KR20210108720A - Fingerprint High Speed Matching System for Mass Fingerprint Identification - Google Patents

Abstract

The present invention relates to a high speed fingerprint matching system for mass fingerprint identification which rapidly confirms an identity of a subject by rapidly acquiring data from a plurality of matching servers through a high speed 1 vs N (1 vs multiple) fingerprint matching method to be applied to a distributed environment. The high speed fingerprint matching system comprises: an identity confirmation detector to confirm an identity associated with one work among polices, immigration, soldiers, finance, medical treatment, or social welfare; a web server to transmit a matching request signal including input fingerprint information to an application programming interface (API) server when an identify confirming request is input together with an input fingerprint from the identity confirmation detector, and for receiving a matching result from a matching DB to transmit the received matching result to the identity confirmation detector; an API server to register the matching request signal received from the web server in the matching DB; and a matching server for reading the matching request signal from the matching DB and performing matching by applying 1 vs N (1 vs multiple) matching algorithm and 1 vs 1 matching algorithm, selecting candidates matched with the input fingerprint, and generating and storing the selected candidates list in the matching DB.

Description

Fingerprint High Speed Matching System for Mass Fingerprint Identification

The present invention obtains data at a high speed from a plurality of matching servers and quickly identifies the subject through a high-speed 1vsN (one-to-many) fingerprint matching method, and a large amount of fingerprints applicable to a distributed environment It relates to a fingerprint high-speed matching system for identification.

It takes a lot of time to process a fingerprint search for a large number of fingerprints, such as millions of fingerprints or tens of millions of fingerprints. For example, when performing 10,000 fingerprint scans per second for 12 million fingerprints, it takes a lot of time to perform a large-scale fingerprint scan that takes 1200 seconds (20 minutes) to perform a fingerprint scan for all objects. . In addition, it takes a lot of time to obtain a large amount of fingerprint data. Obtaining data from a file or DB that stores fingerprint feature point data requires considerable time, memory and disk space, such as time to search the contents and network time, and the time to obtain data is shorter than the processing time of the fingerprint matching algorithm. It usually costs more.

Therefore, in order to confirm the identity of the subject within 60 seconds in the matching system after requesting a fingerprint search, the fingerprint data for the search is quickly obtained, and a plurality of fingerprints are retrieved from a number of servers through an algorithm that processes the fingerprint search at a high speed. How to do it is desired

Therefore, the present invention obtains data at a high speed from a plurality of matching servers and quickly identifies the subject through a high-speed 1vsN (one-to-many) fingerprint matching method, and high-speed fingerprint matching applicable to a distributed environment propose a system.

In the present invention, a matching processing unit (MPU, Matching Processing Unit) that performs a matching operation to perform a fingerprint search operation requested by a plurality of servers, and a matching control unit (MPU) that collects and manages the matching request operation and results, and manages the MPUs ( MCU, Matching Control Unit).

As a prior art, Japanese Patent Laid-Open No. 2003-30661 relates to a fingerprint combination device that matches one input fingerprint with a plurality of registered fingerprints, but the matching method is not fast because the existing method is applied as it is.

The task to be achieved by the present invention is to obtain data at a high speed from a number of matching servers and quickly identify the subject through a high-speed 1vsN (one-to-many) fingerprint matching method, and apply it to a distributed environment It is possible to provide a high-speed fingerprint matching system for the identification of a large number of fingerprints.

Another object of the present invention is to obtain and manage fingerprint data at high speed for a large number of target fingerprints, and to have a structure of an MCU (Matching Control Unit) and a plurality of MPUs (Matching Processing Unit) operating in a distributed environment It is to provide a high-speed fingerprint matching system for large-scale fingerprint identification, which collects the results of matching in a plurality of MPUs and registers the upper results from the collected results.

In order to solve the above problems, the fingerprint matching system of the present invention is a terminal for identification in relation to one of the police, immigration, military, finance, medical care, and social welfare, the identification detection device; When an identification request is input together with an input fingerprint from the identification detection device, a matching request signal including input fingerprint information is transmitted to an API (application programming interface) server, and a matching result is received from the matching DB, the identification detection device to the web server; an API (application programming interface) server that registers the matching request signal received from the web server, in the matching DB, the matching request signal; By reading a matching request signal from the matching DB, matching is performed by applying a 1vsN (1 to many) matching algorithm and a 1vs1 (1 to 1) matching algorithm to select candidates that match the input fingerprint, and generate a list of selected candidates and storing it in the matching DB, a matching server unit;

The matching server unit includes an MCU (Matching Control Unit) and a plurality of MPUs (Matching Processing Units), and the MCU receives the matching request information registered in the matching DB, and generates a result data set for the corresponding matching request, A matching request is sent to a plurality of MPUs (Matching Processing Units), and the MPU records a bundle of requested feature information and target feature information in a matching queue based on the matching request information received from the MCU, and the matching thread The matching information is obtained from the matching queue, the matching is performed, the results are recorded, and when processing for one matching request is completed, the MPU 120 collects the matching results and sorts the results in the order of the highest score, The result of the upper predetermined number of cases is transmitted to the MCU as a list of candidates for each MPU.

When the MCU receives the MPU-specific candidate list as the MPU matching result, it records it in the matching result data set, and the matching result data set receives the MPU-specific matching results from all MPUs and rearranges them to generate a candidate list, record

The MCU has a Network Manager in charge of network processing, a Job Manager that processes received requests, and a Request Manager that checks whether there is a new fingerprint search request and sends a matching request to the MPUs. ), the result set that collects the result data and records the result in the DB (database) when all the results are received, and the matching (data access object) DAO (Data Access Object), MPU in charge of DB (database) processing It includes an MPU client (Client) that sends a request to them.

The MPU is a network manager that handles network processing, a job manager that handles received requests, a matching thread that performs matching tasks in the form of threads, and a system that manages the entire target fingerprint data. Matching data set (Matching Data Set), including MCU client (Client) that transmits the result to the MCU.

The matching server unit, in direct management of key points in the matching system, bundles and processes key point data in units processed by 1vsN matching algorithm and 1vs1 matching algorithm, creates a memory manager that processes location in Key/Value form, and provides basic data to the memory manager After loading the fingerprint data, it is processed by reading it.

The fingerprint high-speed matching system for mass fingerprint identification of the present invention obtains data from a number of matching servers at a high speed and uses a high-speed 1vsN (one-to-many) fingerprint matching method to quickly identify the subject. It can be checked, can be applied to a distributed environment, and can identify a large number of fingerprints.

That is, the present invention is capable of obtaining and managing fingerprint data at high speed for a large number of target fingerprints, and has an algorithm that performs 1 vs. multiple fingerprint matching for a large number of target fingerprints at high speed, in a distributed environment. It is designed in the structure of an operating MCU (Matching Control Unit) and multiple MPUs (Matching Processing Units).

Accordingly, the high-speed fingerprint matching system of the present invention is configured to quickly secure fingerprint data for search and search for a plurality of fingerprints in a plurality of servers through an algorithm that processes the fingerprint search at a high speed, so that after the user requests a fingerprint search, In the matching system, the identity of the subject can be confirmed within 60 seconds.

In addition, the present invention can be applied to the development of a prototype test bed environment.

1 is an explanatory diagram schematically illustrating a fingerprint high-speed matching system for mass fingerprint identification of the present invention.
FIG. 2 is a configuration diagram of a matching server unit of the fingerprint high-speed matching system of FIG. 1 .
FIG. 3 is a flowchart for explaining the operation of the fingerprint high-speed matching system of FIG. 1 .

The high-speed fingerprint matching system for mass fingerprint identification of the present invention will be described in detail below with reference to the accompanying drawings.

<Configuration of high-speed fingerprint matching system>

1 is an explanatory diagram schematically illustrating a fingerprint high-speed matching system for mass fingerprint identification of the present invention, and FIG. 2 is a configuration diagram of a matching server unit of the high-speed fingerprint matching system of FIG. 1 .

Identification from various identification detection devices 10 such as local police-related identification unit, immigration identification unit, military security-related identification unit, financial service-related application unit, medical-related patient identification unit, social welfare-related identification unit, etc. When a confirmation request signal is requested from the web server (or main server or application server) 20 through the Internet network, the identity confirmation detection device 10 is configured to perform security authentication through the web server 20 through the application service unit. The API (application programming interface) server 51 of 50 is interlocked, and the identification detection device 10 checks the identity of a specific application (eg, local police-related identification application) through the web server 20 . The matching request signal is transmitted to the API server 51 , and the API server 51 registers the received matching request signal in the matching DB of the application DB 52 . The API server 51 transmits the matching request signal registered in the matching DB to the matching server unit (matching server group) 100 in order, and the matching result from the matching server unit (matching server group) 100 is returned to the API The server 51 receives it and transmits it to the identification detection device 10 through the web server 20 .

Here, the identification detection device 10 is related to a local police-related identification unit, an immigration identification unit, a military security-related identification unit, a financial service-related application unit, a medical-related patient identification unit, a social welfare-related identification unit, etc. It may be a terminal (in some cases, a smartphone) of the person in charge.

When a matching request signal is received through the web server 20, etc. and the matching request signal is registered in the matching DB by the API server 51, the MCU 110 automatically reads the new matching request information registered in the matching DB. and a result data set for the corresponding matching request is generated, and then the matching request is transmitted to the MPU 120 . The MPU 120 records a bundle of requested feature information and target feature information in a matching queue based on the received matching request information, and the matching thread obtains the corresponding matching information uploaded to the matching queue, performs matching, and records the result. When the processing of one matching request is completed in this way, the MPU 120 collects the matching results, sorts the results in the order of the highest score, and then delivers the results of the top N cases (eg, 100 cases) to the MCU 110 . . When the MCU 110 receives the matching result from the MPU 120 , it records it in the matching result data set, and the matching result data set receives the matching result from all MPUs 120 , rearranges the result, and then records it in the matching DB.

The matching server unit 100 includes a matching control unit (MCU) 110 and a plurality of matching processing units (MPU, Matching Processing Unit) 120 . Here, since the core function of the fingerprint high-speed matching system is performed by the matching server unit 100, the fingerprint high-speed matching system is largely composed of a single matching control unit (MCU, Matching Control Unit) 110 and a plurality of matching processing units (MPU, MPU, Matching Processing Unit) 120 may be said to be configured.

The MCU 110 has a network manager in charge of processing the network, a job manager that processes the received request, and a request manager that checks whether there is a new fingerprint search request and sends a matching request to the MPUs. Manager), a result set that collects result data and records the results in DB (database) when all results are received, and Matching DAO (Data Access Object), MPU in charge of DB (database) processing Including the MPU client (Client) and the like to send a request to (120).

The MPU 120 includes a network manager in charge of network processing, a job manager for processing received requests, a matching thread for performing a matching operation in the form of a thread, and the entire target fingerprint data. It includes a matching data set that manages , and an MCU client that transmits the result to the MCU.

3 is a flowchart for explaining the operation of the fingerprint high-speed matching system of FIG. 1 .

In the matching request step, when a matching request signal is received through the web server 20, etc., and the matching request signal is registered in the matching DB of the application DB 52 by the API server 51 (S110), the API server 51 ), the MCU 110 reads the new matching request information registered in the matching DB (S120), generates a result data set for the corresponding matching request (S130), and then asks the MPUs 120 for a matching request (S130). is transmitted (S140).

In the matching step, the MPU 120 records a bundle of requested feature point information and target feature point information based on the matching request information received in a matching queue (S150), and the matching thread is the matching information that is uploaded to the matching queue. (ie, requested feature point information and target feature point information), perform matching, and record the result (S160). Here, the matching thread performs matching based on the requested feature point.

In the MPU-specific candidate list selection step, when the processing for one matching request in the matching step is completed, the MPU 120 collects the matching results and sorts the results in the order of the highest score (S170), and the upper predetermined number (that is, , the top N cases (eg, 100 cases)), as a list of candidates for each MPU, is transmitted to the MCU 110 (S180).

In the final candidate list selection step, the MCU 110 receives the MPU-specific candidate list as the matching result from the MPU 120, records it in the matching result data set, and the matching result data set is MPU-specific in all MPUs 120 When a matching result (ie, a list of candidates for each MPU) is received, the result is rearranged to generate a list of candidates (S200), and it is recorded in the matching DB (S210).

The web server 20 requests a matching result from the matching DB, inquires the candidate list selected in the final candidate list selection step (S220), and transmits it to the identification detection device 10 for display.

<Data Fetch>

In general, as a type of data fetch method, fingerprint data is stored in the DB, the stored fingerprint data is directly obtained through a DB query, and the obtained data is stored as a file and then processed based on the file. There are a number of methods, a method of using the obtained data as a memory cache using an in-memory store such as Redis, and a method of accumulating and utilizing the obtained data within the development program. The method of directly obtaining data for fingerprint matching through a DB query is the most intuitive and general method, but has a disadvantage in that it takes a considerable amount of time to fetch the DB. In addition, in the case of methods such as file, in-memory store, and direct program management, the time to fetch data through DB query is relatively short, but it is confirmed that the fingerprint data stored in the DB and the fingerprint data handled in files are completely consistent. The downside is that it is difficult to do. In this case, it is necessary to first obtain the first data from the DB and load it into a file, etc., and additional processing is required to match the sync with the DB data. In the case of direct management of key points in the matching system, the key point data is bundled and processed by the matching algorithm, and a memory manager that processes the location in a key/value format is created, basic data is loaded in the memory manager, and then the fingerprint data is read and processed. constructed in this way. However, in the case of directly managing the key points in this form, the size of the entire key point data that can be read is limited according to the size of the entire physical memory. In the present invention, as a result of testing in each fetch method, the general method obtained from DB Fetch was slow enough to be adopted, and the method using a file or in-memory store was also relatively fast compared to DB, but it was difficult to achieve sufficient speed. In the matching system of the present invention, that is, data patching in the matching request stage is basically made to directly create a memory manager and directly manage the feature point data in the memory manager.

<Matching Algorithm Processing>

In the case of the existing 1vs1 matching algorithm, the accuracy is very high, but there is a disadvantage that it takes a lot of time. In the case of 1vsN matching algorithm, filter processing is primarily performed on the entire data by the highest candidate group, and 1vs1 is performed on only the highest candidate group to perform matching on the same target at a relatively faster speed, but 1vs1 matching is omitted. can occur

In the case of processing with the 1vsN matching algorithm, the highest candidate group is first created based on the index of one search-required characteristic point and a plurality of matching target characteristic points (information generated by indexing the correlation within the fingerprint characteristic point), and then only the highest candidate group is targeted to perform a one-to-one comparison. In this way, one-to-one comparison, which is a time-consuming task, can be reduced, and matching can be performed at a much faster speed. As a verification of this, as a result of testing the matching algorithm of the present invention in multi-thread or multi-process processing, it was confirmed that matching proceeds faster when multi-thread or multi-process is used. In general, it is desirable to create multi-threads or multi-processes as many as the number of cores of the CPU of the test computer.

Table 1 shows the results of comparing the matching speed between the system of the present invention and the conventional AFIS system.

Here, the AFIS format is the result of fetching data from the DB and then performing matching by the 1vs1 algorithm, and the format of the present invention is the result of performing matching by the 1vsN algorithm based on the data already fetched and loaded in the memory.

As a result of comparison, it can be seen that the fingerprint search is processed 800 times faster than when performing the fingerprint search in the same way as AFIS. Even if it is assumed that one MPU server handles approximately 10 million fingerprints, the result can be calculated in approximately 12.5 seconds. Even assuming a number of other additional time-consuming factors, the target result can be achieved within 60 seconds, and if a higher speed is required, further improvement is possible through parameter adjustment or tuning of the algorithm. However, in the case of processing in this form, there is a disadvantage that one device cannot process more than the total amount of physical memory of the device in which the MPU operates because the fingerprint data to be searched is loaded into the actual memory and processed. About 40.55 GiB in all cases) However, if sufficient memory is secured or a sufficiently large number of distributed environments are established, it can be expected to process fingerprint searches at a considerable speed.

<MCU Configuration>

The MCU is based on Global, which is responsible for initialization and access of major objects such as JobManager, MatchingThread, and ResultSet, which handle the entire operation of the MCU. There is a JobManager that handles . JobManager creates a JobThread for the received request to process feature point addition, removal, and update tasks and processing the matching result. MatchingThread periodically checks whether a new matching task exists. If a new task exists, it creates an MpuResulSet, prepares the result to be received from the MPU in advance, and sends a matching request to the MPU. When the MPU delivers the matching result, the JobThread records the result in the MpuResultSet. When the result arrives from all MPUs, the result is sorted, and then the matching result is recorded in the DB using Dao.

< MPU configuration>

MPU is based on Global, which is responsible for initialization and access of major objects such as JobManager, MatchingThread, etc., which handle the entire task of the MPU. There is a JobManager that handles it. JobManager creates a JobThread for the received request and registers matching data in MtxQueue for the processing of feature point addition, removal and update tasks and matching requests, so that MatchingThread handles the matching. MpuDataSet is a structure that stores all target fingerprint data in memory, and does not use additional memory for operations such as memory copy during 1vsN (multiple) matching. to optimize performance for matching. When a specific matching task is finished, the corresponding result is registered in the MCU using MCUClient.

In the fingerprint high-speed matching system of the present invention, in the case of direct management of key points in the matching system, a memory manager is created that bundles key point data in units processed by the matching algorithm, and processes the location in a key/value format, and provides basic data to the memory manager. It is configured in such a way that the fingerprint data is read and processed after loading.

As a result of testing in each fetch method, the general method obtained from DB fetch was slow enough to be difficult to adopt, and the method using a file or in-memory store was also relatively fast compared to DB, but it was difficult to achieve sufficient speed, so the matching system of the present invention Basically, it was decided to create a direct memory manager and manage keypoint data directly in the memory manager.

In the present invention, in the case of processing with the 1vsN matching algorithm, the highest candidate group is first created based on the index of one search-required feature point and a plurality of matching target feature points (information generated by indexing the correlation within the fingerprint feature point), and then As a result, a one-to-one comparison was performed on only the highest candidate group.

As a result of testing the matching algorithm of the present invention in multi-thread or multi-process processing, it was confirmed that matching proceeds faster when using multi-thread or multi-process. As a result of testing under each condition, the 1vsN algorithm and multi-process It was confirmed that it is optimal to perform matching using threads.

In the present invention, the MCU is basically responsible for the initialization and access of major objects such as JobManager, MatchingThread, ResultSet, etc., which process the entire operation of the MCU, and the MPU is responsible for the overall operation of the MPU, such as JobManager, MatchingThread, etc. Based on Global, which is responsible for initialization and access of main objects, Main is responsible for network processing in the server for overall initialization. In addition, since MCU and MPU have complex parts in maintenance using multiple languages, they can be developed in the form of a socket server after removing unnecessary parts and simplifying them for easy maintenance by integrating functions with C++.

As described above, although the present invention has been described with reference to the limited examples and drawings, the present invention is not limited to the above examples, which are various modifications and Transformation is possible. Therefore, the spirit of the present invention should be understood only by the claims described below, and all equivalents or equivalent modifications thereof will fall within the scope of the spirit of the present invention.

10: identification detection device 20: web server
50: application service unit 51: API server
52: application DB 53: matching server unit
110: MCU 120: MPU

Claims (6)

An identification detection device, which is a terminal for identification in relation to one of the police, immigration, military, finance, medical, and social welfare;
When an identification request is input together with an input fingerprint from the identification detection device, a matching request signal including input fingerprint information is transmitted to an API (application programming interface) server, and a matching result is received from the matching DB, the identification detection device to the web server;
API (application programming interface) server for registering the matching request signal received from the web server, the matching DB, the matching request signal;
By reading a matching request signal from the matching DB, matching is performed by applying a 1vsN (1 to many) matching algorithm and a 1vs1 (1 to 1) matching algorithm to select candidates that match the input fingerprint, and generate a list of selected candidates and storing it in the matching DB, a matching server unit;
A fingerprint matching system comprising a. According to claim 1,
The matching server unit includes an MCU (Matching Control Unit) and multiple MPUs (Matching Processing Units),
The MCU receives the matching request information registered in the matching DB, generates a result data set for the corresponding matching request, and transmits the matching request to a plurality of MPUs (Matching Processing Units),
Based on the matching request information received from the MCU, the MPU records the requested feature point information and the target feature point information bundle in a matching queue, and the matching thread obtains the matching information uploaded to the matching queue and performs matching. When the processing for one matching request is completed, the MPU 120 collects the matching results, sorts the results in the order of the highest score, and sends the results of the upper predetermined number of cases to the MCU as a list of candidates for each MPU. Characterized in delivering, the fingerprint matching system. 3. The method of claim 2,
When the MCU receives the MPU-specific candidate list as the MPU matching result, it records it in the matching result data set, and the matching result data set receives the MPU-specific matching results from all MPUs and rearranges them to generate a candidate list, A fingerprint matching system, characterized in that it records. 4. The method of claim 3,
The MCU has a Network Manager in charge of network processing, a Job Manager that processes received requests, and a Request Manager that checks whether there is a new fingerprint search request and sends a matching request to the MPUs. ), the result set that collects the result data and records the result in the DB (database) when all the results are received, and the matching (data access object) DAO (Data Access Object), MPU in charge of DB (database) processing The fingerprint matching system, characterized in that it comprises an MPU client (Client) for sending a request to them. 4. The method of claim 3,
The MPU is a network manager that handles network processing, a job manager that handles received requests, a matching thread that performs matching tasks in the form of threads, and a system that manages the entire target fingerprint data. Matching data set (Matching Data Set), characterized in that it comprises an MCU client (Client) for transmitting the result to the MCU, fingerprint matching system. The method of claim 3, wherein the matching server unit,
In the direct management of key points in the matching system, the 1vsN matching algorithm and 1vs1 matching algorithm bundle and process the key point data in units processed, create a memory manager that processes the location in the key/value format, load the basic data into the memory manager, and then A fingerprint matching system, characterized in that it is made in a way that reads and processes fingerprint data.

Images