CN113886081A - Station multi-face-brushing array face library segmentation method based on load balancing - Google Patents

Station multi-face-brushing array face library segmentation method based on load balancing Download PDF

Info

Publication number
CN113886081A
CN113886081A CN202111151261.XA CN202111151261A CN113886081A CN 113886081 A CN113886081 A CN 113886081A CN 202111151261 A CN202111151261 A CN 202111151261A CN 113886081 A CN113886081 A CN 113886081A
Authority
CN
China
Prior art keywords
face
load
node
terminal
index
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202111151261.XA
Other languages
Chinese (zh)
Inventor
吴娟
王宏博
林磊
陆赛杰
张亦然
徐健洲
张义鑫
何跃齐
李道全
刘光杰
王耀
司光字
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing Metro Construction Co ltd
Nanjing University of Information Science and Technology
Beijing Urban Construction Design and Development Group Co Ltd
Original Assignee
Nanjing Metro Construction Co ltd
Nanjing University of Information Science and Technology
Beijing Urban Construction Design and Development Group Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing Metro Construction Co ltd, Nanjing University of Information Science and Technology, Beijing Urban Construction Design and Development Group Co Ltd filed Critical Nanjing Metro Construction Co ltd
Priority to CN202111151261.XA priority Critical patent/CN113886081A/en
Publication of CN113886081A publication Critical patent/CN113886081A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/46—Multiprogramming arrangements
    • G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
    • G06F9/5083—Techniques for rebalancing the load in a distributed system
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/46—Multiprogramming arrangements
    • G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
    • G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
    • G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
    • G06F9/505—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals considering the load
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/46—Multiprogramming arrangements
    • G06F9/54—Interprogram communication
    • G06F9/546—Message passing systems or structures, e.g. queues
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2209/00—Indexing scheme relating to G06F9/00
    • G06F2209/54—Indexing scheme relating to G06F9/54
    • G06F2209/548—Queue

Landscapes

  • Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention provides a station multi-face-brushing-array face library segmentation method based on load balancing, which belongs to the technical field of urban rail transit intellectualization, firstly, by utilizing the operation information of each terminal device in a terminal array, the CPU utilization rate and the memory occupancy rate of each terminal device and the network bandwidth utilization rate of the nodes are obtained by periodically collecting the load operation condition of each terminal node, then obtaining the load weight of each terminal node according to the three load indexes and an entropy weight method, and quantifying the processing capacity of each end node, dividing a face database deployed in a station into small face databases according to the proportion of load weight values, distributing the small face databases to different terminal devices, cooperatively completing feature comparison by a plurality of face recognition terminals, and finally uniformly gathering recognition results and returning the recognition results to the terminal devices uploading face features. Therefore, the response time is shortened, the speed of passing the gate of the passenger is increased, and the operation of multi-terminal collaborative warehouse-splitting calculation on one group of gates is realized.

Description

Station multi-face-brushing array face library segmentation method based on load balancing
Technical Field
The invention belongs to the technical field of urban rail transit intellectualization, and particularly relates to a station multi-face-brushing array face library segmentation method based on load balancing.
Background
In recent years, with the rapid development of deep learning, biometric feature technologies typified by face recognition have come into our lives. The face recognition mainly comprises five calculation operations of face image acquisition and preprocessing, face detection, face alignment, face feature extraction and feature comparison, and in practical application, the face recognition technology is generally divided into two recognition modes of 1:1 and 1: N. The face identification lockage payment under the mode of 1:1 is relatively simple, when a passenger enters a station, the gate interacts with the gate through a unique personal uID in other forms such as Bluetooth, NFC or RFID, meanwhile, the face information of the passenger is collected through a camera above the gate and is subjected to 1:1 identification matching with the face information locally stored by a passenger mobile phone, the passenger is locked, finally, OD closing is carried out on the uID through a clearing system to form a transaction record, then, automatic fee deduction transaction is initiated, in practice, the cost for obtaining extra information bound with the identity of the user is high, and the passive radio frequency information such as the Bluetooth and the RFID still needs extra equipment support; the obtaining mode of NFC and the like is almost indistinguishable from the traditional IC card and has no advantages.
The AFC technology based on face recognition has the expected technical advantages that the requirement that a person waiting for passing a gate to check tickets does not need to be preset is that the technology needs to be in a 1: N mode, but in the mode, the time consumed by face feature comparison and the size of a face library are in a linear relation, the larger the face library is, the longer the comparison time is, the smaller the face library is, and the shorter the comparison time is, generally, a group of gates at a station can be provided with a plurality of face recognition terminals, but the computing capacity and the storage resource of a single terminal device are limited.
Disclosure of Invention
The purpose of the invention is as follows: the invention provides a station multi-face-brushing-array face library segmentation method based on load balancing, aiming at the characteristics of a 1: N identification mode of a face identification technology under a rail transit scene and the problem that computing resources and storage space of single terminal equipment are limited.
The technical scheme is as follows: the invention relates to a station multi-face-brushing array face library segmentation method based on load balancing, which comprises the following steps of:
step 1: the method comprises the steps of periodically collecting operation information of each terminal device in a multi-face-brushing array of a station, and calculating to obtain a load performance index of each end node;
step 2: quantizing the load weight of each terminal device, namely the magnitude of the processing task capacity by using an entropy weight method and combining the load performance indexes obtained in the step 1;
and step 3: dividing the face library into face libraries according to the weight ratio according to the load weight of each terminal device in the step 2;
and 4, step 4: and (4) transmitting the face library obtained in the step (3) to different terminal devices.
Further, the load performance index in step 1 includes a CPU utilization rate, a memory occupancy rate, and a network bandwidth utilization rate of the node.
Further, in step 2, the terminal device is a face recognition terminal, the face recognition terminal adopts a linux system, the use condition of the CPU is stored in a/proc/stat file of the system, and the CPU use rate U is setcpuThe calculation expression is as follows:
Figure BDA0003287249460000021
wherein idle1，idle2Respectively representing the time of CPU idle at T1 and T2, Tcpu1，Tcpu2The total CPU use time at T1 and T2 after the system is started is represented, and the CPU use rate in a certain time period is obtained by counting the total CPU use time and the idle time in two time nodes.
Further, in step 2, the linux system records the memory usage and the memory occupancy rate U of the current time in the/proc/meminfo file of the systemmemThe calculation expression is as follows:
Figure BDA0003287249460000022
wherein, MemTotal represents the size of the total memory of the terminal node; MemFree represents the size of the free memory space of the terminal node at the current moment; buffers represents the size of the disk buffer, and Cached represents the size of the disk buffer.
Further, in step 2, in the linux system, the usage of the network frame of the node is recorded in a/proc/net/dev file, where statistics of the usage of the network traffic from system boot to the current time and the usage of the network bandwidth U are countednetThe calculation expression is as follows:
Figure BDA0003287249460000031
in the above formula, Receive1，Receive2Indicates the input byte number, Transmit, at T1, T21，Transmit2Representing T1 and T2 byte number output, Net representing the bandwidth size of the network port of the terminal node, obtaining the transmission rate of the network port in the delta T time period by counting the byte number change of Receive and Transmit at two time points, and further obtaining the network bandwidth utilization rate Unet。
Furthermore, the load weight of each terminal device in step 2 determines the task processing capability of the node, and the value is obtained by integrating the total resource amount of the cluster node and the node calculation capability; the load weight is updated by adopting a dynamic balance algorithm, load performance indexes are periodically collected, and then the load weight of each node is obtained by carrying out quantitative calculation on the load performance indexes of each node in the cluster, so that the size of a node processing task is dynamically distributed;
the entropy weight method in the step 2 comprises the following steps:
step 2.1: selecting m terminal nodes, n load performance indexes and using xij(i 1, 2.. 7., m; j 1, 2.. 7., n) tableA value of the jth index of the ith node;
step 2.2: the normalization processing of the indexes, because each index has different metering units, the absolute value of the index needs to be converted into a relative value, and the normalization processing of different indexes solves the homogenization problem of different indexes, and further calculates the comprehensive index by using the values processed by the indexes; this operation is also called heterogeneous index homogenization; because the load performance index comprises a positive index and a negative index, the calculation formula is as follows:
the forward direction index is as follows:
Figure BDA0003287249460000032
negative direction index:
Figure BDA0003287249460000033
wherein, for the forward direction index pxijThe larger the numerical value is, the better the evaluation result is, and for the negative index nxijThe lower the value, the better the evaluation result;
step 2.3: calculating the proportion P of the j-th load performance index in the ith terminal nodeijThe calculation formula is as follows:
Figure BDA0003287249460000034
step 2.4: calculating an entropy value e of a load performance indicatorjThe calculation formula is as follows:
Figure BDA0003287249460000041
step 2.5: calculating the weight w of each indexjThe calculation formula is as follows:
Figure BDA0003287249460000042
step 2.6: calculating the weight score s of each nodeiThe calculation formula is as follows:
Figure BDA0003287249460000043
further, in step 3, by calculating the load weight of each terminal device, the face library deployed in the station is divided according to the proportion, so that the face library is divided into a small library according to the proportion of the load weight and distributed to different terminal devices, and the proportion of each terminal is alphajThe calculation expression is as follows:
Figure BDA0003287249460000044
wherein s isjRepresenting the load weight, s, of the jth terminal device in the multi-face-brushing arrayjThe entropy weight method in the step 2 and the load performance index in the step 1 are jointly calculated.
The invention principle is as follows: all terminal devices on a group of gates form a cluster, a large face library is divided into small face libraries and then sent to different terminal devices, so that one-time complete face recognition and passing is performed, face detection and feature extraction are performed by the current terminal device, face features are broadcasted to different terminal devices on a group of gates through a local area network, and feature comparison calculation operation is completed by all terminal arrays together, so that feature comparison response time can be shortened, and passing speed of passengers is accelerated.
Has the advantages that: compared with the prior art, the invention has the following beneficial effects: 1) the running information of the face recognition terminal is periodically collected, so that the load running condition of each terminal device can be dynamically mastered, and the running condition of each terminal device can be adjusted according to the actual situation, so that the life is simpler, more convenient and faster; 2) an entropy weight method of objective weighting is selected, and the objectivity of each load weight is ensured to the maximum extent; 3) the face library is segmented by utilizing the quantized load weight of each terminal device, so that the resource utilization efficiency of each terminal device in the end array can be effectively improved; 4) the multiple arrays cooperate with the terminal to perform face recognition, and faster gate passing service is provided.
Drawings
FIG. 1 is a flow chart of a segmentation method of a multi-face-brushing array human face library in a station.
Detailed Description
The technical solution of the present invention is described in detail below, but the scope of the present invention is not limited to the embodiments.
A station multi-face-brushing array face library segmentation method based on load balancing comprises the following steps:
step 1: the method comprises the steps of periodically collecting operation information of each terminal device in a multi-face-brushing array of a station, and calculating to obtain a load performance index of each end node;
step 2: quantizing the load weight of each terminal device, namely the magnitude of the processing task capacity by using an entropy weight method and combining the load performance indexes obtained in the step 1;
and step 3: dividing the face library into small face libraries according to the weight ratio of each terminal device in the step 2;
and 4, step 4: and transmitting the small face libraries to different terminal devices.
The load performance index in step 1 mainly includes three parts of CPU utilization, memory occupancy, and node network bandwidth utilization. The face recognition terminals used by the invention all adopt a linux system, the use condition of a CPU is saved in a/proc/stat file of the system, and the CPU use ratio UcpuThe calculation expression is as follows:
Figure BDA0003287249460000051
wherein idle1，idle2Respectively representing the time of CPU idle at T1 and T2, Tcpu1，Tcpu2Showing the total CPU usage at T1, T2 after the system is turned onIn the meantime, the CPU utilization rate in a certain time period can be obtained by counting the total CPU utilization time and the idle time in the two time nodes. The system/proc/meminfo file records the memory use condition at the current time and the memory occupancy rate UmemThe calculation expression is as follows:
Figure BDA0003287249460000052
wherein, MemTotal represents the size of the total memory of the terminal node; MemFree represents the size of the free memory space of the terminal node at the current moment; buffers represents the size of the disk buffer, and Cached represents the size of the disk buffer. The network frame use condition of the node is recorded in a/proc/net/dev file, and the statistics of the file are the use condition of network flow from system startup to the current time and the network bandwidth use ratio UnetThe calculation expression is as follows:
Figure BDA0003287249460000061
in the above formula, Receive1，Receive2Indicates the input byte number, Transmit, at T1, T21，Transmit2Representing T1 and T2 byte number output, Net representing the bandwidth of the network port of the terminal node, and obtaining the transmission rate of the network port in the delta T time period by counting the byte number change of Receive and Transmit at two time points, and further obtaining the network bandwidth utilization rate Unet。
The load weight in step 2 determines the task processing capability of the node, and the value is obtained by integrating the total resource amount of the cluster node and the node calculation capability.
And updating the load weight by adopting a dynamic balancing algorithm, periodically collecting load performance indexes, and then carrying out quantitative calculation on the load performance indexes of each node in the cluster to obtain the load weight of each node, thereby dynamically distributing the size of the node processing task.
The entropy weight method in the step 2 mainly comprises the following steps:
step 2.1: selecting m terminal nodes, n load performance indexes and using xijAnd (i 1, 2.. multidot.m, j 1, 2.. multidot.n) represents the value of the j index of the i node.
Step 2.2: the normalization of the indexes needs to convert the absolute values of the indexes into relative values because each index has different metering units, and the normalization of different indexes solves the homogenization problem of different indexes, so that the comprehensive indexes can be calculated by using the values processed by the indexes. This operation is also called heterogeneous index homogenization. Because the load performance index comprises a positive index and a negative index, the calculation formula is as follows:
the forward direction index is as follows:
Figure BDA0003287249460000062
negative direction index:
Figure BDA0003287249460000063
wherein, for the forward direction index pxijThe larger the numerical value is, the better the evaluation result is, and for the negative index nxijThe lower the value, the better the evaluation result.
Step 2.3: calculating the proportion P of the j-th load performance index in the ith terminal nodeijThe calculation formula is as follows:
Figure BDA0003287249460000064
step 2.4: calculating an entropy value e of a load performance indicatorjThe calculation formula is as follows:
Figure BDA0003287249460000071
step 2.5: calculating the weight w of each indexjWhich calculatesThe formula is as follows:
Figure BDA0003287249460000072
step 2.6: calculating the weight score s of each nodeiThe calculation formula is as follows:
Figure BDA0003287249460000073
in step 3, by calculating the load weight of each terminal device, the face library deployed in the station can be divided according to the proportion, so that the face library is divided into a small library according to the proportion of the load weight and distributed to different terminal devices, and the proportion alpha of each terminal isjThe calculation expression is as follows:
Figure BDA0003287249460000074
wherein s isjRepresenting the load weight, s, of the jth terminal device in the multi-face-brushing arrayjThe entropy weight method in the step 2 and the load performance index in the step 1 are jointly calculated.
Example 1:
it is to be noted that the following detailed description is exemplary and is intended to provide further explanation of the disclosure, unless otherwise specified, and all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments according to the present application.
Example (b):
FIG. 1 is a flow chart of a segmentation method of a multi-face-brushing array human face library in a station. In the method for segmenting the station multi-face-brushing array face database based on load balancing in the embodiment, the load operation information of each terminal node in the end array is periodically adopted to obtain the CPU utilization rate and the memory occupancy rate of each terminal node and the network bandwidth utilization rate of the node, then the load weight of each terminal node is obtained according to the three load indexes and an entropy weight method, the processing capacity of each terminal node is further quantified, and finally the face database deployed in the station is divided into small face databases according to the proportion of the load weights. The method comprises the following specific steps:
step 1: the method comprises the steps of periodically collecting operation information of each terminal device in a multi-face-brushing array of a station, and calculating to obtain a load performance index of each end node;
step 2: quantizing the load weight of each terminal device, namely the magnitude of the processing task capacity by using an entropy weight method and combining the load performance indexes obtained in the step 1;
and step 3: dividing the face library into small face libraries according to the weight ratio of each terminal device in the step 2;
and 4, step 4: and transmitting the small face libraries to different terminal devices.
The step 1 specifically comprises the following steps: the testing environment comprises three face recognition terminal devices, wherein one face recognition terminal device serves as a collaborative sub-library computing equalizer, and the remaining two face recognition terminal devices serve as terminal cluster nodes. The three face recognition terminals are all in a local area network, and load node information of the terminal equipment is regularly collected every 30 minutes, so that the face library is dynamically distributed to different terminals.
The load performance index includes a CPU usage rate, a memory occupancy rate, and a network bandwidth usage rate of the node.
CPU utilization UcpuCalculating the formula:
Figure BDA0003287249460000081
the idle1 and the idle2 respectively represent the idle time of the CPU in T1 and T2, the Tcpu1 and the Tcpu2 represent the total use time of the CPU in T1 and T2 after the system is started, and the CPU use rate in a certain time period can be obtained by counting the total use time and the idle time of the CPU in two time nodes. Memory occupancy rate UmemThe calculation expression is as followsThe following steps:
Figure BDA0003287249460000082
MemTotal represents the size of the total memory of the terminal node; MemFree represents the size of the free memory space of the terminal node at the current moment; buffers represents the size of the disk buffer, and Cached represents the size of the disk buffer.
Network bandwidth utilization rate UnetThe calculation expression is as follows:
Figure BDA0003287249460000083
the method comprises the steps that a Receive1, a Receive2 represent T1, the input byte number is in T2, Transmit1 and Transmit2 represent T1, the output byte number is in T2, Net represents the bandwidth size of a network port of a terminal node, the transmission rate of the network port in a time period can be obtained by counting the change of the byte number of the sum of two time points, and the network bandwidth utilization rate can be further obtained.
The step 2 specifically comprises the following steps: firstly, a face registration program is utilized to send a face recognition request to terminal equipment, stat, meninfo and dev files of the three terminal equipment are initialized randomly, load performance of the equipment of the three terminals is obtained, so that library splitting operation of a face library is carried out randomly, then the terminal equipment obtains face images collected by Rtsp cameras, and finally time calculated by terminal array collaborative library splitting and time consumed by a single terminal equipment for completing polling of all the face libraries are calculated respectively for different numbers of face libraries.
The entropy weight method mainly comprises the following steps:
selecting m terminal nodes, n load performance indexes and using xijAnd (i 1, 2.. multidot.m, j 1, 2.. multidot.n) represents the value of the j index of the i node.
The normalization of the indexes needs to convert the absolute values of the indexes into relative values because each index has different metering units, and the normalization of different indexes solves the homogenization problem of different indexes, so that the comprehensive indexes can be calculated by using the values processed by the indexes. This operation is also called heterogeneous index homogenization. Because the load performance index comprises a positive index and a negative index, the calculation formula is as follows:
the forward direction index is as follows:
Figure BDA0003287249460000091
negative direction index:
Figure BDA0003287249460000092
wherein, for the forward direction index pxijThe larger the numerical value is, the better the evaluation result is, and for the negative index nxijThe lower the value, the better the evaluation result.
Calculating the proportion P of the j-th load performance index in the ith terminal nodeijThe calculation formula is as follows:
Figure BDA0003287249460000093
calculating an entropy value e of a load performance indicatorjThe calculation formula is as follows:
Figure BDA0003287249460000094
calculating the weight w of each indexjThe calculation formula is as follows:
Figure BDA0003287249460000095
calculating the weight score s of each nodejThe calculation formula is as follows:
Figure BDA0003287249460000101
the step 3 specifically comprises the following steps: during identification, socket communication is used inside, the face library deployed in the station can be divided according to the proportion by calculating the load weight of each terminal device, and the proportion alpha of each terminal isjThe calculation expression is as follows:
Figure BDA0003287249460000102
the step 4 specifically comprises the following steps: and dividing the face library into a small library according to the load weight ratio, distributing the small library to different terminal equipment, and returning the recognition result to the face recognition terminal requesting recognition to finish final face recognition.

Claims (7)

1. A station multi-face-brushing array face library segmentation method based on load balancing is characterized by comprising the following steps:
step 1: the method comprises the steps of periodically collecting operation information of each terminal device in a multi-face-brushing array of a station, and calculating to obtain a load performance index of each end node;
step 2: quantizing the load weight of each terminal device, namely the magnitude of the processing task capacity by using an entropy weight method and combining the load performance indexes obtained in the step 1;
and step 3: dividing the face library into face libraries according to the weight ratio according to the load weight of each terminal device in the step 2;
and 4, step 4: and (4) transmitting the face library obtained in the step (3) to different terminal devices.
2. The method for segmenting the station multi-face-brushing array face library based on load balancing according to claim 1, wherein the load performance indexes in the step 1 include a CPU utilization rate, a memory occupancy rate and a network bandwidth utilization rate of nodes.
3. The method for segmenting the station multi-face-brushing array human face library based on load balancing according to claim 1,in step 2, the terminal device is a face recognition terminal, the face recognition terminal adopts a linux system, the use condition of a CPU is stored in a/proc/stat file of the system, and the CPU use ratio U is usedcpuThe calculation expression is as follows:
Figure FDA0003287249450000011
wherein idle1，idle2Respectively representing the time of CPU idle at T1 and T2, Tcpu1，Tcpu2The total CPU use time at T1 and T2 after the system is started is represented, and the CPU use rate in a certain time period is obtained by counting the total CPU use time and the idle time in two time nodes.
4. The method for segmenting the station multi-face-brushing array human face library based on load balancing as claimed in claim 3, wherein in the step 2, the linux system records the memory usage condition and the memory occupancy rate U of the system in a/proc/meminfo file at the current momentmemThe calculation expression is as follows:
Figure FDA0003287249450000012
wherein, MemTotal represents the size of the total memory of the terminal node; MemFree represents the size of the free memory space of the terminal node at the current moment; buffers represents the size of the disk buffer, and Cached represents the size of the disk buffer.
5. The method for segmenting the station multi-face-brushing array face library based on load balancing as claimed in claim 4, wherein in step 2, the network frame usage of the linux system and the node is recorded in a/proc/net/dev file, and the file statistics includes the usage of network traffic from system startup to the current time and the network bandwidth usage rate UnetThe calculation expression is as follows:
Figure FDA0003287249450000021
in the above formula, Receive1，Receive2Indicates the input byte number, Transmit, at T1, T21，Transmit2Representing T1 and T2 byte number output, Net representing the bandwidth size of the network port of the terminal node, obtaining the transmission rate of the network port in the delta T time period by counting the byte number change of Receive and Transmit at two time points, and further obtaining the network bandwidth utilization rate Unet。
6. The method for segmenting the station multi-face-brushing array human face library based on load balancing according to claim 5, wherein the load weight of each terminal device in the step 2 determines the capability of the node to process tasks, and the value is obtained by integrating the total resource amount of cluster nodes and the node calculation capability; the load weight is updated by adopting a dynamic balance algorithm, load performance indexes are periodically collected, and then the load weight of each node is obtained by carrying out quantitative calculation on the load performance indexes of each node in the cluster, so that the size of a node processing task is dynamically distributed;
the entropy weight method in the step 2 comprises the following steps:
step 2.1: selecting m terminal nodes, n load performance indexes and using xij(i 1, 2.. multidot.m, j 1, 2.. multidot.n) represents the value of the j index of the i node;
step 2.2: the normalization processing of the indexes, because each index has different metering units, the absolute value of the index needs to be converted into a relative value, and the normalization processing of different indexes solves the homogenization problem of different indexes, and further calculates the comprehensive index by using the values processed by the indexes; this operation is also called heterogeneous index homogenization; because the load performance index comprises a positive index and a negative index, the calculation formula is as follows:
the forward direction index is as follows:
Figure FDA0003287249450000022
negative direction index:
Figure FDA0003287249450000023
wherein, for the forward direction index pxijThe larger the numerical value is, the better the evaluation result is, and for the negative index nxijThe lower the value, the better the evaluation result;
step 2.3: calculating the proportion P of the j-th load performance index in the ith terminal nodeijThe calculation formula is as follows:
Figure FDA0003287249450000031
step 2.4: calculating an entropy value e of a load performance indicatorjThe calculation formula is as follows:
Figure FDA0003287249450000032
step 2.5: calculating the weight w of each indexjThe calculation formula is as follows:
Figure FDA0003287249450000033
step 2.6: calculating the weight score s of each nodeiThe calculation formula is as follows:
Figure FDA0003287249450000034
7. the station multi-face-brushing-array face library segmentation method based on load balancing as claimed in claim 6The method is characterized in that in the step 3, the face library deployed in the station is divided according to the proportion by calculating the load weight of each terminal device, so that the face library is divided into a small library according to the proportion of the load weight and distributed to different terminal devices, and the proportion alpha of each terminal isjThe calculation expression is as follows:
Figure FDA0003287249450000035
wherein s isjRepresenting the load weight, s, of the jth terminal device in the multi-face-brushing arrayjThe entropy weight method in the step 2 and the load performance index in the step 1 are jointly calculated.
CN202111151261.XA 2021-09-29 2021-09-29 Station multi-face-brushing array face library segmentation method based on load balancing Pending CN113886081A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111151261.XA CN113886081A (en) 2021-09-29 2021-09-29 Station multi-face-brushing array face library segmentation method based on load balancing

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111151261.XA CN113886081A (en) 2021-09-29 2021-09-29 Station multi-face-brushing array face library segmentation method based on load balancing

Publications (1)

Publication Number Publication Date
CN113886081A true CN113886081A (en) 2022-01-04

Family

ID=79007984

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111151261.XA Pending CN113886081A (en) 2021-09-29 2021-09-29 Station multi-face-brushing array face library segmentation method based on load balancing

Country Status (1)

Country Link
CN (1) CN113886081A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114863541A (en) * 2022-07-06 2022-08-05 南京熊猫电子股份有限公司 Station multi-face terminal equipment cooperation method supporting degradation application

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114863541A (en) * 2022-07-06 2022-08-05 南京熊猫电子股份有限公司 Station multi-face terminal equipment cooperation method supporting degradation application

Similar Documents

Publication Publication Date Title
Hauswald et al. A hybrid approach to offloading mobile image classification
CN102222213B (en) Distributed vision computing method based on open type Web Service framework
US10692089B2 (en) User classification using a deep forest network
US20210312587A1 (en) Distributed image analysis method and system, and storage medium
CN111553213A (en) Real-time distributed identity-aware pedestrian attribute identification method in mobile edge cloud
CN111898578A (en) Crowd density acquisition method and device, electronic equipment and computer program
CN113886081A (en) Station multi-face-brushing array face library segmentation method based on load balancing
CN110909817B (en) Distributed clustering method and system, processor, electronic device and storage medium
CN112668632A (en) Data processing method and device, computer equipment and storage medium
CN114066620A (en) Client information processing method and device based on client portrait
CN117114623A (en) Intelligent management method and system for monitoring equipment in park
CN116561445A (en) Wedding and love object bidirectional matching method, device and storage medium based on coupling selection condition
CN110489175A (en) Service processing method, device, server and storage medium
CN113239879B (en) Federal model training and certificate detection methods, devices, equipment and media
CN110532071B (en) Multi-application scheduling system and method based on GPU
CN111767419A (en) Picture searching method, device, equipment and computer readable storage medium
CN115546846A (en) Image recognition processing method and device, electronic equipment and storage medium
CN113849454A (en) Computing system
CN116012067A (en) Resource allocation method, apparatus, computer, readable storage medium, and program product
CN113343004A (en) Object recognition method and device, storage medium and electronic device
CN113128278A (en) Image identification method and device
Wang et al. Edge-Cloud Collaboration Architecture for Efficient Web-Based Cognitive Services
CN113094530B (en) Image data retrieval method and device, electronic equipment and storage medium
CN117812185B (en) Control method and system of intelligent outbound system
US20210311795A1 (en) Method and apparatus for allocating database server

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination