CN108446649A - Method and device for alarm - Google Patents
Method and device for alarm Download PDFInfo
- Publication number
- CN108446649A CN108446649A CN201810256338.1A CN201810256338A CN108446649A CN 108446649 A CN108446649 A CN 108446649A CN 201810256338 A CN201810256338 A CN 201810256338A CN 108446649 A CN108446649 A CN 108446649A
- Authority
- CN
- China
- Prior art keywords
- video
- behavior
- history
- feature vector
- obtains
- 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
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the present application discloses the method and device for alarm.One specific implementation mode of this method includes：Image recognition is carried out to the video of acquisition, obtains people information；The video clip of the corresponding personage of people information is imported to Activity recognition model trained in advance, obtains the behavior label of corresponding personage, behavior label is arranged for the behavior of personage to be identified, and for behavior in above-mentioned Activity recognition model；It is non-security in response to above-mentioned behavior label, then sends out alarm signal.This embodiment improves the accuracys of behavior in identification video, can find non-security behavior in time, and send out alarm signal.
Description
Technical field
The invention relates to field of computer technology, and in particular to the method and device for alarm.
Background technology
With the development of science and technology, the application range of electronic equipment is also increasingly wider.Electronic equipment can be arranged at crossing or
The positions such as shop, collection vehicle or pedestrian's video or image.When accident etc. occurs, staff can be adopted by electronic equipment
The video or image of collection make analysis to accident.
Invention content
The purpose of the embodiment of the present application is to propose the method and device for alarm.
In a first aspect, the embodiment of the present application provides a kind of method for alarm, this method includes：To the video of acquisition
Image recognition is carried out, people information is obtained；The video clip of the corresponding personage of people information is imported behavior trained in advance to know
Other model obtains the behavior label of corresponding personage, and above-mentioned Activity recognition model is row for the behavior of personage to be identified
For behavior label is arranged；It is non-security in response to above-mentioned behavior label, then sends out alarm signal.
In some embodiments, the above-mentioned video to acquisition carries out image recognition, obtains people information, including：In response to
There are character images in video, extract character image；Identify the people information of the corresponding personage of character image, above-mentioned people information
Including at least one of following：Gender, height, dressing color.
In some embodiments, above-mentioned Activity recognition model includes convolutional neural networks, Recognition with Recurrent Neural Network and full connection
Layer.
In some embodiments, the above-mentioned video clip by the corresponding personage of people information imports behavior trained in advance and knows
Other model obtains the behavior label of corresponding personage, including：Above-mentioned video clip is input to above-mentioned convolutional neural networks, is obtained
The feature vector of each frame image of above-mentioned video clip, wherein above-mentioned convolutional neural networks are for characterizing video clip and video
Each frame image feature vector between correspondence；The feature vector of each frame image of above-mentioned video clip is input to
Recognition with Recurrent Neural Network is stated, the feature vector of above-mentioned video clip is obtained, wherein above-mentioned Recognition with Recurrent Neural Network is for characterizing piece of video
The feature vector of correspondence between the feature vector and the feature vector of video clip of each frame image of section, video clip is used
Incidence relation between the feature vector of each frame image of characterization video clip；The feature vector of above-mentioned video clip is inputted
To above-mentioned full articulamentum, obtain above-mentioned behavior label, wherein above-mentioned full articulamentum be used to characterize the feature vector of video clip with
Correspondence between behavior label.
In some embodiments, training obtains above-mentioned Activity recognition model as follows：Obtain multiple record someone
The historical behavior label corresponding to each history video in the history video of object behavior and above-mentioned multiple history videos, wherein
Whether historical behavior label is safe for identifying personage's behavior；Using each history video in above-mentioned multiple history videos as defeated
Enter, using the historical behavior label corresponding to each history video in above-mentioned multiple history videos as output, training obtains
State Activity recognition model.
In some embodiments, above-mentioned each history video using in above-mentioned multiple history videos, will be above-mentioned as input
For historical behavior label corresponding to each history video in multiple history videos as output, training obtains above-mentioned Activity recognition
Model, including：Execute following training step：Each history video in above-mentioned multiple history videos is sequentially input to initial row
For identification model, the prediction history behavior label corresponding to each history video in above-mentioned multiple history videos is obtained, it will be upper
State the prediction history behavior label corresponding to each history video in multiple history videos and going through corresponding to the history video
History behavior label is compared, and obtains the recognition accuracy of above-mentioned initial behavior identification model, determines that above-mentioned recognition accuracy is
It is no to be more than default accuracy rate threshold value, if more than above-mentioned default accuracy rate threshold value, then using above-mentioned initial behavior identification model as instruction
Practice the Activity recognition model completed.
In some embodiments, above-mentioned each history video using in above-mentioned multiple history videos, will be above-mentioned as input
For historical behavior label corresponding to each history video in multiple history videos as output, training obtains above-mentioned Activity recognition
Model, including：In response to being not more than above-mentioned default accuracy rate threshold value, the parameter of above-mentioned initial behavior identification model is adjusted, and after
It is continuous to execute above-mentioned training step.
Second aspect, the embodiment of the present application provide a kind of device for alarm, which includes：People information identifies
Unit carries out image recognition for the video to acquisition, obtains people information；Behavior label acquiring unit, for believing personage
The video clip for ceasing corresponding personage imports Activity recognition model trained in advance, obtains the behavior label of corresponding personage, above-mentioned
Behavior label is arranged for the behavior of personage to be identified, and for behavior in Activity recognition model；Alarm Unit, in response to
Above-mentioned behavior label is non-security, then sends out alarm signal.
In some embodiments, above-mentioned people information recognition unit includes：Personal image extraction subelement, in response to
There are character images in video, extract character image；People information identifies subelement, for identification the corresponding personage of character image
People information, above-mentioned people information includes at least one of following：Gender, height, dressing color.
In some embodiments, above-mentioned Activity recognition model includes convolutional neural networks, Recognition with Recurrent Neural Network and full connection
Layer.
In some embodiments, above-mentioned behavior label acquiring unit includes：Image feature vector obtains subelement, and being used for will
Above-mentioned video clip is input to above-mentioned convolutional neural networks, obtains the feature vector of each frame image of above-mentioned video clip, wherein
Correspondence between the feature vector for each frame image that above-mentioned convolutional neural networks are used to characterize video clip and video；Video
Feature vector obtains subelement, for the feature vector of each frame image of above-mentioned video clip to be input to above-mentioned cycle nerve net
Network obtains the feature vector of above-mentioned video clip, wherein above-mentioned Recognition with Recurrent Neural Network is used to characterize each frame image of video clip
Feature vector and the feature vector of video clip between correspondence, the feature vector of video clip is for characterizing piece of video
Incidence relation between the feature vector of each frame image of section；Behavior label obtains subelement, is used for above-mentioned video clip
Feature vector is input to above-mentioned full articulamentum, obtains above-mentioned behavior label, wherein above-mentioned full articulamentum is for characterizing video clip
Feature vector and behavior label between correspondence.
In some embodiments, above-mentioned apparatus includes Activity recognition model training unit, above-mentioned Activity recognition model training
Unit includes：Historical information obtains subelement, multiple record the history video for having personage's behavior for obtaining and above-mentioned multiple goes through
The historical behavior label corresponding to each history video in history video, wherein historical behavior label is for identifying personage's behavior
It is whether safe；Activity recognition model training subelement, for using each history video in above-mentioned multiple history videos as defeated
Enter, using the historical behavior label corresponding to each history video in above-mentioned multiple history videos as output, training obtains
State Activity recognition model.
In some embodiments, above-mentioned Activity recognition model training subelement includes：Activity recognition model training module is used
It is sequentially input to initial behavior identification model in by each history video in above-mentioned multiple history videos, obtains above-mentioned multiple go through
The prediction history behavior label corresponding to each history video in history video, by each history in above-mentioned multiple history videos
Prediction history behavior label corresponding to video is compared with the historical behavior label corresponding to the history video, is obtained above-mentioned
The recognition accuracy of initial behavior identification model, determines whether above-mentioned recognition accuracy is more than default accuracy rate threshold value, if more than
Above-mentioned default accuracy rate threshold value, then the Activity recognition model completed above-mentioned initial behavior identification model as training.
In some embodiments, above-mentioned Activity recognition model training subelement further includes：Parameter adjustment module, for responding
In no more than above-mentioned default accuracy rate threshold value, the parameter of above-mentioned initial behavior identification model is adjusted, and continue to execute above-mentioned training
Step.
The third aspect, the embodiment of the present application provide a kind of server, including：One or more processors；Memory is used
In the one or more programs of storage, camera, for obtaining image and/or video；When said one or multiple programs are above-mentioned
When one or more processors execute so that said one or multiple processors execute the side for alarm of above-mentioned first aspect
Method.
Fourth aspect, the embodiment of the present application provide a kind of computer-readable medium, are stored thereon with computer program,
It is characterized in that, which realizes the method for alarm of above-mentioned first aspect when being executed by processor.
Method and device provided by the embodiments of the present application for alarm carries out image recognition to the video of acquisition first,
Obtain people information；Then, the video clip of the corresponding personage of people information is imported to Activity recognition model trained in advance, is obtained
To the behavior label of corresponding personage；Finally alarm signal is sent out when behavior label is non-security.The application can be to video bag
The behavior contained is analyzed, and behavior label is arranged for behavior；Alarm signal is sent out when behavior label is non-security.It improves
The accuracy for identifying behavior in video, can find non-security behavior, and send out alarm signal in time.
Description of the drawings
By reading a detailed description of non-restrictive embodiments in the light of the attached drawings below, the application's is other
Feature, objects and advantages will become more apparent upon：
Fig. 1 is that this application can be applied to exemplary system architecture figures therein；
Fig. 2 is the flow chart according to one embodiment of the method for alarm of the application；
Fig. 3 is the flow chart according to one embodiment of the Activity recognition model training method of the application；
Fig. 4 is the schematic diagram according to an application scenarios of the method for alarm of the application；
Fig. 5 is the structural schematic diagram according to one embodiment of the device for alarm of the application；
Fig. 6 is adapted for the structural schematic diagram of the computer system of the server for realizing the embodiment of the present application.
Specific implementation mode
The application is described in further detail with reference to the accompanying drawings and examples.It is understood that this place is retouched
The specific embodiment stated is used only for explaining related invention, rather than the restriction to the invention.It also should be noted that in order to
Convenient for description, is illustrated only in attached drawing and invent relevant part with related.
It should be noted that in the absence of conflict, the features in the embodiments and the embodiments of the present application can phase
Mutually combination.The application is described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.
Fig. 1 shows the exemplary of the device for the method for alarm or for alarm that can apply the embodiment of the present application
System architecture 100.
As shown in Figure 1, system architecture 100 may include terminal device 101,102,103, network 104 and server 105.
Network 104 between terminal device 101,102,103 and server 105 provide communication link medium.Network 104 can be with
Including various connection types, such as wired, wireless communication link or fiber optic cables etc..
User can be interacted by network 104 with server 105 with using terminal equipment 101,102,103, to receive or send out
It delivers letters breath etc..Various video processing applications can be installed on terminal device 101,102,103, such as image processing application, regarded
Frequency acquisition applications, video control application, Video segmentation application, video send tool, video frequency tracking application etc..
Terminal device 101,102,103 can be hardware, can also be software.When terminal device 101,102,103 is hard
Can be the various electronic equipments that there is camera and support image and video acquisition when part, including but not limited to monitoring is taken the photograph
As head, IP Camera, camera violating the regulations etc..When terminal device 101,102,103 is software, above-mentioned institute may be mounted at
In the electronic equipment enumerated.Multiple softwares or software module (such as providing Distributed Services) may be implemented into it, also may be used
To be implemented as single software or software module.It is not specifically limited herein.
Server 105 can be to provide the server of various services, such as be regarded to 101,102,103 offer of terminal device
Frequency carries out the server of data processing.Server can carry out the processing such as analyzing to data such as the videos that receives, when finding to regard
There are alarm signal can be sent out when non-security behavior in frequency.
It should be noted that the method for alarm that the embodiment of the present application is provided generally is executed by server 105, phase
Ying Di, the device for alarm are generally positioned in server 105.
It should be noted that server can be hardware, can also be software.When server is hardware, may be implemented
At the distributed server cluster that multiple servers form, individual server can also be implemented as.It, can when server is software
To be implemented as multiple softwares or software module (such as providing Distributed Services), single software or software can also be implemented as
Module.It is not specifically limited herein.
It should be understood that the number of the terminal device, network and server in Fig. 1 is only schematical.According to realization need
It wants, can have any number of terminal device, network and server.
With continued reference to Fig. 2, the flow 200 of one embodiment of the method for alarm according to the application is shown.It should
Method for alarm includes the following steps：
Step 201, image recognition is carried out to the video of acquisition, obtains people information.
In the present embodiment, being used for the executive agent (such as server 105 shown in FIG. 1) of the method for alarm can pass through
Wired connection mode or radio connection receive video from terminal device 101,102,103.Wherein, terminal device 101,
102,103 the video of real-time/non-real time acquisition can be sent to server 105.People information is used for the personage in video
It is described.It should be pointed out that above-mentioned radio connection can include but is not limited to 3G/4G connections, WiFi connections, bluetooth
Connection, WiMAX connections, Zigbee connections, UWB (ultra wideband) connections and other currently known or exploitations in the future
Radio connection.
The existing camera mounted on positions such as crossing or shops can acquire the video information in setting range.Work people
Member can carry out a degree of control to camera, such as can be：The video acquisition angle of camera is controlled, with
Obtain the video data in particular video frequency acquisition range.The equipment such as the light compensating lamp that can also be carried by camera, to a certain degree
On to acquisition video when brightness control.When generation accident, staff can transfer the phase obtained before
Video is closed, to analyze etc. accident.But this analyzing processing to accident just carries out after generation accident, tool
Having time hysteresis quality can only play accident the effects that analyzing reason or process, not be avoided that the generation of accident.
For this purpose, the executive agent of the present embodiment can obtain the acquisition of terminal device 101,102,103 real-time or non real-time
Video.Later, each frame image that executive agent can include to video carries out the processing such as image recognition, from each frame image
The people information for including to video.
In some optional realization methods of the present embodiment, the above-mentioned video to acquisition carries out image recognition, obtains people
Object information, may comprise steps of：
The first step extracts character image in response to there are character images in video.
May be recorded in video has personage, it is also possible to not have record personal.In general, to the analysis of video be all to personage or
The analysis carried out with the relevant information of personage.Therefore, when, there are when character image, executive agent can extract accordingly in video
Character image.
Second step, the people information of the corresponding personage of identification character image.
After obtaining character image, image procossing is carried out to character image, the personage of the corresponding character image can be recognized
People information.Wherein, above-mentioned people information may include at least one of following for the personage in video to be described：
Number of person, gender, height, dressing color etc..
Step 202, the video clip of the corresponding personage of people information is imported to Activity recognition model trained in advance, is obtained
The behavior label of corresponding personage.
The data volume of video is usually larger.For the ease of analyzing specific personage, executive agent can be from video
In extract the video clip of corresponding personage.Later, the video clip of personage can be imported row trained in advance by executive agent
For identification model, the behavior label of corresponding personage is obtained.Wherein, the above-mentioned Activity recognition model of the application can be used for personage
Behavior be identified, and for behavior be arranged behavior label.Behavior can be a variety of, such as can be on foot, quarrel, fight
Deng.Behavior label is for classifying to behavior by certain standard.For example, standard can be standard according to whether safety, it is right
Behavior is divided.The content of corresponding behavior label can be safety behavior or non-security behavior；Standard can also be personage
The content of relationship, corresponding behavior label can be friends, family's relationship, strange relationship etc..
In some optional realization methods of the present embodiment, the Activity recognition model of the present embodiment can be artificial neuron
Network may include convolutional neural networks, Recognition with Recurrent Neural Network and full articulamentum.Correspondingly, above-mentioned that people information is corresponding
The video clip of personage imports Activity recognition model trained in advance, obtains the behavior label of corresponding personage, may include following
Step：
Above-mentioned video clip is input to above-mentioned convolutional neural networks by the first step, obtains each frame figure of above-mentioned video clip
The feature vector of picture.
In the present embodiment, video clip can be inputted convolutional neural networks by executive agent, obtained video clip and included
Each frame image feature vector.Video clip is made of multiple image, and the feature vector of each frame image can be used for describing each
Feature possessed by frame image.For example, can there is the feature vector of corresponding " taking in one's arms " action, " sealing mouth " to act in the frame image
The feature vector etc. of feature vector and " running " action.The feature vector of image can pass through the label for marking each action
The modes such as point indicate.Executive agent can train the feature of each frame image of characterization video clip and video in several ways
The convolutional neural networks of correspondence between vector.The feature vector of above-mentioned image can pass through setting flag on the image
Point, and realized in plane coordinates or the coordinate value of space coordinate with mark point.
In the present embodiment, convolutional neural networks can be a kind of feedforward neural network, its artificial neuron can ring
The surrounding cells in a part of coverage area are answered, have outstanding performance for large-scale image procossing.In general, the base of convolutional neural networks
This structure includes two layers, and one is characterized extract layer, and the input of each neuron is connected with the local acceptance region of preceding layer, and carries
Take the feature of the part.After the local feature is extracted, its position relationship between other feature is also decided therewith；
The second is Feature Mapping layer, each computation layer of network is made of multiple Feature Mappings, and each Feature Mapping is a plane, is put down
The weights of all neurons are equal on face.Here, executive agent can be defeated by the input side of video clip from convolutional neural networks
Enter, passes through the processing of the parameter of each layer in convolutional neural networks successively, and exported from the outlet side of convolutional neural networks, output
The information of side output is the feature vector of each frame image of video clip.
As an example, executive agent can be based on each frame figure to great amount of samples video clip and Sample video segment
The feature vector of picture is counted and is generated the spy for each frame image for being stored with multiple Sample video segments and Sample video segment
The mapping table of the correspondence of vector is levied, and using the mapping table as convolutional neural networks.In this way, executive agent can
Video clip and multiple Sample video segments in the mapping table to be compared successively.If in the mapping table
Some Sample video segment and the video clip are same or similar, then can be by the Sample video segment in the mapping table
Each frame image feature vector as the video clip each frame image feature vector.
As another example, executive agent can obtain each frame figure of Sample video segment and Sample video segment first
The feature vector of picture；Then using Sample video segment as input, the feature vector of each frame image of Sample video segment is made
For output, pair between the feature vector for each frame image for obtaining to characterize Sample video segment and Sample video segment is trained
The convolutional neural networks that should be related to.In this way, executive agent can input video clip from the input side of convolutional neural networks, according to
The processing of the secondary parameter by each layer in convolutional neural networks, and exported from the outlet side of convolutional neural networks, outlet side is defeated
The information gone out is the feature vector of each frame image of video clip.
The feature vector of each frame image of above-mentioned video clip is input to above-mentioned Recognition with Recurrent Neural Network, obtained by second step
The feature vector of above-mentioned video clip.
In the present embodiment, the feature vector of each frame image of video clip can be input to cycle nerve by executive agent
Network, to obtain the feature vector of video clip.Wherein, the feature vector of video clip can be used for characterizing video clip
Incidence relation between the feature vector of each frame image.The combination of eigenvectors of each frame image at multiple moment is got up, it can be with
Constitute the feature vector of video clip.For example, the feature vector that " the taking in one's arms " of above-mentioned each frame image is acted, " sealing mouth " act
Feature vector and the feature vector etc. of " running " action be combined according to chronological order, corresponding video can be obtained
The feature vector of segment.The feature vector of the video clip can be the feature vector for characterizing coherent multiple actions.It will
Multiple combination of actions, which are got up, can correspond to certain behavior.Therefore, the feature vector of video clip is it is also assumed that be various figures
Feature vector mobilism expression, can include information more more than the feature vector of each frame image.Executive agent can lead to
It crosses between the feature vector and the feature vector of video clip for each frame image that various ways training can characterize video clip
The Recognition with Recurrent Neural Network of incidence relation.
In the present embodiment, Recognition with Recurrent Neural Network is a kind of artificial neural network of node orientation connection cyclization.This net
The substantive characteristics of network is the feedback link of the existing inside between processing unit has feedforward to connect again, and internal state can be shown
Dynamic time sequence behavior.
As an example, executive agent can be based on the feature vector and sample of each frame image to great amount of samples video clip
The feature vector of video clip counted and generate the feature vector of each frame image for being stored with multiple Sample video segments with
The mapping table of the correspondence of the feature vector of Sample video segment, and using the mapping table as cycle nerve net
Network.In this way, executive agent can calculate the feature vector of each frame image of video clip and multiple samples in the mapping table
The parameters such as the Euclidean distance between the feature vector of each frame image of this video clip.If some sample in the mapping table
Euclidean distance between the feature vector of each frame image of video clip and the feature vector of each frame image of video clip is less than
Preset distance threshold, can be using the feature vector of the Sample video segment in the mapping table as the video clip
Feature vector.
As another example, executive agent can obtain the feature vector and sample of each frame image of Sample video segment
The feature vector of video clip；Then using the feature vector of each frame image of Sample video segment as input, by Sample video
The feature vector of segment obtains to characterize the feature vector and sample of each frame image of Sample video segment as output, training
The Recognition with Recurrent Neural Network of correspondence between the feature vector of video clip.In this way, executive agent can be by video clip
The feature vector of each frame image is inputted from the input side of Recognition with Recurrent Neural Network, passes through the ginseng of each layer in Recognition with Recurrent Neural Network successively
Several processing, and exported from the outlet side of Recognition with Recurrent Neural Network, the information of outlet side output is the feature vector of video clip.
Third walks, and the feature vector of above-mentioned video clip is input to above-mentioned full articulamentum, obtains above-mentioned behavior label.
In the present embodiment, the feature vector of video clip can be input to full articulamentum by executive agent, to obtain
The behavior label of video clip.Wherein, can include a plurality of types of behaviors such as walk, quarrel, fighting in video clip.It is different
Behavior can correspond to different behavior labels.For example, by the feature vector of " taking in one's arms " action of above-mentioned each frame image, " sealing
The feature vector of mouth " action and the feature vector etc. of " running " action are combined according to chronological order, can be obtained pair
The feature vector for the video clip answered.The feature vector of the video clip can correspond to the behaviors such as " kidnaping and selling people ", " robbery ".This
When, can be able to be the behavior label of " non-security " to the feature vector set content of the video clip.Executive agent can lead to
Cross the full articulamentum for the correspondence that various ways training can characterize between the feature vector of video clip and behavior label.
In the present embodiment, all node phases of each node of full articulamentum and the output layer of Recognition with Recurrent Neural Network
Even, the feature vector of the video for Recognition with Recurrent Neural Network output layer is exported integrates.Due to the characteristic that it is connected entirely, one
As the parameter of full articulamentum be also most.Meanwhile the feature vector of video is carried out in the parameter using full articulamentum linear
After transformation, the result of linear transformation can be converted plus a nonlinear activation function, to introduce non-linear factor,
To enhance the ability to express of Activity recognition model.Wherein, excitation function can be softmax functions.Softmax functions are artificial
Common a kind of excitation function in neural network, in this not go into detail.
As an example, executive agent can based on to great amount of samples video clip feature vector and behavior label into
Row counts and generates the feature vector pass corresponding with the correspondence of sample behavior label for being stored with multiple Sample video segments
It is table, and using the mapping table as full articulamentum.In this way, executive agent can calculate the feature vector of video clip and be somebody's turn to do
Euclidean distance between the feature vector of multiple Sample video segments in mapping table.If some in the mapping table
Euclidean distance between the feature vector of Sample video segment and the feature vector of the video clip is less than preset distance threshold,
Then using the corresponding sample behavior label of the Sample video segment in the mapping table as the behavior label of the video clip.
As another example, executive agent can obtain the feature vector and sample behavior mark of Sample video segment first
Label.Then, using the feature vector of Sample video segment as input, using sample behavior label as output, training obtains can
Characterize the full articulamentum of the correspondence between the feature vector and sample behavior label of Sample video segment.In this way, executing master
Body can input the feature vector of video clip from the input side of full articulamentum, by the parameter and excitation function of full articulamentum
Processing, and from the outlet side of full articulamentum export, outlet side output information be behavior label.
Step 203, it is non-security in response to above-mentioned behavior label, then sends out alarm signal.
Seen from the above description, the behavior label substance that behavior label model obtains can be there are many situation.When behavior mark
When the content of label is non-security, executive agent can send out alarm signal in time, to prevent the generation of accident, or subtract
The loss etc. of minor accident.For example, trailing a child when terminal device 101,102,103 collects a certain adult, and cover small
When child is embraced by the mouth of child, the behavior label model on executive agent can analyze the video clip of the adult,
Show that the behavior of the adult is non-security behavior label.Later, executive agent can send out alarm signal in time, so as to phase
Pass personnel are monitored or arrest to the adult in time.
With further reference to Fig. 3, it illustrates according to one embodiment of the Activity recognition model training method of the application
Flow 300.The flow 300 of behavior identification model training method, includes the following steps：
Step 301, obtaining multiple records has each of the history video of personage's behavior and above-mentioned multiple history videos to go through
Historical behavior label corresponding to history video.
In the present embodiment, executive agent (such as the clothes shown in FIG. 1 of Activity recognition model training method operation thereon
Business device 105) multiple each history video institutes for recording and having in the history video and multiple history videos of personage's behavior can be obtained
Corresponding historical behavior label.
In the present embodiment, executive agent can obtain multiple history videos for recording and having personage's behavior, and be this field
Technical staff plays.Those skilled in the art can be rule of thumb to each history video labeling history in multiple history videos
Behavior label.Wherein, whether safe the historical behavior label of the present embodiment can be used for identifying personage's behavior in history video.
For example, when the video content of history video is the behavior fought, quarreled etc. with propensity to violence, technical staff can be history
Video set content is non-security historical behavior label.The content of historical behavior label can be different interior as needed
Hold, depending on actual needs.
Step 302, each history video in above-mentioned multiple history videos is sequentially input to initial behavior identification model,
Obtain the prediction history behavior label corresponding to each history video in above-mentioned multiple history videos.
In the present embodiment, there are the history video of personage's behavior, executive agent that can incite somebody to action based on acquired multiple records
Each history video in multiple history videos is sequentially input to initial behavior identification model, to obtain in multiple history videos
Each history video corresponding to prediction history behavior label.Here, executive agent can be by each history video from initial
The input side of Activity recognition model inputs, and passes through the processing of the parameter of each layer in initial behavior identification model successively, and from first
It begins as the outlet side output of identification model, the information that outlet side exports is the prediction history behavior corresponding to the history video
Label.Wherein, initial behavior identification model can be unbred Activity recognition model or the Activity recognition that training is not completed
Model, each layer are provided with initiation parameter, and initiation parameter can be by constantly in the training process of behavior identification model
Adjustment.
Initial behavior identification model can train in accordance with the following steps to be obtained：
The first step obtains the sample of the Sample video segment and above-mentioned multiple and different behavior types of multiple and different behavior types
The sample behavior label of the Sample video segment of each behavior type in video clip.
Can be that the Sample video segment setting comprising typical behavior type corresponds to when the initial behavior identification model of training
Sample behavior label.Such as can be that Sample video segment contents include abducting child behavior, and the behavior includes to child's
" sealing mouth " acts and " quickly walking about " action.Can be then the Sample video segment set content be " kidnaping and selling people " or " non-peace
Sample behavior label entirely ".
Second step, using machine learning method, by each row in the Sample video segment of above-mentioned multiple and different behavior types
It is the Sample video segment conduct input of type, by each behavior class in the Sample video segment of above-mentioned multiple and different behavior types
The sample behavior label of the Sample video segment of type obtains initial behavior identification model as output, training.
It, can be by each row after obtaining the sample behavior label of above-mentioned Sample video segment and corresponding Sample video segment
It is the Sample video segment conduct input of type, using the sample behavior label of the Sample video segment of behavior type as defeated
Go out, training obtains initial behavior identification model.Initial behavior identification model can be based on the model realizations such as deep learning model, and
And can have and include the structures such as convolutional neural networks, Recognition with Recurrent Neural Network and full articulamentum.
Step 303, by corresponding to each history video in above-mentioned multiple history videos prediction history behavior label with
Historical behavior label corresponding to the history video is compared, and obtains the recognition accuracy of above-mentioned initial behavior identification model.
In the present embodiment, based on corresponding to each history video in the obtained multiple history videos of step 302
Prediction history behavior label, executive agent can be by the prediction history rows corresponding to each history video in multiple history videos
It is compared with the historical behavior label corresponding to the history video for label, to obtain the identification of initial behavior identification model
Accuracy rate.Specifically, if the prediction history behavior label corresponding to a history video and the history corresponding to the history video
Behavior label is same or similar, then initial behavior identification model identification is correct；If the prediction history corresponding to a history video
Behavior label is different or not close from the historical behavior label corresponding to the history video, then initial behavior identification model identification is wrong
Accidentally.Here, executive agent can calculate the ratio for identifying correct number and total sample number, and as initial behavior identification model
Recognition accuracy.
Step 304, determine whether above-mentioned recognition accuracy is more than default accuracy rate threshold value.
In the present embodiment, the recognition accuracy based on the obtained initial behavior identification model of step 303, executive agent
The recognition accuracy of initial behavior identification model can be compared with default accuracy rate threshold value, if more than default accuracy rate threshold
Value, thens follow the steps 305；If more than default accuracy rate threshold value, 306 are thened follow the steps.
Step 305, the Activity recognition model above-mentioned initial behavior identification model completed as training.
In the present embodiment, the case where the recognition accuracy of initial behavior identification model is more than default accuracy rate threshold value
Under, illustrate that the training of behavior identification model is completed.At this point, executive agent can be completed initial behavior identification model as training
Activity recognition model.
Step 306, the parameter of above-mentioned initial behavior identification model is adjusted.
In the present embodiment, the case where the recognition accuracy of initial behavior identification model is not more than default accuracy rate threshold value
Under, executive agent can adjust the parameter of initial behavior identification model, and return to step 302, until training and can know
In other video until the Activity recognition model of the behavior of personage.
It is a schematic diagram according to the application scenarios of the method for alarm of the present embodiment with continued reference to Fig. 4, Fig. 4.
In the application scenarios of Fig. 4, terminal device 101 collects the video that two spadgers fight.Later, terminal device 101 will be adopted
The video collected is sent to server 105 by network 104.Server 105 carries out image recognition to video, determines in the video
Including people information.For example, people information can be " 2 people, man ".Then, server 105 believes the correspondence personage in video
The video clip of the personage of breath imports Activity recognition model, can be confirmed that personage's behavior is to fight, and it is non-security to obtain content
Behavior label.Finally, server 105 sends out alarm signal according to behavior label.Related personnel can be true according to alarm signal
Determine personage position, and is prevented.
The method that above-described embodiment of the application provides carries out image recognition to the video of acquisition first, obtains personage's letter
Breath；Then, the video clip of the corresponding personage of people information is imported to Activity recognition model trained in advance, obtains corresponding personage
Behavior label；Finally alarm signal is sent out when behavior label is non-security.The application can to behavior that video includes into
Row analysis, and behavior label is set for behavior；Alarm signal is sent out when behavior label is non-security.It improves in identification video
The accuracy of behavior can find non-security behavior, and send out alarm signal in time.
With further reference to Fig. 5, as the realization to method shown in above-mentioned each figure, this application provides a kind of for alarm
One embodiment of device, the device embodiment is corresponding with embodiment of the method shown in Fig. 2, which specifically can be applied to
In various electronic equipments.
As shown in figure 5, the device 500 for alarm of the present embodiment may include：People information recognition unit 501, row
For label acquiring unit 502 and Alarm Unit 503.Wherein, people information recognition unit 501 to the video of acquisition for carrying out figure
As identification, people information is obtained；Behavior label acquiring unit 502 is used to import the video clip of the corresponding personage of people information
Trained Activity recognition model in advance, obtains the behavior label of corresponding personage, and above-mentioned Activity recognition model is used for the row to personage
To be identified, and behavior label is set for behavior；Alarm Unit 503 be used in response to above-mentioned behavior label be it is non-security, then
Send out alarm signal.
In some optional realization methods of the present embodiment, above-mentioned people information recognition unit 501 may include：Personage
Image zooming-out subelement (not shown) and people information identify subelement (not shown).Wherein, personal image extraction
Subelement is used to, in response to there are character images in video, extract character image；People information identifies subelement people for identification
The people information of the corresponding personage of object image, above-mentioned people information may include at least one of following：Gender, height, dressing face
Color etc..
In some optional realization methods of the present embodiment, above-mentioned Activity recognition model includes convolutional neural networks, follows
Ring neural network and full articulamentum.
In some optional realization methods of the present embodiment, above-mentioned behavior label acquiring unit 502 may include：Image
Feature vector obtains subelement (not shown), video feature vector obtains subelement (not shown) and behavior label obtains
Take subelement (not shown).Wherein, it is above-mentioned for above-mentioned video clip to be input to obtain subelement for image feature vector
Convolutional neural networks obtain the feature vector of each frame image of above-mentioned video clip, wherein above-mentioned convolutional neural networks are used for table
Levy the correspondence between the feature vector of each frame image of video clip and video；Video feature vector obtains subelement and is used for
The feature vector of each frame image of above-mentioned video clip is input to above-mentioned Recognition with Recurrent Neural Network, obtains the spy of above-mentioned video clip
Sign vector, wherein above-mentioned Recognition with Recurrent Neural Network is used to characterize the feature vector and video clip of each frame image of video clip
Correspondence between feature vector, the feature vector of video clip are used to characterize the feature vector of each frame image of video clip
Between incidence relation；Behavior label obtains subelement, above-mentioned connects entirely for the feature vector of above-mentioned video clip to be input to
Connect layer, obtain above-mentioned behavior label, wherein above-mentioned full articulamentum be used to characterize the feature vector of video clip and behavior label it
Between correspondence.
In some optional realization methods of the present embodiment, the device 500 for alarm may include Activity recognition mould
Type training unit (not shown), above-mentioned Activity recognition model training unit include：Historical information obtains subelement (in figure not
Show) and Activity recognition model training subelement (not shown).Wherein, it is multiple for obtaining to obtain subelement for historical information
Record has the historical behavior mark corresponding to each history video in the history video and above-mentioned multiple history videos of personage's behavior
Label, wherein whether historical behavior label is safe for identifying personage's behavior；Activity recognition model training subelement is used for will be above-mentioned
Each history video in multiple history videos is as input, corresponding to each history video in above-mentioned multiple history videos
Historical behavior label as output, training obtain above-mentioned Activity recognition model.
In some optional realization methods of the present embodiment, above-mentioned Activity recognition model training subelement may include：
Activity recognition model training module (not shown), for each history video in above-mentioned multiple history videos is defeated successively
Enter to initial behavior identification model, obtains the prediction history behavior corresponding to each history video in above-mentioned multiple history videos
Label, by prediction history behavior label and history video institute corresponding to each history video in above-mentioned multiple history videos
Corresponding historical behavior label is compared, and is obtained the recognition accuracy of above-mentioned initial behavior identification model, is determined above-mentioned identification
Whether accuracy rate is more than default accuracy rate threshold value, if more than above-mentioned default accuracy rate threshold value, then by above-mentioned initial Activity recognition mould
The Activity recognition model that type is completed as training.
In some optional realization methods of the present embodiment, above-mentioned Activity recognition model training subelement can also wrap
It includes：Parameter adjustment module (not shown), in response to being not more than above-mentioned default accuracy rate threshold value, adjusting above-mentioned initial row
For the parameter of identification model, and continue to execute above-mentioned training step.
The present embodiment additionally provides a kind of server, including：One or more processors；Memory, for storing one
Or multiple programs, camera, for obtaining image and/or video；When said one or multiple programs are by said one or multiple
When processor executes so that said one or multiple processors execute the above-mentioned method for alarm.
The present embodiment additionally provides a kind of computer-readable medium, is stored thereon with computer program, which is handled
Device realizes the above-mentioned method for alarm when executing.
Below with reference to Fig. 6, it illustrates the computer systems 600 suitable for the server for realizing the embodiment of the present application
Structural schematic diagram.Server shown in Fig. 6 is only an example, should not be to the function and use scope band of the embodiment of the present application
Carry out any restrictions.
As shown in fig. 6, computer system 600 includes central processing unit (CPU) 601, it can be read-only according to being stored in
Program in memory (ROM) 602 or be loaded into the program in random access storage device (RAM) 603 from storage section 608 and
Execute various actions appropriate and processing.In RAM 603, also it is stored with system 600 and operates required various programs and data.
CPU 601, ROM 602 and RAM 603 are connected with each other by bus 604.Input/output (I/O) interface 605 is also connected to always
Line 604.
It is connected to I/O interfaces 605 with lower component：Importation 606 including keyboard, mouse etc.；It is penetrated including such as cathode
The output par, c 607 of spool (CRT), liquid crystal display (LCD) etc. and loud speaker etc.；Storage section 608 including hard disk etc.；
And the communications portion 609 of the network interface card including LAN card, modem etc..Communications portion 609 via such as because
The network of spy's net executes communication process.Driver 610 is also according to needing to be connected to I/O interfaces 605.Detachable media 611, such as
Disk, CD, magneto-optic disk, semiconductor memory etc. are mounted on driver 610, as needed in order to be read from thereon
Computer program be mounted into storage section 608 as needed.Camera 612 is also according to needing to be connected to I/O interfaces 605.
Particularly, in accordance with an embodiment of the present disclosure, it may be implemented as computer above with reference to the process of flow chart description
Software program.For example, embodiment of the disclosure includes a kind of computer program product comprising be carried on computer-readable medium
On computer program, which includes the program code for method shown in execution flow chart.In such reality
It applies in example, which can be downloaded and installed by communications portion 609 from network, and/or from detachable media
611 are mounted.When the computer program is executed by central processing unit (CPU) 601, limited in execution the present processes
Above-mentioned function.
It should be noted that the above-mentioned computer-readable medium of the application can be computer-readable signal media or meter
Calculation machine readable storage medium storing program for executing either the two arbitrarily combines.Computer readable storage medium for example can be --- but not
Be limited to --- electricity, magnetic, optical, electromagnetic, infrared ray or semiconductor system, device or device, or arbitrary above combination.Meter
The more specific example of calculation machine readable storage medium storing program for executing can include but is not limited to：Electrical connection with one or more conducting wires, just
It takes formula computer disk, hard disk, random access storage device (RAM), read-only memory (ROM), erasable type and may be programmed read-only storage
Device (EPROM or flash memory), optical fiber, portable compact disc read-only memory (CD-ROM), light storage device, magnetic memory device,
Or above-mentioned any appropriate combination.In this application, can be any include computer readable storage medium or storage journey
The tangible medium of sequence, the program can be commanded the either device use or in connection of execution system, device.And at this
In application, computer-readable signal media may include in a base band or as the data-signal that a carrier wave part is propagated,
Wherein carry computer-readable program code.Diversified forms may be used in the data-signal of this propagation, including but unlimited
In electromagnetic signal, optical signal or above-mentioned any appropriate combination.Computer-readable signal media can also be that computer can
Any computer-readable medium other than storage medium is read, which can send, propagates or transmit and be used for
By instruction execution system, device either device use or program in connection.Include on computer-readable medium
Program code can transmit with any suitable medium, including but not limited to：Wirelessly, electric wire, optical cable, RF etc. or above-mentioned
Any appropriate combination.
Flow chart in attached drawing and block diagram, it is illustrated that according to the system of the various embodiments of the application, method and computer journey
The architecture, function and operation in the cards of sequence product.In this regard, each box in flowchart or block diagram can generation
A part for a part for one module, program segment, or code of table, the module, program segment, or code includes one or more uses
The executable instruction of the logic function as defined in realization.It should also be noted that in some implementations as replacements, being marked in box
The function of note can also occur in a different order than that indicated in the drawings.For example, two boxes succeedingly indicated are actually
It can be basically executed in parallel, they can also be executed in the opposite order sometimes, this is depended on the functions involved.Also it to note
Meaning, the combination of each box in block diagram and or flow chart and the box in block diagram and or flow chart can be with holding
The dedicated hardware based system of functions or operations as defined in row is realized, or can use specialized hardware and computer instruction
Combination realize.
Being described in unit involved in the embodiment of the present application can be realized by way of software, can also be by hard
The mode of part is realized.Described unit can also be arranged in the processor, for example, can be described as：A kind of processor packet
Include people information recognition unit, behavior label acquiring unit and Alarm Unit.Wherein, the title of these units is under certain conditions
The restriction to the unit itself is not constituted, for example, Alarm Unit is also described as " sending out when behavior label is non-security
Go out the unit of alarm signal ".
As on the other hand, present invention also provides a kind of computer-readable medium, which can be
Included in device described in above-described embodiment；Can also be individualism, and without be incorporated the device in.Above-mentioned calculating
Machine readable medium carries one or more program, when said one or multiple programs are executed by the device so that should
Device：Image recognition is carried out to the video of acquisition, obtains people information；The video clip of the corresponding personage of people information is imported
Trained Activity recognition model in advance, obtains the behavior label of corresponding personage, and above-mentioned Activity recognition model is used for the row to personage
To be identified, and behavior label is set for behavior；It is non-security in response to above-mentioned behavior label, then sends out alarm signal.
Above description is only the preferred embodiment of the application and the explanation to institute's application technology principle.People in the art
Member should be appreciated that invention scope involved in the application, however it is not limited to technology made of the specific combination of above-mentioned technical characteristic
Scheme, while should also cover in the case where not departing from foregoing invention design, it is carried out by above-mentioned technical characteristic or its equivalent feature
Other technical solutions of arbitrary combination and formation.Such as features described above has similar work(with (but not limited to) disclosed herein
Can technical characteristic replaced mutually and the technical solution that is formed.
Claims (16)
1. a kind of method for alarm, including：
Image recognition is carried out to the video of acquisition, obtains people information；
The video clip of the corresponding personage of people information is imported to Activity recognition model trained in advance, obtains the row of corresponding personage
For label, behavior label is arranged for the behavior of personage to be identified, and for behavior in the Activity recognition model；
It is non-security in response to the behavior label, then sends out alarm signal.
2. according to the method described in claim 1, wherein, the video of described pair of acquisition carries out image recognition, obtains people information,
Including：
In response to there are character images in video, character image is extracted；
Identify that the people information of the corresponding personage of character image, the people information include at least one of following：Gender, height,
Dressing color.
3. according to the method described in claim 1, wherein, the Activity recognition model includes convolutional neural networks, cycle nerve
Network and full articulamentum.
4. according to the method described in claim 3, wherein, the video clip by the corresponding personage of people information imports advance
Trained Activity recognition model obtains the behavior label of corresponding personage, including：
The video clip is input to the convolutional neural networks, obtain the feature of each frame image of the video clip to
Amount, wherein the convolutional neural networks are corresponding between video clip and the feature vector of each frame image of video for characterizing
Relationship；
The feature vector of each frame image of the video clip is input to the Recognition with Recurrent Neural Network, obtains the video clip
Feature vector, wherein the Recognition with Recurrent Neural Network be used for characterize video clip each frame image feature vector and piece of video
Correspondence between the feature vector of section, the feature vector of video clip are used to characterize the feature of each frame image of video clip
Incidence relation between vector；
The feature vector of the video clip is input to the full articulamentum, obtains the behavior label, wherein described to connect entirely
Layer is connect for characterizing the correspondence between the feature vector of video clip and behavior label.
5. according to the method described in claim 1, wherein, training obtains the Activity recognition model as follows：
It obtains corresponding to each history video that multiple records have in the history video and the multiple history video of personage's behavior
Historical behavior label, wherein historical behavior label for identify personage's behavior whether safety；
Using each history video in the multiple history video as input, by each history in the multiple history video
For historical behavior label corresponding to video as output, training obtains the Activity recognition model.
6. according to the method described in claim 5, wherein, each history video using in the multiple history video as
Input, using the historical behavior label corresponding to each history video in the multiple history video as output, training obtains
The Activity recognition model, including：
Execute following training step：Each history video in the multiple history video is sequentially input to initial Activity recognition
Model obtains the prediction history behavior label corresponding to each history video in the multiple history video, will be the multiple
The prediction history behavior label corresponding to each history video in history video and the historical behavior corresponding to the history video
Label is compared, and is obtained the recognition accuracy of the initial behavior identification model, is determined whether the recognition accuracy is more than
Default accuracy rate threshold value is then completed the initial behavior identification model as training if more than the default accuracy rate threshold value
Activity recognition model.
7. according to the method described in claim 6, wherein, each history video using in the multiple history video as
Input, using the historical behavior label corresponding to each history video in the multiple history video as output, training obtains
The Activity recognition model, including：
In response to being not more than the default accuracy rate threshold value, the parameter of the initial behavior identification model is adjusted, and continue to execute
The training step.
8. a kind of device for alarm, including：
People information recognition unit carries out image recognition for the video to acquisition, obtains people information；
Behavior label acquiring unit, for the video clip of the corresponding personage of people information to be imported to Activity recognition trained in advance
Model obtains the behavior label of corresponding personage, and the Activity recognition model is behavior for the behavior of personage to be identified
Setting behavior label；
Alarm Unit then sends out alarm signal for being non-security in response to the behavior label.
9. device according to claim 8, wherein the people information recognition unit includes：
Personal image extraction subelement, in response to there are character images in video, extracting character image；
People information identifies subelement, and the people information of the corresponding personage of character image, the people information include for identification
At least one of below：Gender, height, dressing color.
10. device according to claim 8, wherein the Activity recognition model includes convolutional neural networks, cycle nerve
Network and full articulamentum.
11. device according to claim 10, wherein the behavior label acquiring unit includes：
Image feature vector obtains subelement, for the video clip to be input to the convolutional neural networks, obtains described
The feature vector of each frame image of video clip, wherein the convolutional neural networks are used to characterize each of video clip and video
Correspondence between the feature vector of frame image；
Video feature vector obtains subelement, for the feature vector of each frame image of the video clip to be input to described follow
Ring neural network obtains the feature vector of the video clip, wherein the Recognition with Recurrent Neural Network is for characterizing video clip
The feature vector of correspondence between the feature vector and the feature vector of video clip of each frame image, video clip is used for table
Levy the incidence relation between the feature vector of each frame image of video clip；
Behavior label obtains subelement and obtains institute for the feature vector of the video clip to be input to the full articulamentum
State behavior label, wherein the full articulamentum is used to characterize the corresponding pass between the feature vector of video clip and behavior label
System.
12. device according to claim 8, wherein described device includes Activity recognition model training unit, the behavior
Identification model training unit includes：
Historical information obtains subelement, has the history video of personage's behavior and the multiple history video for obtaining multiple records
In each history video corresponding to historical behavior label, wherein historical behavior label is for identifying whether personage's behavior is pacified
Entirely；
Activity recognition model training subelement is used for using each history video in the multiple history video as input, will
For historical behavior label corresponding to each history video in the multiple history video as output, training obtains the behavior
Identification model.
13. device according to claim 12, wherein the Activity recognition model training subelement includes：
Activity recognition model training module, for sequentially inputting each history video in the multiple history video to initial
Activity recognition model obtains the prediction history behavior label corresponding to each history video in the multiple history video, will
Corresponding to the prediction history behavior label and the history video corresponding to each history video in the multiple history video
Historical behavior label is compared, and is obtained the recognition accuracy of the initial behavior identification model, is determined the recognition accuracy
Whether be more than default accuracy rate threshold value, if more than the default accuracy rate threshold value, then using the initial behavior identification model as
The Activity recognition model that training is completed.
14. device according to claim 13, wherein the Activity recognition model training subelement further includes：
Parameter adjustment module, in response to being not more than the default accuracy rate threshold value, adjusting the initial behavior identification model
Parameter, and continue to execute the training step.
15. a kind of server, including：
One or more processors；
Memory, for storing one or more programs；
Camera, for obtaining image and/or video；
When one or more of programs are executed by one or more of processors so that one or more of processors
Perform claim requires any method in 1 to 7.
16. a kind of computer-readable medium, is stored thereon with computer program, which is characterized in that the program is executed by processor
In Shi Shixian such as claim 1 to 7 it is any as described in method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810256338.1A CN108446649A (en) | 2018-03-27 | 2018-03-27 | Method and device for alarm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810256338.1A CN108446649A (en) | 2018-03-27 | 2018-03-27 | Method and device for alarm |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108446649A true CN108446649A (en) | 2018-08-24 |
Family
ID=63196750
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810256338.1A Pending CN108446649A (en) | 2018-03-27 | 2018-03-27 | Method and device for alarm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108446649A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109697815A (en) * | 2019-01-24 | 2019-04-30 | 广州市天河区保安服务公司 | Anti-theft communication network alarming method, appliance arrangement and storage medium |
| CN110046278A (en) * | 2019-03-11 | 2019-07-23 | 北京奇艺世纪科技有限公司 | Video classification methods, device, terminal device and storage medium |
| CN110795703A (en) * | 2019-09-20 | 2020-02-14 | 华为技术有限公司 | Data anti-theft method and related product |
| CN111079477A (en) * | 2018-10-19 | 2020-04-28 | 北京奇虎科技有限公司 | Monitoring analysis method and monitoring analysis system |
| CN111222370A (en) * | 2018-11-26 | 2020-06-02 | 浙江宇视科技有限公司 | Case studying and judging method, system and device |
| CN111881946A (en) * | 2020-07-09 | 2020-11-03 | 珠海格力电器股份有限公司 | Safety monitoring method and device, storage medium, electronic equipment and air conditioner |
| CN112667883A (en) * | 2019-10-16 | 2021-04-16 | 腾讯科技（深圳）有限公司 | Resource information pushing method and device, storage medium and computer equipment |
| CN112911204A (en) * | 2019-12-03 | 2021-06-04 | 宇龙计算机通信科技(深圳)有限公司 | Monitoring method, monitoring device, storage medium and electronic equipment |
| CN114724080A (en) * | 2022-03-31 | 2022-07-08 | 慧之安信息技术股份有限公司 | Construction site intelligent safety identification method and device based on security video monitoring |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102542301A (en) * | 2011-12-28 | 2012-07-04 | 浙江大学 | Early-stage drowning behavior detection method based on videos |
| CN105574133A (en) * | 2015-12-15 | 2016-05-11 | 苏州贝多环保技术有限公司 | Multi-mode intelligent question answering system and method |
| CN105631415A (en) * | 2015-12-25 | 2016-06-01 | 中通服公众信息产业股份有限公司 | Video pedestrian recognition method based on convolution neural network |
| CN106096568A (en) * | 2016-06-21 | 2016-11-09 | 同济大学 | A kind of pedestrian's recognition methods again based on CNN and convolution LSTM network |
| CN106446015A (en) * | 2016-08-29 | 2017-02-22 | 北京工业大学 | Video content access prediction and recommendation method based on user behavior preference |
| CN106682050A (en) * | 2015-11-24 | 2017-05-17 | 北京中科汇联科技股份有限公司 | System and method capable of achieving intelligent questioning and answering |
| CN107045623A (en) * | 2016-12-30 | 2017-08-15 | 厦门瑞为信息技术有限公司 | A kind of method of the indoor dangerous situation alarm based on human body attitude trace analysis |
| CN107291232A (en) * | 2017-06-20 | 2017-10-24 | 深圳市泽科科技有限公司 | A kind of somatic sensation television game exchange method and system based on deep learning and big data |
| CN107330392A (en) * | 2017-06-26 | 2017-11-07 | 司马大大（北京）智能系统有限公司 | Video scene annotation equipment and method |
| CN107832708A (en) * | 2017-11-09 | 2018-03-23 | 云丁网络技术(北京)有限公司 | A kind of human motion recognition method and device |
-
2018
- 2018-03-27 CN CN201810256338.1A patent/CN108446649A/en active Pending
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102542301A (en) * | 2011-12-28 | 2012-07-04 | 浙江大学 | Early-stage drowning behavior detection method based on videos |
| CN106682050A (en) * | 2015-11-24 | 2017-05-17 | 北京中科汇联科技股份有限公司 | System and method capable of achieving intelligent questioning and answering |
| CN105574133A (en) * | 2015-12-15 | 2016-05-11 | 苏州贝多环保技术有限公司 | Multi-mode intelligent question answering system and method |
| CN105631415A (en) * | 2015-12-25 | 2016-06-01 | 中通服公众信息产业股份有限公司 | Video pedestrian recognition method based on convolution neural network |
| CN106096568A (en) * | 2016-06-21 | 2016-11-09 | 同济大学 | A kind of pedestrian's recognition methods again based on CNN and convolution LSTM network |
| CN106446015A (en) * | 2016-08-29 | 2017-02-22 | 北京工业大学 | Video content access prediction and recommendation method based on user behavior preference |
| CN107045623A (en) * | 2016-12-30 | 2017-08-15 | 厦门瑞为信息技术有限公司 | A kind of method of the indoor dangerous situation alarm based on human body attitude trace analysis |
| CN107291232A (en) * | 2017-06-20 | 2017-10-24 | 深圳市泽科科技有限公司 | A kind of somatic sensation television game exchange method and system based on deep learning and big data |
| CN107330392A (en) * | 2017-06-26 | 2017-11-07 | 司马大大（北京）智能系统有限公司 | Video scene annotation equipment and method |
| CN107832708A (en) * | 2017-11-09 | 2018-03-23 | 云丁网络技术(北京)有限公司 | A kind of human motion recognition method and device |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111079477A (en) * | 2018-10-19 | 2020-04-28 | 北京奇虎科技有限公司 | Monitoring analysis method and monitoring analysis system |
| CN111222370A (en) * | 2018-11-26 | 2020-06-02 | 浙江宇视科技有限公司 | Case studying and judging method, system and device |
| CN109697815A (en) * | 2019-01-24 | 2019-04-30 | 广州市天河区保安服务公司 | Anti-theft communication network alarming method, appliance arrangement and storage medium |
| CN110046278B (en) * | 2019-03-11 | 2021-10-15 | 北京奇艺世纪科技有限公司 | Video classification method and device, terminal equipment and storage medium |
| CN110046278A (en) * | 2019-03-11 | 2019-07-23 | 北京奇艺世纪科技有限公司 | Video classification methods, device, terminal device and storage medium |
| CN110795703A (en) * | 2019-09-20 | 2020-02-14 | 华为技术有限公司 | Data anti-theft method and related product |
| WO2021052201A1 (en) * | 2019-09-20 | 2021-03-25 | 华为技术有限公司 | Data theft prevention method and related product |
| CN110795703B (en) * | 2019-09-20 | 2024-04-16 | 华为技术有限公司 | Data theft prevention method and related product |
| CN112667883A (en) * | 2019-10-16 | 2021-04-16 | 腾讯科技（深圳）有限公司 | Resource information pushing method and device, storage medium and computer equipment |
| CN112667883B (en) * | 2019-10-16 | 2023-12-26 | 腾讯科技（深圳）有限公司 | Resource information pushing method and device, storage medium and computer equipment |
| CN112911204A (en) * | 2019-12-03 | 2021-06-04 | 宇龙计算机通信科技(深圳)有限公司 | Monitoring method, monitoring device, storage medium and electronic equipment |
| CN111881946A (en) * | 2020-07-09 | 2020-11-03 | 珠海格力电器股份有限公司 | Safety monitoring method and device, storage medium, electronic equipment and air conditioner |
| CN114724080A (en) * | 2022-03-31 | 2022-07-08 | 慧之安信息技术股份有限公司 | Construction site intelligent safety identification method and device based on security video monitoring |
| CN114724080B (en) * | 2022-03-31 | 2023-10-27 | 慧之安信息技术股份有限公司 | Building site intelligent safety identification method and device based on security video monitoring |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108446649A (en) | Method and device for alarm | |
| CN108154196B (en) | Method and apparatus for exporting image | |
| CN108171207A (en) | Face identification method and device based on video sequence | |
| CN108509941A (en) | Emotional information generation method and device | |
| CN107590807A (en) | Method and apparatus for detection image quality | |
| CN107168952A (en) | Information generating method and device based on artificial intelligence | |
| CN108229376A (en) | For detecting the method and device of blink | |
| CN109446990A (en) | Method and apparatus for generating information | |
| CN107578034A (en) | information generating method and device | |
| CN108510472A (en) | Method and apparatus for handling image | |
| CN108509892A (en) | Method and apparatus for generating near-infrared image | |
| CN108229485A (en) | For testing the method and apparatus of user interface | |
| CN108197318A (en) | Face identification method, device, robot and storage medium | |
| CN108491823A (en) | Method and apparatus for generating eye recognition model | |
| CN108549848A (en) | Method and apparatus for output information | |
| CN109189544A (en) | Method and apparatus for generating dial plate | |
| CN108133197A (en) | For generating the method and apparatus of information | |
| CN108062416B (en) | Method and apparatus for generating label on map | |
| CN108491825A (en) | information generating method and device | |
| CN107832721A (en) | Method and apparatus for output information | |
| CN108509994A (en) | character image clustering method and device | |
| CN108171208A (en) | Information acquisition method and device | |
| CN108446659A (en) | Method and apparatus for detecting facial image | |
| CN108364029A (en) | Method and apparatus for generating model | |
| CN109697392A (en) | Draw the method and device of target object thermodynamic chart |
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 |
Application publication date: 20180824 |
|
| RJ01 | Rejection of invention patent application after publication |