CN109753848A

CN109753848A - Execute the methods, devices and systems of face identifying processing

Info

Publication number: CN109753848A
Application number: CN201711070074.2A
Authority: CN
Inventors: 谢忠贤; 浦世亮; 周明耀; 翁祖源
Original assignee: Hangzhou Hikvision Digital Technology Co Ltd
Current assignee: Hangzhou Hikvision Digital Technology Co Ltd
Priority date: 2017-11-03
Filing date: 2017-11-03
Publication date: 2019-05-14
Anticipated expiration: 2037-11-03
Also published as: CN109753848B

Abstract

The disclosure is directed to a kind of methods, devices and systems for executing face identifying processing, belong to technical field of face recognition.The described method includes: receiving the image of picture pick-up device acquisition, facial image to be identified in image is determined；Resource inquiry request is sent at least one identification server；At least one face word bank mark that each identification server there are vacant GPU is sent is received respectively；At least one the face word bank mark sent according to each identification server there are vacant GPU, the identification server that recognition of face processing is carried out using everyone face library is determined respectively, respectively to each identification server determined, the recognition of face request for carrying corresponding face word bank mark and facial image to be identified is sent；The processing result for the recognition of face processing that each identification server is sent is received, processing result, determines final process result based on the received.Using the disclosure, the benchmark face amount of images used can be improved.

Description

Execute the methods, devices and systems of face identifying processing

Technical field

The disclosure is directed to technical field of face recognition, especially with respect to a kind of method for executing face identifying processing, dress It sets and system.

Background technique

With the development of face recognition technology, public security organ assists the example layer solved a case to go out not using face recognition technology Thoroughly, the efficiency solved a case is improved, has ensured the security of the lives and property of the people.

The GPU (Graphics Processing Unit, graphics processor) relied primarily in identification server carries out people Face identifying processing.Technical staff can pre-establish face database, for example, can establish the face of the suspect ordered to arrest The face database of library or Missing Persons.It is then possible to the face database be downloaded in GPU, to carry out subsequent processing.Pass through peace Camera (the monitoring camera head of electronic eyes, railway station on such as road) mounted in each corner in city, continuous collecting image, The facial image (facial image i.e. to be identified) in image is extracted, based on unappropriated GPU to the facial image and GPU extracted Benchmark face image in the face database of middle storage compares.Wherein, when GPU is carrying out recognition of face processing, as locate In occupied state；Conversely, GPU is in free state.If the benchmark face image to match is found, by lookup result Public security system is fed back to alarm.

In implementing the present disclosure, inventor discovery the prior art has at least the following problems:

Since the data capacity that GPU can store is limited, the data volume for the face database that can be stored is limited, when face database When data volume is greater than the data capacity that GPU can store, GPU can not store complete face database.In order to continue Recognition of face processing, it has to the data volume of face database is reduced, thus, cause recognition of face to handle the base that can be used Quasi- facial image is less.

Summary of the invention

The disclosure provides a kind of methods, devices and systems for executing face identifying processing, can solve recognition of face processing The less problem of the benchmark face image that can be used.The technical solution is as follows:

According to the first aspect of the embodiments of the present disclosure, a kind of method for executing face identifying processing is provided, the method is answered For face identification system, the face identification system includes dispatch server and multiple identification servers, the identification service Device includes at least one graphics processor GPU, which comprises

The dispatch server receives the image of picture pick-up device acquisition, determines the facial image to be identified in described image；

The dispatch server sends resource inquiry request at least one identification server；

The dispatch server receives at least one face that each identification server there are vacant GPU is sent respectively The mark of word bank；

At least one face that the dispatch server is sent according to each identification server there are vacant GPU The mark of word bank determines the identification server that recognition of face processing is carried out using everyone face library respectively, respectively to determining Each identification server, send carry corresponding face word bank mark and the facial image to be identified recognition of face Request；

The dispatch server receives the processing result for the recognition of face processing that each identification server is sent, according to Received processing result determines the final process result that recognition of face processing is carried out to the facial image to be identified.

Optionally, the dispatch server receives the processing knot for the recognition of face processing that each identification server is sent Fruit, processing result, determines the final process result that recognition of face processing is carried out to the facial image to be identified based on the received, Include:

The dispatch server receives the processing result for the recognition of face processing that each identification server is sent, In, the processing result include at least one face information and the corresponding benchmark face image of each face information and it is described to Identify facial image matching degree, the face information include in the mark of benchmark face image, benchmark face image at least A kind of information；

The dispatch server obtains the corresponding highest preset number of matching degree in received all processing results Face information, as the final process result for carrying out recognition of face processing to the facial image to be identified.

Optionally, the dispatch server receives the image of picture pick-up device acquisition, determines the people to be identified in described image Before face image, further includes:

The dispatch server draws the face database according to the data capacity of single GPU and the data volume of face database It is divided into multiple face word banks, wherein the data volume of the face word bank is not more than the data capacity of the single GPU；

The dispatch server determines the people of GPU and the GPU storage of each identification server in the face identification system The corresponding relationship in face library；

The face that the dispatch server stores the GPU and GPU of identification server each in the face identification system The corresponding relationship of word bank is sent respectively to corresponding identification server.

Optionally, the dispatch server determines that the GPU and GPU of each identification server in the face identification system are deposited The corresponding relationship of the face word bank of storage, comprising:

The dispatch server determines that the total number for the available GPU that all identification servers include and the face are known The number for the available GPU that each identification server includes in other system；

The dispatch server determines that storage is each according to the total number of the available GPU and the number of face word bank The number of the GPU of face word bank, wherein the sum of described number of GPU for storing everyone face library is equal to the available GPU Total number, and it is described store everyone face library GPU number between difference be not more than the first preset threshold；

According to identification clothes each in the number of the GPU for storing everyone face library and the face identification system The number for the available GPU that business device includes, determines in the face identification system and stores each face in each identification server The number of the GPU of word bank, wherein store the GPU in everyone face library in the face identification system in each identification server Number between difference be not more than the second preset threshold；

According to the number for the GPU for storing everyone face library in identification server each in the face identification system, divide The corresponding relationship of the face word bank of GPU and the GPU storage of each identification server in the face identification system is not determined.

Optionally, the method also includes:

The dispatch server receives the processing that is used for that each identification server is sent and obtains corresponding processing result GPU mark with it is corresponding identify server mark；

The dispatch server is based on the mark for being used to handle the GPU for obtaining corresponding processing result and corresponding knowledge The mark of other server determines the identification server to break down or the GPU to break down；

The dispatch server redefines institute based on the identification server not broken down and the GPU not broken down The corresponding relationship for stating the face word bank of GPU and the GPU storage of the update of each identification server in face identification system, sends out respectively Give corresponding identification server.

According to the second aspect of an embodiment of the present disclosure, a kind of method for executing face identifying processing is provided, the method is answered For face identification system, the face identification system includes dispatch server and multiple identification servers, the identification service Device includes at least one graphics processor GPU, which comprises

When the identification server receives the resource inquiry request that the dispatch server is sent, if local exist Unappropriated GPU determines that the unappropriated GPU is corresponding then according to the corresponding relationship of the face word bank of GPU and GPU storage At least one face word bank；

The identification server sends the mark of at least one face word bank to the dispatch server；

The mark in the target person face library at least one face word bank described in receiving and carry when the identification server When knowing the recognition of face request with facial image to be identified, the corresponding unappropriated target GPU in target person face library is determined, Based on the target GPU, using target person face library, face identifying processing is executed to the facial image to be identified；

The processing result that the recognition of face is handled is sent to the dispatch server by the identification server.

Optionally, described to be based on the target GPU, using target person face library, to the facial image to be identified Execute face identifying processing, comprising:

Based on the target GPU, determine the facial image to be identified with it is pre-stored in target person face library The matching degree of each benchmark face image；

The corresponding face information of benchmark face image and corresponding matching degree that match degree is greater than the preset threshold are obtained, is made For the processing result of face identifying processing, the face information includes in the mark of benchmark face image, benchmark face image At least one information.

Optionally, described when the identification server receives the resource inquiry request of dispatch server transmission, if Locally there are unappropriated GPU, then according to the corresponding relationship of the face word bank of GPU and GPU storage, determine described unappropriated Before at least one corresponding face word bank of GPU, further includes:

When the identification server receives the face word bank of the GPU that the dispatch server is sent and GPU storage Corresponding relationship when, the corresponding relationship of the GPU and GPU face word bank stored is stored, the corresponding relationship is obtained Each of face library, be respectively stored into corresponding GPU.

Optionally, the method also includes:

The identification server is when the processing result for handling the recognition of face is sent to the dispatch server, also Send the mark of the target GPU and the mark of the identification server；

When the identification server receives face of GPU and the GPU storage for the update that the dispatch server is sent When the corresponding relationship in library, the corresponding relationship of the face word bank of GPU based on the update and GPU storage, to the GPU of storage with The corresponding relationship of the face word bank of GPU storage is updated；

The face word bank that stores in GPU is removed, the corresponding of face word bank that the GPU and GPU of the update are stored is obtained and closes Each of system face library, is respectively stored into corresponding GPU.

According to the third aspect of an embodiment of the present disclosure, a kind of dispatch server is provided, the dispatch server is applied to people Face identifying system, the face identification system include the dispatch server and multiple identification servers, the identification server Including at least one graphics processor GPU, the dispatch server includes:

First receiving module determines the face figure to be identified in described image for receiving the image of picture pick-up device acquisition Picture；

First sending module, for sending resource inquiry request at least one identification server；

Second receiving module, for receive respectively it is each there are vacant GPU identification server send at least one The mark of face word bank；

Second sending module, for according to it is described it is each there are vacant GPU identification server send at least one The mark of face word bank determines the identification server that recognition of face processing is carried out using everyone face library, respectively to true respectively The each identification server made sends and carries the mark of corresponding face word bank and the face of the facial image to be identified Identification request；

First determining module, for receiving the processing result for the recognition of face processing that each identification server is sent, Processing result based on the received determines the final process result that recognition of face processing is carried out to the facial image to be identified.

Optionally, first determining module is used for:

Receive the processing result for the recognition of face processing that each identification server is sent, wherein the processing result Including at least one face information and the corresponding benchmark face image of each face information and the facial image to be identified Matching degree, the face information include at least one of the mark of benchmark face image, benchmark face image information；

In received all processing results, the corresponding highest preset number face information of matching degree is obtained, as The final process result of recognition of face processing is carried out to the facial image to be identified.

Optionally, the dispatch server further include:

Division module, for according to the data capacity of single GPU and the data volume of face database, the face database to be drawn It is divided into multiple face word banks, wherein the data volume of the face word bank is not more than the data capacity of the single GPU；

Second determining module, for determining GPU and the GPU storage of each identification server in the face identification system The corresponding relationship of face word bank；

Third sending module, the people for storing the GPU and GPU of identification server each in the face identification system The corresponding relationship in face library is sent respectively to corresponding identification server.

Optionally, second determining module is used for:

Optionally, the dispatch server further include:

Third receiving module obtains corresponding processing knot for receiving the processing that is used for that each identification server is sent The mark of the mark of the GPU of fruit and corresponding identification server；

Third determining module, for based on it is described for handle the mark of the GPU for obtaining corresponding processing result with it is corresponding Identification server mark, determine the identification server to break down or the GPU to break down；

Module is redefined, for redefining based on the identification server not broken down and the GPU not broken down The corresponding relationship of the face word bank of GPU and the GPU storage of the update of each identification server in the face identification system, respectively It is sent to corresponding identification server.

According to a fourth aspect of embodiments of the present disclosure, a kind of identification server is provided, the identification server application is in people Face identifying system, the face identification system include dispatch server and multiple identification servers, the identification server Including at least one graphics processor GPU, the identification server includes:

Determining module, for not accounted for if locally existed when receiving the resource inquiry request of dispatch server transmission GPU determines that the unappropriated GPU is corresponding at least then according to the corresponding relationship of the face word bank of GPU and GPU storage One people's face library；

First sending module, for sending the mark of at least one face word bank to the dispatch server；

Processing module receives the mark for carrying the target person face library at least one described face word bank for working as When recognition of face with facial image to be identified is requested, the corresponding unappropriated target GPU in target person face library, base are determined In the target GPU, using target person face library, face identifying processing is executed to the facial image to be identified；

Second sending module, the processing result for handling the recognition of face are sent to the dispatch server.

Optionally, the identification server further include:

First memory module, for when face for receiving the GPU that the dispatch server is sent and GPU storage When the corresponding relationship in library, the corresponding relationship of the GPU and GPU face word bank stored is stored, obtains the corresponding pass Each of system face library, is respectively stored into corresponding GPU.

Optionally, second sending module is also used to be sent in the processing result for handling the recognition of face described When dispatch server, the mark of the target GPU and the mark of the identification server are also sent；

The identification server, further includes:

Update module, for the face word bank when GPU and the GPU storage for receiving the update that the dispatch server is sent Corresponding relationship when, the corresponding relationship of the face word bank of GPU based on the update and GPU storage, to the GPU and GPU of storage The corresponding relationship of the face word bank of storage is updated；

Second memory module obtains GPU and the GPU storage of the update for removing the face word bank stored in GPU Each of corresponding relationship of face word bank face library, is respectively stored into corresponding GPU.

According to a fifth aspect of the embodiments of the present disclosure, a kind of face identification system is provided, the face identification system includes Dispatch server and multiple identification servers, the identification server include at least one graphics processor GPU, in which:

The dispatch server, the dispatch server as described in the third aspect；

The identification server, the identification server as described in fourth aspect.

According to a sixth aspect of an embodiment of the present disclosure, a kind of dispatch server is provided, the dispatch server includes processing Device and memory are stored at least one instruction in the memory, and described instruction is loaded by the processor and executed with reality The now method of execution face identifying processing as described in relation to the first aspect.

According to the 7th of embodiment of the present disclosure aspect, a kind of computer readable storage medium is provided, in the storage medium It is stored at least one instruction, described instruction is loaded by processor and executed to realize that execution face as described in relation to the first aspect is known The method of other places reason.

According to the eighth aspect of the embodiment of the present disclosure, a kind of identification server is provided, the identification server includes processing Device and memory are stored at least one instruction in the memory, and described instruction is loaded by the processor and executed with reality The now method of the execution face identifying processing as described in second aspect.

According to the 9th of the embodiment of the present disclosure the aspect, a kind of computer readable storage medium is provided, which is characterized in that described At least one instruction is stored in storage medium, described instruction is loaded as processor and executed to realize as described in second aspect The method for executing face identifying processing.

The technical scheme provided by this disclosed embodiment can include the following benefits:

In the embodiment of the present disclosure, dispatch server receives the image of picture pick-up device acquisition, determines the people to be identified in image Face image sends resource inquiry request at least one identification server, receives that each there are the identification of vacant GPU clothes respectively Be engaged in device send at least one face word bank mark, according to it is each there are the identification server of vacant GPU transmission at least The mark in one people's face library determines the identification server that recognition of face processing is carried out using everyone face library, respectively respectively To each identification server determined, the mark and the facial image to be identified for carrying corresponding face word bank are sent Then recognition of face request receives the processing result for the recognition of face processing that each identification server is sent, locates based on the received Reason is as a result, determine the final process result for carrying out recognition of face processing to facial image to be identified.In this way, face database is divided into Multiple face word banks, and face can be carried out to facial image to be identified using everyone face library based on multiple identification servers Identifying processing, it is thus possible to improve the quantity of the benchmark face image used when recognition of face processing.

It should be understood that above general description and following detailed description be only it is exemplary and explanatory, not The disclosure can be limited.

Detailed description of the invention

The drawings herein are incorporated into the specification and forms part of this specification, and shows the implementation for meeting the disclosure Example, and together with specification for explaining the principles of this disclosure.In the accompanying drawings:

Fig. 1 is a kind of method flow diagram for executing face identifying processing shown according to an exemplary embodiment；

Fig. 2 is that a kind of dispatch server shown according to an exemplary embodiment determines the identification for carrying out recognition of face processing The process schematic of server；

Fig. 3 is the process schematic that a kind of dispatch server shown according to an exemplary embodiment determines final result；

Fig. 4 is a kind of schematic device of dispatch server shown according to an exemplary embodiment；

Fig. 5 is a kind of schematic device of dispatch server shown according to an exemplary embodiment；

Fig. 6 is a kind of schematic device of dispatch server shown according to an exemplary embodiment；

Fig. 7 is a kind of schematic device for identifying server shown according to an exemplary embodiment；

Fig. 8 is a kind of schematic device for identifying server shown according to an exemplary embodiment；

Fig. 9 is a kind of schematic device for identifying server shown according to an exemplary embodiment；

Figure 10 is a kind of structural schematic diagram of dispatch server shown according to an exemplary embodiment；

Figure 11 is a kind of structural schematic diagram for identifying server shown according to an exemplary embodiment.

Through the above attached drawings, it has been shown that the specific embodiment of the disclosure will be hereinafter described in more detail.These attached drawings It is not intended to limit the scope of this disclosure concept by any means with verbal description, but is by referring to specific embodiments Those skilled in the art illustrate the concept of the disclosure.

Specific embodiment

Example embodiments are described in detail here, and the example is illustrated in the accompanying drawings.Following description is related to When attached drawing, unless otherwise indicated, the same numbers in different drawings indicate the same or similar elements.Following exemplary embodiment Described in embodiment do not represent all implementations consistent with this disclosure.On the contrary, they be only with it is such as appended The example of the consistent device and method of some aspects be described in detail in claims, the disclosure.

An exemplary embodiment of the present disclosure provides a kind of method for executing face identifying processing, this method is applied to face Identifying system, face identification system include dispatch server and multiple identification servers.

Wherein, identification server may include the portions such as at least one graphics processor GPU, processor, memory, transceiver Part.Graphics processor GPU can be used for carrying out the processing such as facial image identification.Processor can be CPU (Central Processing Unit, central processing unit) etc., it is determined for the occupancy situation of GPU, determines that unappropriated GPU is corresponding The processing such as at least one face word bank.Memory can be RAM (Random Access Memory, random access memory Device), Flash (flash memory) etc., data needed for can be used for storing the data received, treatment process generate in treatment process Data etc., such as GPU and GPU storage face word bank corresponding relationship, face word bank.Transceiver can be used for and dispatch Server carries out data transmission, for example, receiving the resource inquiry request of dispatch server transmission, sending at least to dispatch server The mark in one people's face library, the recognition of face request for receiving dispatch server transmission send recognition of face to dispatch server The processing result of processing, transceiver may include antenna, match circuit, modem etc..

Dispatch server may include the components such as processor, memory, transceiver.Processor, can be for CPU etc., can be with For determining facial image to be identified, determining the corresponding relationship of face word bank that the GPU and GPU of each identification server are stored etc. Processing.Memory can be RAM, Flash etc., data needed for can be used for storing the data received, treatment process, place The data etc. generated during reason, such as facial image to be identified, face database, face word bank, each identification server GPU with The corresponding relationship etc. of the face word bank of GPU storage.Transceiver can be used for carrying out data transmission with identification server, for example, to Identify that server sends resource inquiry request, the mark for receiving at least one face word bank that identification server is sent, to identification The processing result of the recognition of face processing of recognition of face request, reception identification server transmission that server is sent, transceiver can To include antenna, match circuit, modem etc..

As shown in Figure 1, the process flow of this method may include following step:

In a step 101, dispatch server receives the image of picture pick-up device acquisition, determines the face figure to be identified in image Picture.

Wherein, picture pick-up device can be mounted in the electronic eyes above road, monitoring camera head in railway station etc., this is taken the photograph It is handled as equipment acquired image can be transmitted directly to the relevant departments such as public security organ.

In an implementation, picture pick-up device may be mounted at each corner in city, acquire on urban road or each traffic Image in hinge (such as passenger station, railway station, airport), is monitored city.Picture pick-up device can be by collected view Frequency image is sent to dispatch server, for example, the monitoring camera in railway station is by the video image in collected railway station It is sent to dispatch server.The image that picture pick-up device is also possible to capture is sent to dispatch server, for example, being mounted on The electronic eyes of road side is detecting that driving passes through, or detects in acquired image there are when facial image, can be with Acquired image is captured, the image captured is sent to dispatch server.

Dispatch server can carry out Face datection, detection after the image for receiving picture pick-up device transmission to the image After human face region into image, the image interception of the human face region can be come out, as facial image to be identified.For The case where collected video image is sent to dispatch server by above-mentioned picture pick-up device, server can be to the every of video image One frame image carries out Face datection, is also possible to choose a frame image every preset number frame image and carry out Face datection, into And the facial image to be identified in the image selected can be intercepted out；The figure that will be captured for above-mentioned picture pick-up device As the case where being sent to dispatch server, server directly can carry out Face datection to the image captured, and in turn, can incite somebody to action Facial image to be identified in the image captured intercepts out.

Optionally, before step 101, face database can be divided into multiple face word banks by dispatch server, and to every A identification server carries out the resource distribution of face word bank, and corresponding processing can be such that dispatch server according to the number of single GPU According to the data volume of capacity and face database, face database is divided into multiple face word banks；Dispatch server determines recognition of face system The corresponding relationship of the face word bank of GPU and the GPU storage of each identification server in system；Dispatch server is by face identification system In each identification server GPU and GPU storage face word bank corresponding relationship, be sent respectively to corresponding identification and service Device.

Wherein, the data volume of face word bank is not more than the data capacity of single GPU, can not store face to avoid GPU The case where library.

In an implementation, technical staff can pre-establish face database, which is made of each benchmark face image.Base Quasi- facial image is handled for subsequent recognition of face, to find the facial image to match with benchmark face image.In order to find Fugitive suspect finds missing crew, which can be the face database of the suspect ordered to arrest, or The face database of Missing Persons.Dispatch server can be according to the data capacity of single GPU and the data volume of face database, by people Face library is divided into multiple face word banks, and the data volume in everyone face library is not more than the data capacity that single GPU can be stored, The sum of the data volume in owner's face library is equal to the data volume of face database.Server can also be previously stored with each identification service The mark of device, each mark for identifying the GPU on server, can be stored in the mapping table of identification server and GPU In.In turn, dispatch server can determine each GPU difference in each identification server according to the face word bank marked off The result determined is stored in the corresponding relationship of the face word bank of GPU and GPU storage, the correspondence by the face word bank of storage Relationship can be the form of mapping table.Then, dispatch server will determine respectively lower each identification server GPU and The corresponding relationship of the face word bank of GPU storage, is sent to corresponding identification server.

Optionally, the processing for dividing face word bank can be such that dispatch server, into a method, determines face in a use The quotient of the data capacity of the data volume in library and single GPU, the number in owner's face library as face database put down face database It is divided into number individual's face library.

In an implementation, the attribute of each GPU can be identical, it can the data capacity of storage is identical.Face database is built After standing well, data volume is also decided.In turn, the data capacity and face database of the available single GPU of dispatch server Data volume.In the present embodiment, the data volume of face database is bigger than the data capacity of single GPU.By the data volume of face database divided by The data capacity of single GPU, if obtained quotient is integer, using quotient as the number in owner's face library of face database Mesh；If aliquant, no matter remainder be it is how many, be all based on into a method and the integer part of quotient added 1, by last total Number as the face word bank that face database divides.In turn, face database can be divided into number individual's face library, and each The data volume of face word bank can be equal, and the difference being also possible between the data volume of each word bank is less than preset threshold.For example, If the data capacity that single GPU can store is 5GB, the data volume of face database is 15GB, then can divide the face database For 3 people's face libraries, the data volume in everyone face library can be 5GB, the data capacity no more than GPU；If single GPU can Data capacity with storage is 5GB, and the data volume of face database is 16GB, then the face database can be divided into 4 people's face libraries, The data volume in everyone face library can be 4GB, the data capacity no more than GPU.Based on above-mentioned processing, it is ensured that face The data volume in library is not more than the data capacity of single GPU.

Optionally, in order to enable each identification server oncurrent processing ability with higher, in each identification server The face word bank of distribution is evenly distributed, and corresponding processing can be such that dispatch server determines that all identification servers include The number for the available GPU that each identification server includes in the total number and face identification system of available GPU；Scheduling clothes Device be engaged according to the total number of available GPU and the number of face word bank, determines the number for storing the GPU in everyone face library；Root The available GPU for including according to identification server each in the number and face identification system of the GPU for storing everyone face library Number, determine in face identification system it is each identification server in store everyone face library GPU number；According to face The number for storing the GPU in everyone face library in identifying system in each identification server, determines in face identification system respectively The corresponding relationship of the face word bank of GPU and the GPU storage of each identification server.

Dispatch server determines available GPU, can be idle GPU, or new human face analysis can be supported to appoint Be engaged in GPU handled etc..For example, human face analysis mission requirements whether can be can satisfy according to the resource utilization of GPU come really Whether the fixed GPU can be used.

Wherein, the total number that the sum of the number of GPU in everyone face library is equal to available GPU is stored, and is stored each Difference between the number of the GPU of face word bank is not more than the first preset threshold.Each identification server in face identification system Difference between the number of the middle GPU for storing everyone face library is not more than the second preset threshold.Above-mentioned first preset threshold and Second preset threshold can be 1.

In an implementation, server can be previously stored with each identification mark of server, on each identification server The mark of GPU can be stored in the mapping table of identification server and GPU, as shown in table 1.

Table 1

Wherein, identify server be identified as 1,2,3, GPU be identified as GPU_nm。GPU_nmSubscript n be that the GPU is corresponding Identification server mark, subscript m be the GPU it is corresponding identification server in number.For example, in identification server 1 It include 8 GPU, GPU₁₅The GPU for being 5 for number in identification server 1.

The total number and each identification server for the available GPU that the available identification server of dispatch server includes The number for the available GPU for including, the corresponding relationship of identification server and GPU as shown in Table 1 are available to identification service The total number for the available GPU that device includes is 24, and the quantity for the available GPU that each identification server includes is 8.? In practical application, the total number of general GPU may be relatively more, and the number of face word bank comparatively may be fewer, In this case, everyone face library can be stored in simultaneously in multiple GPU.

Server can be according to the number of the face word bank obtained in the above process and the total number of available GPU, really In fixed all identification servers, the number of the GPU in everyone face library can store.Store the number of the GPU in everyone face library The sum of mesh is the total number of available GPU, and the difference stored between the number of the GPU in everyone face library is default no more than first Threshold value.For example, the number of face word bank is 3 (such as face word banks 1, face word bank 2, face if the first preset threshold is 1 Word bank 3), the total number of available GPU is 24, then can obtain by the total number of available GPU divided by the number of face word bank Number to the GPU for storing everyone face library is 8.I.e. in all identification servers, there are 8 GPU to can store face Word bank 1 has 8 GPU to can store face word bank 2, has 8 GPU to can store face word bank 3.If the sum of available GPU The number of the aliquant face word bank of mesh, if the number of face word bank is 3, the total number of available GPU is 23, available It is 7 that the total number of GPU, which obtains quotient divided by the number of face word bank, remainder 2.Then obtain storing the GPU's in everyone face library Number can be to have 8 GPU to can store face word bank 1, have 8 GPU to can store face word bank 2, have 7 GPU that can deposit Store up face word bank 3.Wherein it is possible to be to randomly select remainder individual's face library, its quotient is added 1, obtains the people that storage selects The number of the GPU in face library, and the number for storing the GPU of unselected face word bank out is quotient；It is also possible to according to pre- If rule choose remainder individual face library, such as according to face word bank preset order selection remainder individual face library, by its quotient Number plus 1 obtains the number of the GPU for the face word bank that storage selects, and stores the number of the GPU of unselected face word bank out Mesh is quotient.In this way, the difference between the number of the GPU of any two storage face word bank is not more than 1, i.e., it is pre- no more than first If threshold value.

In turn, dispatch server can be according to the number and each knowledge of the GPU in everyone the face library of storage under determining The number for the available GPU that other server includes determines the number that the GPU in everyone face library is stored in each identification server Mesh.As shown in table 2, the difference between the number for the GPU for storing everyone face library in each identification server is no more than second Preset threshold (second preset threshold is 1 in table 2), and the sum of the number of storage in everyone face library is equal in the above process really The number of the GPU in everyone the face library of storage fixed.

Table 2

It wherein, include 8 GPU in each identification server, the number for storing the GPU in everyone face library is 8 It is a.It identifies in server 1, the GPU for storing face word bank 1 has 3, and the GPU for storing face word bank 2 has 3, is used for The GPU of storage face word bank 3 has 2, and the sum of number is the number for the GPU for identifying that server 1 includes.Identify server 2,3 with Identify that server 1 is similar, details are not described herein again.The GPU for storing face word bank 1 has 3 in identification server 1, takes in identification There are 3 in business device 2, there are 2 in identification server 3, the sum of number is the number of the GPU of all storage face word banks 1.Storage The GPU of face word bank 2,3 is similar with the storage GPU of face word bank 1, and details are not described herein again.In each identification server, storage is every Difference between the number of the GPU in personal face library is not more than 1, that is, is not more than the second preset threshold.

It, can be according to depositing in advance after the number for the GPU for storing everyone face library in each identification server is decided Mark, each mark for identifying the available GPU on server for containing each identification server, determine that each GPU can respectively With the face word bank of storage, the corresponding relationship of the face word bank of GPU and the GPU storage as each identification server.It can be Each identification server is randomly selected corresponding according to the number for the GPU for storing everyone face library in each identification server The GPU of number, for storing corresponding face word bank；It is also possible to choose respective number GPU according to default rule, uses In the corresponding face word bank of storage.The corresponding relationship of the face word bank of GPU and GPU storage can be the form of mapping table. In turn, the corresponding relationship of the face word bank of GPU and the GPU storage of each identification server can be sent to corresponding identification Server.

Optionally, identification server can be by the storage of everyone face library into corresponding GPU, and corresponding processing can be as Under: when identifying that server receives the corresponding relationship of the face word bank of GPU and the GPU storage of dispatch server transmission, to GPU It is stored with the corresponding relationship of the face word bank of GPU storage, obtains each of corresponding relationship face library, be respectively stored into In corresponding GPU.

In an implementation, pair of dispatch server face word bank of GPU and the GPU storage of each identification server under determination It should be related to, after being sent to each corresponding identification server, each identification server can receive the GPU and GPU of its own The corresponding relationship of the face word bank of storage, and the corresponding relationship is stored.In turn, each identification server can be according to this Corresponding relationship, to each GPU, by the storage of corresponding face word bank into GPU.

In a step 102, dispatch server sends resource inquiry request at least one identification server.

In an implementation, it after dispatch server intercepts out facial image to be identified, needs for facial image to be identified to be sent to Identify that server carries out recognition of face processing, and facial image to be identified needs to pass through and the benchmark face in owner's face library Image is matched, and recognition of face processing could be completed.And everyone face library is distributed in the corresponding GPU of each identification server In, dispatch server needs to obtain the face word bank that can be used when carrying out recognition of face processing in each identification server, really Everyone fixed face library can use in which identification server respectively.Therefore, dispatch server intercepts out face to be identified After image, resource inquiry request can be sent to each identification server, face knowledge is carried out to facial image to be identified to determine The identification server of other places reason.

In step 103, when identifying that server receives the resource inquiry request of dispatch server transmission, if local There are unappropriated GPU, then according to the corresponding relationship of the face word bank of GPU and GPU storage, determine that unappropriated GPU is corresponding At least one face word bank.

In an implementation, it when each identification server receives the resource inquiry request of dispatch server transmission, can detecte Whether the GPU for including in local is carrying out recognition of face processing, i.e. whether detection GPU is occupied.If identified in server In the presence of the GPU for not carrying out recognition of face processing, that is, there is unappropriated GPU, then according to the face of the GPU of storage and GPU storage The corresponding relationship of word bank determines the mark in the corresponding face word bank of each unappropriated GPU and the people's face library.For example, identification Server 1 detects that unappropriated GPU has GPU₁₁、GPU₁₅、GPU₁₆, according to the corresponding pass of GPU and the face word bank that GPU is stored It is (corresponding relationship as shown in Figure 2) to determine that identification server 1 can be in face word bank 1, face word bank 2, face word bank 3 Middle progress recognition of face processing；Identification server 2 detects that unappropriated GPU has GPU₂₄、GPU₂₅, stored according to GPU and GPU Face word bank corresponding relationship (corresponding relationship as shown in Figure 2), determine identification server 2 can be in face word bank 1, people Recognition of face processing is carried out in face library 2；Identification server 3 detects that unappropriated GPU has GPU₃₂、GPU₃₃, according to GPU with The corresponding relationship (corresponding relationship as shown in Figure 2) of the face word bank of GPU storage determines that identification server 3 can be in face Recognition of face processing is carried out in word bank 2, face word bank 3.

At step 104, identification server sends the mark at least one people's face library to dispatch server.

It in an implementation, can will be true after the mark of face word bank that each identification server respectively can be used under determining The mark of face word bank under recognizing is sent to dispatch server.The mark for the face word bank that one identification server is sent may have It is multiple, indicate that multiple face word banks can be used to carry out recognition of face processing in the identification server；One identification server hair The mark of the face word bank sent may only one, indicate that the people's face library can be used to carry out face knowledge in the identification server Other places reason；The mark for the face word bank that one identification server is sent may be sky, indicate that the identification server is not vacant GPU can to facial image to be identified carry out recognition of face processing.

In step 105, dispatch server receive respectively it is each there are the identification server of vacant GPU send at least The mark in one people's face library.

In step 106, dispatch server according to it is each there are vacant GPU identification server send at least one The mark of face word bank determines the identification server that recognition of face processing is carried out using everyone face library, respectively to true respectively The each identification server made sends the mark and the recognition of face of facial image to be identified for carrying corresponding face word bank Request.

In an implementation, the mark that dispatch server can receive the face word bank that each identification server is sent in turn can To determine all identification servers for carrying out recognition of face processing using everyone face library.For example, being determined according in step 103 The mark of face word bank can obtain, can be carried out in face word bank 1 recognition of face processing server have identification server 1, Identify server 2, the server that recognition of face processing can be carried out in face word bank 2 has identification server 1, identification server 2, server 3 is identified, the server that recognition of face processing can be carried out in face word bank 3 has identification server 1, identification service Device 3.It is then possible to for everyone face library, select the identification server that can carry out recognition of face processing, can be with Machine selection is also possible to according to scheduled rule selection.For example, can choose identification server 2 as in face word bank 1 into The identification server of pedestrian's face identifying processing can choose identification server 1 and carry out at recognition of face as in face word bank 2 The identification server of reason can choose identification server 3 as the identification service for carrying out recognition of face processing in face word bank 3 Device.It, can be every to what is selected respectively after choosing the identification server for carrying out recognition of face processing for everyone face library A identification server sends the recognition of face request of the mark and facial image to be identified that carry corresponding face word bank.Separately Outside, for non-selected identification server, it can be sent to it the message ended processing, for notifying the identification server, No longer carry out the subsequent processing of this recognition of face.Dispatch server determines the mistake for carrying out the identification server of recognition of face processing Journey is as shown in Figure 2.

In step 107, when identification server receives the target person face library carried at least one face word bank Mark when being requested with the recognition of face of facial image to be identified, determine the corresponding unappropriated target GPU in target person face library, Based on target GPU, using target person face library, face identifying processing is executed to facial image to be identified.

In an implementation, if identification server receives the recognition of face request of dispatch server transmission, the recognition of face Request carry face word bank mark and facial image to be identified, the then face that can be stored according to the GPU and GPU of storage it is sub The corresponding relationship in library determines the corresponding GPU of mark of the face word bank carried in recognition of face request, can be a GPU, It can be multiple GPU.Then, in the GPU determined, unappropriated GPU is determined.If unappropriated GPU have it is multiple, can To select one of GPU, random selection can be, be also possible to be selected according to pre-defined rule.In turn, the GPU can be based on (i.e. target GPU) searches the benchmark face image to match with facial image to be identified in the face word bank of GPU storage, Face identifying processing is executed to facial image to be identified.If identification server receives at the end of dispatch server transmission The message of reason can then terminate the relevant treatment of this recognition of face, wait the message of recognition of face next time.

Optionally, the corresponding processing for carrying out recognition of face processing can be such that based on target GPU, determine people to be identified The matching degree of pre-stored each benchmark face image in face image and target person face library；It obtains matching degree and is greater than default threshold The corresponding face information of benchmark face image of value and corresponding matching degree, the processing result as recognition of face processing.

Wherein, face information includes at least one of the mark of benchmark face image, benchmark face image information.

In an implementation, in the identification server for carrying out recognition of face processing, can be based on determining during above under Target GPU compares each benchmark face image in face word bank that facial image to be identified is stored with target GPU, Calculate the matching degree of facial image to be identified Yu each benchmark face image.In turn, it can be determined that facial image to be identified and every Whether the matching degree of a benchmark face image is greater than preset threshold, can be by the matching if match degree is greater than the preset threshold Spend the processing result handled with corresponding benchmark face image as recognition of face.If it is judged that each matching degree is not more than Processing result can then be expressed as not finding the benchmark face image to match by preset threshold.

In step 108, the processing result that recognition of face is handled is sent to dispatch server by identification server.

In an implementation, processing result can be sent to scheduling clothes by each identification server for having carried out recognition of face processing Business device, so that dispatch server arranges each processing result.

Optionally, processing result, the mark of target GPU and the identification server that identification server handles recognition of face Mark, be sent to dispatch server.

Wherein, the corresponding GPU of mark of target GPU is the GPU for carrying out the face identifying processing.

In step 109, dispatch server receives the processing result for the recognition of face processing that each identification server is sent, Processing result based on the received determines the final process result that recognition of face processing is carried out to facial image to be identified.

In an implementation, dispatch server can receive the people that each identification server for having carried out recognition of face processing is sent The processing result of face identifying processing can arrange all processing results in turn, can be and carry out to all processing results Integration, using all processing results as the final process result for carrying out recognition of face processing to facial image to be identified；It can also Being screened to all processing results, select with the most matched benchmark face image of facial image to be identified, as right Facial image to be identified carries out the final process result of recognition of face processing.Dispatch server determines the process of final result as schemed Shown in 3.

Optionally, the highest face information picture of matching degree can be obtained in all processing results as final process knot Fruit, corresponding processing can be such that dispatch server receives the processing knot for the recognition of face processing that each identification server is sent Fruit；Dispatch server obtains the highest preset number face information of matching degree, as right in received all processing results Facial image to be identified carries out the final process result of recognition of face processing.

Wherein, processing result include at least one face information and the corresponding benchmark face image of each face information with The matching degree of facial image to be identified, face information include at least one in the mark of benchmark face image, benchmark face image Kind information.

In an implementation, dispatch server can receive the people that each identification server for having carried out recognition of face processing is sent The processing result of face identifying processing can wrap containing at least one face information and each face information in each processing result The matching degree of corresponding benchmark face image and facial image to be identified.Dispatch server may determine that in all processing results Whether the number of face information is greater than preset number.If the number of the benchmark face image in all processing results is greater than default Number can then be ranked up the matching degree in all processing results, obtain the highest preset number benchmark people of matching degree The corresponding face information of face image, for example, the corresponding face information of the highest 5 benchmark face images of matching degree is obtained, as The final process result of recognition of face processing is carried out to facial image to be identified.If face information in all processing results Number is not more than preset number, then carries out people using all face informations in all processing results as to facial image to be identified The final process result of face identifying processing.If being the benchmark face figure for not finding and matching in all processing results Picture, then facial image to be identified may not be the facial image to be found being stored in face database.

If in final process result including face information, facial image to be identified may be face figure to be found The final process result can be sent to relevant departments by picture, dispatch server, to notify relevant departments that may have found and base The corresponding target person of quasi- facial image.For example, being adjusted if face database is the face database of the suspect ordered to arrest Degree server final process result can be sent to public security system, alarm, notice public security department may have found with most The corresponding suspect of face information in whole processing result；If face database is the face database of Missing Persons, dispatch Final process result can be sent to public security system by server, be alarmed, notice public security department may have found with finally The corresponding missing crew of face information in processing result.

Optionally, if identification server or GPU break down, the GPU for storing everyone face library is carried out again Distribution, corresponding processing can be such that dispatch server receives that each identification server sends for handle obtain it is corresponding The mark of the mark of the GPU of processing result and corresponding identification server；Dispatch server is based on obtaining for processing corresponding The mark of the mark of the GPU of processing result and corresponding identification server determines the identification server to break down or event occurs The GPU of barrier；Dispatch server redefines face knowledge based on the identification server not broken down and the GPU not broken down The corresponding relationship of the face word bank of GPU and the GPU storage of the update of each identification server, is sent respectively to correspond in other system Identification server.

In an implementation, it identifies that server or GPU may break down and stop operating, sends and request in dispatch server When, corresponding feedback cannot be made.For example, dispatch server sends resource inquiry request to the identification server to break down When, the mark of the corresponding face word bank of the identification server cannot be received；Alternatively, dispatch server is sent out to identification server When sending recognition of face to request, and sending failure for carrying out the GPU of the face identifying processing in the identification server, dispatch service Device will cannot receive the processing result of corresponding recognition of face processing.Corresponding GPU is carried when dispatch server receives Mark with it is corresponding identify server mark processing result when, can recorde a work log, for indicating the knowledge The GPU in other server completes recognition of face processing.

The case where for identification server fail, dispatch server reaches default etc. after sending resource inquiry request When duration, the mark for sending the corresponding face word bank of identification server of failure is not received also, then dispatch server can be with The identification server for being no unappropriated GPU by the identification server-tag carries out the identification of recognition of face processing in selection The identification server will not be chosen when server, it, can be with when completing to carry out recognition of face processing to facial image to be identified The mark and the corresponding mark for identifying server that are used for processing and obtain the GPU of corresponding processing result that will be received, write-in One work log is recorded.The case where breaking down for GPU, dispatch server reach after sending recognition of face request When default waiting time, the corresponding processing result in owner's face library is not received also, then dispatch server can will receive For handle the GPU for obtaining corresponding processing result mark with it is corresponding identify server mark, a job is written Log is recorded, and gos to step 102, re-starts recognition of face processing for facial image to be identified.

Dispatch server is sent to by the corresponding relationship of the face word bank of GPU and the GPU storage of each identification server After corresponding identification server, when often reaching predetermined period duration, work log can be counted, determine each identification At the record moment of the corresponding the last item work log of each GPU in server, judge the note from the last item work log Record whether moment to current time is more than duration threshold value, if it exceeds duration threshold value, then be labeled as breaking down by corresponding GPU GPU.If all GPU are marked as the GPU to break down in an identification server, which can be serviced Device is labeled as the identification server to break down.

In turn, dispatch server can be in the mark and each identification server of pre-stored each identification server GPU in, the mark of the identification server to break down or the mark of the GPU to break down are deleted, do not broken down Identification server mark and the mark of GPU that does not break down.Dispatch server can be according to the identification that do not break down The mark of server and the mark for the GPU not broken down redefine the update of each identification server not broken down GPU and GPU storage face word bank corresponding relationship, pair of the face word bank of this process and above-mentioned determining GPU and GPU storage The process that should be related to is similar, and details are not described herein again.In turn, dispatch server can be by face of the GPU of update and GPU storage The corresponding relationship in library is sent respectively to corresponding identification server.

Optionally, identification server can receive the corresponding relationship of the face word bank of GPU and the GPU storage of update, and can To be updated to the face word bank stored in each GPU, corresponding processing be can be such that when identification server receives scheduling When the corresponding relationship of the face word bank of GPU and the GPU storage for the update that server is sent, GPU's and GPU storage based on update The corresponding relationship of face word bank is updated the corresponding relationship of the GPU and GPU of storage the face word bank stored；It removes in GPU The face word bank of storage obtains each of the corresponding relationship of face word bank of GPU and GPU storage of update face library, point It Cun Chu not be into corresponding GPU.

In an implementation, the corresponding of the face word bank that stores of GPU and GPU of the update of dispatch server transmission is each received The identification server of relationship, can GPU based on update and GPU storage face word bank corresponding relationship, to the GPU of storage with The corresponding relationship of the face word bank of GPU storage is updated.Also, in each identification server, for each GPU, Ke Yiqing Except the face word bank stored in GPU, it is then possible to according to the corresponding relationship of the face word bank of the GPU of update and GPU storage, it will Corresponding face word bank storage is into GPU.

Disclosure another exemplary embodiment provides a kind of dispatch server, and the dispatch server is known applied to face Other system, the face identification system include the dispatch server and multiple identification servers, and the identification server includes At least one graphics processor GPU, as shown in figure 4, the dispatch server includes:

First receiving module 401 determines the face to be identified in described image for receiving the image of picture pick-up device acquisition Image；

First sending module 402, for sending resource inquiry request at least one identification server；

Second receiving module 403, at least one sent for receiving each identification server there are vacant GPU respectively The mark in personal face library；

Second sending module 404, at least one for being sent according to each identification server there are vacant GPU The mark in personal face library determines the identification server that recognition of face processing is carried out using everyone face library respectively, respectively to The each identification server determined sends the mark for carrying corresponding face word bank and the people of the facial image to be identified Face identification request；

First determining module 405, for receiving the processing knot for the recognition of face processing that each identification server is sent Fruit, processing result, determines the final process result that recognition of face processing is carried out to the facial image to be identified based on the received.

Optionally, first determining module 405 is used for:

Optionally, as shown in figure 5, the dispatch server further include:

Division module 406, for according to the data capacity of single GPU and the data volume of face database, by the face database It is divided into multiple face word banks, wherein the data volume of the face word bank is not more than the data capacity of the single GPU；

Second determining module 407, for determining that the GPU and GPU of each identification server in the face identification system are deposited The corresponding relationship of the face word bank of storage；

Third sending module 408, for storing the GPU and GPU of identification server each in the face identification system Face word bank corresponding relationship, be sent respectively to corresponding identification server.

Optionally, second determining module 407 is used for:

Optionally, as shown in fig. 6, the dispatch server further include:

Third receiving module 409 obtains corresponding place for receiving the processing that is used for that each identification server is sent Manage identifying and the mark of corresponding identification server for the GPU of result；

Third determining module 410, for based on it is described for handle the mark of the GPU for obtaining corresponding processing result with it is right The mark for the identification server answered, determines the identification server to break down or the GPU to break down；

Module 411 is redefined, for based on the identification server not broken down and the GPU not broken down, again Determine the corresponding relationship of the face word bank of GPU and the GPU storage of the update of each identification server in the face identification system, It is sent respectively to corresponding identification server.

The another exemplary embodiment of the disclosure provides a kind of identification server, and the identification server application is known in face Other system, the face identification system include dispatch server and multiple identification servers, and the identification server includes At least one graphics processor GPU, as shown in fig. 7, the identification server includes:

Determining module 701, for when receiving the resource inquiry request of dispatch server transmission, if local exist not The GPU of occupancy determines that the unappropriated GPU is corresponding extremely then according to the corresponding relationship of the face word bank of GPU and GPU storage Few people's face library；

First sending module 702, for sending the mark of at least one face word bank to the dispatch server；

Processing module 703 carries the target person face library at least one described face word bank for working as to receive When mark and the recognition of face of facial image to be identified are requested, the corresponding unappropriated target in target person face library is determined GPU is based on the target GPU, using target person face library, executes at recognition of face to the facial image to be identified Reason；

Second sending module 704, the processing result for handling the recognition of face are sent to the dispatch server.

Optionally, as shown in figure 8, the identification server further include:

First memory module 705, for as the people for receiving the GPU that the dispatch server is sent and GPU storage When the corresponding relationship in face library, the corresponding relationship of the GPU and GPU face word bank stored is stored, it is described right to obtain It each of should be related to face library, be respectively stored into corresponding GPU.

Optionally, second sending module 704 is also used to be sent in the processing result for handling the recognition of face When the dispatch server, the mark of the target GPU and the mark of the identification server are also sent；

As shown in figure 9, the identification server, further includes:

Update module 706, for the face when GPU and the GPU storage for receiving the update that the dispatch server is sent When the corresponding relationship of word bank, the corresponding relationship of the face word bank of GPU based on the update and GPU storage, to the GPU of storage with The corresponding relationship of the face word bank of GPU storage is updated；

Second memory module 707, for removing the face word bank stored in GPU, the GPU and GPU for obtaining the update are deposited Each of corresponding relationship of face word bank of storage face library, is respectively stored into corresponding GPU.

The another exemplary embodiment of the disclosure provides a kind of face identification system, and the face identification system includes scheduling Server and multiple identification servers, the identification server include at least one graphics processor GPU, in which:

The dispatch server determines the face to be identified in described image for receiving the image of picture pick-up device acquisition Image sends resource inquiry request at least one identification server, receives that each there are the identification services of vacant GPU respectively Device send at least one face word bank mark, according to it is described it is each there are the identification server of vacant GPU transmission extremely The mark in few people's face library determines the identification server that recognition of face processing is carried out using everyone face library respectively, point Not to each identification server determined, the mark for carrying corresponding face word bank and the facial image to be identified are sent Recognition of face request, receive it is described it is each identification server send recognition of face processing processing result, based on the received Processing result determines the final process result that recognition of face processing is carried out to the facial image to be identified；

The identification server, for when receiving the resource inquiry request that the dispatch server is sent, if this There are unappropriated GPU on ground, then according to the corresponding relationship of the face word bank of GPU and GPU storage, determine the unappropriated GPU At least one corresponding face word bank, Xiang Suoshu dispatch server send the mark of at least one face word bank, work as reception To the mark and the recognition of face of facial image to be identified for carrying the target person face library at least one described face word bank When request, the corresponding unappropriated target GPU in target person face library is determined, be based on the target GPU, use the target Face word bank executes face identifying processing to the facial image to be identified；The processing result hair that the recognition of face is handled Give the dispatch server.

About the dispatch server and identification server in above-described embodiment, wherein modules execute the specific side of operation Formula is described in detail in the embodiment of the method, and no detailed explanation will be given here.

In embodiment disclosed above, dispatch server receives the image of picture pick-up device acquisition, determines to be identified in image Facial image sends resource inquiry request at least one identification server, receives that each there are the identifications of vacant GPU respectively Server send at least one face word bank mark, according to it is each there are the identification server of vacant GPU transmission extremely The mark in few people's face library determines the identification server that recognition of face processing is carried out using everyone face library respectively, point Not to each identification server determined, the mark for carrying corresponding face word bank and the facial image to be identified are sent Recognition of face request, then, receive it is each identification server send recognition of face processing processing result, based on the received Processing result determines the final process result that recognition of face processing is carried out to facial image to be identified.In this way, face database is divided For multiple face word banks, and can based on multiple identification servers using everyone face library to facial image to be identified progress people Face identifying processing, it is thus possible to improve the quantity of the benchmark face image used when recognition of face processing.

It should be understood that dispatch server provided by the above embodiment and identification server are executing face identifying processing When, only the example of the division of the above functional modules, in practical application, it can according to need and divide above-mentioned function With being completed by different functional modules, i.e., the internal structure of dispatch server and identification server is divided into different function moulds Block, to complete all or part of the functions described above.In addition, dispatch server provided by the above embodiment and identification service Device and the embodiment of the method for executing face identifying processing belong to same design, and specific implementation process is detailed in embodiment of the method, this In repeat no more.

The another exemplary embodiment of the disclosure shows a kind of structural schematic diagram of dispatch server.Referring to Fig.1 0, scheduling Server 1000 includes processing component 1022, further comprises one or more processors, and by 1032 generations of memory The memory resource of table, can be by the instruction of the execution of processing component 1022, such as application program for storing.In memory 1032 The application program of storage may include it is one or more each correspond to one group of instruction module.In addition, processing group Part 1022 is configured as executing instruction, the method to execute above-mentioned execution face identifying processing.

Dispatch server 1000 can also include that a power supply module 1026 is configured as executing dispatch server 1000 Power management, a wired or wireless network interface 1050 are configured as dispatch server 1000 being connected to network and one Input and output (I/O) interface 1058.Dispatch server 1000 can be operated based on the operating system for being stored in memory 1032, example Such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM or similar.

Dispatch server 1000 may include having a kind of non-transitorycomputer readable storage medium and one or one A above program, one of them or more than one program are stored in non-transitorycomputer readable storage medium, and are passed through Configuration includes for carrying out following grasp to execute the one or more programs by one or more than one processor The instruction of work:

Optionally, the method also includes:

The another exemplary embodiment of the disclosure shows a kind of structural schematic diagram for identifying server.Referring to Fig.1 1, identification Server 1100 includes processing component 1122, further comprises one or more processors, and by 1132 generations of memory The memory resource of table, can be by the instruction of the execution of processing component 1122, such as application program for storing.In memory 1132 The application program of storage may include it is one or more each correspond to one group of instruction module.In addition, processing group Part 1122 is configured as executing instruction, the method to execute above-mentioned execution face identifying processing.

Identification server 1100 can also include that a power supply module 1126 is configured as executing identification server 1100 Power management, a wired or wireless network interface 1150 are configured as to identify that server 1100 is connected to network and one Input and output (I/O) interface 1158.Identification server 1100 can be operated based on the operating system for being stored in memory 1132, example Such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM or similar.

Identification server 1100 may include having a kind of non-transitorycomputer readable storage medium and one or one A above program, one of them or more than one program are stored in non-transitorycomputer readable storage medium, and are passed through Configuration includes for carrying out following grasp to execute the one or more programs by one or more than one processor The instruction of work:

Optionally, the method also includes:

In the embodiment of the present disclosure, when identifying that server receives the resource inquiry request of dispatch server transmission, if Locally there are unappropriated GPU, then according to the corresponding relationship of the face word bank of GPU and GPU storage, determine unappropriated GPU pairs At least one the face word bank answered sends the mark at least one people's face library to dispatch server, when identification server receives The recognition of face of mark and facial image to be identified to the target person face library carried at least one face word bank is requested When, it determines the corresponding unappropriated target GPU in target person face library, knowledge is treated using target person face library based on target GPU Others executes face identifying processing at face image, identifies that the processing result that recognition of face is handled is sent to dispatch service by server Device.In this way, face database is divided into multiple face word banks, and everyone face library pair can be used based on multiple identification servers Facial image to be identified carries out recognition of face processing, it is thus possible to improve the benchmark face image used when recognition of face processing Quantity.

Those skilled in the art will readily occur to its of the disclosure after considering specification and practicing disclosure disclosed herein Its embodiment.This application is intended to cover any variations, uses, or adaptations of the disclosure, these modifications, purposes or Person's adaptive change follows the general principles of this disclosure and including the undocumented common knowledge in the art of the disclosure Or conventional techniques.The description and examples are only to be considered as illustrative, and the true scope and spirit of the disclosure are by following Claim is pointed out.

It should be understood that the present disclosure is not limited to the precise structures that have been described above and shown in the drawings, and And various modifications and changes may be made without departing from the scope thereof.The scope of the present disclosure is only limited by the accompanying claims.

Claims

1. a kind of method for executing face identifying processing, which is characterized in that the method is applied to face identification system, the people Face identifying system includes dispatch server and multiple identification servers, and the identification server includes at least one graphics processor GPU, which comprises

The dispatch server receives at least one face word bank that each identification server there are vacant GPU is sent respectively Mark；

At least one face word bank that the dispatch server is sent according to each identification server there are vacant GPU Mark, determine the identification server that recognition of face processing is carried out using everyone face library respectively, it is every to what is determined respectively A identification server, the recognition of face for sending the mark and the facial image to be identified that carry corresponding face word bank are asked It asks；

The dispatch server receives the processing result for the recognition of face processing that each identification server is sent, according to reception Processing result, determine the final process result that recognition of face processing is carried out to the facial image to be identified.

2. the method according to claim 1, wherein the dispatch server receives each identification server The processing result of the recognition of face processing of transmission, processing result, determines and carries out to the facial image to be identified based on the received The final process result of recognition of face processing, comprising:

The dispatch server receives the processing result for the recognition of face processing that each identification server is sent, wherein institute State processing result include at least one face information and the corresponding benchmark face image of each face information with it is described to be identified The matching degree of facial image, the face information include at least one of benchmark face image, the mark of benchmark face image Information；

The dispatch server obtains the corresponding highest preset number face of matching degree in received all processing results Information, as the final process result for carrying out recognition of face processing to the facial image to be identified.

3. the method according to claim 1, wherein the dispatch server receives the figure of picture pick-up device acquisition Picture, before determining the facial image to be identified in described image, further includes:

The face database is divided by the dispatch server according to the data capacity of single GPU and the data volume of face database Multiple face word banks, wherein the data volume of the face word bank is not more than the data capacity of the single GPU；

The dispatch server determines face of GPU and the GPU storage of each identification server in the face identification system The corresponding relationship in library；

The face word bank that the dispatch server stores the GPU and GPU of identification server each in the face identification system Corresponding relationship, be sent respectively to corresponding identification server.

4. according to the method described in claim 3, it is characterized in that, the dispatch server determines in the face identification system The corresponding relationship of the face word bank of GPU and the GPU storage of each identification server, comprising:

The dispatch server determine all identification servers available GPU for including total number and the recognition of face system The number for the available GPU that each identification server includes in system；

The dispatch server determines according to the total number of the available GPU and the number of face word bank and stores each face The number of the GPU of word bank, wherein the sum of described number of GPU for storing everyone face library is total equal to the available GPU Number, and the difference between the number of the GPU for storing everyone face library is not more than the first preset threshold；

According to each identification server in the number of the GPU for storing everyone face library and the face identification system The number for the available GPU for including determines in the face identification system and stores everyone face library in each identification server GPU number, wherein in the face identification system it is each identification server in store everyone face library GPU number Difference between mesh is not more than the second preset threshold；

It is true respectively according to the number for the GPU for storing everyone face library in identification server each in the face identification system The corresponding relationship of the face word bank of GPU and the GPU storage of each identification server in the fixed face identification system.

5. according to the described in any item methods of claim 3-4, which is characterized in that the method also includes:

What the dispatch server received that each identification server sends obtains the GPU of corresponding processing result for handling Mark with it is corresponding identify server mark；

The dispatch server is taken based on the mark for handling the GPU for obtaining corresponding processing result with corresponding identification The mark of business device, determines the identification server to break down or the GPU to break down；

The dispatch server redefines the people based on the identification server not broken down and the GPU not broken down The corresponding relationship of the face word bank of GPU and the GPU storage of the update of each identification server, is sent respectively in face identifying system Corresponding identification server.

6. a kind of method for executing face identifying processing, which is characterized in that the method is applied to face identification system, the people Face identifying system includes dispatch server and multiple identification servers, and the identification server includes at least one graphics processor GPU, which comprises

When the identification server receives the resource inquiry request that the dispatch server is sent, do not accounted for if locally existed GPU determines that the unappropriated GPU is corresponding at least then according to the corresponding relationship of the face word bank of GPU and GPU storage One people's face library；

When the identification server receive carry the mark in the target person face library at least one described face word bank with When the recognition of face request of facial image to be identified, determines the corresponding unappropriated target GPU in target person face library, be based on The target GPU executes face identifying processing to the facial image to be identified using target person face library；

7. according to the method described in claim 6, it is characterized in that, it is described be based on the target GPU, use the target face Word bank executes face identifying processing to the facial image to be identified, comprising:

Based on the target GPU, determine pre-stored each in the facial image to be identified and target person face library The matching degree of benchmark face image；

The corresponding face information of benchmark face image and corresponding matching degree that match degree is greater than the preset threshold are obtained, as people The processing result of face identifying processing, the face information include in the mark of benchmark face image, benchmark face image at least A kind of information.

8. according to the method described in claim 6, it is characterized in that, described when the identification server receives dispatch server When the resource inquiry request of transmission, if local, there are unappropriated GPU, according to pair of the face word bank of GPU and GPU storage It should be related to, before determining at least one corresponding face word bank of the unappropriated GPU, further includes:

When the identification server receives pair of the face word bank of the GPU that the dispatch server is sent and GPU storage When should be related to, the corresponding relationship of the GPU and GPU face word bank stored is stored, is obtained in the corresponding relationship Everyone face library, is respectively stored into corresponding GPU.

9. according to the method described in claim 8, it is characterized in that, the method also includes:

The identification server is also sent when the processing result for handling the recognition of face is sent to the dispatch server The mark of the mark of the target GPU and the identification server；

When the identification server receives the face word bank of GPU and the GPU storage for the update that the dispatch server is sent When corresponding relationship, the corresponding relationship of the face word bank of GPU and GPU storage based on the update deposits the GPU and GPU of storage The corresponding relationship of the face word bank of storage is updated；

The face word bank stored in GPU is removed, in the corresponding relationship for obtaining the face word bank of GPU and the GPU storage of the update Everyone face library, be respectively stored into corresponding GPU.

10. a kind of dispatch server, which is characterized in that the dispatch server is applied to face identification system, and the face is known Other system includes the dispatch server and multiple identification servers, and the identification server includes at least one graphics processor GPU, the dispatch server include:

First receiving module determines the facial image to be identified in described image for receiving the image of picture pick-up device acquisition；

Second receiving module, at least one face sent for receiving each identification server there are vacant GPU respectively The mark of word bank；

Second sending module, at least one face for being sent according to each identification server there are vacant GPU The mark of word bank determines the identification server that recognition of face processing is carried out using everyone face library respectively, respectively to determining Each identification server, send carry corresponding face word bank mark and the facial image to be identified recognition of face Request；

First determining module, for receiving the processing result for the recognition of face processing that each identification server is sent, according to Received processing result determines the final process result that recognition of face processing is carried out to the facial image to be identified.

11. dispatch server according to claim 10, which is characterized in that first determining module is used for:

Receive the processing result for the recognition of face processing that each identification server is sent, wherein the processing result includes The matching of at least one face information and each face information corresponding benchmark face image and the facial image to be identified Degree, the face information includes at least one of the mark of benchmark face image, benchmark face image information；

In received all processing results, the corresponding highest preset number face information of matching degree is obtained, as to institute State the final process result that facial image to be identified carries out recognition of face processing.

12. dispatch server according to claim 10, which is characterized in that the dispatch server further include:

Division module, for according to the data capacity of single GPU and the data volume of face database, the face database to be divided into Multiple face word banks, wherein the data volume of the face word bank is not more than the data capacity of the single GPU；

Second determining module, for determining the face of GPU and the GPU storage of each identification server in the face identification system The corresponding relationship of word bank；

Third sending module, for the face of GPU and the GPU storage of identification server each in the face identification system is sub The corresponding relationship in library is sent respectively to corresponding identification server.

13. dispatch server according to claim 12, which is characterized in that second determining module is used for:

14. the described in any item dispatch servers of 2-13 according to claim 1, which is characterized in that the dispatch server also wraps It includes:

Third receiving module obtains corresponding processing result for handling for receive that each identification server sends The mark of the mark of GPU and corresponding identification server；

Third determining module, for based on the mark for being used to handle the GPU for obtaining corresponding processing result and corresponding knowledge The mark of other server determines the identification server to break down or the GPU to break down；

Module is redefined, it is described for redefining based on the identification server not broken down and the GPU not broken down The corresponding relationship of the face word bank of GPU and the GPU storage of the update of each identification server, sends respectively in face identification system To corresponding identification server.

15. a kind of identification server, which is characterized in that the identification server application is known in face identification system, the face Other system includes dispatch server and multiple identification servers, and the identification server includes at least one graphics processor GPU, the identification server include:

Determining module, for when receiving the resource inquiry request of dispatch server transmission, there are unappropriated if local GPU, then according to GPU and GPU storage face word bank corresponding relationship, determine the unappropriated GPU it is corresponding at least one Face word bank；

Processing module, for when receive the mark for carrying the target person face library at least one described face word bank with to When identifying the recognition of face request of facial image, determines the corresponding unappropriated target GPU in target person face library, be based on institute Target GPU is stated, using target person face library, face identifying processing is executed to the facial image to be identified；

16. identification server according to claim 15, which is characterized in that it is described to be based on the target GPU, using described Target person face library executes face identifying processing to the facial image to be identified, comprising:

17. identification server according to claim 15, which is characterized in that the identification server further include:

First memory module, for when the face word bank for receiving the GPU that the dispatch server is sent and GPU storage When corresponding relationship, the corresponding relationship of the GPU and GPU face word bank stored is stored, is obtained in the corresponding relationship Everyone face library, be respectively stored into corresponding GPU.

18. identification server according to claim 17, which is characterized in that second sending module is also used to by institute The processing result of recognition of face processing is stated when being sent to the dispatch server, also sends the mark of the target GPU and described Identify the mark of server；

The identification server, further includes:

Update module, pair for the face word bank when GPU and the GPU storage for receiving the update that the dispatch server is sent When should be related to, the corresponding relationship of the face word bank of GPU and GPU storage based on the update stores the GPU and GPU of storage The corresponding relationship of face word bank be updated；

Second memory module obtains the face of GPU and the GPU storage of the update for removing the face word bank stored in GPU Each of corresponding relationship of word bank face library, is respectively stored into corresponding GPU.

19. a kind of face identification system, which is characterized in that the face identification system includes dispatch server and multiple identifications clothes Business device, the identification server includes at least one graphics processor GPU, in which:

The dispatch server, the dispatch server as described in any claim in the claim 10-14；

The identification server, the identification server as described in any claim in the claim 15-18.

20. a kind of dispatch server, which is characterized in that the dispatch server includes processor and memory, the memory In be stored at least one instruction, described instruction is loaded by the processor and is executed to realize such as any institute of claim 1 to 5 The method for the execution face identifying processing stated.

21. a kind of identification server, which is characterized in that the identification server includes processor and memory, the memory In be stored at least one instruction, described instruction is loaded by the processor and is executed to realize such as any institute of claim 6 to 9 The method for the execution face identifying processing stated.