
The adaptive mixed Gaussian context update algorithm refer to document (Bhandarkar, S.M., Fujiyoshi,
Patil,R.S.,“An efficient background updating scheme for real-time traffic
monitoring,”The7th International IEEE Conference:Intelligent Transportation
Systems,859-864(2004).)。
Step 2:Canny operator edge detections are carried out to moving target residual plot, extract the rough profile of moving target;
Rough profile is clustered using average drifting Mean shift algorithms, in conjunction with moving target residual plot, reservation belongs to human body
Class, and the human body parts clustered out are added in rough profile, to obtain more complete human body contour outline, this is more complete
Human body contour outline be bianry image, that is, obtain objective contour bianry image.
This step first goes out the rough profile of moving target using Canny operators as the template extraction of edge detection.But
The high-low threshold value parameter of Canny operators is by being manually set, and to different scenes, high-low threshold value does not have adaptivity;In addition,
Canny operators inevitably extract the edge of background image, or even can will should belong to the part misidentification of movement human
To be background, human body head-and-shoulder area does not plan a successor after causing edge detection.Therefore, the present invention uses average drifting Mean shift
Algorithm clusters image, and the rough profile of moving target is supplemented and corrected.Average drifting Mean shift algorithms
Essence be to calculate the offset mean value m of sampled point xh(x), offset mean value mh(x) shown in calculating such as formula (1),
Wherein, xiFor ith sample point, | | (x-xi)h-1||2=| | (x-xi)||2HiFor Mahalanobis
(Mahalanobis) distance, g (x) are kernel function;Sampled point x is moved into mh(x) distance obtains point x', and is new with x'
Starting point continues to move to, until meeting certain iterated conditional.
The specific calculating process that rough profile is clustered using average drifting Mean shift algorithms described in this step
For:
Step 2.1:Select the tomography part of human head and shoulder in the rough profile of moving target after edge detection as cluster
Region S can select human body head to keep the outline portion that cluster goes out more complete and coherent when selecting cluster areas S
Relatively larger region is as cluster areas S where shoulder tomography part.Initial ranging area is arbitrarily selected in cluster areas S
Domain, the initial search area are using point O as border circular areas that the center of circle, radius are bandwidth h;
Step 2.2:The drift mean value m of sampled point x in circle O is calculated according to formula (1)h(x), mh(x) density at place should be greater than
Density at the O of the center of circle;
Step 2.3:Calculate center of circle O and drift mean value mh(x) difference, as mean shift vectors Mh(x), i.e. Mh(x)=mh
(x)-x, the mean shift vectors always point towards the increased direction of pixel probability density in cluster areas S;
Step 2.4:By mean shift vectors Mh(x) compared with the threshold epsilon of setting, if | | Mh(x) | | < ε are set up, then are changed
In generation, terminates, and sampled point x is the point clustered out, and the region that these points clustered out are formed is the class clustered out;If | | Mh(x)
| | < ε are invalid, then the drift mean value m that will be found outh(x) it is assigned to center of circle O, returns to step 2.2, until | | Mh(x) | | <
ε is set up, iteration stopping.
The average drifting Mean shift algorithms refer to document (Cheng, Y., Z., " Mean Shift, mode
seeking,and clustering”IEEE Trans on Pattern Analysis and Machine
Intelligence,17(8):790-799(1995).)。
Step 3:According to human head and shoulder wide high proportion and objective contour bianry image, human head and shoulder skeleton pattern is established,
The feature vector for extracting human head and shoulder skeleton pattern, using the feature vector of human head and shoulder skeleton pattern as the defeated of BP neural network
Enter value, establish the correspondence of human head and shoulder Outline Feature Vector and moving target, clusters out multiple human head and shoulder skeleton patterns,
Human bioequivalence is carried out, corresponding movement human target is obtained and moves non-human target.
It is described to establish human head and shoulder skeleton pattern, that is, the parameter of human head and shoulder skeleton pattern is obtained, as shown in Fig. 2, including:
Head width Hw, the height H of head the widest part to the crownh, the width N of neckw, the height N of neck to the crownh;The width of shoulder
Sw, the height S of shoulder to the crownh.Obtain above-mentioned parameter method be:
The objective contour bianry image that step 2 obtains is projected into every trade, drop shadow curve is carried out smoothly, to obtain capable throwing
Shadow histogram, by the pixel value deposit row projection array Line of respective coordinates in row projection histogram;Simultaneously to objective contour two
It is worth image to project into ranks, drop shadow curve is carried out smooth, obtain corresponding row projection histogram, and by row projection histogram pair
The pixel value of coordinate is answered to be stored in row projection array Row.
Scan line projects the point A that first pixel value is 255 in obtained row projection array Line, array Line successively
The position on the corresponding objective contour bianry image crown.It is continued to scan on by A points, until finding first pixel value in array Line
It is neck width N for the corresponding coordinate of point B, point B of minimum minw, this time point B respective columns projection array Row midpoint B''s
Coordinate is the height N of neckh.The corresponding seat of point C, point C that a pixel value is maximum max is found between point A and point B
Mark is head width Hw, the coordinate of this time point C respective columns projection array Row midpoint C' is the height H on headh.And human body shoulder
The width S in portionwGeneral and human body width is equal.It is gained knowledge according to human body, the height S of general shoulders of human bodyhFor head width Hw's
2.5~3 times, present embodiment takes 2.5 times, i.e. Sh=2.5Hw.It is possible thereby to establish the head-shoulder contour of human body according to above-mentioned parameter
Model, as shown in Figure 3.
The present invention uses the 7 invariant moments group bp=[M of Hu1,M2,M3,M4,M5,M6,M7] spy as human head and shoulder profile
Sign vector.Shown in the calculating of Hu squares such as formula (2), wherein ηpq(p, q=0,1,2,3) is normalization central moment.For the ease of meter
It calculates and compares, present embodiment will respectively take absolute value in formula (2) and take half power, will be less than 0.00001
Value be approximately 0, then 7 obtained vector is human head and shoulder Outline Feature Vector.
M1=η20+η02
M2=(η20-η02)2+4η11 2
M3=(η30-3η12)2+(3η21-η03)2
M4=(η30+η12)2+(η21+η03)2
M5=(η30-3η12)(η30+η12)[(η30+η12)2-3(η21+η03)2]+(3η21-η03)(η21+η03) (2)
[3(η30+η12)2-(η21+η03)2]
M6=(η20-η02)[(η30+η02)2-(η21+η03)2]+4η11(η30-η12)(η21+η03)
M7=(3 η21-η02)(η30+η12)[(η30+η12)2-3(η21+η03)]-(η30-3η12)(η21+η03)
[3(η30+η12)2-(η21+η02)2]
The BP neural network technology refers to document (NICOLAOUCA, EGBERTAL, LACHERRC, etc., " Human
shape recognition using the method of moment and artificial neural networks,”
IJCNN99.International Joint Conference on Washington DC:IEEE Computer
Society,3147-3151(1999).)。
Step 4:Real-time moving target is carried out to the movement human target that step 3 identifies using particle filter algorithm
Tracking obtains the position coordinates and its movement locus of moving target central point, i.e., to human body target into line trace.
The particle filter refers to document (human body target tracking [J] the computers of Ran Wang Ran based on particle filter
Using with software .vol.25, no.12,2008.).
Beneficial effects of the present invention can be further illustrated by following experiment:
The embodiment of the present invention is using Matlab2012b as experiment porch, and the training sample of BP neural network is from NICTAP pedestrian
It is obtained in database, test sample is the video sequence of BASLER-CCD acquisitions, amounts to 920 frames, has 4~6 in every frame image
Moving target, including human body target and inhuman target, video resolution are 660 × 492.Fig. 4 show in the video sequence
300 frame original images.
According to step 1 of the present invention, using adaptive mixed Gaussian context update algorithm to the video sequence of acquisition
It is handled, detects that the residual plot of moving target, residual plot are as shown in Figure 5.
Below by taking the white human body target of the rightmost side in original image shown in Fig. 4 as an example, step 2 is obtained more complete
The process of human body contour outline illustrate.The process as shown in fig. 6, first with Canny operators in residual plot shown in Fig. 5 most
The white human body Objective extraction on right side goes out the rough profile of moving target;Then, poly- using average drifting Mean shift algorithms
Class goes out the tomography part in head and shoulder, and obtains the profile of head and shoulder tomography part, that is, clusters profile.By the cluster of head and shoulder tomography part
Profile is added in the rough profile of edge detection, obtains more complete human body contour outline, completes the integrity profile of moving target
Extraction.
According to described in step 3, in conjunction with the head and shoulder wide high proportion of human body shown in Fig. 2, human head and shoulder skeleton pattern, root are established
Feature vector bp=[the M of human head and shoulder profile are extracted according to formula (2)1,M2,M3,M4,M5,M6,M7], as BP nerve nets
The input of network.The training of neural network is carried out using pedestrian's database of NICTAP.Training sample is divided into two classes:Human body and inhuman
Body.It is 7 to input the number of plies, corresponding 7 human body head-shoulder contour feature vector bp=[M1,M2,M3,M4,M5,M6,M7], hidden layer is according to warp
Formula M=2N+1 settings are tested, are 15, wherein N is the input number of plies, and the output number of plies is 1.
After the training for completing neural network, the moving target in the video sequence of acquisition is tested, human body is completed and knows
Not.Fig. 7 is part human body recognition effect figure.Wherein, 1 indicate that the target identification is human body, 2 indicate that the target identification is inhuman
Body.
According to described in step 4, the movement human target that step 3 identifies is transported in real time using particle filter algorithm
Tracking of maneuvering target.Fig. 8 is movement human target following result figure.Wherein, frame expression be identified as human body and carry out moving target with
Track, cross indicate the center position of moving target.
This experiment proposes the method for the present invention and N.Dalal et al. to realize human bioequivalence based on HOG features and SVM classifier
Method compare, include the comparison of recognition rate and accuracy, comparison result is as shown in table 1.As it can be seen from table 1
The present invention reduces calculation amount while ensure that accuracy.
The comparison of table 1 inventive algorithm and HOG+SVM algorithms
Algorithm | Recognition rate | Accuracy |
HOG+SVM algorithms | 1.1fps | 87.7% |
Inventive algorithm | 2.9fps | 85.6% |
Claims (2)
1. a kind of pedestrian detection and tracking based on head-shoulder contour and BP neural network, which is characterized in that including following step
Suddenly:
Step 1:The moving target in video sequence is detected using adaptive mixed Gaussian context update algorithm, is obtained
Obtain the residual plot of moving target;
Step 2:Canny operator edge detections are carried out to moving target residual plot, extract the rough profile of moving target;Using
Average drifting Mean shift algorithms cluster rough profile, in conjunction with moving target residual plot, retain the class for belonging to human body,
And the human body parts clustered out are added in rough profile, obtain objective contour bianry image;
Step 3:According to human head and shoulder wide high proportion and objective contour bianry image, human head and shoulder skeleton pattern is established;Extraction
The feature vector of human head and shoulder skeleton pattern, using the feature vector of human head and shoulder skeleton pattern as the input of BP neural network
Value, establishes the correspondence of human head and shoulder Outline Feature Vector and moving target, clusters out multiple human head and shoulder skeleton patterns, into
Row human bioequivalence obtains movement human target and moves non-human target;
Step 4:Real-time moving target tracking is carried out to the movement human target that step 3 identifies using particle filter algorithm;
Used described in step 2 the process that average drifting Mean shift algorithms cluster rough profile for:
Step 2.1:Select the tomography part of human head and shoulder in the rough profile of moving target as cluster areas, in cluster areas
Arbitrary selection initial search area, which is using point O as the center of circle, the border circular areas of radius h;
Step 2.2:The drift mean value m of sampled point x in circle O is calculated according to formula (1)h(x),
In formula (1), xiFor ith sample point, | | (x-xi)h-1||2=| | (x-xi)||2HiFor Mahalanobis generalised distance, g
(x) it is kernel function;
Step 2.3:Calculate center of circle O and drift mean value mh(x) difference obtains mean shift vectors Mh(x);
Step 2.4:By mean shift vectors Mh(x) compared with the threshold epsilon of setting, if | | Mh(x) | | < ε are set up, then sampled point
X is the point clustered out;If | | Mh(x) | | < ε are invalid, then the drift mean value m that will be found outh(x) it is assigned to center of circle O, return is held
Row step 2.2, until | | Mh(x) | | < ε are set up.
2. the pedestrian detection as described in claim 1 based on head-shoulder contour and BP neural network and tracking, which is characterized in that
The process that human head and shoulder skeleton pattern is established described in step 3 is:
The objective contour bianry image that step 2 obtains is projected into every trade, drop shadow curve is carried out smoothly, it is straight to obtain row projection
Fang Tu, by the pixel value deposit row projection array of respective coordinates in row projection histogram;Simultaneously to objective contour bianry image into
Ranks project, and are carried out to drop shadow curve smooth, obtain corresponding row projection histogram, and by row projection histogram respective coordinates
Pixel value deposit row projection array;Scan line projects obtained row and projects array successively, will first pixel value be wherein 255
Point A position of the coordinate position as the crown;It is continued to scan on by A points, first pixel is found until being expert in projection array
Value is the point B of minimum, using the corresponding coordinates of point B as neck width;The coordinate of point B respective columns projection array midpoint B' is made
For neck height;The point C that pixel value is maximum is found between point A and point B, using the corresponding coordinates of point C as head width;
Using the coordinate of point C respective columns projection array midpoint C' as height of head;Head is obtained according to height of head and human body proportion relationship
Portion's width;
Using seven of Hu not feature vectors of the bending moment as human head and shoulder profile in step 3, the calculation of seven not bending moments
As shown in formula (2),
In formula (2), ηpq(p, q=0,1,2,3) is normalization central moment.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410302238.XA CN105320917B (en) | 2014-06-27 | 2014-06-27 | A kind of pedestrian detection and tracking based on head-shoulder contour and BP neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410302238.XA CN105320917B (en) | 2014-06-27 | 2014-06-27 | A kind of pedestrian detection and tracking based on head-shoulder contour and BP neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105320917A CN105320917A (en) | 2016-02-10 |
CN105320917B true CN105320917B (en) | 2018-09-04 |
Family
ID=55248280
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410302238.XA Expired - Fee Related CN105320917B (en) | 2014-06-27 | 2014-06-27 | A kind of pedestrian detection and tracking based on head-shoulder contour and BP neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105320917B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107329131A (en) * | 2017-08-08 | 2017-11-07 | 电子科技大学 | A kind of radar dim target detection tracking of utilization particle filter |
Families Citing this family (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105631484A (en) * | 2016-03-10 | 2016-06-01 | 江苏信息职业技术学院 | Electronic countermeasure radiation source identification method based on equivalent fuzzy neural network |
CN107730534B (en) * | 2016-08-09 | 2020-10-23 | 深圳光启合众科技有限公司 | Target object tracking method and device |
CN106803083B (en) * | 2017-02-04 | 2021-03-19 | 北京旷视科技有限公司 | Pedestrian detection method and device |
CN106934380A (en) * | 2017-03-19 | 2017-07-07 | 北京工业大学 | A kind of indoor pedestrian detection and tracking based on HOG and MeanShift algorithms |
CN107580279A (en) * | 2017-08-10 | 2018-01-12 | 深圳腾芈技术服务有限公司 | Vehicle whistle control method, device and computer-readable recording medium |
CN108241849B (en) * | 2017-08-28 | 2021-09-07 | 北方工业大学 | Human body interaction action recognition method based on video |
CN108830152B (en) * | 2018-05-07 | 2020-12-29 | 北京红云智胜科技有限公司 | Pedestrian detection method and system combining deep learning network and artificial features |
CN109446895B (en) * | 2018-09-18 | 2022-04-08 | 中国汽车技术研究中心有限公司 | Pedestrian identification method based on human head features |
CN110070074B (en) * | 2019-05-07 | 2022-06-14 | 安徽工业大学 | Method for constructing pedestrian detection model |
CN112765284B (en) * | 2021-01-21 | 2024-06-21 | 广州羊城通有限公司 | Method and device for determining relevant places of users |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102214309A (en) * | 2011-06-15 | 2011-10-12 | 北京工业大学 | Special human body recognition method based on head and shoulder model |
CN103632146A (en) * | 2013-12-05 | 2014-03-12 | 南京理工大学 | Head-shoulder distance based human body detection method |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR100951890B1 (en) * | 2008-01-25 | 2010-04-12 | 성균관대학교산학협력단 | Method for simultaneous recognition and pose estimation of object using in-situ monitoring |
-
2014
- 2014-06-27 CN CN201410302238.XA patent/CN105320917B/en not_active Expired - Fee Related
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102214309A (en) * | 2011-06-15 | 2011-10-12 | 北京工业大学 | Special human body recognition method based on head and shoulder model |
CN103632146A (en) * | 2013-12-05 | 2014-03-12 | 南京理工大学 | Head-shoulder distance based human body detection method |
Non-Patent Citations (1)
Title |
---|
基于智能视频监控的人流量统计***研究;时升云;《中国优秀硕士学位论文全文数据库》;20130315;第21-24页、35-39页 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107329131A (en) * | 2017-08-08 | 2017-11-07 | 电子科技大学 | A kind of radar dim target detection tracking of utilization particle filter |
CN107329131B (en) * | 2017-08-08 | 2020-08-28 | 电子科技大学 | Radar weak target detection tracking method using particle filtering |
Also Published As
Publication number | Publication date |
---|---|
CN105320917A (en) | 2016-02-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105320917B (en) | A kind of pedestrian detection and tracking based on head-shoulder contour and BP neural network | |
Sun et al. | Research on the hand gesture recognition based on deep learning | |
CN104268583B (en) | Pedestrian re-recognition method and system based on color area features | |
CN103942577B (en) | Based on the personal identification method for establishing sample database and composite character certainly in video monitoring | |
CN108288033B (en) | A kind of safety cap detection method based on random fern fusion multiple features | |
CN102799901B (en) | Method for multi-angle face detection | |
CN107368778A (en) | Method for catching, device and the storage device of human face expression | |
CN107330371A (en) | Acquisition methods, device and the storage device of the countenance of 3D facial models | |
Ibraheem et al. | Comparative study of skin color based segmentation techniques | |
CN103049751A (en) | Improved weighting region matching high-altitude video pedestrian recognizing method | |
CN111079518B (en) | Ground-falling abnormal behavior identification method based on law enforcement and case handling area scene | |
CN107292252A (en) | A kind of personal identification method of autonomous learning | |
CN106570490B (en) | A kind of pedestrian's method for real time tracking based on quick clustering | |
CN103310194A (en) | Method for detecting head and shoulders of pedestrian in video based on overhead pixel gradient direction | |
Zhang et al. | A swarm intelligence based searching strategy for articulated 3D human body tracking | |
CN106204594A (en) | A kind of direction detection method of dispersivity moving object based on video image | |
CN108537181A (en) | A kind of gait recognition method based on the study of big spacing depth measure | |
CN109271932A (en) | Pedestrian based on color-match recognition methods again | |
CN106204223A (en) | Image processing method and system | |
CN105825168A (en) | Golden snub-nosed monkey face detection and tracking algorithm based on S-TLD | |
CN106845456A (en) | A kind of method of falling over of human body monitoring in video monitoring system | |
Jian-zheng et al. | Head movement recognition based on LK algorithm and Gentleboost | |
CN109815887B (en) | Multi-agent cooperation-based face image classification method under complex illumination | |
CN111626107A (en) | Human-shaped contour analysis and extraction method oriented to smart home scene | |
Głowacki et al. | Reconstructing evolving tree structures in time lapse sequences by enforcing time-consistency |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20180904 Termination date: 20200627 |