CN107832728A - A kind of judge based on video makes a phone call Activity recognition method - Google Patents
A kind of judge based on video makes a phone call Activity recognition method Download PDFInfo
- Publication number
- CN107832728A CN107832728A CN201711168589.6A CN201711168589A CN107832728A CN 107832728 A CN107832728 A CN 107832728A CN 201711168589 A CN201711168589 A CN 201711168589A CN 107832728 A CN107832728 A CN 107832728A
- Authority
- CN
- China
- Prior art keywords
- phone call
- personage
- head
- shoulder
- recognition method
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
Abstract
Made a phone call Activity recognition method the invention provides a kind of judge based on video, specifically include the personage's head and shoulder region identified using the textural characteristics of personage's head and shoulder in image in area-of-interest；Personage's head and shoulder region is identified using deep learning disaggregated model, judges present image with the presence or absence of behavior of making a phone call；Energy of making a phone call is calculated according to judged result, and finally judged.The present invention utilizes high definition court's trial video, is automatically found judge position, then carries out making a phone call to identify.Due to having used the thought of energy accumulation, this method has accuracy height, loss low and the characteristic of strong robustness.
Description
Technical field
The invention belongs to video detection technology field, know more particularly, to a kind of judge based on video behavior of making a phone call
Other method.
Background technology
During opening a court session, it is both that itself professional not strong performance causes to the seriousness of court again that judge, which takes phone,
Destroy, it could even be possible to influenceing the fairness of trial by phone.Judge, which should strictly observe, " forbids people's electricity of being combatted smuggling during opening a court session
Court's discipline of words ".With the progress of court's electronization upgrading, present most courts record court's trial using picture pick-up device
Process.After having high definition court's trial video, carry out making a phone call to be identified as in order to possible for people using computer generation.
The content of the invention
In view of this, a kind of Activity recognition method the present invention is directed to propose judge based on video makes a phone call, to utilize
High definition court's trial video, is automatically found judge position, carries out making a phone call to identify.
To reach above-mentioned purpose, the technical proposal of the invention is realized in this way：
A kind of judge based on video makes a phone call Activity recognition method, specifically comprises the following steps：
(1) personage's head and shoulder region in image in area-of-interest is identified using the textural characteristics of personage's head and shoulder；
(2) personage's head and shoulder region is identified using deep learning disaggregated model, judges that present image whether there is and beat
Phone behavior；
(3) energy of making a phone call is calculated according to judged result, and finally judged.
Further, specifically included in the step (1), personage's head and shoulder Sample Storehouse is established, according to known personage's head and shoulder sample
This HOG features, establish personage's head and shoulder model using DPM algorithms, then use head and shoulder in the area-of-interest of video to be detected
Model is detected, and obtains the head and shoulder scope of personage in video.
Further, specifically included in the step (2), foundation is made a phone call and Sample Storehouse of not making a phone call, utilizes Sample Storehouse
Disaggregated model training is carried out using deep learning, the head and shoulder range image of personage is identified by deep learning, judges to work as
Preceding image is with the presence or absence of behavior of making a phone call.
Further, specifically included in the step (3), energy scores are carried out to human target according to multiple judged result
Whether accumulation, then finally making a phone call to judge according to energy scores to personage.
Further, the utilization DPM algorithms are established personage's head and shoulder model and specifically included
(11) image to be checked is carried out piecemeal by DPM algorithms first, then to each piece of extraction HOG feature；
(12) each block of calculating obtains the weight of each block to the distance at center, and the HOG of these Weights is special
Sign is trained and detected by SVM cascade grader, finally gives personage's head and shoulder regional location.
Relative to prior art, a kind of judge based on video of the present invention Activity recognition method of making a phone call has
Following advantage：The present invention utilizes high definition court's trial video, is automatically found judge position, then carries out making a phone call to identify；The present invention by
In the thought for having used energy accumulation, have that accuracy is high, loss is low and the characteristic of strong robustness.
Brief description of the drawings
The accompanying drawing for forming the part of the present invention is used for providing a further understanding of the present invention, schematic reality of the invention
Apply example and its illustrate to be used to explain the present invention, do not form inappropriate limitation of the present invention.In the accompanying drawings：
Fig. 1 is that a kind of judge based on video described in the embodiment of the present invention makes a phone call Activity recognition method flow diagram；
Fig. 2 is the HOG feature extraction block diagrams described in the embodiment of the present invention；
Fig. 3 is the division schematic diagram of the rectangle HOG blocks described in the embodiment of the present invention；
Fig. 4 is the DPM flow charts described in the embodiment of the present invention.
Embodiment
It should be noted that in the case where not conflicting, the feature in embodiment and embodiment in the present invention can phase
Mutually combination.
Describe the present invention in detail below with reference to the accompanying drawings and in conjunction with the embodiments.
Made a phone call Activity recognition method the invention provides a kind of judge based on video, overall flow figure of the invention
As shown in figure 1, key step is as follows：
First, personage's head and shoulder region is found in the picture, carries out the training of personage's head and shoulder model in the present invention using DPM algorithms
And detection.
Image to be checked is carried out piecemeal by 1.DPM algorithms first, then to each piece of extraction HOG feature, gradient direction Nogata
Figure (Histogram of Oriented Gradient, HOG) is characterized in a kind of normal in computer vision and image processing field
Character description method, it is as shown in Figure 2 that it extracts characteristic procedure：
(1) by image normalization：
Wherein max and min is respectively the minimum and maximum pixel value of image, and x is normalization preceding pixel value, and x* is normalization
Pixel value afterwards.
(2) gradient of normalized image is calculated using first differential.
I, j are the coordinate of current pixel,Represent gradient, each corresponding 8 Grad of point.
(3) the direction weight projection based on gradient magnitude.As shown in figure 3, one block (Block) of the division of rectangle HOG blocks
It is made up of 4 units (Cell), a unit is made up of 64 pixels.The distribution of 0~180 degree is divided into 9 deflections
Degree, each orientation angle scope can correspond to a Bin, and the gradient obtained in step (2) is independently done into ladder in each Cell
Directional statistics are spent, obtain a 9 dimensional vector v (x1, x2 ..., x9), wherein x1 correspondingly projects to terraced in 0~20 degree of direction bin
The number of degree, for each Block, obtain 36 dimensional characteristics vector vs (x1, x2 ..., x36).
(4) HOG characteristic vectors normalize.
Wherein ε is the constant value of a very little, and it is 0 to avoid denominator.
(5) characteristic vector generation final HOG.Finally we can obtain an a*b*c data by above step
The vector of composition, wherein a represent the number of Bin in each Cell, and b represents the number of Cell in each Block, and c represents block
Middle Block number.
2. each block is calculated again to the distance at center to obtain the weight of each block, the HOG of these Weights is special
Sign is trained and detected by SVM cascade grader.DPM algorithm flows are as shown in Figure 4.Finally give personage's head and shoulder region position
Put.
2nd, personage's head and shoulder area image is passed into deep learning disaggregated model to be identified, and carries out energy accumulation, obtained
Final result.Main the following steps：
1. personage's head and shoulder area image is passed into deep learning disaggregated model to be identified.
Three kinds are shared to each frame recognition result：Left hand is made a phone call CALL_LEFT), the right hand makes a phone call (CALL_
RIGHT), do not make a phone call (NO_CALL).The recognition result of confidence score (Conf) is obtained by deep learning model, in order that
Result it is more credible, we add threshold value (Thresh), work as Conf>It is considered that result is effective when Thresh.
2. the result obtained in upper step is constantly accumulated into acquisition energy scores, and final alarm knot is obtained according to energy scores
Fruit.
(1) two field picture is often handled, score (Score) fixation of target region (TargetRegion) subtracts one
The result of individual value (MinusValue), this operation and deep learning is unrelated, and limitation Score minimum values are 0, i.e. Score=Max
(Score-MinusValue, 0).
(2) two field picture is often handled, when result is CALL_LEFT or CALL_RIGHT, target region
(TargetRegion) score (Score) increases a value (AddValue), and limitation Score maximums are 100, i.e. Score
=Min (Score+AddValue, 100).
(3) final result whether alarm mainly according to judge energy scores (Score) whether exceed setting score threshold
(AlarmThresh).Each frame can all judge whether Score is more than AlarmThresh, work as Score>AlarmThresh is
During Ture, it is believed that when the region where Area Objects has behavior of making a phone call and starts to alarm.Once Score>AlarmThresh is
False, alarm eliminate.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all essences in the present invention
God any modification, equivalent substitution and improvements made etc., should be included in the scope of the protection with principle.
Claims (5)
- A kind of Activity recognition method 1. judge based on video makes a phone call, it is characterised in that：Specifically comprise the following steps：(1) personage's head and shoulder region in image in area-of-interest is identified using the textural characteristics of personage's head and shoulder；(2) personage's head and shoulder region is identified using deep learning disaggregated model, judges that present image whether there is and make a phone call Behavior；(3) energy of making a phone call is calculated according to judged result, and finally judged.
- The Activity recognition method 2. a kind of judge based on video according to claim 1 makes a phone call, it is characterised in that：Institute State and specifically included in step (1), establish personage's head and shoulder Sample Storehouse, according to the HOG features of known personage's head and shoulder sample, utilize DPM Algorithm establishes personage's head and shoulder model, is then detected in the area-of-interest of video to be detected using head and shoulder model, depending on The head and shoulder scope of personage in frequency.
- The Activity recognition method 3. a kind of judge based on video according to claim 2 makes a phone call, it is characterised in that：Institute State and specifically included in step (2), foundation is made a phone call and Sample Storehouse of not making a phone call, is classified using Sample Storehouse using deep learning Model training, the head and shoulder range image of personage is identified by deep learning, judges that present image whether there is and make a phone call Behavior.
- The Activity recognition method 4. a kind of judge based on video according to claim 3 makes a phone call, it is characterised in that：Institute State and specifically included in step (3), energy scores accumulation is carried out to human target according to multiple judged result, then according to can measure Divide and finally whether personage is being made a phone call to judge.
- The Activity recognition method 5. a kind of judge based on video according to claim 2 makes a phone call, it is characterised in that：Institute State and establish personage's head and shoulder model using DPM algorithms and specifically include(11) image to be checked is carried out piecemeal by DPM algorithms first, then to each piece of extraction HOG feature；(12) each block of calculating obtains the weight of each block to the distance at center, and the HOG features of these Weights are led to Cross SVM cascade grader to be trained and detect, finally give personage's head and shoulder regional location.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711168589.6A CN107832728A (en) | 2017-11-21 | 2017-11-21 | A kind of judge based on video makes a phone call Activity recognition method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711168589.6A CN107832728A (en) | 2017-11-21 | 2017-11-21 | A kind of judge based on video makes a phone call Activity recognition method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107832728A true CN107832728A (en) | 2018-03-23 |
Family
ID=61653168
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711168589.6A Pending CN107832728A (en) | 2017-11-21 | 2017-11-21 | A kind of judge based on video makes a phone call Activity recognition method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107832728A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109684990A (en) * | 2018-12-20 | 2019-04-26 | 天津天地伟业信息系统集成有限公司 | A kind of behavioral value method of making a phone call based on video |
| CN111178436A (en) * | 2019-12-30 | 2020-05-19 | 深圳信息职业技术学院 | Data processing method and device, computer equipment and storage medium |
| CN114241521A (en) * | 2021-12-13 | 2022-03-25 | 北京华夏电通科技股份有限公司 | Method, device and equipment for identifying court trial video picture normal area |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140270367A1 (en) * | 2013-03-14 | 2014-09-18 | Nec Laboratories America, Inc. | Selective Max-Pooling For Object Detection |
| CN105488490A (en) * | 2015-12-23 | 2016-04-13 | 天津天地伟业数码科技有限公司 | Judge dressing detection method based on video |
| CN105868690A (en) * | 2016-03-11 | 2016-08-17 | 博康智能信息技术有限公司 | Method and apparatus for identifying mobile phone use behavior of driver |
| CN105913022A (en) * | 2016-04-11 | 2016-08-31 | 深圳市飞瑞斯科技有限公司 | Handheld calling state determining method and handheld calling state determining system based on video analysis |
| CN106056071A (en) * | 2016-05-30 | 2016-10-26 | 北京智芯原动科技有限公司 | Method and device for detection of driver' behavior of making call |
-
2017
- 2017-11-21 CN CN201711168589.6A patent/CN107832728A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140270367A1 (en) * | 2013-03-14 | 2014-09-18 | Nec Laboratories America, Inc. | Selective Max-Pooling For Object Detection |
| CN105488490A (en) * | 2015-12-23 | 2016-04-13 | 天津天地伟业数码科技有限公司 | Judge dressing detection method based on video |
| CN105868690A (en) * | 2016-03-11 | 2016-08-17 | 博康智能信息技术有限公司 | Method and apparatus for identifying mobile phone use behavior of driver |
| CN105913022A (en) * | 2016-04-11 | 2016-08-31 | 深圳市飞瑞斯科技有限公司 | Handheld calling state determining method and handheld calling state determining system based on video analysis |
| CN106056071A (en) * | 2016-05-30 | 2016-10-26 | 北京智芯原动科技有限公司 | Method and device for detection of driver' behavior of making call |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109684990A (en) * | 2018-12-20 | 2019-04-26 | 天津天地伟业信息系统集成有限公司 | A kind of behavioral value method of making a phone call based on video |
| CN109684990B (en) * | 2018-12-20 | 2023-05-30 | 天津天地伟业信息系统集成有限公司 | Video-based phone call behavior detection method |
| CN111178436A (en) * | 2019-12-30 | 2020-05-19 | 深圳信息职业技术学院 | Data processing method and device, computer equipment and storage medium |
| CN114241521A (en) * | 2021-12-13 | 2022-03-25 | 北京华夏电通科技股份有限公司 | Method, device and equipment for identifying court trial video picture normal area |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105631455B (en) | A kind of image subject extracting method and system | |
| CN106127148B (en) | A kind of escalator passenger's anomaly detection method based on machine vision | |
| CN105046245B (en) | Video human face method of determination and evaluation | |
| CN105488490A (en) | Judge dressing detection method based on video | |
| CN106682641A (en) | Pedestrian identification method based on image with FHOG- LBPH feature | |
| CN105844242A (en) | Method for detecting skin color in image | |
| CN101620673A (en) | Robust face detecting and tracking method | |
| CN107832728A (en) | A kind of judge based on video makes a phone call Activity recognition method | |
| CN106446779A (en) | Method and apparatus for identifying identity | |
| TWI415032B (en) | Object tracking method | |
| CN107358152A (en) | Living body identification method and system | |
| CN111401310B (en) | Kitchen sanitation safety supervision and management method based on artificial intelligence | |
| CN106934353A (en) | A kind of method of the recognition of face and active tracing for robot of supporting parents | |
| CN109215010A (en) | A kind of method and robot face identification system of picture quality judgement | |
| CN107392089A (en) | Eyebrow movement detection method and device and living body identification method and system | |
| CN106881716A (en) | Human body follower method and system based on 3D cameras robot | |
| CN107330370A (en) | Forehead wrinkle action detection method and device and living body identification method and system | |
| CN105184771A (en) | Adaptive moving target detection system and detection method | |
| CN109684919A (en) | A kind of shuttlecock faulty serving method of discrimination based on machine vision | |
| CN112101260B (en) | Method, device, equipment and storage medium for identifying safety belt of operator | |
| CN107358155A (en) | Method and device for detecting ghost face action and method and system for recognizing living body | |
| CN108549872B (en) | Visual attention fusion method suitable for quality evaluation of redirected image | |
| CN108108740B (en) | Active millimeter wave human body image gender identification method | |
| US11263762B2 (en) | Image recognition method and apparatus, and computer-readable storage medium | |
| CN107358154A (en) | Head motion detection method and device and living body identification method and system |
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: 20180323 |