CN109117771A - Incident of violence detection system and method in a kind of image based on anchor node - Google Patents
Incident of violence detection system and method in a kind of image based on anchor node Download PDFInfo
- Publication number
- CN109117771A CN109117771A CN201810863149.0A CN201810863149A CN109117771A CN 109117771 A CN109117771 A CN 109117771A CN 201810863149 A CN201810863149 A CN 201810863149A CN 109117771 A CN109117771 A CN 109117771A
- Authority
- CN
- China
- Prior art keywords
- sample
- image
- anchor node
- violence
- data set
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/44—Event detection
Abstract
The present invention relates to video image processing technologies, it discloses incident of violence detection methods in a kind of image based on anchor node, automatically analysis detection is carried out to the act of violence in video flowing, improves the real-time and accuracy of detection, and adapt to a variety of different monitoring environment.Method includes the following steps: a. is used to obtain sample characteristics according to all feature vectors of image data set, and anchor node is sought according to the sample characteristics；B. according to the corresponding Hash codes of sample in the similarity calculation data set between anchor node and the sample of image data set；C. for carrying out image prediction using trained SVM model using the corresponding Hash codes of each sample of image data set as input, to judge in image with the presence or absence of incident of violence.In addition, being suitable for plurality of application scenes the invention also discloses incident of violence detection system in a kind of image based on anchor node.
Description
Technical field
The present invention relates to video image processing technologies, and in particular to incident of violence detection in a kind of image based on anchor node
System and method.
Background technique
With a large amount of uses of monitoring system, there is fulminant growth in video data.The effect of monitoring system is to carry out
Target detection and unusual checking.With the sharp increase of data, the mode that traditional dependence manually monitors is more tired
Difficulty, and inefficiency.Therefore, become hot spot by the research of the monitoring system of artificial intelligence.Wherein, for the violence row of people
For detection be very important research direction.
Due to act of violence movement compared with simply run, slip a line for than it is complicated very much, so how to carry out act of violence
Detection is also the difficult point of correlative study.Currently, traditional act of violence detection mainly uses the side based on artificial design features
Method also has certain defects although recognition accuracy is higher, such as: real-time effect cannot be reached, the shadow vulnerable to noise
Ring etc..
Summary of the invention
The technical problems to be solved by the present invention are: proposing incident of violence detection system in a kind of image based on anchor node
And method, analysis detection is carried out to the act of violence in video flowing automatically, improves the real-time and accuracy of detection, and adapt to
A variety of different monitoring environment.
The technical proposal adopted by the invention to solve the above technical problems is that: violence thing in a kind of image based on anchor node
Part detection system, comprising: anchor node extraction module, Hash codes computing module and image prediction module；
The anchor node extraction module, for obtaining sample characteristics, and root according to all feature vectors of image data set
Anchor node is sought according to the sample characteristics；
The Hash codes computing module, for according to the similarity calculation number between anchor node and the sample of image data set
According to the corresponding Hash codes of the sample of concentration；
Described image prediction module, for using the corresponding Hash codes of each sample of image data set as input, using instruction
The SVM model perfected carries out image prediction, to judge in image with the presence or absence of incident of violence.
As advanced optimizing, the anchor node extraction module is specifically used for, and is existed using human skeleton feature extracting method
The joint point data of personage in image is obtained on image data set, then the artis of adjacent two interframe is relatively obtained into each joint
The displacement vector of point finds out anchor node using clustering algorithm as sample characteristics.
As advanced optimizing, the Hash codes computing module is specifically used for, and calculates anchor node and image data set is all
Sample between approximate similarity matrix, then similarity matrix is simulated by approximate similarity matrix, and seek replacing phase
Like the companion matrix of degree matrix, the characteristic value and feature vector of companion matrix are then calculated, finally according to the characteristic value and spy
Levy the corresponding Hash codes of sample in vector calculating data set.
As advanced optimizing, the trained SVM model of the use carries out image prediction, to judge whether deposit in image
In incident of violence, specifically include:
The corresponding Hash codes of each sample of image data set to be predicted are input in trained SVM model, if sample
This output result is 1, then is determined as violence frame, if the output result of sample is 0, is determined as nonviolent frame, last basis
Whether violence frame proportion exceeds certain proportion to determine whether there are incidents of violence.
As advanced optimizing, the anchor node extraction module, Hash codes computing module and image prediction module are by portion
It is deployed on the same server；Alternatively, the anchor node extraction module, Hash codes computing module are deployed in the same server
On, and described image prediction module is deployed on another server.
In addition, being based on above system, the invention also provides incident of violence detection sides in a kind of image based on anchor node
Method comprising following steps:
A. it for obtaining sample characteristics according to all feature vectors of image data set, and is sought according to the sample characteristics
Anchor node；
B. according to the corresponding Kazakhstan of sample in the similarity calculation data set between anchor node and the sample of image data set
Uncommon code；
C. for using the corresponding Hash codes of each sample of image data set as input, using trained SVM model into
Row image prediction, to judge in image with the presence or absence of incident of violence.
As advanced optimizing, step a is specifically included:
The joint point data of personage in image is obtained on image data set using human skeleton feature extracting method, then will
The artis of adjacent two interframe relatively obtains the displacement vector of each artis as sample characteristics, finds out anchor using clustering algorithm
Node.
As advanced optimizing, step b is specifically included:
The approximate similarity matrix between anchor node and all sample of image data set is calculated, then passes through approximate similarity
Matrix simulates similarity matrix, and seeks the companion matrix instead of similarity matrix, then calculates the characteristic value of companion matrix
And feature vector, the corresponding Hash codes of sample in data set are finally calculated according to the characteristic value and feature vector.
As advanced optimizing, in step c, the method for training SVM model is:
Using training image data set as image data set, step a and b are executed, using SVM model to the Hash codes of acquisition
It is trained, obtains trained SVM model.
As advanced optimizing, step c is specifically included:
The corresponding Hash codes of each sample of image data set to be predicted are input in trained SVM model, if sample
This output result is 1, then is determined as violence frame, if the output result of sample is 0, is determined as nonviolent frame, last basis
Whether violence frame proportion exceeds certain proportion to determine whether there are incidents of violence.
The beneficial effects of the present invention are:
By the image to existing point of good class carry out processing be extracted as sample after, by the side of these sample machine learning
Method trains model, then is predicted after processing to image to be predicted using model.
For a variety of different types of monitoring places, this method can automatically analyze video flowing, if detection
To the generation of act of violence, immediately triggers warning device and alarm, administrative staff can then be notified at the first time, be gone forward side by side
The corresponding processing of row, has real-time；And when handling image, using the anchor node that finds out as the feature distinguished to
Amount, to improve the high efficiency and accuracy of act of violence prediction.
Detailed description of the invention
Fig. 1 is the incident of violence detection method flow chart in the present invention；
Fig. 2 is that the incident of violence in embodiment 1 detects non real-time distributed treatment schematic diagram；
Fig. 3 is that the act of violence in embodiment 2 detects real-time centralized processing schematic diagram.
Specific embodiment
The present invention is directed to propose incident of violence detection system and method in a kind of image based on anchor node, automatically to video
Act of violence in stream carries out analysis detection, improves the real-time and accuracy of detection, and adapt to a variety of different monitoring rings
Border.In the present invention, by the image to existing point of good class carry out processing be extracted as sample after, by these sample engineerings
The method of habit trains model, then is predicted after processing to image to be detected using model.
For ease of understanding, the technical terms being likely to occur in the present invention are explained first:
1. anchor node: be usually according to certain clustering algorithm or other asked in many ordinary nodes than more prominent features
Out, the presence the same as beacon.
2.SVM: i.e. support vector machines is the learning model for having supervision, it is therefore an objective to find one in machine learning field
A function can most Interval datas by different labels open.
3. similarity matrix: similarity matrix is the square matrix of a n*n, is housed between n sample to indicate every two
Similarity between sample distance in other words.
4. Hash codes: Hash codes are obtained by hash algorithm as a result, Hash codes are not that completely uniquely, it is a kind of
Algorithm, allow the object of same class according to oneself different feature as far as possible have different Hash codes, but do not indicate different pairs
As Hash codes are entirely different.
Incident of violence detection system in the image based on anchor node in the present invention, comprising: anchor node extraction module, Hash
Three parts of code computing module and image prediction module, wherein the function of various pieces is as follows:
The anchor node extraction module, for obtaining sample characteristics, and root according to all feature vectors of image data set
Anchor node is sought according to the sample characteristics；Specifically: first with human body framework characteristic extracting method on raw image data collection
The joint point data of personage in image is obtained, then the displacement that the artis of adjacent two interframe relatively obtains each artis is sweared
Amount, then using artis displacement vector as sample characteristics, a small amount of anchor node is found out using clustering algorithm.
The Hash codes computing module, for according to the similarity calculation number between anchor node and the sample of image data set
According to the corresponding Hash codes of the sample of concentration；Specifically: firstly, calculating the two according to all samples of anchor node and data set
Between approximate similarity matrix Z；Then, each sample can be estimated by calculating after having obtained approximate similarity matrix Z
Between similarity matrix A；It can be obtained by the feature vector and characteristic value that calculate the Laplacian Matrix L of similarity matrix A again
To Hash codes.But it in order to reduce space consuming, is changed to calculate the minor matrix M of a substitution, then calculate the characteristic value of the matrix
And feature vector, finally calculate input of the corresponding Hash codes of each sample of the data set as next step.
Described image prediction module, for using the corresponding Hash codes of each sample of image data set as input, using instruction
The SVM model perfected carries out image prediction, to judge in image with the presence or absence of incident of violence.Specifically: by image to be predicted
The corresponding Hash codes of each sample of data set are input in trained SVM model, if the output result of sample is 1, are determined
It is determined as nonviolent frame if the output result of sample is 0 for violence frame, finally whether is exceeded according to violence frame proportion
Certain proportion is to determine whether there are incidents of violence.
Referring to Fig. 1, incident of violence detection system in the image based on anchor node in the present invention the following steps are included:
1. inputting training image data set；
In this step, training image data set derives from categorized good image pattern, including the view containing incident of violence
Frequency image and video image without incident of violence.
2. obtaining the joint point data of human skeleton feature；
In this step, the artis of personage in image is obtained on image data set using human skeleton feature extracting method
Data.
3. it is poor that more adjacent two frame obtains frame；
In this step, then the displacement vector that the artis of adjacent two interframe is relatively obtained each artis is special as sample
Sign.
4. calculating anchor node；
In this step, anchor node is found out using clustering algorithm according to sample characteristics.
5. obtaining approximate similarity matrix Z；
In this step, it is similar that the approximation between anchor node and all samples of image data set is calculated by distance calculating method
Spend matrix.
6. the minor matrix M of substitution is calculated by approximation similarity matrix Z；
In this step, similarity matrix A is simulated by approximate similarity matrix Z, and is sought instead of similarity matrix A's
Minor matrix M is as companion matrix.
7. calculating characteristic value and feature vector；
In this step, the characteristic value and feature vector of companion matrix M are calculated.
8. calculating Hash codes；
In this step, the corresponding Hash codes of sample in data set are calculated according to the characteristic value and feature vector.
9. training sample obtains SVM model；
In this step, the Hash codes by acquisition are as training SVM model is inputted, to obtain trained SVM model.
10. obtaining the Hash codes of prediction video image；
In this step, video image to be predicted is inputted, executing step 2-10 can be obtained the Hash codes of prediction video image；
11. being predicted using SVM model forecast image sample Hash codes；
In this step, the Hash codes for obtaining prediction video image are input in trained SVM model, thus prognostic chart
As whether frame is violence frame (i.e. containing the frame of act of violence in image)；
12. exporting result according to the accounting of violence frame；
In this step, if the totalframes that violence frame accounts for input picture is more than certain proportion (can be with preset in advance), determine
To have incident of violence in video, otherwise, it is determined that being free of incident of violence.
Embodiment 1:
The scene of the present embodiment description is: the Distributed Application scene handled in real time is not needed, such as the camera shooting in several streets
After head capture image, image transmitting is stored into image file into a host, its corresponding a small amount of camera shooting by the host process
The image of head simultaneously obtains prediction result, while result is temporarily stored in local, every fixed several hours, will temporarily store
Prediction result on uploaded in respective database respectively, and provide inquiry function by Reverse Proxy for client computer
Can, the presence of the imperceptible multiple databases of client computer can directly access the data of a large amount of cameras, as shown in Fig. 2, it has
Steps are as follows for body:
Before the system deployment, on collection network or the acquisition of other approach is largely used to trained image data set, right
The data set arranges, and accurately classifies to all image datas and to every class plus corresponding label, such as a kind of image is sudden and violent
Power event image, another kind of image are non-incident of violence images.After dividing good class, using human body framework characteristic extracting method to all
Image handle while keeping classifying, and using the human synovial point data of obtained each frame as sample, passes through clustering algorithm
Anchor node is acquired in these samples, and anchor node needs long-term keep to use when system operation；Then pass through the step in Fig. 1
Suddenly the corresponding Hash codes of each sample are obtained while still maintaining classification auxiliary moment that is constant, while will being used to ask Hash codes matrix
Battle array separates the convenient Hash codes for calculating forecast sample later of long-term preservation；Finally by Hash codes of all training sets and its right
The classification answered be used as svm model training sample obtain trained svm model, this svm model long-term preservation is used for be
It is predicted when system operation.
After deployment system, each camera capturing video in real time, while the image transmission that will be captured is taken the photograph to this
As on corresponding server.On each server, all store passed through the trained svm model of training dataset with
And anchor node companion matrix, each server upon receiving the image, in order to reduce server stress, are first temporarily stored data
It is every to cross one section of regular time (such as several hours) in local, the video of one of camera accumulation is taken out, is concentrated to the section
Image is handled, and obtains corresponding Hash codes and using svm model to obtained Hash codes by anchor node and companion matrix
It is predicted.
After prediction, the testing result predicted this period and used video streaming are uploaded to well
The corresponding database of book server is stored for a long time for customer inquiries.
It is stored in respectively due to result in different databases, looking into for a Reverse Proxy subscribing client is set
Ask request then into corresponding database access detection as a result, can simultaneously provide such as statistical data convenient function at times.
Embodiment 2:
The scene of the present embodiment description is: the part Code in Hazardous Special Locations for needing to focus on image in real time, due to region one
As smaller camera it is integrally less, it is very high to real-time monitoring incident of violence demand, therefore, it is necessary to these images centralization in real time
Prediction.Each video stream server only connects a small amount of camera, these servers are only responsible for image store and are carried out on a small quantity to it
Processing obtains Hash codes, these Hash codes are sent to the same detection service device after being labeled source, quick by the server
Detection simultaneously be stored in database, client computer can be polled database, give a warning when inquiring incident of violence.
Referring to Fig. 3, the specific steps are as follows:
Before the system deployment, on collection network or the acquisition of other approach is largely used to trained image data set, right
The data set arranges, and accurately classifies to all image datas and to every class plus corresponding label, such as a kind of image is sudden and violent
Power event image, another kind of image are non-incident of violence images.After dividing good class, extracted using human body framework characteristic to all images
Handle while keeping classifying, using the human synovial point data of obtained each frame as sample, by clustering algorithm at this
Anchor node is acquired in a little samples, anchor node needs long-term keep to use when system operation；Then it is obtained by the step in Fig. 1
It takes the corresponding Hash codes of each sample while still maintaining companion matrix point that is constant, while will being used to ask Hash codes matrix of classifying
Separate out the Hash codes for carrying out after long-term preservation is convenient to calculate forecast sample；Finally by Hash codes of all training sets and its corresponding
The training sample that classification is used as svm model obtains trained svm model, this svm model long-term preservation is used for system fortune
It is predicted when row.
After the system deployment, each camera capturing video in real time, while the image transmission that will be captured was to should
On the corresponding server of camera.On each server, anchor node companion matrix is all stored, each server is receiving
After image, in order to reduce predictive server pressure, every excessively a bit of time (such as half a minute), to the video data meter received
Corresponding Hash codes are calculated, data volume these Hash codes much smaller compared to video are only transferred to predictive server down.
Predictive server is connected to all video stream servers in the region, is housed before system operation on the server
With regard to trained svm model, predictive server predicts the Hash codes sequence received using the model, further according to violence
Event frame proportion judges the property of this section of video and is saved in database.Client computer is by ceaselessly carrying out database
Whether poll check there is incident of violence, incident of violence occurs if inquired, alarms.
In above two example scheme, database is to the storage mode of testing result referring to following table:
Incident of violence based on anchor node detects storage table
There are five fields in total for the table, are respectively as follows:
Id, i.e. serial number, the number of the video flowing to label detection；
TStamp, i.e. timestamp, to identify the time for carrying out act of violence detection to video flowing；
CameraId, i.e. camera ID, the number of the camera to identify video flowing source；
HashCode, i.e. Hash codes are used according to the similarity between the anchor node of extraction and the sample of image data set
Hash algorithm carries out calculating acquisition to sample, to the input as prediction model；
Whether isViolent is violence, to identify the incident of violence testing result of video flowing, indicate video flowing for 1
Containing incident of violence, indicate that video flowing is free of incident of violence for 0.
The violence of the video flowing of some corresponding camera is inquired according to cameraId in the above-mentioned storage table of clients poll
Behavioral value is as a result, according to the value of isViolent field to determine whether there are incident of violence, if the value of isViolent field
Incident of violence is indicated for 1, by the corresponding warning device of automatic trigger, before administrator can notify the operator on duty of corresponding area
It is checked and is handled toward corresponding location.
Since above-mentioned storage table has recorded the Hash codes hashCode calculated every time for video flowing and act of violence
Testing result isViolent, our SVM model can extract a large amount of historical data from database and be trained, i.e., not
The training sample of disconnected abundant SVM model, to keep SVM prediction more and more accurate.
Claims (10)
1. incident of violence detection system in a kind of image based on anchor node, which is characterized in that
It include: anchor node extraction module, Hash codes computing module and image prediction module；
The anchor node extraction module, for obtaining sample characteristics according to all feature vectors of image data set, and according to institute
It states sample characteristics and seeks anchor node；
The Hash codes computing module, for according to the similarity calculation data set between anchor node and the sample of image data set
In the corresponding Hash codes of sample；
Described image prediction module, for using the corresponding Hash codes of each sample of image data set as input, using training
SVM model carry out image prediction, to judge in image with the presence or absence of incident of violence.
2. incident of violence detection system in a kind of image based on anchor node as described in claim 1, which is characterized in that described
Anchor node extraction module is specifically used for, and obtains personage in image on image data set using human skeleton feature extracting method
Joint point data, then the displacement vector that the artis of adjacent two interframe is relatively obtained each artis make as sample characteristics
Anchor node is found out with clustering algorithm.
3. incident of violence detection system in a kind of image based on anchor node as described in claim 1, which is characterized in that described
Hash codes computing module is specifically used for, and calculates the approximate similarity matrix between anchor node and all sample of image data set,
Similarity matrix is simulated by approximate similarity matrix again, and seeks the companion matrix instead of similarity matrix, is then calculated
It is corresponding finally to calculate the sample in data set according to the characteristic value and feature vector for the characteristic value and feature vector of companion matrix
Hash codes.
4. incident of violence detection system in a kind of image based on anchor node as described in claim 1, which is characterized in that described
Image prediction is carried out using trained SVM model, to judge to specifically include in image with the presence or absence of incident of violence:
The corresponding Hash codes of each sample of image data set to be predicted are input in trained SVM model, if sample
Exporting result is 1, then is determined as violence frame, if the output result of sample is 0, is determined as nonviolent frame, finally according to violence
Frame proportion whether exceeds certain proportion to determine whether there are incidents of violence.
5. incident of violence detection system in a kind of image based on anchor node as described in claim 1-4 any one, special
Sign is that the anchor node extraction module, Hash codes computing module and image prediction module are deployed in the same server
On；Alternatively, the anchor node extraction module, Hash codes computing module are deployed on the same server, and described image is pre-
Module is surveyed to be deployed on another server.
6. incident of violence detection method in a kind of image based on anchor node, which comprises the following steps:
A. for obtaining sample characteristics according to all feature vectors of image data set, and anchor section is sought according to the sample characteristics
Point；
B. according to the corresponding Hash codes of sample in the similarity calculation data set between anchor node and the sample of image data set；
C. for carrying out figure using trained SVM model using the corresponding Hash codes of each sample of image data set as input
As prediction, to judge in image with the presence or absence of incident of violence.
7. incident of violence detection method in a kind of image based on anchor node as claimed in claim 6, which is characterized in that step
A is specifically included:
The joint point data of personage in image is obtained on image data set using human skeleton feature extracting method, then will be adjacent
The artis of two interframe relatively obtains the displacement vector of each artis as sample characteristics, finds out anchor section using clustering algorithm
Point.
8. incident of violence detection method in a kind of image based on anchor node as claimed in claim 6, which is characterized in that step
B is specifically included:
The approximate similarity matrix between anchor node and all sample of image data set is calculated, then passes through approximate similarity matrix
It simulates similarity matrix, and seeks the companion matrix instead of similarity matrix, then calculate characteristic value and the spy of companion matrix
Vector is levied, the corresponding Hash codes of sample in data set are finally calculated according to the characteristic value and feature vector.
9. incident of violence detection method in a kind of image based on anchor node as claimed in claim 6, which is characterized in that step
In c, the method for training SVM model is:
Using training image data set as image data set, step a and b are executed, is carried out using Hash codes of the SVM model to acquisition
Training, obtains trained SVM model.
10. incident of violence detection method in a kind of image based on anchor node as described in claim 6-9 any one, special
Sign is that step c is specifically included:
The corresponding Hash codes of each sample of image data set to be predicted are input in trained SVM model, if sample
Exporting result is 1, then is determined as violence frame, if the output result of sample is 0, is determined as nonviolent frame, finally according to violence
Frame proportion whether exceeds certain proportion to determine whether there are incidents of violence.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810863149.0A CN109117771B (en) | 2018-08-01 | 2018-08-01 | System and method for detecting violence events in image based on anchor nodes |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810863149.0A CN109117771B (en) | 2018-08-01 | 2018-08-01 | System and method for detecting violence events in image based on anchor nodes |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109117771A true CN109117771A (en) | 2019-01-01 |
| CN109117771B CN109117771B (en) | 2022-05-27 |
Family
ID=64862516
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810863149.0A Active CN109117771B (en) | 2018-08-01 | 2018-08-01 | System and method for detecting violence events in image based on anchor nodes |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109117771B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110147729A (en) * | 2019-04-16 | 2019-08-20 | 深圳壹账通智能科技有限公司 | User emotion recognition methods, device, computer equipment and storage medium |
| CN110363131A (en) * | 2019-07-08 | 2019-10-22 | 上海交通大学 | Anomaly detection method, system and medium based on human skeleton |
| CN110503081A (en) * | 2019-08-30 | 2019-11-26 | 山东师范大学 | Act of violence detection method, system, equipment and medium based on inter-frame difference |
| CN111475674A (en) * | 2020-04-22 | 2020-07-31 | 深圳市艾伯信息科技有限公司 | Deep learning model training data set construction method and system for violent behavior detection |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120197896A1 (en) * | 2008-02-25 | 2012-08-02 | Georgetown University | System and method for detecting, collecting, analyzing, and communicating event-related information |
| WO2013169644A1 (en) * | 2012-05-07 | 2013-11-14 | Google Inc. | Detection of unauthorized content in live multiuser composite streams |
| WO2015106808A1 (en) * | 2014-01-15 | 2015-07-23 | Nec Europe Ltd. | Method and system for crowd detection in an area |
| CN105512289A (en) * | 2015-12-07 | 2016-04-20 | 郑州金惠计算机系统工程有限公司 | Image retrieval method based on deep learning and Hash |
| CN105913559A (en) * | 2016-04-06 | 2016-08-31 | 南京华捷艾米软件科技有限公司 | Motion sensing technique based bank ATM intelligent monitoring method |
| CN106897670A (en) * | 2017-01-19 | 2017-06-27 | 南京邮电大学 | A kind of express delivery violence sorting recognition methods based on computer vision |
| CN107067020A (en) * | 2016-12-30 | 2017-08-18 | 腾讯科技（上海）有限公司 | Image identification method and device |
| CN107273835A (en) * | 2017-06-07 | 2017-10-20 | 南京航空航天大学 | Act of violence intelligent detecting method based on video analysis |
| CN107368820A (en) * | 2017-08-03 | 2017-11-21 | 中国科学院深圳先进技术研究院 | One kind becomes more meticulous gesture identification method, device and equipment |
-
2018
- 2018-08-01 CN CN201810863149.0A patent/CN109117771B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120197896A1 (en) * | 2008-02-25 | 2012-08-02 | Georgetown University | System and method for detecting, collecting, analyzing, and communicating event-related information |
| WO2013169644A1 (en) * | 2012-05-07 | 2013-11-14 | Google Inc. | Detection of unauthorized content in live multiuser composite streams |
| WO2015106808A1 (en) * | 2014-01-15 | 2015-07-23 | Nec Europe Ltd. | Method and system for crowd detection in an area |
| CN105512289A (en) * | 2015-12-07 | 2016-04-20 | 郑州金惠计算机系统工程有限公司 | Image retrieval method based on deep learning and Hash |
| CN105913559A (en) * | 2016-04-06 | 2016-08-31 | 南京华捷艾米软件科技有限公司 | Motion sensing technique based bank ATM intelligent monitoring method |
| CN107067020A (en) * | 2016-12-30 | 2017-08-18 | 腾讯科技（上海）有限公司 | Image identification method and device |
| CN106897670A (en) * | 2017-01-19 | 2017-06-27 | 南京邮电大学 | A kind of express delivery violence sorting recognition methods based on computer vision |
| CN107273835A (en) * | 2017-06-07 | 2017-10-20 | 南京航空航天大学 | Act of violence intelligent detecting method based on video analysis |
| CN107368820A (en) * | 2017-08-03 | 2017-11-21 | 中国科学院深圳先进技术研究院 | One kind becomes more meticulous gesture identification method, device and equipment |
Non-Patent Citations (8)
| Title |
|---|
| HAO, Y等: "Unsupervised t-Distributed Video Hashing and its Deep Hashing Extension", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 * |
| TOBIAS SENST等: "Crowd Violence Detection Using Global Motion-Compensated Lagrangian Features and Scale-Sensitive Video-Level Representation", 《IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY》 * |
| 丁春辉: "基于深度学习的暴力检测及人脸识别方法研究", 《中国博士学位论文全文数据库_信息科技辑》 * |
| 冯伟国: "视频监控中运动对象提取与海量对象快速检索", 《中国博士学位论文全文数据库_信息科技辑》 * |
| 唐文杰: "公安警务领域视频图像侦查关键技术综述", 《中国公共安全》 * |
| 方正涛: "基于视频的孤寡老人室内异常行为检测研究", 《中国优秀硕士学位论文全文数据库_信息科技辑》 * |
| 朱明凌: "面向智能视频监控系统的运动目标检测与异常行为分析", 《中国优秀硕士学位论文全文数据库_信息科技辑》 * |
| 陈龙等: "自助银行服务中暴力异常事件检测", 《计算机科学与探索》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110147729A (en) * | 2019-04-16 | 2019-08-20 | 深圳壹账通智能科技有限公司 | User emotion recognition methods, device, computer equipment and storage medium |
| CN110363131A (en) * | 2019-07-08 | 2019-10-22 | 上海交通大学 | Anomaly detection method, system and medium based on human skeleton |
| CN110503081A (en) * | 2019-08-30 | 2019-11-26 | 山东师范大学 | Act of violence detection method, system, equipment and medium based on inter-frame difference |
| CN111475674A (en) * | 2020-04-22 | 2020-07-31 | 深圳市艾伯信息科技有限公司 | Deep learning model training data set construction method and system for violent behavior detection |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109117771B (en) | 2022-05-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ranjan et al. | Learning to count everything | |
| CN109117771A (en) | Incident of violence detection system and method in a kind of image based on anchor node | |
| CN115272037B (en) | Smart city based on Internet of things regional public security management early warning method and system | |
| CN110796074B (en) | Pedestrian re-identification method based on space-time data fusion | |
| CN104573111A (en) | Method for structured storage and pre-retrieval of pedestrian data in surveillance videos | |
| CN103324919B (en) | Video monitoring system and data processing method thereof based on recognition of face | |
| CN113963315A (en) | Real-time video multi-user behavior recognition method and system in complex scene | |
| CN109902681B (en) | User group relation determining method, device, equipment and storage medium | |
| CN105809711A (en) | Video tracking based pig movement associated big data extracting method and system | |
| CN112488071A (en) | Method, device, electronic equipment and storage medium for extracting pedestrian features | |
| CN111325153B (en) | Student behavior feature intelligent analysis method based on multidimensional data | |
| CN116977937A (en) | Pedestrian re-identification method and system | |
| CN109086717A (en) | Act of violence detection system and method based on human skeleton and motor message feature | |
| CN111898418A (en) | Human body abnormal behavior detection method based on T-TINY-YOLO network | |
| CN108875712A (en) | A kind of act of violence detection system and method based on ViF descriptor | |
| Zhang et al. | A Multiple Instance Learning and Relevance Feedback Framework for Retrieving Abnormal Incidents in Surveillance Videos. | |
| Lee et al. | Deep-Clustering Based Plant Disease Segmentation Network | |
| Pillai et al. | Fine-Tuned EfficientNetB4 Transfer Learning Model for Weather Classification | |
| CN110990634A (en) | Method for searching personnel in video based on face recognition in K12 education stage | |
| Lee | A smart camera network with SVM classifiers for crowd event recognition | |
| CN111191620A (en) | Method for constructing human-object interaction detection data set | |
| CN110738167A (en) | pedestrian identification method based on multi-domain spatial attribute correlation analysis | |
| CN116229518B (en) | Bird species observation method and system based on machine learning | |
| Meena et al. | Aerial imaging based sea lion count using modified U-net architecture | |
| KR102643254B1 (en) | Apparatus and method for calculating shipping information of livestock populations, and shipping information calculating system including the apparatus |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |