CN109871771A - A kind of method and system of the automatic detection human body based on single-view videos - Google Patents
A kind of method and system of the automatic detection human body based on single-view videos Download PDFInfo
- Publication number
- CN109871771A CN109871771A CN201910051948.2A CN201910051948A CN109871771A CN 109871771 A CN109871771 A CN 109871771A CN 201910051948 A CN201910051948 A CN 201910051948A CN 109871771 A CN109871771 A CN 109871771A
- Authority
- CN
- China
- Prior art keywords
- human
- picture
- human body
- training
- module
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Landscapes
- Image Analysis (AREA)
Abstract
The method and system of the present invention relates to a kind of automatic detection human body based on single-view videos, method includes the following steps: S1: human testing training, in sport video, the region detection of all human bodies will be come out and is labeled in a picture, S2: human bioequivalence training, in sport video, by adjacent video frames by the same frame choose body mark out come, while by different human bodies mark out come；S3: characteristics of human body is obtained using human bioequivalence algorithm model and human body tracking algorithm model and is gathered, several crowds in picture are detected using human testing algorithm model；S4: characteristics of human body is gathered and is compared with several crowds detected in picture, calculate the similarity of human body in picture, the system includes human detection module, human bioequivalence module, algorithm training module and human tracking module, the present invention detects and calculates the similarity of crowd in picture by human testing algorithm model and human bioequivalence algorithm model, searches the target body of frame choosing.
Description
Technical field
The present invention relates to computer vision fields, and in particular to a kind of side of the automatic detection human body based on single-view videos
Method and system.
Background technique
Very big variation often occurs in a very short period of time for position of human body and form in Sports Video.?
In some professional motion races (such as U.S.'s NBA league matches), the shooting seat in the plane of multi-point, multi-angle can be set up, sportsman
Dress also specification (having number plate, feature is obvious) has evaded position of human body form in this way and has mutated and be stranded to visual identity bring
It is difficult.Being laid with camera equally to set up multimachine position without the race of image of Buddha profession in public sports buildings (is on the one hand because multimachine position is more
Angle laying cost performance is low, is on the other hand that the construction conditions of venue itself have limitation), the dressing of public sport people is also compared
Relatively arbitrarily, irregular to follow, so, Visual identification technology and method in professional race can not be applicable in this project.
Summary of the invention
The method and system of the object of the present invention is to provide a kind of automatic detection human body based on single-view videos solve single
Automatic identification searches for the problem of some personage in the Sports Video of visual angle.
The purpose of the present invention is be achieved through the following technical solutions:
A method of the automatic detection human body based on single-view videos, method includes the following steps: S1: human testing
Training, the region detection of all human bodies in sport video, will be come out and be labeled in a picture, repetition training obtains
To human testing algorithm model；
S2: human bioequivalence training in sport video, target body frame is elected, by will be same in adjacent video frames
One frame choose body mark out come, mark out as model training positive sample, while by different human bodies as model training
Negative sample inputs two pictures, compares human body on picture and whether human body is consistent, the similarity result of two pictures of return,
Repetition training obtains human bioequivalence algorithm model；
S3: certain picture center selects the human body for wishing to search in video to user in video, is calculated using human bioequivalence
Method model and human body tracking algorithm model obtain characteristics of human body's set, using human testing algorithm model to each frame figure of video
Piece carries out human testing, detects several crowds in picture；
S4: characteristics of human body is gathered and is compared with several crowds detected in picture, human bioequivalence algorithm mould is utilized
Type calculates the similarity of human body in picture.
In preferred embodiments, in the step S2, model training positive sample includes the front, side, back of human body
Face, rotation and light change the labeled data in a variety of situations.
In preferred embodiments, the track algorithm model in the step S3 is all in 5s after choosing picture
It is tracked out in picture by the personage that frame selects, the form different to personage is compared, and selects the characteristic set of the personage, instead
Refreshment is practiced, and track algorithm model is obtained.
In preferred embodiments, is carried out by person detecting, is detected for each frame picture of video in the step S4
All people's body region is then based on human bioequivalence algorithm model and character features set, carries out identification meter to all people's body
It calculates, provides the similarity of each human region, every picture is chosen the highest human region of similarity and returned as a result.
In preferred embodiments, a similarity threshold is arranged to all several crowds detected in the step S4
Value, is shown to user more than the threshold value, the result as people search.
A kind of system of the automatic detection human body based on single-view videos, including human detection module, human bioequivalence module,
Algorithm training module and human tracking module, the human detection module are used to detect all people's body region in picture, and will
Detection data is sent into algorithm training module and obtains human testing algorithm model；The human bioequivalence module is for comparing on picture
Whether human body and frame body of choosing are consistent, and provide similarity, and data feeding algorithm training module is obtained human bioequivalence algorithm
Model；The human tracking module is used to track the personage selected in all pictures chosen after picture in 5s by frame, and will count
Track algorithm model is obtained according to algorithm training module is sent into.
The invention has the benefit that detecting by human testing algorithm model and human bioequivalence algorithm model and calculating figure
The similarity of crowd in piece searches the target body of frame choosing.
Detailed description of the invention
Below according to attached drawing, invention is further described in detail.
Fig. 1 is the flow chart of the method for the automatic detection human body described in the embodiment of the present invention based on single-view videos；
Fig. 2 is the flow chart of the system of the automatic detection human body described in the embodiment of the present invention based on single-view videos.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.The present invention being usually described and illustrated herein in the accompanying drawings is implemented
The component of example can be arranged and be designed with a variety of different configurations.Therefore, below to the reality of the invention provided in the accompanying drawings
The detailed description for applying example is not intended to limit the range of claimed invention, but is merely representative of selected implementation of the invention
Example.Based on the embodiments of the present invention, obtained by those of ordinary skill in the art without making creative efforts
Every other embodiment, shall fall within the protection scope of the present invention.
The embodiment of the present invention is described below in detail, examples of the embodiments are shown in the accompanying drawings, wherein from beginning to end
Same or similar label indicates same or similar element or element with the same or similar functions.Below with reference to attached
The embodiment of figure description is exemplary, it is intended to is used to explain the present invention, and is not considered as limiting the invention.
The present invention is further illustrated with specific embodiment below with reference to accompanying drawings.
As shown in Figure 1, a kind of method of automatic detection human body based on single-view videos of the embodiment of the present invention, this method
The following steps are included:
S1: human testing training is gone forward side by side in sport video by the region detection of all human bodies is come out in a picture
Rower note, the region of all human bodies and each position of human body, such as head, drive, repetition training, construct a human testing
CNN neural network model is trained using the human body labeled data collection under moving scene, obtains being suitable for sports scene
Under human testing algorithm model；Data in human testing algorithm model are that have in video to extract in a large amount of pictures of each frame
Out, physical characteristic data can be used is extracted by gradient orientation histogram (the H O G) method that Dalal et al. is proposed, should
Method extracts the gradient orientation histogram of image local area by the description of gradient and directional spreding situation to topography
As detection feature.Characteristic value may include resemblance value (such as size, profile, color) and motion characteristic value (such as walk, run,
It stands, bends over, squat down).
S2: human bioequivalence training in sport video, target body frame is elected, by will be same in adjacent video frames
One frame choose body mark out come, mark out as model training positive sample, while by different human bodies as model training
Negative sample constructs a CNN neural network model, inputs two pictures, whether the human body compared on picture is consistent with human body, returns
The similarity result for returning two pictures is trained CNN neural network, repetition training by the positive and negative sample data of mark,
Obtain human bioequivalence algorithm model.
Since the state of target body is all different in every frame, characteristic may be not suitable for continuing conduct at first
Characteristic parameter, system are also required to constantly learn to update, and are applicable in more situations.
S3: certain picture center selects the human body for wishing to search in video to user in video, is calculated using human bioequivalence
Method model and human body tracking algorithm model obtain characteristics of human body's set, using human testing algorithm model to each frame figure of video
Piece carries out human testing, detects several crowds in picture；
S4: characteristics of human body is gathered and is compared with several crowds detected in picture, human bioequivalence algorithm mould is utilized
Type calculates the similarity of human body in picture.Find the target body of frame choosing.
In the step S2, model training positive sample includes that front, side, the back side, rotation and the light variation of human body are more
Labeled data in the case of kind, with the adaptability of lift scheme.
Track algorithm model in the step S3 is the personage selected in all pictures after choosing picture in 5s by frame
It is tracked out, the form different to personage is compared, and selects the characteristic set of the personage, and repetition training obtains tracking and calculates
Method model.
To each frame picture of video in the step S4, person detecting is carried out, detects all people's body region, then
Based on human bioequivalence algorithm model and character features set, identification calculating is carried out to all people's body, provides each human region
Similarity, every picture chooses the highest human region of similarity and returns as a result.
A similarity threshold is arranged to all several crowds detected in the step S4, more than the display of the threshold value
Result to user, as people search.
As shown in Fig. 2, a kind of system of the automatic detection human body based on single-view videos, including human detection module, people
Body identification module, algorithm training module and human tracking module, the human detection module is for detecting all people in picture
Body region, and will test data feeding algorithm training module and obtain human testing algorithm model；The human bioequivalence module is used for
Whether consistent compare human body on picture and frame body of choosing, and provides similarity, and data feeding algorithm training module is obtained
Human bioequivalence algorithm model；The human tracking module is used for the personage selected in all pictures chosen after picture in 5s by frame
Tracking, and data feeding algorithm training module is obtained into track algorithm model.
Finally, it should be noted that above-described embodiments are merely to illustrate the technical scheme, rather than to it
Limitation；Although the present invention is described in detail referring to the foregoing embodiments, those skilled in the art should understand that:
It can still modify to technical solution documented by previous embodiment, or to part of or all technical features into
Row equivalent replacement；And these modifications or substitutions, it does not separate the essence of the corresponding technical solution various embodiments of the present invention technical side
The range of case.
Claims (6)
1. a kind of method of the automatic detection human body based on single-view videos, it is characterised in that: method includes the following steps:
S1: the region detection of all human bodies will be come out rower of going forward side by side in a picture in sport video by human testing training
Note, repetition training obtain human testing algorithm model；
S2: human bioequivalence training in sport video, target body frame is elected, by will be same in adjacent video frames
Frame choose body mark out come, mark out as model training positive sample, while by different human bodies as the negative sample of model training
This, inputs two pictures, compares human body on picture and whether human body is consistent, the similarity result of two pictures of return, repeatedly
Training, obtains human bioequivalence algorithm model；
S3: certain picture center selects the human body for wishing to search in video to user in video, utilizes human bioequivalence algorithm mould
Type and human body tracking algorithm model obtain characteristics of human body's set, using human testing algorithm model to each frame picture of video into
Row human testing detects several crowds in picture；
S4: characteristics of human body is gathered and is compared with several crowds detected in picture, human bioequivalence algorithm model meter is utilized
The similarity of human body in nomogram piece.
2. the method for the automatic detection human body according to claim 1 based on single-view videos, it is characterised in that: the step
In rapid S2, model training positive sample includes that front, side, the back side, rotation and the light of human body change the mark in a variety of situations
Data.
3. the method for the automatic detection human body according to claim 1 based on single-view videos, it is characterised in that: the step
Track algorithm model in rapid S3 is that will choose in all pictures after picture in 5s to be tracked out by the personage that frame selects, to personage
Different forms are compared, and select the characteristic set of the personage, and repetition training obtains track algorithm model.
4. the method for the automatic detection human body according to claim 1 based on single-view videos, it is characterised in that: the step
To each frame picture of video in rapid S4, person detecting is carried out, detects all people's body region, is then based on human bioequivalence calculation
Method model and character features set, carry out identification calculating to all people's body, provide the similarity of each human region, every figure
Piece is chosen the highest human region of similarity and is returned as a result.
5. the method for the automatic detection human body according to claim 1 based on single-view videos, it is characterised in that: the step
A similarity threshold is arranged to all several crowds detected in rapid S4, user is shown to more than the threshold value, as people
The result of object search.
6. a kind of system of the automatic detection human body based on single-view videos, it is characterised in that: including human detection module, human body
Identification module, algorithm training module and human tracking module, the human detection module is for detecting all people's body in picture
Region, and will test data feeding algorithm training module and obtain human testing algorithm model；The human bioequivalence module be used for than
It is whether consistent compared with the body of choosing of human body and the frame on picture, and similarity is provided, and data feeding algorithm training module is obtained into people
Body recognizer model；The human tracking module be used for the personage selected in all pictures chosen after picture in 5s by frame with
Track, and data feeding algorithm training module is obtained into track algorithm model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910051948.2A CN109871771A (en) | 2019-01-21 | 2019-01-21 | A kind of method and system of the automatic detection human body based on single-view videos |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910051948.2A CN109871771A (en) | 2019-01-21 | 2019-01-21 | A kind of method and system of the automatic detection human body based on single-view videos |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109871771A true CN109871771A (en) | 2019-06-11 |
Family
ID=66917860
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910051948.2A Pending CN109871771A (en) | 2019-01-21 | 2019-01-21 | A kind of method and system of the automatic detection human body based on single-view videos |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109871771A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112784680A (en) * | 2020-12-23 | 2021-05-11 | 中国人民大学 | Method and system for locking dense contacts in crowded place |
| CN113705510A (en) * | 2021-09-02 | 2021-11-26 | 广州市奥威亚电子科技有限公司 | Target identification tracking method, device, equipment and storage medium |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1072014A1 (en) * | 1998-04-13 | 2001-01-31 | Eyematic Interfaces, Inc. | Face recognition from video images |
| CN103177247A (en) * | 2013-04-09 | 2013-06-26 | 天津大学 | Target detection method fused with multi-angle information |
| CN104392223A (en) * | 2014-12-05 | 2015-03-04 | 青岛科技大学 | Method for recognizing human postures in two-dimensional video images |
| CN105913275A (en) * | 2016-03-25 | 2016-08-31 | 哈尔滨工业大学深圳研究生院 | Clothes advertisement putting method and system based on video leading role identification |
| CN106295524A (en) * | 2016-08-01 | 2017-01-04 | 马平 | A kind of human motion recognition method of view-based access control model word bag |
| CN106384360A (en) * | 2016-09-22 | 2017-02-08 | 北京舜裔科技有限公司 | Interactive video creation method |
| CN108345837A (en) * | 2018-01-17 | 2018-07-31 | 浙江大学 | A kind of pedestrian's recognition methods again based on the study of human region alignmentization feature representation |
| CN108960078A (en) * | 2018-06-12 | 2018-12-07 | 温州大学 | A method of based on monocular vision, from action recognition identity |
-
2019
- 2019-01-21 CN CN201910051948.2A patent/CN109871771A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1072014A1 (en) * | 1998-04-13 | 2001-01-31 | Eyematic Interfaces, Inc. | Face recognition from video images |
| CN103177247A (en) * | 2013-04-09 | 2013-06-26 | 天津大学 | Target detection method fused with multi-angle information |
| CN104392223A (en) * | 2014-12-05 | 2015-03-04 | 青岛科技大学 | Method for recognizing human postures in two-dimensional video images |
| CN105913275A (en) * | 2016-03-25 | 2016-08-31 | 哈尔滨工业大学深圳研究生院 | Clothes advertisement putting method and system based on video leading role identification |
| CN106295524A (en) * | 2016-08-01 | 2017-01-04 | 马平 | A kind of human motion recognition method of view-based access control model word bag |
| CN106384360A (en) * | 2016-09-22 | 2017-02-08 | 北京舜裔科技有限公司 | Interactive video creation method |
| CN108345837A (en) * | 2018-01-17 | 2018-07-31 | 浙江大学 | A kind of pedestrian's recognition methods again based on the study of human region alignmentization feature representation |
| CN108960078A (en) * | 2018-06-12 | 2018-12-07 | 温州大学 | A method of based on monocular vision, from action recognition identity |
Non-Patent Citations (2)
| Title |
|---|
| 常永鑫: "基于视觉认知的目标识别技术研究", 《中国博士学位论文全文数据库 信息科技辑》 * |
| 郭丽纯: "基于视频序列的人体动作识别研究", 《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 |
|---|---|---|---|---|
| CN112784680A (en) * | 2020-12-23 | 2021-05-11 | 中国人民大学 | Method and system for locking dense contacts in crowded place |
| CN112784680B (en) * | 2020-12-23 | 2024-02-02 | 中国人民大学 | Method and system for locking dense contactors in people stream dense places |
| CN113705510A (en) * | 2021-09-02 | 2021-11-26 | 广州市奥威亚电子科技有限公司 | Target identification tracking method, device, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106650620B (en) | A kind of target person identification method for tracing using unmanned plane monitoring | |
| McNally et al. | Golfdb: A video database for golf swing sequencing | |
| Sundberg et al. | Occlusion boundary detection and figure/ground assignment from optical flow | |
| CN106845357B (en) | A kind of video human face detection and recognition methods based on multichannel network | |
| CN109064484B (en) | Crowd movement behavior identification method based on fusion of subgroup component division and momentum characteristics | |
| Kondori et al. | 3D head pose estimation using the Kinect | |
| CN109584213B (en) | Multi-target number selection tracking method | |
| CN107862240B (en) | Multi-camera collaborative face tracking method | |
| KR101035055B1 (en) | System and method of tracking object using different kind camera | |
| CN105868574B (en) | A kind of optimization method of camera track human faces and wisdom health monitor system based on video | |
| Denman et al. | Soft-biometrics: unconstrained authentication in a surveillance environment | |
| Tulyakov et al. | Robust real-time extreme head pose estimation | |
| CN109583373B (en) | Pedestrian re-identification implementation method | |
| CN104298968A (en) | Target tracking method under complex scene based on superpixel | |
| JP7422456B2 (en) | Image processing device, image processing method and program | |
| CN104517095A (en) | Head division method based on depth image | |
| CN109871771A (en) | A kind of method and system of the automatic detection human body based on single-view videos | |
| CN111881749A (en) | Bidirectional pedestrian flow statistical method based on RGB-D multi-modal data | |
| Gao et al. | Non-goal scene analysis for soccer video | |
| Sha et al. | Swimmer localization from a moving camera | |
| KR102434397B1 (en) | Real time multi-object tracking device and method by using global motion | |
| CN107103301B (en) | Method and system for matching discriminant color regions with maximum video target space-time stability | |
| Shah et al. | Multi-camera handoff for person re-identification | |
| Singh et al. | A multi-gait dataset for human recognition under occlusion scenario | |
| CN108520208A (en) | Localize face recognition method |
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: 20190611 |