CN107274432A - A kind of common scene intelligent video monitoring method of view-based access control model conspicuousness and depth own coding - Google Patents
A kind of common scene intelligent video monitoring method of view-based access control model conspicuousness and depth own coding Download PDFInfo
- Publication number
- CN107274432A CN107274432A CN201710434834.7A CN201710434834A CN107274432A CN 107274432 A CN107274432 A CN 107274432A CN 201710434834 A CN201710434834 A CN 201710434834A CN 107274432 A CN107274432 A CN 107274432A
- Authority
- CN
- China
- Prior art keywords
- network
- input
- encoder
- depth
- sample
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/254—Analysis of motion involving subtraction of images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The present invention relates to the common scene intelligent video monitoring method of a kind of view-based access control model conspicuousness and depth own coding, including：Single frames decomposition is carried out to video, movable information is extracted using vision significance, then the light stream of consecutive frame moving object is calculated, detection process afterwards is divided into two processes of training and test, in training, the input of own coding is used as using the light stream of training sample, whole autoencoder network is trained by minimizing loss function, in test phase, respectively to train the light stream with test sample to be used as input, extract the encoder in the autoencoder network trained, the feature of input is extracted by dimensionality reduction, then the result after dimensionality reduction is visualized, the visualization scope of training sample is represented using suprasphere, in input test sample, visualized using same method, if the visual result of sample is fallen into the range of suprasphere, then judge that sample is normal；Conversely, falling outside suprasphere scope, judge that sample is abnormal, be achieved in the intelligent monitoring of video.
Description
Technical field
The present invention relates to image processing techniques, more particularly to a kind of view-based access control model conspicuousness and depth own coding it is public
Scene intelligent video frequency monitoring method.
Background technology
In recent years, monitoring device was used in all trades and professions, and the common scene such as modern airport, station, hospital is covered into
Thousand supervising devices up to ten thousand, because video data is numerous, depends merely on Security Personnel's analysis, filter out the normal behaviour under normal scene,
Note abnormalities behavior in time, be a very big workload, and increasing with analysis quantity, the notice of personnel and work are imitated
Rate can produce obvious decline, in order to which people is freed from substantial amounts of analysis and understanding, study a kind of intelligent video monitoring side
Method is significant.
Intelligent monitor system relates generally to the content of three parts：The extraction of movable information in video, that is, extract in video
Moving target, because monitoring system is fixed, so this part is mainly the movable information for extracting foreground target in video；OK
A major challenge in the extraction being characterized, intelligent monitor system, the features such as having uniqueness, robustness, extracts feature；It is abnormal
Behavioral value, is divided into rule-based detection, and whether such as detection target violates some predefined rules, and based on statistics
Detection, i.e., find the pattern of behavior, use pattern knows method for distinguishing and model carries out abnormal behaviour identification in substantial amounts of sample.
Second of existing technology multidigit, use pattern knows method for distinguishing to be identified, but this method precision compares deep learning
Method precision it is low, therefore service precision of the present invention is higher carries out abnormal row based on depth autoencoder network in deep learning
For identification.
The content of the invention
In view of this, it is a primary object of the present invention to provide, a kind of accuracy of detection is high, strong robustness view-based access control model shows
The common scene intelligent video monitoring method of work property and depth own coding, substantially increases accuracy of detection, meanwhile, it can tackle a variety of
Abnormal behaviour identification under scene, robustness is very strong.
In order to achieve the above object, technical scheme proposed by the present invention is：A kind of view-based access control model conspicuousness and depth are self-editing
The common scene intelligent video monitoring method of code, realizes that step is as follows：
Step 1, the video read under common scene, resolve into single frame by video, are then based on difference of Gaussian assemblage zone
Bandpass filter, calculates the vision significance figure of each frame, movable information is extracted with this；
Step 2, on the basis of every frame Saliency maps, calculate consecutive frame light stream, so as to extract the motion of foreground target
Information, obtains motion feature；
Step 3, in the algorithm of anomalous identification comprising train and test two processes, in the training process, calculate training
The vision significance figure of sample simultaneously extracts motion feature, and obtained Optical-flow Feature is converted to column vector as depth own coding net
The input of network, using the dimensionality reduction and the restructuration of decoder of encoder in depth autoencoder network, by minimizing loss letter
Number rebuilds input, trains depth autoencoder network；
Step 4, rebuild after input, training depth autoencoder network by minimizing loss function, extract the depth that trains
The encoder section of autoencoder network is spent as the network in test process, and the aobvious of training sample and test sample is calculated respectively
After work property figure and motion feature, using the Optical-flow Feature of each sample as the input of encoder in depth autoencoder network, pass through
The dimensionality reduction operation of the encoder network, the low-dimensional feature of input can most be represented by being extracted with low-dimensional vector；
Step 5, in three-dimensional coordinate during visual testing encoder network result, represent it with a suprasphere
Distribution after middle training sample dimensionality reduction；
Step 6, the anomalous identification for input test sample, if the visual scope of test sample falls into the model of suprasphere
In enclosing, then judge the test sample as normal sequence；Conversely, falling beyond suprasphere scope, then judge the test sample to be different
Chang Xulie, is achieved in the intelligent monitoring of video under the identification of abnormal behaviour, common scene.
The method of vision significance figure is as follows in the step 1：
Step i) is for a two field picture, and the significance each put in image is defined as：
S (x, y)=| | Iμ-Iwhc(x,y)||
Wherein, IμThe average of input picture each pixel color in Lab space, Iwhc(x, y) is to carry out height to image
After this is fuzzy, each pixel is in the value of Lab space, and S (x, y) represents the significance of each pixel, be the Euclidean of the two away from
From；
Step ii) Gaussian Blur is carried out to image first, two-dimentional gauss of distribution function is：
Wherein, x and y respectively correspond to central point around 8 points transverse and longitudinal coordinate, σ be gauss of distribution function variance, G (x,
Y) it is the fog-level of each pixel；
For coloured image, in R, G, tri- passages of B do convolution operation using Gaussian kernel and original image respectively, will be each
The result of passage merges, as the image after Gaussian Blur, and it is empty that the image and original image after Gaussian Blur are transformed into Lab respectively
Between
Step iii) calculate Gaussian Blur after each pixel of image Lab values IwhcEach pixel of (x, y) and original image
In the average I of Lab space colorμ, the Euclidean distance of the two is calculated, that is, obtains the vision significance figure of original image.
The step 3 using training depth autoencoder network detailed process be：
Normal sample is only included in step i) training samples, in the training process, the adjacent two field picture of training sample is calculated
Optical-flow Feature, column vector is converted to by Optical-flow Feature, as the input of depth autoencoder network, and own coding is one to connect entirely
Mode, allows output as closely as possible equal to the network that the structure of input is input layer-hidden layer-output layer, whole network is by left-half
Encoder and right half part decoder composition, encoder be used for Data Dimensionality Reduction, extraction can most represent the characteristic information of input；
Decoder is with error as small as possible, using the output of encoder as the input of decoder, rebuilds being originally inputted for whole network,
Depth autoencoder network is on the basis of autoencoder network, to be added in encoder network and decoder network several hidden
Layer；
Step ii) using light stream as input X={ x1,x2...xn, the activation primitive of network uses ReLU function f (x)=max
(0, x), wherein, x is the input of activation primitive, i.e. independent variable, and f (x) is the dependent variable of activation primitive, and the first half of network is
Encoder network is output as：Z=f (wX+b), wherein, w is the weight of encoder network, and b is the biasing of encoder network, Z
It is the result after X dimensionality reductions for the output of encoder network, i.e. Z, X characteristic information can be represented；The latter half of network is decoding
Device is output as：Y=f (w'Z+b'), wherein, w' is the weight of decoder network, and b' is the biasing of decoder network, i.e. Y is X
Reconstruction, whole encoder network is expressed as with formulating：Y=f (w'(f (wX+b))+b').
Step iii) loss function uses mean square error：MSE=| | X-Y | |2=| | X-f (w'(f (wX+b))+b') | |,
And minimize loss function to rebuild input, it is exactly the training process by depth autoencoder network, makes mean square error minimum, this
When output be input reconstruction.
The step 4 extracts the encoder section of the depth autoencoder network trained as the network mistake in test process
Cheng Wei：
First, the pretreatment of image is similar with training process by step i), with training sample and the Optical-flow Feature of test sample
The column vector of conversion as network input；
Step ii) it is different from network used in training process, the training that training process is obtained is extracted in test process
Encoder in good depth autoencoder network is acted on as the network of test process using the dimensionality reduction of encoder network, will be defeated
Enter 3 neurons of boil down to, from encoder the characteristics of, these three neurons can comprising input full detail.
In summary, the common scene intelligent video of a kind of view-based access control model conspicuousness and depth own coding of the present invention
Monitoring method, including：Single frames decomposition is carried out to the video under common scene, it is aobvious using vision in the frame of video come is decomposited
Work property extracts movable information, then calculates the light stream of consecutive frame moving object, includes the size and Orientation of movement velocity, afterwards
Detection process is divided into two processes of training and test, in training, using the light stream of training sample as the input of depth own coding,
Entire depth autoencoder network is trained by minimizing loss function, in test phase, respectively with training and test sample
Light stream extracts the encoder in the depth autoencoder network trained as input, is carried by the dimensionality reduction effect of encoder network
The feature of input is taken, according to the characteristic of encoder network, the feature after dimensionality reduction can represent the full detail of input, then visualize
Result after dimensionality reduction, the visualization scope of training sample is represented using suprasphere, in input test sample, uses same side
Method is visualized, if the visual result of sample is fallen into the range of suprasphere, judges that sample is normal；Conversely, falling in suprasphere model
Outside enclosing, judge that sample is abnormal, be achieved in the intelligent monitoring of video.
The advantage of the present invention compared with prior art is：
(1) present invention tentatively extracts movable information based on the identification of abnormal behaviour using vision significance and optical flow method,
Then feature is extracted using the depth self-encoding encoder in deep learning, is trained and detects, because depth self-encoding encoder can be with
Minimize loss function and rebuild input, the dimensionality reduction effect of encoder can extract the low-dimensional feature that can represent to input information, so carrying
The feature taken has very strong robustness, and just because of the robustness of feature, can efficiently carry out very much the knowledge of abnormal behaviour
Not, arithmetic accuracy is improved.Due to representing normal scope using suprasphere, when carrying out abnormal differentiation, it is only necessary to which judgement can
Scope depending on changing result, so judging that speed is fast.
(2) the features such as present invention has accuracy of detection high, strong robustness, can be widely applied to Community Safety protection, hospital,
The safeguard protection of the common scenes such as bank.By using the depth autoencoder network in optical flow method and deep learning, extracting can table
Show the low-dimensional feature of object full detail, judge accurate, strong robustness, due to representing normal scope using suprasphere, entering
During the differentiation of row exception, it is only necessary to the scope of visualization result is judged, so judging that speed is fast.
Brief description of the drawings
Fig. 1 is implementation process figure of the present invention.
Embodiment
It is right below in conjunction with the accompanying drawings and the specific embodiments to make the object, technical solutions and advantages of the present invention clearer
The present invention is described in further detail.
The common scene intelligent video monitoring method of a kind of view-based access control model conspicuousness and depth own coding of the present invention,
Including：Single frames decomposition is carried out to video under common scene, in the frame of video come is decomposited, is extracted and moved using vision significance
Information, then calculates the light stream of consecutive frame moving object, includes the size and Orientation of movement velocity, detection process afterwards is divided into
Two processes of training and test, in training, using the light stream of training sample as the input of depth own coding, are damaged by minimizing
Function is lost to train entire depth autoencoder network, in test phase, respectively to train the light stream with test sample as input,
The encoder in the depth autoencoder network trained is extracted, the feature of input is extracted by the dimensionality reduction effect of encoder network,
According to the characteristic of encoder network, the feature after dimensionality reduction can represent the full detail of input, then visualize the result after dimensionality reduction,
The visualization scope of training sample is represented using suprasphere, in input test sample, is visualized using same method, if sample
This visual result is fallen into the range of suprasphere, then judges that sample is normal；Conversely, falling outside suprasphere scope, sample is judged
This exception, is achieved in the intelligent monitoring of video.
As shown in figure 1, the present invention is implemented as follows step：
Step 1), read common scene under video, video is resolved into single frame, difference of Gaussian assemblage zone is then based on
Bandpass filter, calculates the vision significance figure of each frame, movable information is extracted with this；
Step 2), on the basis of every frame Saliency maps, calculate consecutive frame light stream, so as to extract the motion of foreground target
Information, obtains motion feature；
Step 3), in the algorithm of anomalous identification comprising train and test two processes, in the training process, calculate training
The vision significance figure of sample simultaneously extracts motion feature, and the Optical-flow Feature of obtained every two field picture is converted into column vector as depth
The input of autoencoder network is spent, using the reconstruction of the dimensionality reduction and decoder of encoder in depth autoencoder network, by minimizing
Loss function rebuilds input, trains depth autoencoder network；
Step 4), rebuild after input, training depth autoencoder network by minimizing loss function, extract the depth that trains
The encoder section of autoencoder network is spent as the network in test process, and the aobvious of training sample and test sample is calculated respectively
After work property figure and motion feature, using the Optical-flow Feature of each picture frame sample as the input of encoder in depth autoencoder network,
Operated by the dimensionality reduction of the encoder network, the low-dimensional feature of input can most be represented by being extracted with low-dimensional vector；
Step 5), in three-dimensional coordinate during visual testing encoder network result, represented with a suprasphere
Distribution wherein after training sample dimensionality reduction；
Step 6), for the anomalous identification of input test sample, if the visual scope of test sample falls into suprasphere
In the range of, then judge the test sample as normal sequence；Conversely, fall beyond suprasphere scope, then judge the test sample as
Unusual sequences, are achieved in the intelligent monitoring of video under the identification of abnormal behaviour, common scene.
The step 1) in vision significance figure computational methods it is as follows：
Step i) is for a two field picture, and the significance each put in image is defined as：
S (x, y)=| | Iμ-Iwhc(x,y)||
Wherein, IμThe average of input picture each pixel color in Lab space, Iwhc(x, y) is to carry out height to image
After this is fuzzy, each pixel is in the value of Lab space, and S (x, y) represents the significance of each pixel, be the Euclidean of the two away from
From；
Step ii) Gaussian Blur is carried out to image first, two-dimentional gauss of distribution function is：
Wherein, x and y respectively correspond to central point around 8 points transverse and longitudinal coordinate, σ be gauss of distribution function variance, G (x,
Y) it is the fog-level of each pixel；
In R, G, tri- passages of B do convolution operation using Gaussian kernel and original image respectively, and the result of each passage is closed
And, the image and original image after Gaussian Blur are transformed into Lab space by the as image after Gaussian Blur respectively.
Step iii) calculate Gaussian Blur after each pixel of image Lab values IwhcEach pixel of (x, y) and original image
In the average I of Lab space colorμ, the Euclidean distance of the two is calculated, that is, obtains the vision significance figure of original image.
The step 3) train depth autoencoder network principle as follows：
Normal sample is only included in step i) training samples, in the training process, the adjacent two field picture of training sample is calculated
Optical-flow Feature, column vector is converted to by Optical-flow Feature, as the input of depth autoencoder network, and own coding is one to connect entirely
Mode, allows output as closely as possible equal to the network that the structure of input is input layer-hidden layer-output layer, whole network by encoder and
Decoder is constituted, and encoder is used for Data Dimensionality Reduction, and extraction can most represent the characteristic information of input；Decoder is with mistake as small as possible
Difference, using the output of encoder as the input of decoder, rebuilds being originally inputted for whole network, depth autoencoder network is certainly
On the basis of coding network, several hidden layers are added in encoder network and decoder network；
Step ii) using light stream as input X={ x1,x2...xn, the activation primitive of network uses ReLU function f (x)=max
(0, x), wherein, x is the input of activation primitive, i.e. independent variable, and f (x) is the dependent variable of activation primitive, and the first half of network is
Encoder network is output as：Z=f (wX+b), wherein, w is the weight of encoder network, and b is the biasing of encoder network, Z
It is the result after X dimensionality reductions for the output of encoder network, i.e. Z, X characteristic information can be represented；The latter half of network is decoding
Device is output as：Y=f (w'Z+b'), wherein, w' is the weight of decoder network, and b' is the biasing of decoder network, i.e. Y is X
Reconstruction, whole encoder network is expressed as with formulating：Y=f (w'(f (wX+b))+b').
Step iii) loss function uses mean square error：MSE=| | X-Y | |2=| | X-f (w'(f (wX+b))+b') | |,
And minimize loss function to rebuild input, it is exactly the training process by depth autoencoder network, makes mean square error minimum, this
When output be input reconstruction.
The step 4) encoder section of the depth autoencoder network trained is extracted as the network in test process
Detailed process is：
First, the pretreatment of image is similar with training process by step i), with training sample and the Optical-flow Feature of test sample
The column vector of conversion as network input；
Step ii) it is different from network used in training process, the training that training process is obtained is extracted in test process
Encoder in good depth autoencoder network is acted on as the network of test process using the dimensionality reduction of encoder network, will be defeated
Enter 3 neurons of boil down to, 3 neurons can include the full detail of input.
In summary, presently preferred embodiments of the present invention is these are only, is not intended to limit the scope of the present invention.
Within the spirit and principles of the invention, any modification, equivalent substitution and improvements made etc., should be included in the present invention's
Within protection domain.
Claims (4)
1. the common scene intelligent video monitoring method of a kind of view-based access control model conspicuousness and depth own coding, it is characterised in that real
Existing step is as follows：
Step 1, the video read under common scene, single frame is resolved into by video, is then based on difference of Gaussian combination band logical filter
Ripple device, calculates the vision significance figure of each frame, movable information is extracted with this；
Step 2, on the basis of every frame Saliency maps, calculate the light stream of consecutive frame, so as to extract the movable information of foreground target,
Obtain motion feature；
Step 3, in the algorithm of anomalous identification comprising train and test two processes, in the training process, calculate training sample
Vision significance figure and extract motion feature, obtained Optical-flow Feature is converted to column vector as depth autoencoder network
Input, using the dimensionality reduction and the restructuration of decoder of encoder in depth autoencoder network, by minimizing loss function weight
Input is built, depth autoencoder network is trained；
Step 4, rebuild after input, training depth autoencoder network by minimizing loss function, extract the depth that trains from
The encoder section of coding network calculates the conspicuousness of training sample and test sample respectively as the network in test process
After figure and motion feature, using the Optical-flow Feature of each sample as the input of encoder in autoencoder network, by described self-editing
The dimensionality reduction operation of encoder in code network, the low-dimensional feature of input can most be represented by being extracted with low-dimensional vector；
Step 5, in three-dimensional coordinate during visual testing encoder network result, represent wherein to instruct with a suprasphere
Practice the distribution after sample dimensionality reduction；
Step 6, the anomalous identification for input test sample, if the visual scope of test sample falls into the scope of suprasphere
It is interior, then judge the test sample as normal sequence；Conversely, falling beyond suprasphere scope, then judge the test sample as exception
Sequence, is achieved in the intelligent monitoring of video under the identification of abnormal behaviour, common scene.
2. the common scene intelligent video monitoring of a kind of view-based access control model conspicuousness according to claim 1 and depth own coding
Method, it is characterised in that：In the step 1, the method for calculating the vision significance figure of each frame is as follows：
Step i) is for a two field picture, and the significance each put in image is：
S (x, y)=| | Iμ-Iwhc(x,y)||
Wherein, IμThe average of input picture each pixel color in Lab space, Iwhc(x, y) is to carry out Gaussian mode to image
After paste, each pixel is in the value of Lab space, and S (x, y) represents the significance of each pixel, is the Euclidean distance of the two；
Step ii) Gaussian Blur is carried out to image first, two-dimentional gauss of distribution function is：
<mrow>
<mi>G</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<mrow>
<mn>2</mn>
<msup>
<mi>&pi;&sigma;</mi>
<mn>2</mn>
</msup>
</mrow>
</mfrac>
<msup>
<mi>e</mi>
<mrow>
<mo>-</mo>
<mfrac>
<mrow>
<msup>
<mi>x</mi>
<mn>2</mn>
</msup>
<mo>+</mo>
<msup>
<mi>y</mi>
<mn>2</mn>
</msup>
</mrow>
<mrow>
<mn>2</mn>
<msup>
<mi>&sigma;</mi>
<mn>2</mn>
</msup>
</mrow>
</mfrac>
</mrow>
</msup>
</mrow>
Wherein, x and y correspond to the transverse and longitudinal coordinate of 8 points around central point respectively, and σ is the variance of gauss of distribution function, and G (x, y) is
The fog-level of each pixel；
In R, G, tri- passages of B do convolution operation using Gaussian kernel and original image respectively, the result of each passage are merged, i.e.,
For the image after Gaussian Blur；The image and original image after Gaussian Blur are transformed into Lab space respectively；
Step iii) calculate Gaussian Blur after each pixel of image Lab values IwhcEach pixel of (x, y) and original image exists
The average I of Lab space colorμ, the Euclidean distance of the two is calculated, that is, obtains the vision significance figure of original image.
3. the common scene intelligent video monitoring of a kind of view-based access control model conspicuousness according to claim 1 and depth own coding
Method, it is characterised in that：In the step 3, the process that implements of training depth autoencoder network is：
Normal sample is only included in step i) training samples, in the training process, the light stream of the adjacent two field picture of training sample is calculated
Feature, column vector is converted to by Optical-flow Feature, as the input of depth autoencoder network, and own coding is one with full connection side
Formula, allows output as closely as possible equal to the network that the structure of input is input layer-hidden layer-output layer, whole network is by left-half
The decoder composition of encoder and right half part, encoder is used for Data Dimensionality Reduction, and extraction can most represent the characteristic information of input；Solution
Code device is with error as small as possible, using the output of encoder as the input of decoder, rebuilds being originally inputted for whole network, deep
Degree autoencoder network is that on the basis of autoencoder network, it is hidden in encoder network and decoder network to add several
Layer；
Step ii) using light stream as input X={ x1,x2...xn, the activation primitive of network using ReLU function f (x)=max (0,
X), wherein, x is the input of activation primitive, i.e. independent variable, f (x) is the dependent variable of activation primitive, and the first half of network is to compile
Code device network is output as：Z=f (wX+b), wherein, w is the weight of encoder network, and b is the biasing of encoder network, and Z is
The output of encoder network, i.e. Z are the results after X dimensionality reductions, can represent X characteristic information；The latter half of network is decoder
It is output as：Y=f (w'Z+b'), wherein, w' is the weight of decoder network, and b' is the biasing of decoder network, i.e. Y is X
Rebuild, whole encoder network is expressed as with formulating：Y=f (w'(f (wX+b))+b')；
Step iii) loss function uses mean square error：MSE=| | X-Y | |2=| | X-f (w'(f (wX+b))+b') | |, and it is minimum
Change loss function and rebuild input, be exactly the training process by depth autoencoder network, make mean square error minimum, now defeated
It is the reconstruction of input to go out.
4. the common scene intelligent video monitoring of a kind of view-based access control model conspicuousness according to claim 1 and depth own coding
Method, it is characterised in that：In the step 4, the encoder section of depth autoencoder network that trains is extracted as testing
The detailed process of network in journey is：
First, the pretreatment of image is similar with training process by step i), is changed with training sample and the Optical-flow Feature of test sample
Column vector as network input；
Step ii) it is different from network used in training process, what what extraction training process was obtained in test process trained
Encoder in depth autoencoder network is acted on using the dimensionality reduction of encoder network as the network of test process, input is pressed
3 neurons are condensed to, 3 neurons can include the full detail of input.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710434834.7A CN107274432B (en) | 2017-06-10 | 2017-06-10 | A kind of intelligent video monitoring method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710434834.7A CN107274432B (en) | 2017-06-10 | 2017-06-10 | A kind of intelligent video monitoring method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107274432A true CN107274432A (en) | 2017-10-20 |
| CN107274432B CN107274432B (en) | 2019-07-26 |
Family
ID=60066526
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710434834.7A Active CN107274432B (en) | 2017-06-10 | 2017-06-10 | A kind of intelligent video monitoring method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107274432B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107833208A (en) * | 2017-10-27 | 2018-03-23 | 哈尔滨工业大学 | A kind of hyperspectral abnormity detection method based on changeable weight depth own coding |
| CN108830882A (en) * | 2018-05-25 | 2018-11-16 | 中国科学技术大学 | Video abnormal behaviour real-time detection method |
| CN110009866A (en) * | 2019-04-03 | 2019-07-12 | 天津中航亿达科技有限公司 | A kind of method of video detection temperature anomaly |
| CN111107107A (en) * | 2019-12-31 | 2020-05-05 | 奇安信科技集团股份有限公司 | Network behavior detection method and device, computer equipment and storage medium |
| CN113222883A (en) * | 2020-01-21 | 2021-08-06 | 四零四科技股份有限公司 | Apparatus and method for handling exception detection |
| CN113222926A (en) * | 2021-05-06 | 2021-08-06 | 西安电子科技大学 | Zipper abnormity detection method based on depth support vector data description model |
| CN113592390A (en) * | 2021-07-12 | 2021-11-02 | 嘉兴恒创电力集团有限公司博创物资分公司 | Warehousing digital twin method and system based on multi-sensor fusion |
| CN115714731A (en) * | 2022-09-27 | 2023-02-24 | 中国人民解放军63921部队 | Deep space measurement and control link abnormity detection method based on deep learning self-encoder |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106778921A (en) * | 2017-02-15 | 2017-05-31 | 张烜 | Personnel based on deep learning encoding model recognition methods again |
-
2017
- 2017-06-10 CN CN201710434834.7A patent/CN107274432B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106778921A (en) * | 2017-02-15 | 2017-05-31 | 张烜 | Personnel based on deep learning encoding model recognition methods again |
Non-Patent Citations (4)
| Title |
|---|
| PEI XU ETC.: ""Dynamic Background Learning through Deep Auto-encoder Networks"", 《PROCEEDINGS OF THE 22ND ACM INTERNATIONAL CONFERENCE ON MULTIMEDIA》 * |
| TRAVIS PORTZ ETC.: ""Optical flow in the presence of spatially-varying motion blur"", 《IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
| 凌南平: ""视觉显著性检测方法及其应用研究"", 《万方数据知识服务平台》 * |
| 尹宏鹏等: ""基于视觉的目标检测与跟踪综述"", 《自动化学报》 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107833208A (en) * | 2017-10-27 | 2018-03-23 | 哈尔滨工业大学 | A kind of hyperspectral abnormity detection method based on changeable weight depth own coding |
| CN107833208B (en) * | 2017-10-27 | 2021-07-13 | 哈尔滨工业大学 | Hyperspectral anomaly detection method based on dynamic weight depth self-encoding |
| CN108830882B (en) * | 2018-05-25 | 2022-05-17 | 中国科学技术大学 | Video abnormal behavior real-time detection method |
| CN108830882A (en) * | 2018-05-25 | 2018-11-16 | 中国科学技术大学 | Video abnormal behaviour real-time detection method |
| CN110009866A (en) * | 2019-04-03 | 2019-07-12 | 天津中航亿达科技有限公司 | A kind of method of video detection temperature anomaly |
| CN111107107A (en) * | 2019-12-31 | 2020-05-05 | 奇安信科技集团股份有限公司 | Network behavior detection method and device, computer equipment and storage medium |
| CN111107107B (en) * | 2019-12-31 | 2022-03-29 | 奇安信科技集团股份有限公司 | Network behavior detection method and device, computer equipment and storage medium |
| CN113222883A (en) * | 2020-01-21 | 2021-08-06 | 四零四科技股份有限公司 | Apparatus and method for handling exception detection |
| CN113222883B (en) * | 2020-01-21 | 2024-02-23 | 四零四科技股份有限公司 | Device and method for processing abnormality detection |
| CN113222926A (en) * | 2021-05-06 | 2021-08-06 | 西安电子科技大学 | Zipper abnormity detection method based on depth support vector data description model |
| CN113222926B (en) * | 2021-05-06 | 2023-04-18 | 西安电子科技大学 | Zipper abnormity detection method based on depth support vector data description model |
| CN113592390A (en) * | 2021-07-12 | 2021-11-02 | 嘉兴恒创电力集团有限公司博创物资分公司 | Warehousing digital twin method and system based on multi-sensor fusion |
| CN115714731A (en) * | 2022-09-27 | 2023-02-24 | 中国人民解放军63921部队 | Deep space measurement and control link abnormity detection method based on deep learning self-encoder |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107274432B (en) | 2019-07-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107274432B (en) | A kind of intelligent video monitoring method | |
| CN105956560B (en) | A kind of model recognizing method based on the multiple dimensioned depth convolution feature of pondization | |
| CN108805002B (en) | Monitoring video abnormal event detection method based on deep learning and dynamic clustering | |
| CN108549846B (en) | Pedestrian detection and statistics method combining motion characteristics and head-shoulder structure | |
| CN105046195B (en) | Human bodys' response method based on asymmetric generalized gaussian model | |
| CN107330364A (en) | A kind of people counting method and system based on cGAN networks | |
| CN107506740A (en) | A kind of Human bodys' response method based on Three dimensional convolution neutral net and transfer learning model | |
| CN109271886A (en) | A kind of the human body behavior analysis method and system of examination of education monitor video | |
| CN104281853A (en) | Behavior identification method based on 3D convolution neural network | |
| CN107272655A (en) | Batch process fault monitoring method based on multistage ICA SVDD | |
| CN104036243B (en) | A kind of Activity recognition method based on Optic flow information | |
| CN110232404A (en) | A kind of recognition methods of industrial products surface blemish and device based on machine learning | |
| CN109389057A (en) | A kind of object detecting method based on multiple dimensioned high-level semantics converged network | |
| CN107330360A (en) | A kind of pedestrian's clothing colour recognition, pedestrian retrieval method and device | |
| CN104298974A (en) | Human body behavior recognition method based on depth video sequence | |
| CN109740609A (en) | A kind of gauge detection method and device | |
| CN109145841A (en) | A kind of detection method and device of the anomalous event based on video monitoring | |
| CN107092884A (en) | Rapid coarse-fine cascade pedestrian detection method | |
| CN106971158A (en) | A kind of pedestrian detection method based on CoLBP symbiosis feature Yu GSS features | |
| CN106910204A (en) | A kind of method and system to the automatic Tracking Recognition of sea ship | |
| CN107392142A (en) | A kind of true and false face identification method and its device | |
| CN108229300A (en) | Video classification methods, device, computer readable storage medium and electronic equipment | |
| Yu et al. | Robust median filtering forensics by CNN-based multiple residuals learning | |
| CN105631405B (en) | Traffic video intelligent recognition background modeling method based on Multilevel Block | |
| CN104680189B (en) | Based on the bad image detecting method for improving bag of words |
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 |