KR20190076745A - Method for controlling gate using machine learning - Google Patents

Abstract

According to an embodiment of the present invention, a visitor control method comprises: a visitor image generation process in which a camera photographs a gate defined between two objects facing each other and generates a visitor image of a visitor passing through the gate; a visitor image storage process in which a machine learning server stores the visitor image together with the time of access; an access information storage process in which the machine learning server classifies each visitor in the visitor image, extracts a visitor face image, and stores access information for each read visitor face image; a machine learning process in which the machine learning server determines whether a preset visitor classification condition is met using the access information, and stores a visitor face image corresponding to the visitor classification condition as learning data; a pre-registered person classification process in which the machine learning server classifies a visitor having the visitor face image as a pre-registered person when the visitor face image exists in preset learning data; and a new person classification process in which the machine learning server classifies a visitor having the visitor face image as a new person when the visitor face image does not exist in the preset learning data.

Description

[0001] The present invention relates to a method for controlling access to a moving object,

The present invention relates to a method for controlling an accessing person, which can control accessing persons through machine learning learning.

In the industrial field, there are workplaces where various work processes are performed. Also, the larger the industrial site, the more work will be carried out by many partner companies under the supervision of the supervisor. However, there are the following problems that the personnel management of partner companies is not efficient in such an industrial field.

There is a problem that it is difficult to understand how much personnel should be input for each workplace. According to the supervisor's instructions, the necessary personnel are input into each workplace. In some cases, too many personnel are input to the site where a large number of workers are not needed, while too few persons are input to a site requiring a large number of workers . Therefore, there is a problem in that the work process is delayed due to the readjustment of the input personnel during the work progress.

In addition, in the management of personnel at the industrial site, the management is dependent on the output log of each partner company and the daily work log, or the personnel is identified by the safety manager at the site. Access control by the work log, , Management by health status and history may not be efficiently performed. Especially, in the case of industrial sites where large-scale processes are carried out by the work of several partner companies, it is difficult to grasp whether each of the employees of the partner company correctly put the planned personnel into the industrial site. As a result, disputes may arise between the supervisor and the supplier later on.

In particular, there is a problem that it is difficult to confirm the presence of the passengers at the entrance and the passengers who can enter the open industrial space, not the closed space.

Korean Registration No. 10-1435572

The technical problem of the present invention is to provide an access control means capable of efficiently controlling the passengers through machine learning learning.

The camera of the present invention may include a moving image generating step of generating a moving image of a person who has photographed a gate passing through a gate by photographing a gate defined between two objects facing each other; A machine learning learning server for storing the accessing image together with the access time; An access information storing step of the machine learning learning server classifying each of the passers by the passer-by image, extracting the passer-bye face image, and storing access information for each of the passer-bye face images to be read; A machine learning learning process for determining whether the machine learning learning server meets a preset accessor classification condition using the access information and storing the entrance face image satisfying the entry classification condition as learning data; Wherein the machine learning learning server classifies a passenger having the entrance face image as a registered entry force if the entrance face image exists in preset running data; And a new manpower classifying step of classifying the machine learning learning server as a new manpower if the entrance face image is not present in the preset running data.

The passenger classification condition may include an access frequency classification condition and an entry-time passenger classification condition.

Wherein the access frequency classification condition is a condition that satisfies an entrance face image that exceeds a preset access count and the access time classification type condition satisfies an entrance face image having an entrance / can do.

The passenger image generating process includes a passer-and -resser image, which is an entrance image before passing through a gate between two objects, a passer -by image that is an entrance image passing through a gate between two objects, a gate between two objects, It is possible to generate a viewer image after passing through the viewer image.

The passenger image storage process may include storing the passer-by-passer image, passer-by image, passer-by image, and passer-by image image into a group for each passenger.

The passer-by-passer image extraction process identifies each of the passers-by in the passer-by-passer image as the passer-by-passer face image, extracts the passer-by-passer image as passer- The passengers in the passenger image after the passing can be respectively identified and extracted as passenger face images after pass.

Wherein the passenger classification condition is a condition for a similarity rate between a facial expression of a passer-by-pass facial image, a facial expression of a passer-by-pass facial image, and a facial expression of a passer-by facial image after passage, If the rate is within the predetermined tolerance range, it is determined that the user is a normal person and stored as running data. On the other hand, if the similarity rate is out of the allowable error range, have.

According to the embodiment of the present invention, access control can be efficiently performed by classifying new manpower and existing manpower through machine learning learning. Also, according to the embodiment of the present invention, it is possible to efficiently perform the access control by learning the face pattern of the entrance and exit lines of the passerby passing through the set security zone.

1 is a block diagram of an access control system through machine learning learning according to an embodiment of the present invention;
2 is a view showing a state where a camera is installed according to an embodiment of the present invention.
3 is a view showing a pre-passenger image according to an embodiment of the present invention.
4 is a view showing a passenger image during passage according to an embodiment of the present invention.
FIG. 5 is a view showing a passenger image after passage according to the embodiment of the present invention. FIG.
6 is a view showing a state in which a warning alarm is generated when a new manpower is judged according to an embodiment of the present invention.
FIG. 7 is a flowchart illustrating an access control process through machine learning learning according to an embodiment of the present invention; FIG.

BRIEF DESCRIPTION OF THE DRAWINGS The advantages and features of the present invention, and how to achieve them, will be apparent from the following detailed description of embodiments thereof taken in conjunction with the accompanying drawings. The present invention may, however, be embodied in many different forms and should not be construed as being limited to the exemplary embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete and will fully convey the concept of the invention to those skilled in the art. And the present invention is only defined by the scope of the claims. In the following description, well-known functions or constructions are not described in detail since they would obscure the invention in unnecessary detail.

FIG. 1 is a configuration diagram of a user access control system through machine learning learning according to an embodiment of the present invention. FIG. 2 is a diagram illustrating a camera installed according to an embodiment of the present invention. FIG. 4 is a view showing a passenger image during passage according to an embodiment of the present invention, and FIG. 5 is a view showing a passenger image after passing according to an embodiment of the present invention FIG. 6 is a diagram illustrating a state in which a warning alarm is generated when a new manpower is determined according to an embodiment of the present invention.

The access control system of the present invention includes a camera 100 and a machine learning learning server 200 as shown in Fig.

The camera 100 is a module for photographing a gate defined between two objects facing each other and generating an entrance image of a person who has taken a pass through the gate. That is, the two facing objects may be a facing desk, a flowerpot facing each other, a facing advertising column, a cabinet of a paper box, and the like. Accordingly, when two opposing objects are referred to as a cabinet cabinet, as shown in FIG. 2, the camera 100 is positioned on the upper side of two cabinet cabinets facing each other to photograph the cabinets between the cabinets.

As shown in Fig. 3, the image of the person to be photographed includes a passenger image before passing through the gate between two objects (hereinafter referred to as a passer-and-passer image) and two objects (Hereinafter, referred to as " passer-by image ") passing through the gate between the two objects, as shown in Fig. 5, and a passer- ) May be applied.

Therefore, the camera 100 photographs the entrance image before passing through the gate between the two objects, and calculates the distance between the entrance and exit entrance image and the passenger entrance image passing through the gate between the two objects, After passing through the gate, passenger images can be generated respectively.

The passer-by-pass image (FIG. 3), the passer-by image during pass (FIG. 4), and the passer-by image (FIG. 5) may be photographed at respective moments, The passer-by-passer image, the passer-by image in pass, and the passer-in image in the moving image can be extracted through the moving image photographed up to the passer-by.

The photographed impersonator image can be transmitted to the machine learning running server 200 through a wired communication or a wireless communication. For example, an Ethernet, a universal serial bus, an IEEE 1394, a serial communication And wireless communication methods such as infrared communication, Bluetooth, home radio frequency (RF), and wireless LAN may be used. .

For reference, the camera 100 may include a lens assembly, a filter, a photoelectric conversion module, and an analog / digital conversion module. The lens assembly includes a zoom lens, a focus lens, and a compensation lens. The focus distance of the lens can be shifted under the control of the focus motor MF. The filters include an optical low pass filter and an optical low pass filter. And may include an Infra-Red Cut Filter. An optical low-pass filter removes the high-frequency component of optical noise, while the Infra-Red cut filter blocks the infrared component of incident light. The photoelectric conversion module may include an image pickup device such as a CCD (Charge Coupled Device) or a CMOS (Complementary Metal-Oxide-Semiconductor). The photoelectric conversion module converts light from the optical system (OPS) into an electrical analog signal. The analog / digital conversion module may include a CDS-ADC (Correlation Double Sampler and Analog-to-Digital Converter) device. An analog / digital conversion module (not shown) processes an analog signal from the photoelectric conversion module OEC, removes the high-frequency noise, adjusts the amplitude, and converts the analog signal into a digital signal.

The machine learning learning server 200 has the same configuration as that of a typical web server in terms of hardware and is implemented in various types of languages such as C, C ++, Java, Visual Basic, Visual C, And a program module which functions as a program module. In addition, it can be implemented using a web server program that is variously provided according to operating systems such as dos, window, linux, unix, and macintosh for general server hardware. Typical examples include a Web site used in a Windows environment, Internet Information Server (IIS), and CERN, NCSA, and APPACH used in a UNIX environment.

The machine learning learning server 200 is a server for performing machine learning learning such as deep learning and the like. Here, deep learning is a technique in which a computer uses various data to learn Is a machine learning technology based on an artificial neural network (ANN). Deep Learning is a machine in which a human brain finds patterns in a large number of data and then implements the information processing method of distinguishing objects so that the computer can distinguish objects. Deep learning technology allows computers to recognize, reason, and judge themselves without having to set all criteria.

As described above, the deep learning technique is widely used for voice recognition, image recognition, and photo analysis. The machine learning learning server 200 of the present invention is a system that allows a machine learning run such as lip learning to pass between two opposing objects Run the machine's access pattern of the passengers.

To this end, the machine learning learning server 200 stores the entrance image together with the entrance and exit time in the storage unit. Here, the storage body may be a hard disk drive, an SSD drive (Solid State Drive), a flash memory, a CF card (Compact Flash Card), an SD card (Secure Digital Card) Card, an MMC card (Multi-Media Card), a memory stick, and the like, or may be provided in a separate device.

The machine learning learning server 200 classifies each participant in the participant image, extracts the participant's face image, and stores the entrance information for each participant's face image to be read. Then, using the access information, it is judged whether or not the predetermined accessor classification condition is satisfied, and the image of the face of the passenger which satisfies the passenger classification condition is stored as the running data.

Thereafter, when the entrance face image exists in the preset running data, the machine learning running server 200 classifies the entrance attendant having the entrance face image as the original registration work. On the other hand, when the passenger face image is not present in the preset running data, the machine learning running server 200 classifies the passenger having the passenger face image as a new human attraction.

Therefore, the machine learning running server 200 can analyze the entrance image recognized through the camera 100 through machine learning learning, determine whether it is a pre-registration work force or a new work force, and perform appropriate countermeasures accordingly. For example, when an image of a passenger recognized through the camera 100 is analyzed through machine learning learning, when a new manpower is detected, an access time and an entrance image are stored, and the security manager terminal is notified in real- . Here, the security administrator terminal may be a smart phone, a desktop PC, a tablet PC, a slate PC, a notebook computer, and the like. Of course, it is needless to say that the terminal to which the present invention is applicable is not limited to the above-described type, but may include both terminals capable of communicating with external devices.

Also, the machine learning running server 200 analyzes the entrance image recognized through the camera 100 through the machine learning learning. When the new work attraction is detected, the machine learning learning server 200 generates an alarm warning as shown in FIG. 6, The workforce can be identified.

On the other hand, when the machine learning learning server 200 stores the entrance face image that satisfies the preset entrance classifying condition as the running data, the entrance classifying condition may be the entry / exit number classifying condition, the entrance time class / have.

Herein, the access frequency classification condition is a condition that satisfies the entrance face image that exceeds the preset access count, and the access time time access classification condition is a condition that satisfies the entrance face image having an experience of entering and leaving in a predetermined time zone.

FIG. 7 is a flowchart illustrating an access control process through machine learning learning according to an embodiment of the present invention.

7, the method for controlling a visitor through machine learning learning includes steps S710, S720, S730, S740, (S750), and a new manpower classification process (S760).

The passenger image generation process S710 is a process in which the camera 100 captures a gate defined between two objects opposed to each other and generates a passer-by image of a passenger who has passed through the gate. That is, a specific point set by the administrator is defined as a gate, and a passer-by is passed through to define an entrance image. For this purpose, it is defined as a desk facing each other, an opposing advertisement column facing each other, a gate between the facing pots, a cabinet for facing each other, and the like, and the vicinity of the defined gate area is photographed through the camera 100, . For example, FIG. 2 illustrates a camera 100 that photographs between facing cabinet cabinets.

The passenger image storage process (S720) is a process in which the machine learning running server 200 stores the passenger image together with the access time.

The accessing information storing process S730 is a process in which the machine learning learning server 200 classifies each of the passengers in the passenger image, extracts the passenger face image, and stores access information for each passenger face image to be read. A human face is read out from the entrance image through a known face recognition algorithm to extract the entrance face image as an entrance face image and then the entrance information is stored for each entrance face image to be read. Therefore, entrance and exit information such as the number of times of entrance and exit, and access time are stored for each face image of the passer-by.

In the machine learning learning process (S740), the machine learning learning server 200 determines whether or not the user class information satisfies a predetermined user classification condition by using the access information, and stores the image of the participant's face that satisfies the user classification condition as learning data Process.

Here, the accessor classification condition may correspond to the access frequency classification condition, the access time class importer classification condition, and the like.

Here, the access frequency classification condition is a condition that satisfies the entrance face image that exceeds the preset access frequency. For example, if the preset number of accesses is set to '3 times accesses', access information of a user who has accessed more than 3 times is stored as running data, whereas if the user has accessed 3 times or less The access information of the accessing person is not stored as running data.

The access time class visitor classification condition is a condition that satisfies the passenger face image that has been accessed in the predetermined time zone. For example, access information of a passenger who has been in the session between 9:00 am and 5:00 pm is stored as running data, while access to the session is different from 9:00 am to 5:00 pm The access information of the accessing person is not stored as running data.

In the pre-registered manpower classifying process (S750), when the entrance face image exists in the preset running data, the machine learning running server 200 classifies the entrance person having the entrance face image as the registered entry man.

The new manpower classification process S760 is a process in which the machine learning running server 200 classifies a passenger having an entrance face image as a new manpower when the entrance face image is not present in the preset running data.

Therefore, the machine learning running server 200 can analyze the entrance image recognized through the camera 100 through machine learning learning, determine whether it is a pre-registration work force or a new work force, and perform appropriate countermeasures accordingly. For example, when an image of a passenger recognized through the camera 100 is analyzed through machine learning learning, when a new manpower is detected, an access time and an entrance image are stored, and the security manager terminal is notified in real- .

Also, the machine learning running server 200 analyzes the entrance image recognized through the camera 100 through the machine learning learning. When the new work attraction is detected, the machine learning learning server 200 generates an alarm warning as shown in FIG. 6, The workforce can be identified.

On the other hand, as shown in Fig. 3, the image of a person to be photographed is divided into a passenger image before passing through the gate between two objects (hereinafter referred to as a passer-and-passer image) (Hereinafter referred to as " passer-by image ") passing through a gate between objects, and a passenger image after passing through a gate between two objects as shown in Fig. 5 ') Can be applied.

Therefore, the camera 100 photographs the entrance image before passing through the gate between the two objects, and calculates the distance between the entrance and exit entrance image and the passenger entrance image passing through the gate between the two objects, After passing through the gate, passenger images can be generated respectively.

Accordingly, the passenger image storage process (S720) can store the passer-by-passer image (FIG. 3), the passer-by image (FIG. 4), and the passer- have. The passer-by-passer image extraction process identifies each passer-by of the passer-by-passer image, extracts the passer-by-passer face image, extracts the passer-by-passer image as passer- Each of the passers on the image can be identified and extracted as an image of the passer-by of the passer.

Accordingly, the passenger classification condition may be a condition for the similarity rate between the facial expression of the face image of the passer-by before the pass, the facial expression of the passer-by image during pass, and the facial expression of the passer-by image.

Accordingly, the machine learning learning process (S740) for storing the face image of the passenger who meets the passenger classification condition as the running data is performed by the facial expression pattern of the passer-player image, the face pattern of the passer- And the similarity rate between the patterns is used as the running data.

Generally, in case of an intruder intruding into a security area, a tense facial expression is displayed in front of a security camera 100 such as a CCTV camera, but after having passed through a CCTV camera, an intruder looks relieved. Therefore, the present invention stores the data as running data by using this feature.

For example, in the case of a legally legitimate person, since it is not nervous in front of the security camera 100 such as a CCTV camera, the face expression pattern of the passerby image, the face pattern of the passerby image during passage, Similarity rates between facial patterns are similar within tolerance limits.

On the other hand, in case of an intruder intruding into the security area, since a confused facial expression is displayed in front of the security camera 100 such as a CCTV camera and a safe expression is made after passing through the CCTV camera, The similarity rate between the pattern, the facial pattern of the passerby image during passage, and the facial expression pattern of the passerby image after passing is out of the tolerance range.

Therefore, if the similarity rate is within the predetermined tolerance range, it is determined that the user is a normal person and stored as running data. On the other hand, if the similarity rate is out of the allowable error range, it is determined that the user is unauthorized.

The embodiments of the present invention described above are selected and presented in order to facilitate the understanding of those skilled in the art from a variety of possible examples. The technical idea of the present invention is not necessarily limited to or limited to these embodiments Various changes, modifications, and other equivalent embodiments are possible without departing from the spirit of the present invention.

S710: Creating a passenger image
S720: Image storage
S730: Store access information
S740: Machine learning learning
S750: Classification of registered workers
S760: New manpower classification

Claims (7)

A camera captures a gate defined between two objects facing each other and generates a viewer image by capturing an image of a person passing through the gate;
A machine learning learning server for storing the accessing image together with the access time;
An access information storing step of the machine learning learning server classifying each of the passers by the passer-by image, extracting the passer-bye face image, and storing access information for each of the passer-bye face images to be read;
A machine learning learning process for determining whether the machine learning learning server meets a preset accessor classification condition using the access information and storing the entrance face image satisfying the entry classification condition as learning data;
Wherein the machine learning learning server classifies a passenger having the entrance face image as a registered entry force if the entrance face image exists in preset running data; And
Wherein the machine learning learning server classifies a passenger having the entrance face image as a new manpower if the entrance face image does not exist in the preset running data;
A method of controlling access to a person through machine learning learning.
The method according to claim 1,
Access control method through machine learning learning including access frequency classification condition and access time classifier condition.
The method of claim 2,
Wherein the access frequency classification condition is a condition that satisfies an entrance face image that exceeds a preset access count,
Wherein the access time class visitor classification condition is a condition that satisfies an entrance face image having an experience of entering and leaving in a predetermined time zone.
The method as claimed in claim 1,
Passer and passer image before passing through the gate between two objects, passer image during passing which is the passer image passing through the gate between two objects, passer image after pass through the gate between two objects A method of access control through machine learning learning which generates an entrance image after passing.
5. The method of claim 4,
Wherein the passer-by-passer image, the passer-by image during passage, and the passer-by image are stored together as one group for each passenger and are stored together.
5. The method according to claim 4,
And identifies the passers on the passer-by-passer image, extracts them as passer-by-passer face images,
Identifies each of the passers in the passer-by image during the pass, extracts it as the passer-face image during pass,
And a machine learning run in which passengers in the passenger image after the pass are respectively identified and extracted as passenger face images.
The method according to claim 6,
The condition for the similarity rate between the facial expression of the face image of the passer before the pass, the facial expression of the passer-face image during pass, and the facial expression of the passer-byant image after passing,
The machine learning learning process determines that the similarity rate is within a predetermined tolerance range and stores the learning data as running data. If the similarity rate is out of the allowable error range, The method according to claim 1,

Images