CN106982359B - A kind of binocular video monitoring method, system and computer readable storage medium - Google Patents
A kind of binocular video monitoring method, system and computer readable storage medium Download PDFInfo
- Publication number
- CN106982359B CN106982359B CN201710286635.6A CN201710286635A CN106982359B CN 106982359 B CN106982359 B CN 106982359B CN 201710286635 A CN201710286635 A CN 201710286635A CN 106982359 B CN106982359 B CN 106982359B
- Authority
- CN
- China
- Prior art keywords
- convolutional neural
- neural networks
- video image
- terminal device
- tensorflow
- 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.)
- Active
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N7/00—Television systems
- H04N7/18—Closed-circuit television [CCTV] systems, i.e. systems in which the video signal is not broadcast
- H04N7/181—Closed-circuit television [CCTV] systems, i.e. systems in which the video signal is not broadcast for receiving images from a plurality of remote sources
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24137—Distances to cluster centroïds
- G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Signal Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to field of intelligent monitoring, a kind of binocular video monitoring method, system and computer readable storage medium are provided, to improve in video monitoring to the accuracy of target identification and efficiency.The described method includes: binocular camera acquires video image and is sent to terminal device；The first convolutional neural networks in video image input terminal equipment are carried out the preliminary identification of target by terminal device；If the result tentatively identified is in video image there are suspicious object, video image is uploaded to Cloud Server by terminal device；The second convolutional neural networks that video image inputs on Cloud Server are carried out again identifying that for target by Cloud Server；If the result again identified that is to determine in video image that the result that Cloud Server would again identify that is back to terminal device, is handled by terminal device is further there are suspicious object.The efficiency that technical solution provided by the invention identifies when on the one hand improving video monitoring；The accuracy rate identified when on the other hand improving video monitoring.
Description
Technical field
The invention belongs to field of intelligent monitoring more particularly to a kind of binocular video monitoring methods, system and computer-readable
Storage medium.
Background technique
Video monitoring system is commonly used in safety-security area, can effectively realize the early warning to illegal invasion, protects the people
The security of the lives and property, and the important supplementary means to keep the peace.In the image recognition of video monitoring system, the back of image
Scape complexity, ambient light and device pixel limitation cause large effect to imaging effect, in addition, video monitoring system needs
It is quickly and accurately identified in the case where identifying that target is blocked, the difficulty for designing algorithm increases.
Undoubtedly, the key of video monitoring is the extraction to clarification of objective to be identified, because target to be identified
It is characterized in that a target to be identified is different from the important logo of another target to be identified.In current video monitoring, to mesh
Target identification, algorithm often rely on the feature of artificial selection.
However, the data that video monitoring system is related to are often magnanimity, therefore, aforesaid way is difficult to from mass data
Study obtains an effective classifier, to improve the accuracy and efficiency to target identification.
Summary of the invention
The purpose of the present invention is to provide a kind of binocular video monitoring method, system and computer readable storage medium, with
Improve the accuracy and efficiency in video monitoring to target identification.
First aspect present invention provides a kind of binocular video monitoring method, which comprises
Binocular camera acquisition video image is simultaneously sent to terminal device；
The video image is inputted the first convolutional neural networks on the terminal device and carries out mesh by the terminal device
Target tentatively identifies；
If the result tentatively identified is there are suspicious object in the video image, the terminal device will be described
Video image is uploaded to Cloud Server；
The video image is inputted the second convolutional neural networks on the Cloud Server and carries out mesh by the Cloud Server
Target again identifies that；
If the result again identified that is to determine there are suspicious object in the video image, the Cloud Server will
The result again identified that is back to the terminal device, is handled by the terminal device is further.
Second aspect of the present invention provides a kind of binocular video monitoring system, and the system comprises binocular camera, terminals to set
Standby and Cloud Server, the terminal device include preliminary identification module and uploading module, and the Cloud Server includes again identifying that
Module and result return module；
The binocular camera, for acquiring video image and being sent to terminal device；
The preliminary identification module, for the video image to be inputted to the first convolution nerve net on the terminal device
The preliminary identification of network progress target；
The uploading module, if the result tentatively identified for the preliminary identification module is to deposit in the video image
In suspicious object, then the video image is uploaded to Cloud Server；
It is described to again identify that module, for the video image to be inputted to the second convolution nerve net on the Cloud Server
Network carries out again identifying that for target；
The result return module, if being to determine the video for the result again identified that for again identifying that module
There are suspicious objects in image, then the result again identified that are back to the terminal device, are done by the terminal device
Further processing.
Third aspect present invention provides a kind of binocular video monitoring system, including memory, processor and is stored in institute
The computer program that can be run in memory and on the processor is stated, the processor executes real when the computer program
Existing following steps:
Binocular camera acquisition video image is simultaneously sent to terminal device；
The video image is inputted the first convolutional neural networks on the terminal device and carries out mesh by the terminal device
Target tentatively identifies；
If the result tentatively identified is there are suspicious object in the video image, the terminal device will be described
Video image is uploaded to Cloud Server；
The video image is inputted the second convolutional neural networks on the Cloud Server and carries out mesh by the Cloud Server
Target again identifies that；
If the result again identified that is to determine there are suspicious object in the video image, the Cloud Server will
The result again identified that is back to the terminal device, is handled by the terminal device is further.
Fourth aspect present invention provides a kind of computer readable storage medium, and the computer-readable recording medium storage has
Computer program, the computer program perform the steps of when being executed by processor
Binocular camera acquisition video image is simultaneously sent to terminal device；
The video image is inputted the first convolutional neural networks on the terminal device and carries out mesh by the terminal device
Target tentatively identifies；
If the result tentatively identified is there are suspicious object in the video image, the terminal device will be described
Video image is uploaded to Cloud Server；
The video image is inputted the second convolutional neural networks on the Cloud Server and carries out mesh by the Cloud Server
Target again identifies that；
If the result again identified that is to determine there are suspicious object in the video image, the Cloud Server will
The result again identified that is back to the terminal device, is handled by the terminal device is further.
It was found from aforementioned present invention technical solution, on the one hand, due to the first convolutional neural networks and the second convolution nerve net
Network can be obtained by training in advance, therefore, can quickly be mentioned automatically when being identified using trained neural network to target
Clarification of objective is taken, the efficiency identified when improving video monitoring；On the other hand, suspicious object is just determined by identifying twice,
The accuracy rate identified when therefore improving video monitoring.
Detailed description of the invention
Fig. 1 is the implementation process schematic diagram for the intelligent video multi-point monitoring method that the embodiment of the present invention one provides；
Fig. 2 is the structural schematic diagram of intelligent video multipoint monitoring system provided by Embodiment 2 of the present invention；
Fig. 3 is the structural schematic diagram for the intelligent video multipoint monitoring system that the embodiment of the present invention three provides；
Fig. 4 is the structural schematic diagram for the intelligent video multipoint monitoring system that the embodiment of the present invention four provides.
Specific embodiment
In order to which the purpose of the present invention, technical solution and beneficial effect is more clearly understood, below in conjunction with attached drawing and implementation
Example, the present invention will be described in further detail.It should be appreciated that specific embodiment described herein is only used to explain this hair
It is bright, it is not intended to limit the present invention.
The embodiment of the present invention provides a kind of binocular video monitoring method, which comprises binocular camera acquires video
Image is simultaneously sent to terminal device；The video image is inputted the mind of the first convolution on the terminal device by the terminal device
The preliminary identification of target is carried out through network；If the result tentatively identified is in the video image there are suspicious object,
The video image is uploaded to Cloud Server by the terminal device；The video image is inputted the cloud by the Cloud Server
The second convolutional neural networks on server carry out again identifying that for target；If the result again identified that is to determine the view
There are suspicious objects in frequency image, then the result again identified that is back to the terminal device by the Cloud Server, by
The terminal device is further to be handled.The embodiment of the present invention also provides corresponding binocular video monitoring system and computer can
Read storage medium.It is described in detail separately below.
Attached drawing 1 is please referred to, is the implementation process schematic diagram for the binocular video monitoring method that the embodiment of the present invention one provides, it is main
Include the following steps S101 to step S105, detailed description are as follows:
S101, binocular camera acquisition video image are simultaneously sent to terminal device.
In embodiments of the present invention, binocular camera, two for imitating the mankind are used using the headend equipment of video monitoring
Eyes acquire video image and are sent to terminal device.
The first convolutional neural networks in video image input terminal equipment are carried out the preliminary of target by S102, terminal device
Identification.
In embodiments of the present invention, terminal device can be a microcomputer, for example, raspberry pie 3B type microcomputer,
It is deployed with the first convolutional neural networks thereon.As an embodiment of the present invention, the first convolutional neural networks, which can be, is based on
The small-sized convolutional neural networks of TensorFlow, wherein TensorFlow is a Machine learning tools, and great advantage is not
Advanced mathematical model and optimization algorithm needed for needing user to grasp deployment depth neural network, to greatly drag down engineering
The threshold of habit.It is then artificial neural network as convolutional neural networks (CNN, Convolutional Neural Networks)
One kind, it has also become the research hotspot of current speech analysis and field of image recognition, its weight are shared network structure and are allowed to more
Similar to biological neural network, the complexity of network model is reduced, reduces the quantity of weight.Based on the small of TensorFlow
Type convolutional neural networks refer to the lesser convolutional neural networks of a kind of scale obtained using TensorFlow.
S103, if the result tentatively identified is in video image there are suspicious object, terminal device will be on video image
Reach Cloud Server.
The second convolutional neural networks that video image inputs on Cloud Server are carried out target again by S104, Cloud Server
Identification.
In embodiments of the present invention, the second convolutional neural networks can be the large-scale convolutional Neural net based on TensorFlow
Network, refer in TensorFlow and convolutional neural networks concept and previous embodiment therein based on the small-sized of TensorFlow
The TensorFlow of convolutional neural networks is identical with convolutional neural networks concept, the difference is that based on the big of TensorFlow
Type convolutional neural networks will be far longer than the small-sized convolutional neural networks based on TensorFlow in scale or level.
It should be noted that in embodiments of the present invention, video image can be acquired in binocular camera and be sent to end
Before end equipment, according to the convolutional neural networks model based on TensorFlow, the small-sized convolution based on TensorFlow is obtained
Neural network and large-scale convolutional neural networks based on TensorFlow.Specifically, according to the convolution mind based on TensorFlow
Through network model, the small-sized convolutional neural networks based on TensorFlow and the large-scale convolutional Neural based on TensorFlow are obtained
Network can be realized with S1 as follows and S2:
S1 builds the convolutional neural networks model based on TensorFlow.
The convolutional neural networks model based on TensorFlow includes the first convolutional layer, the first maximum pond layer, first
Local acknowledgement's normalization layer, the second convolutional layer, the second local acknowledgement normalization layer, the first full connection linearly activated based on amendment
Layer, the second full articulamentum and softmax_linear linearly activated based on amendment, wherein first convolutional layer is for realizing volume
Long-pending and rectified linear activation, can be used a filter (i.e. convolution kernel) specifically to filter video
Each zonule of image, to obtain the characteristic value of these zonules, during hands-on, the value of convolution kernel is to learn
It is acquired during practising；First maximum pond layer (max pooling) is a kind of down-sampled operation, which is in each spy
In fixed zonule, maximum value can be chosen as output valve；First partial response normalization layer is returned for realizing local acknowledgement
One changes；Second convolutional layer is also for realizing convolution sum rectified linear activation；Second local acknowledgement's normalizing
Changing layer is also for realizing local acknowledgement's normalization；Second maximum pond layer (max pooling) is a kind of down-sampled operation, should
Operation is can to choose maximum value as output valve in each specific zonule；Above layers are based on eventually by first
It corrects the full articulamentum linearly activated and second and is connected to softmax_linear based on the full articulamentum linearly activated is corrected,
Softmax_linear is substantially a softmax classifier, for carrying out linear transformation to export logits.
In embodiments of the present invention, the convolutional neural networks model based on TensorFlow can carry out the convolution of N-dimensional classification
The method that neural network uses is multinomial logistic regression, and is called softmax recurrence, and Softmax is returned in the defeated of network
It attached a softmax nonlinearity on layer out, and calculate normalized predicted value and the 1-hot of label
The cross entropy of encoding, during regularization, we can be based on all Variable Learning application weight attenuation losses
The objective function of the convolutional neural networks model of TensorFlow be ask intersect entropy loss and all weight attenuation terms and.
S2, using the TensorFlow cluster server of cloud computing to the convolutional neural networks model based on TensorFlow
It is trained, to obtain the small-sized convolutional neural networks based on TensorFlow and the large-scale convolutional Neural based on TensorFlow
Network.
After convolutional neural networks model buildings based on TensorFlow are good, the TensorFlow collection of cloud computing can use
Group's server is trained the convolutional neural networks model based on TensorFlow, to obtain based on the small-sized of TensorFlow
Convolutional neural networks and large-scale convolutional neural networks based on TensorFlow.The TensorFlow cluster server of cloud computing is held
The figure of a series of task of row, these task executions TensorFlow calculates, and each task can be associated with the one of TensorFlow
A service, the service are calculated for creating TensorFlow session and execution figure.The TensorFlow cluster server of cloud computing
One or more operations can also be divided, each operation may include one or more tasks.In cloud computing
In TensorFlow cluster server, a usual task run on one machine, if the machine supports more GPU equipment,
Multiple tasks can be run on this machine, run in which GPU equipment by application program controlling task；Common depth
Habit training pattern is data parallel, i.e. TensorFlow task is enterprising in different small lot data sets using identical model
Row training, then on parameter server more new model shared parameter.
Since the TensorFlow cluster server of cloud computing has the advantages that distributed system, utilize cloud computing
TensorFlow cluster server the convolutional neural networks model based on TensorFlow is trained, can not only shorten
The time of convolutional neural networks model of the training based on TensorFlow, and the convolutional neural networks model after training exists
Accuracy rate is also higher when video identification.
S105, if the result again identified that is to determine in video image that Cloud Server will be known again there are suspicious object
Other result is back to terminal device, is handled by terminal device is further.
The result again identified that terminal device can be returned according to Cloud Server is further to be handled, for example, if can
Doubtful target is dangerous person, then sounds an alarm immediately, reminds Security Personnel.
It was found from the exemplary binocular video monitoring method of above-mentioned attached drawing 1, on the one hand, due to the first convolutional neural networks and
Two convolutional neural networks can be obtained by training in advance, therefore, when being identified using trained neural network to target
The automatic rapidly extracting clarification of objective of energy, the efficiency identified when improving video monitoring；On the other hand, by identifying twice just really
The accuracy rate identified when determining suspicious object, therefore improving video monitoring.
Attached drawing 2 is please referred to, is the structural schematic diagram of binocular video monitoring system provided by Embodiment 2 of the present invention.In order to just
In explanation, only parts related to embodiments of the present invention are shown for attached drawing 2.The exemplary binocular video monitoring system of attached drawing 2 is main
Including binocular camera 201, terminal device 202 and Cloud Server 203, terminal device 201 includes preliminary identification module 204 and upper
Transmission module 205, Cloud Server 203 include again identifying that module 206 and result return module 207, in which:
Binocular camera 201, for acquiring video image and being sent to terminal device 202；
Preliminary identification module 204, for carrying out the first convolutional neural networks in video image input terminal equipment 202
The preliminary identification of target；
Uploading module 205, if the result tentatively identified for preliminary identification module 204 is that there are suspicious in video image
Video image is then uploaded to Cloud Server 203 by target；
Again identify that module 206, the second convolutional neural networks for inputting video image on Cloud Server 203 carry out
Target again identifies that；
Result return module 207, if for again identifying that the result of module 206 again identified that in determining video image
There are suspicious object, then the result that would again identify that is back to terminal device 202, is handled by terminal device 202 is further.
In the exemplary system of above-mentioned attached drawing 2, the first convolutional neural networks can be the small-sized convolution based on TensorFlow
Neural network, the second convolutional neural networks can be the large-scale convolutional neural networks based on TensorFlow.
The exemplary binocular video monitoring system of attached drawing 2 can also include obtaining module 301, and the present invention is real as shown in Fig. 3
The binocular video monitoring system of the offer of example three is provided.Module 301 is obtained to acquire video image for binocular camera 201 and be sent to
Before terminal device 202, according to the convolutional neural networks model based on TensorFlow, obtain based on the small-sized of TensorFlow
Convolutional neural networks and large-scale convolutional neural networks based on TensorFlow.
The exemplary acquisition module 301 of attached drawing 3 can also include model buildings unit 401 and training unit 402, such as attached drawing 4
The intelligent video multipoint monitoring system that the shown embodiment of the present invention four provides, in which:
Model buildings unit 401, for building the convolutional neural networks model based on TensorFlow, wherein be based on
The convolutional neural networks model of TensorFlow includes the first convolutional layer, the first maximum pond layer, first partial response normalization
Full articulamentum that layer, the second convolutional layer, the second local acknowledgement normalization layer, first are linearly activated based on amendment, second are based on repairing
The full articulamentum and softmax_linear of linear positive activation；
Training unit 402, for the TensorFlow cluster server using cloud computing to the volume based on TensorFlow
Product neural network model is trained, to obtain the small-sized convolutional neural networks based on TensorFlow and be based on TensorFlow
Large-scale convolutional neural networks.
It should be noted that the contents such as information exchange, implementation procedure between each module/unit of above-mentioned apparatus, due to
Embodiment of the present invention method is based on same design, and bring technical effect is identical as embodiment of the present invention method, particular content
It can be found in the narration in embodiment of the present invention method, details are not described herein again.
The embodiment of the present invention also provides a kind of binocular video monitoring system, including memory, processor and is stored in
In reservoir and the computer program that can run on a processor, the processor perform the steps of double when executing computer program
Mesh camera acquisition video image is simultaneously sent to terminal device；Terminal device is by the first volume in video image input terminal equipment
Product neural network carries out the preliminary identification of target；If the result tentatively identified is that there are suspicious object, terminals in video image
Video image is uploaded to Cloud Server by equipment；Video image is inputted the second convolution nerve net on Cloud Server by Cloud Server
Network carries out again identifying that for target；If the result again identified that is to determine that there are suspicious object, Cloud Servers in video image
The result that would again identify that is back to terminal device, is handled by terminal device is further.
The embodiment of the present invention also provides a kind of computer readable storage medium, which has meter
Calculation machine program performs the steps of binocular camera acquisition video image and is sent to when computer program is executed by processor
Terminal device；The first convolutional neural networks in video image input terminal equipment are carried out the preliminary knowledge of target by terminal device
Not；If the result tentatively identified is in video image there are suspicious object, video image is uploaded to cloud service by terminal device
Device；The second convolutional neural networks that video image inputs on Cloud Server are carried out again identifying that for target by Cloud Server；If again
The result of secondary identification is to determine that there are suspicious objects in video image, then the result that Cloud Server would again identify that is back to terminal
Equipment is handled by terminal device is further.
It is apparent to those skilled in the art that for convenience of description and succinctly, only with above-mentioned each function
Can unit, module division progress for example, in practical application, can according to need and by above-mentioned function distribution by different
Functional unit, module are completed, i.e., the internal structure of described device is divided into different functional unit or module, more than completing
The all or part of function of description.Each functional unit in embodiment, module can integrate in one processing unit, can also
To be that each unit physically exists alone, can also be integrated in one unit with two or more units, it is above-mentioned integrated
Unit both can take the form of hardware realization, can also realize in the form of software functional units.In addition, each function list
Member, the specific name of module are also only for convenience of distinguishing each other, the protection scope being not intended to limit this application.Above system
The specific work process of middle unit, module, can refer to corresponding processes in the foregoing method embodiment, and details are not described herein.
In the above-described embodiments, it all emphasizes particularly on different fields to the description of each embodiment, is not described in detail or remembers in some embodiment
The part of load may refer to the associated description of other embodiments.
Those of ordinary skill in the art may be aware that list described in conjunction with the examples disclosed in the embodiments of the present disclosure
Member and algorithm steps can be realized with the combination of electronic hardware or computer software and electronic hardware.These functions are actually
It is implemented in hardware or software, the specific application and design constraint depending on technical solution.Professional technician
Each specific application can be used different methods to achieve the described function, but this realization is it is not considered that exceed
The scope of the present invention.
In embodiment provided by the present invention, it should be understood that disclosed device and method can pass through others
Mode is realized.For example, system embodiment described above is only schematical, for example, the division of the module or unit,
Only a kind of logical function partition, there may be another division manner in actual implementation, such as multiple units or components can be with
In conjunction with or be desirably integrated into another system, or some features can be ignored or not executed.Another point, it is shown or discussed
Mutual coupling or direct-coupling or communication connection can be through some interfaces, the INDIRECT COUPLING of device or unit or
Communication connection can be electrical property, mechanical or other forms.
The unit as illustrated by the separation member may or may not be physically separated, aobvious as unit
The component shown may or may not be physical unit, it can and it is in one place, or may be distributed over multiple
In network unit.It can select some or all of unit therein according to the actual needs to realize the mesh of this embodiment scheme
's.
It, can also be in addition, the functional units in various embodiments of the present invention may be integrated into one processing unit
It is that each unit physically exists alone, can also be integrated in one unit with two or more units.Above-mentioned integrated list
Member both can take the form of hardware realization, can also realize in the form of software functional units.
If the integrated unit is realized in the form of SFU software functional unit and sells or use as independent product
When, it can store in a computer readable storage medium.Based on this understanding, the technical solution of the embodiment of the present invention
Substantially all or part of the part that contributes to existing technology or the technical solution can be with software product in other words
Form embody, which is stored in a storage medium, including some instructions use so that one
Computer equipment (can be personal computer, server or the network equipment etc.) or processor (processor) execute this hair
The all or part of the steps of bright each embodiment the method for embodiment.And storage medium above-mentioned include: USB flash disk, mobile hard disk,
Read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic
The various media that can store program code such as dish or CD.
Embodiment described above is merely illustrative of the technical solution of the present invention, rather than its limitations；Although referring to aforementioned reality
Applying example, invention is explained in detail, those skilled in the art should understand that: it still can be to aforementioned each
Technical solution documented by embodiment is modified or equivalent replacement of some of the technical features；And these are modified
Or replacement, the spirit and scope for technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution should all
It is included within protection scope of the present invention.
Claims (10)
1. a kind of binocular video monitoring method, which is characterized in that the described method includes:
Binocular camera acquisition video image is simultaneously sent to terminal device；
The video image is inputted the first convolutional neural networks on the terminal device and carries out target by the terminal device
Preliminary identification；
If the result tentatively identified is in the video image there are suspicious object, the terminal device is by the video
Image is uploaded to Cloud Server；
The video image is inputted the second convolutional neural networks on the Cloud Server and carries out target by the Cloud Server
It again identifies that, first convolutional neural networks and nervus opticus network use identical model via distributed type assemblies server
Acquisition is trained on different small lot data sets, second convolutional neural networks are greater than described in scale or level
First convolutional neural networks；
If the result again identified that is to determine there are suspicious object in the video image, the Cloud Server will be described
The result again identified that is back to the terminal device, is handled by the terminal device is further.
2. the method as described in claim 1, which is characterized in that first convolutional neural networks are based on TensorFlow's
Small-sized convolutional neural networks, second convolutional neural networks are the large-scale convolutional neural networks based on TensorFlow.
3. method according to claim 2, which is characterized in that the binocular camera acquires video image and is sent to terminal
Before equipment, the method also includes:
According to the convolutional neural networks model based on TensorFlow, the small-sized convolutional Neural based on TensorFlow is obtained
Network and large-scale convolutional neural networks based on TensorFlow.
4. method as claimed in claim 3, which is characterized in that convolutional neural networks mould of the basis based on TensorFlow
Type obtains the small-sized convolutional neural networks based on TensorFlow and the large-scale convolutional Neural net based on TensorFlow
Network, comprising:
Build the convolutional neural networks model based on TensorFlow, the convolutional neural networks model based on TensorFlow
Return including the first convolutional layer, the first maximum pond layer, first partial response normalization layer, the second convolutional layer, the second local acknowledgement
Full articulamentum that one change layer, first are linearly activated based on amendment, second based on the full articulamentum that linearly activates of amendment and
softmax_linear；
Using cloud computing TensorFlow cluster server to the convolutional neural networks model based on TensorFlow into
Row training, to obtain the small-sized convolutional neural networks based on TensorFlow and the mind of the large-scale convolution based on TensorFlow
Through network.
5. a kind of binocular video monitoring system, which is characterized in that the system comprises binocular camera, terminal device and cloud services
Device, the terminal device include preliminary identification module and uploading module, and the Cloud Server includes again identifying that module and result
Return module；
The binocular camera, for acquiring video image and being sent to terminal device；
The preliminary identification module, for by the video image input the first convolutional neural networks on the terminal device into
The preliminary identification of row target；
The uploading module, if the result tentatively identified for the preliminary identification module can to exist in the video image
Target is doubted, then the video image is uploaded to Cloud Server；
It is described to again identify that module, for by the video image input the second convolutional neural networks on the Cloud Server into
Row target again identifies that, first convolutional neural networks and nervus opticus network use phase via distributed type assemblies server
Same model is trained acquisition on different small lot data sets, and second convolutional neural networks are in scale or level
Greater than first convolutional neural networks；
The result return module, if being to determine the video image for the result again identified that for again identifying that module
In there are suspicious objects, then the result again identified that is back to the terminal device, is done by the terminal device into one
The processing of step.
6. system as claimed in claim 5, which is characterized in that first convolutional neural networks are based on TensorFlow's
Small-sized convolutional neural networks, second convolutional neural networks are the large-scale convolutional neural networks based on TensorFlow.
7. system as claimed in claim 6, which is characterized in that the system also includes:
Module is obtained, acquires video image for the binocular camera and before being sent to terminal device, according to being based on
The convolutional neural networks model of TensorFlow obtains the small-sized convolutional neural networks based on TensorFlow and is based on
The large-scale convolutional neural networks of TensorFlow.
8. system as claimed in claim 7, which is characterized in that the acquisition module includes:
Model buildings unit, it is described to be based on TensorFlow for building the convolutional neural networks model based on TensorFlow
Convolutional neural networks model include that the first convolutional layer, the first maximum pond layer, first partial response normalize layer, the second convolution
Full articulamentum that layer, the second local acknowledgement normalization layer, first are linearly activated based on amendment, second are linearly activated based on amendment
Full articulamentum and softmax_linear；
Training unit, for the TensorFlow cluster server using cloud computing to the convolution mind based on TensorFlow
It is trained through network model, to obtain the small-sized convolutional neural networks based on TensorFlow and be based on TensorFlow
Large-scale convolutional neural networks.
9. a kind of binocular video monitoring system, including memory, processor and storage are in the memory and can be described
The computer program run on processor, which is characterized in that the processor realizes following step when executing the computer program
It is rapid:
Binocular camera acquisition video image is simultaneously sent to terminal device；
The video image is inputted the first convolutional neural networks on the terminal device and carries out target by the terminal device
Preliminary identification；
If the result tentatively identified is in the video image there are suspicious object, the terminal device is by the video
Image is uploaded to Cloud Server；
The video image is inputted the second convolutional neural networks on the Cloud Server and carries out target by the Cloud Server
It again identifies that, first convolutional neural networks and nervus opticus network use identical model via distributed type assemblies server
Acquisition is trained on different small lot data sets, second convolutional neural networks are greater than described in scale or level
First convolutional neural networks；
If the result again identified that is to determine there are suspicious object in the video image, the Cloud Server will be described
The result again identified that is back to the terminal device, is handled by the terminal device is further.
10. a kind of computer readable storage medium, the computer-readable recording medium storage has computer program, and feature exists
In the computer program performs the steps of when being executed by processor
Binocular camera acquisition video image is simultaneously sent to terminal device；
The video image is inputted the first convolutional neural networks on the terminal device and carries out target by the terminal device
Preliminary identification；
If the result tentatively identified is in the video image there are suspicious object, the terminal device is by the video
Image is uploaded to Cloud Server；
The video image is inputted the second convolutional neural networks on the Cloud Server and carries out target by the Cloud Server
It again identifies that, first convolutional neural networks and nervus opticus network use identical model via distributed type assemblies server
Acquisition is trained on different small lot data sets, second convolutional neural networks are greater than described in scale or level
First convolutional neural networks；
If the result again identified that is to determine there are suspicious object in the video image, the Cloud Server will be described
The result again identified that is back to the terminal device, is handled by the terminal device is further.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710286635.6A CN106982359B (en) | 2017-04-26 | 2017-04-26 | A kind of binocular video monitoring method, system and computer readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710286635.6A CN106982359B (en) | 2017-04-26 | 2017-04-26 | A kind of binocular video monitoring method, system and computer readable storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106982359A CN106982359A (en) | 2017-07-25 |
| CN106982359B true CN106982359B (en) | 2019-11-05 |
Family
ID=59341496
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710286635.6A Active CN106982359B (en) | 2017-04-26 | 2017-04-26 | A kind of binocular video monitoring method, system and computer readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106982359B (en) |
Families Citing this family (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109685088A (en) * | 2017-10-18 | 2019-04-26 | 上海仪电（集团）有限公司中央研究院 | Narrow band communication intelligent image analysis system based on cloud separation convolutional neural networks |
| CN108229379A (en) * | 2017-12-29 | 2018-06-29 | 广东欧珀移动通信有限公司 | Image-recognizing method, device, computer equipment and storage medium |
| CN108549852B (en) * | 2018-03-28 | 2020-09-08 | 中山大学 | Specific scene downlink person detector automatic learning method based on deep network enhancement |
| CN108920995A (en) * | 2018-04-08 | 2018-11-30 | 华中科技大学 | Intelligent security guard video monitoring method and its system and monitor terminal |
| CN108805184B (en) * | 2018-05-28 | 2020-07-31 | 广州英卓电子科技有限公司 | Image recognition method and system for fixed space and vehicle |
| CN108960209B (en) * | 2018-08-09 | 2023-07-21 | 腾讯科技（深圳）有限公司 | Identity recognition method, identity recognition device and computer readable storage medium |
| CN109068145A (en) * | 2018-08-15 | 2018-12-21 | 上海鸢安智能科技有限公司 | Distribution type intelligent video analysis system, method, apparatus, equipment and storage medium |
| CN110855934A (en) * | 2018-08-21 | 2020-02-28 | 北京嘀嘀无限科技发展有限公司 | Fatigue driving identification method, device and system, vehicle-mounted terminal and server |
| CN109413213A (en) * | 2018-12-20 | 2019-03-01 | 泉州装备制造研究所 | Cluster monitoring system based on system on chip |
| CN109670488A (en) * | 2019-01-31 | 2019-04-23 | 湖北省基础地理信息中心（湖北省北斗卫星导航应用技术研究院） | A kind of video data exemplary dynamic object detection method and system |
| CN111669540A (en) * | 2019-03-07 | 2020-09-15 | 上海思桂信息技术有限公司 | Monitoring system and method based on AI technology |
| CN110415159A (en) * | 2019-06-11 | 2019-11-05 | 汇盈讯科智能科技(佛山市)有限责任公司 | A kind of control system and its control method based on image recognition repeater |
| CN110675449B (en) * | 2019-09-02 | 2020-12-08 | 山东科技大学 | Binocular camera-based offshore flow detection method |
| CN111291748B (en) * | 2020-01-15 | 2020-12-11 | 广州玖峰信息科技有限公司 | Cascade distributed artificial intelligence case number identification system |
| CN113470653A (en) * | 2020-03-31 | 2021-10-01 | 华为技术有限公司 | Voiceprint recognition method, electronic equipment and system |
| CN111785286A (en) * | 2020-05-22 | 2020-10-16 | 南京邮电大学 | Home CNN classification and feature matching combined voiceprint recognition method |
| EP3929801A1 (en) * | 2020-06-25 | 2021-12-29 | Axis AB | Training of an object recognition neural network |
| CN115294507B (en) * | 2022-10-10 | 2022-12-16 | 四川川西数据产业有限公司 | Dynamic video data identification method based on fruit appearance |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102779274A (en) * | 2012-07-19 | 2012-11-14 | 冠捷显示科技（厦门）有限公司 | Intelligent television face recognition method based on binocular camera |
| CN105528754A (en) * | 2015-12-28 | 2016-04-27 | 湖南师范大学 | Old people information service system based on dual neural network behavior recognition model |
| CN105976400A (en) * | 2016-05-10 | 2016-09-28 | 北京旷视科技有限公司 | Object tracking method and device based on neural network model |
| CN106295521A (en) * | 2016-07-29 | 2017-01-04 | 厦门美图之家科技有限公司 | A kind of gender identification method based on multi output convolutional neural networks, device and the equipment of calculating |
| CN106548127A (en) * | 2015-09-18 | 2017-03-29 | 松下电器（美国）知识产权公司 | Image-recognizing method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101659578B1 (en) * | 2015-09-01 | 2016-09-23 | 삼성전자주식회사 | Method and apparatus for processing magnetic resonance imaging |
-
2017
- 2017-04-26 CN CN201710286635.6A patent/CN106982359B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102779274A (en) * | 2012-07-19 | 2012-11-14 | 冠捷显示科技（厦门）有限公司 | Intelligent television face recognition method based on binocular camera |
| CN106548127A (en) * | 2015-09-18 | 2017-03-29 | 松下电器（美国）知识产权公司 | Image-recognizing method |
| CN105528754A (en) * | 2015-12-28 | 2016-04-27 | 湖南师范大学 | Old people information service system based on dual neural network behavior recognition model |
| CN105976400A (en) * | 2016-05-10 | 2016-09-28 | 北京旷视科技有限公司 | Object tracking method and device based on neural network model |
| CN106295521A (en) * | 2016-07-29 | 2017-01-04 | 厦门美图之家科技有限公司 | A kind of gender identification method based on multi output convolutional neural networks, device and the equipment of calculating |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106982359A (en) | 2017-07-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106982359B (en) | A kind of binocular video monitoring method, system and computer readable storage medium | |
| KR102641116B1 (en) | Method and device to recognize image and method and device to train recognition model based on data augmentation | |
| Ale et al. | Deep learning based plant disease detection for smart agriculture | |
| KR102545128B1 (en) | Client device with neural network and system including the same | |
| CN110532996A (en) | The method of visual classification, the method for information processing and server | |
| CN109816009A (en) | Multi-tag image classification method, device and equipment based on picture scroll product | |
| Wu et al. | Applications of deep learning for smart water networks | |
| CN110321910A (en) | Feature extracting method, device and equipment towards cloud | |
| CN110222718B (en) | Image processing method and device | |
| CN111008640A (en) | Image recognition model training and image recognition method, device, terminal and medium | |
| CN109558892A (en) | A kind of target identification method neural network based and system | |
| CN110210513A (en) | Data classification method, device and terminal device | |
| CN109034206A (en) | Image classification recognition methods, device, electronic equipment and computer-readable medium | |
| CN112529146B (en) | Neural network model training method and device | |
| CN110503076A (en) | Video classification methods, device, equipment and medium based on artificial intelligence | |
| CN111368926B (en) | Image screening method, device and computer readable storage medium | |
| CN108875767A (en) | Method, apparatus, system and the computer storage medium of image recognition | |
| CN104063686A (en) | System and method for performing interactive diagnosis on crop leaf segment disease images | |
| CN109255296A (en) | A kind of daily Human bodys' response method based on depth convolutional neural networks | |
| CN106778910A (en) | Deep learning system and method based on local training | |
| KR20180123810A (en) | Data enrichment processing technology and method for decoding x-ray medical image | |
| CN111738074B (en) | Pedestrian attribute identification method, system and device based on weak supervision learning | |
| CN110197116A (en) | A kind of Human bodys' response method, apparatus and computer readable storage medium | |
| Gao et al. | A mobile application for plant recognition through deep learning | |
| CN113536970A (en) | Training method of video classification model and related device |
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 |