KR20210093705A - AI image processing camera and camera system for multi-event analysis and control - Google Patents

Abstract

A camera for multi-event analysis and control according to one embodiment of the present invention includes a camera for acquiring images, wherein the camera includes: an object recognition unit for recognizing at least one object to obtain an object image for the recognized object; an access authentication unit that receives the object image and compare the received the object image with a pre-stored registered image from the database to perform authentication, approve access to the authenticated object and control a gate; and a communication unit that receives a pre-stored registration image from the database to perform authentication of a recognized object, or communicates with the access control system in the access authentication unit and transmits a control signal to control the opening/closing of a gate for an authenticated object.

Description

AI image processing camera and camera system for multi-event analysis and control}

The present invention relates to an AI image processing camera system, and more particularly, to an AI image processing camera system for multi-event analysis and control.

Human tracking and face recognition technology is one of the fields being studied in many application systems such as mobile robots, surveillance systems, and human-robot interaction. Although it is difficult to track a person and recognize a face in real time using current vision technology, research on this is being actively conducted along with the development of pattern recognition and image processing techniques due to the development of computer performance.

Tracking and recognizing people in a surveillance system environment must be built by integrating technologies from multiple fields into one. A technology for tracking multiple people, a technology for tracking faces, and a technology for recognizing faces are required.

In general, a fixed camera is used to track a large number of people. As a method of detecting and tracking a person from a fixed camera, a background modeling technique that is very suitable for a real-time environment is widely used. In addition, since it is based on tracking a large number of people, various methods for resolving the location tracking and temporary occlusion of people are being studied together.

A fixed camera is suitable for tracking a large number of people, but since it does not provide the face size and resolution enough to perform face recognition, a PTZ camera is generally used together. Due to the use of multiple cameras, including fixed ones, several algorithms have been developed that take into account the relationship between these cameras. For face tracking, various methods using skin color information and face shape information have been proposed, and for real-time tracking, an algorithm using face color distribution is common.

Facial recognition has been studied very extensively. It can be broadly classified into 2D recognition, 3D recognition, and face recognition in video. For high recognition rate, face normalization is important in 2D face recognition. In general, normalization is performed based on the position of the two eyes on the face, but this causes a lot of problems.

Active Appearance Model (AAM) has been proposed for a more sophisticated regularization method and facial feature point extraction. Because AAM can extract sophisticated facial feature information, it is widely used in tracking and recognition and shows good performance.

However, there is a problem that prior knowledge of the initial face shape must be given. Such prior knowledge can be given manually or by using a frontal face detection method. However, even if the initialization problem is solved in this way, a problem arises when applying it to face tracking.

If the movement of the face is large, re-initialization is required because the result of the previous face shape cannot be used as prior information, but front face detection cannot be used in this case. In order to obtain a high recognition rate using AAM, accurate registration must be guaranteed.

On the other hand, anti-passback is a structural method for access control and has a function that must be authenticated when entering an area by installing an access control device on the inside/outside of the door. When entering using a card, if you follow the person in front of you without recognizing the card, the door will not open when you leave and an anti-passback event will occur.

Anti-passback is divided into Hard APB and Soft APB. In case of anti-passback violation, Hard APB generates an anti-passback event and cannot enter, while Soft APB generates an anti-passback event and allows access.

Currently, there is no artificial intelligence camera system that can perform object recognition using an artificial intelligence algorithm with a camera and independently control access control from multiple events to each individual event after recognizing multiple objects.

Therefore, in order to solve the above-mentioned problem, research on an AI image processing camera system for multi-event analysis and control is required.

Korean Patent No. 10-1961891 (registered on March 19, 2019)

It is an object of the present invention to provide a camera and a camera system for multi-event analysis and control that can control access control by individual events after recognizing a plurality of objects with a single camera.

A camera for multi-event analysis and control according to an embodiment of the present invention includes a camera for acquiring an image, wherein the camera recognizes at least one object and an object for acquiring an object image for the recognized object recognition unit; an access authentication unit for receiving the object image, comparing it with a registered image stored in advance from a database, performing authentication, authorizing access to the authenticated object, and controlling a gate; A communication unit that receives a pre-stored registration image from a database to perform authentication of a recognized object, or transmits a control signal to control the opening/closing of a gate for an authenticated object by communicating with the access control system in the access authentication unit include

A system for analyzing and controlling multiple events according to another embodiment of the present invention includes: a camera for acquiring an image; and a management server that receives an image from the camera and controls an access control system connected to the camera and a communication network to permit access control, wherein the management server recognizes at least one object from the image, an object recognition unit for obtaining an object image of a recognized object; an access authentication unit for receiving the object image, comparing it with a registered image stored in advance from a database, performing authentication, authorizing access to the authenticated object, and controlling a gate; A communication unit that receives a pre-stored registration image from a database to perform authentication of a recognized object, or transmits a control signal to control the opening/closing of a gate for an authenticated object by communicating with the access control system in the access authentication unit include more

In the above, with respect to the plurality of objects recognized by the object recognition unit, the event management unit for processing the event for each object in real time; further includes.

In the above, characterized in that the event management unit performs an event for individually authenticating a plurality of objects, an event for independently opening and controlling a gate in which the object is located among a plurality of gates, and a notification event for gate control at the same time in real time do.

The method further includes: a heartbeat analyzer that extracts a skin region from the image transmitted from the camera and detects a heartbeat signal based on a change in skin color of the skin region.

In the above, the object recognition unit is characterized in that the recognition of the plurality of objects by using at least one of the SSD algorithm, the YOLO algorithm, and the AdaBoost algorithm.

The camera system capable of processing individual events after recognizing a plurality of objects with a single camera of the present invention can recognize a plurality of objects with a single camera, and has the advantage of independently controlling access control by individual events.

In addition, multiple faces are recognized from the extracted result, and it is possible to check whether the recognized face is a person permitted to enter, and after checking whether access is permitted, each controller can be independently controlled through an individual access control controller. .

In addition, there is an advantage in that it is possible to periodically check health management and status by providing heart rate information for an authenticated object.

1 is a block diagram showing the configuration of a camera for multi-event analysis and control according to an embodiment of the present invention.
FIG. 2 is a block diagram showing the configuration of the camera of FIG. 1 in detail.
3 is a block diagram showing the configuration of a camera system for multi-event analysis and control according to another embodiment of the present invention.
4 is a block diagram showing the configuration of the management server of FIG. 3 in detail.
5 is a flowchart illustrating a process of an image processing method for multi-event analysis and control according to an embodiment of the present invention.
6 is a diagram for explaining a process of controlling a plurality of events in the present invention.

Hereinafter, specific embodiments of the present invention will be described in detail with reference to the drawings. However, the spirit of the present invention is not limited to the embodiments presented, and those skilled in the art who understand the spirit of the present invention may add, change, delete, etc. other components within the scope of the same spirit, through addition, change, deletion, etc. Other embodiments included within the scope of the invention may be easily proposed, but this will also be included within the scope of the invention. In addition, components having the same function within the scope of the same idea shown in the drawings of each embodiment will be described using the same reference numerals.

1 is a block diagram showing the configuration of a camera for multi-event analysis and control according to an embodiment of the present invention, and FIG. 2 is a block diagram showing the configuration of the camera of FIG. 1 in detail.

The camera 100 for multi-event analysis and control of the present invention is provided to recognize a plurality of objects, and controls the access control system 300 connected through the camera 100 and the communication network 200 to enter and exit. provided to allow control.

As shown in FIG. 2 , the camera 100 of the present invention recognizes and authenticates a plurality of objects and controls access control. As shown in FIG. 2 , the object recognition unit 110 , the object recognition unit 120 , and the heart rate analyzer 130 . ), an event management unit 140 , and a communication unit 150 may be further included.

The object recognition unit 110 simultaneously recognizes a plurality of objects (eg, people, animals, identifiable objects, etc.) from the image of the camera 100 , obtains an object image for each object, and sends the image to the object recognition unit 120 . can provide

The object recognition unit 110 may use an object recognition algorithm such as You Only Look Once (YOLO) or Single Shot Multibox Detector (SSD) to simultaneously recognize a plurality of objects.

YOLO is an algorithm that can improve training and detection speed by approaching object recognition with regression and simplifying the overall structure. is output This tensor divides the image into a grid form to express each region, and through this, the region object is recognized.

In addition, SSD is an algorithm that recognizes objects using feature maps of various sizes to generate candidate regions. The size of the feature map obtained from the CNN model decreases as the convolutional layer progresses, and the SSD All the feature maps extracted in this process are used in the inference process to recognize objects, and objects can be recognized more accurately than YOLO.

In addition, in order to increase the object recognition rate when recognizing objects with the camera 100 , AdaBoost, Support Vector Machine (SVM), Linear Disciminant Analysis (LDA), Principal Component Analysis (Principal Component Analysis) : PCA) may be further included.

In particular, when the Adaboost algorithm is used, similar objects can be recognized as the same object by analyzing the feature points of the object. As the number of feature points increases, the object can be recognized by distinguishing it from similar objects better. It may be a state in which the same person wears glasses, a state in a hat, or a person with a change in expression.

All of these algorithmic techniques identify the area to be recognized based on the appearance, and the area around the feature point of a face or object is detected using a model trained by a set of captured images to be used for training, and various surrounding constraints are applied. Since this is overcome through training, as a result, the recognition accuracy and reliability of face or object feature points can be increased.

The object recognition unit 120 receives the object image of the recognized object from the object recognition unit 110, compares it with the registered image for each object previously stored in the database, determines whether the object is the same, and approves access when the same object authentication succeeds. The gate 310 may be controlled.

In particular, by independently comparing object images for a plurality of objects, authentication and access permission for each object are performed.

The heartbeat analyzer 130 may extract the skin region from the image transmitted from the camera 100 and determine the skin region based on the value of Hue, which is the H channel, in the HSV model of the skin image.

Here, the HSV model expresses the color of an image with three components: Hue (hue), Saturation (saturation), and Value (brightness).

In this case, when the skin image captured by the camera 100 is the RGB model, the heartbeat analyzer 130 may convert the skin image of the RGB model into the HSV model.

In addition, the heartbeat analyzer 130 may detect a heartbeat signal through a change in the G (Green) channel in the RGB model of the camera 100 image, or detect a heartbeat signal through a change in skin color from the skin region of the HSV model. there is.

Here, the RGB model is the most basic color model and expresses color as a combination of three components: Red, Green, and Blue.

Also, the heartbeat analyzer 130 may include a band-pass filter configured as an RLC circuit to pass a specific frequency band to remove noise included in the heartbeat signal, and a low-pass filter instead of the band-pass filter. It can also be configured as a combination of a filter and a high-pass filter.

Also, the heartbeat analyzer 130 may apply an adaptive filter, such as a Kalman filter, to be less sensitive to a change in skin tone or color due to a change in lighting.

The event management unit 140 may process an event for each object for a plurality of objects recognized by the object recognition unit 110 in real time.

That is, when multiple events occur for each object, multiple events must be simultaneously performed. For this purpose, event control in a multi-threaded manner can be performed.

Multi-thread is a method in which a plurality of handlers are simultaneously processed for a plurality of event queues, and can be processed faster than a first-in, first-out (FIFO) method in which sequential processing is performed with priority.

In addition, the event occurring in the present invention is an event for individually authenticating a plurality of objects, an event for independently opening and controlling the gate 310 in which the object is located among the plurality of gates 310, and gate 310 control (opening). /closed) can be a notification event.

In addition, there are various events that can occur according to various places. For example, as shown in FIG. 6 , in the case of a golf course, a swing motion notification event of the golf swing device (event 1), when a golf member deviates from a set line, the location is deviated as a danger zone A number of notifications or authentication events may be included, including a warning or notification event (event 2) for , and a notification event (event 3) for guiding the start of golf practice when an authorized golf user is authenticated.

The communication unit 150 receives a registration image from the camera 100 or a database located in a local (remote location) to perform authentication of an object in the camera 100 image, or communicates with the access control system 300 to access the authenticated object. It is provided to control the opening/closing of the gate 310 for the gate 310 or to receive a remote control signal through an external manager terminal.

In addition, the communication unit 150 includes a plurality of devices (not shown) such as a golf swing device, a display device, and a manager terminal in addition to the access control system 300 shown in FIG. 1 for transmitting and receiving control signals during event control by the event management unit 140 . city) can be connected.

3 is a block diagram showing the configuration of a camera system for analyzing and controlling multiple events according to another embodiment of the present invention, and FIG. 4 is a block diagram showing the configuration of the management server of FIG. 3 in detail.

According to another embodiment of the present invention, the camera system for multi-event analysis and control may include at least one management server 400 capable of receiving an image from the camera 100 and performing object recognition and authentication. .

The management server 400 is cloud-based, and a plurality of management servers 400 are linked with the camera 100 through the communication network 200 and distributed among the plurality of servers to quickly perform object recognition and authentication functions.

The management server 400 includes an object recognition unit 410, an object recognition unit 420, a heartbeat analyzer 430, an event management unit 440, and a communication unit as shown in FIG. 4 to perform object recognition and authentication functions. 450 and a database 460 may be further included.

The object recognition unit 410 , the object recognition unit 420 , the heartbeat analyzer 430 , the event management unit 440 , and the communication unit 450 are the camera 100 according to the embodiment described above with reference to FIGS. 1 and 2 . ) can perform the same function as each built-in part, so the description is omitted.

The database 460 stores a plurality of registered images for each object in order to determine whether the object image is the same object authenticated by the object recognition unit 420 .

In addition, the database 460 may store a plurality of registered images for a specific object to increase the recognition rate of the same object through machine learning-based learning in the object recognition unit 420 .

5 is a flowchart illustrating a process of an image processing method for multi-event analysis and control according to an embodiment of the present invention.

A plurality of objects may be recognized from an image captured by the camera 100 by using an object recognition algorithm (S301).

A series of processes such as recognition of a plurality of objects, authentication, and event processing may be performed by a camera or a management server, as in the above-described embodiment.

For a plurality of recognized objects, the object image is compared with a plurality of pre-stored registered images for access approval of the access control system 300 to determine whether the object is the same authenticated object (S302).

If it is determined that the same object is authenticated, the authentication of the corresponding object is successful, and the control can be performed to open the gate 310 in which the corresponding object is located (S303).

In addition, by extracting skin information by analyzing the skin image from the object image recognized by the camera 100, a change in the skin can be detected, and heart rate information can be estimated and calculated (S304).

Here, the heart rate information may include heart rate, cardiac output, stress index, concentration level, and the like.

The calculated heart rate information is visually provided through a display device (not shown) (S305).

If authentication fails in step S302, a warning broadcast guiding authentication failure may be broadcast, and an external manager may be notified through the control room display device.

100 ; camera
110, 410; object recognition unit
120, 420; Access authentication department
130, 430; heart rate analyzer
140, 440; Event Management Department
150, 450 ; communication department
200 ; communications network
300 ; access control system
310; gate
400 ; management server
460; database

Claims (10)

A camera for acquiring an image,
The camera is
an object recognition unit for recognizing at least one object and obtaining an object image for the recognized object;
an access authentication unit for receiving the object image, comparing it with a registered image stored in advance from a database, performing authentication, authorizing access to the authenticated object, and controlling a gate;
A communication unit that receives a pre-stored registration image from a database to perform authentication of a recognized object, or transmits a control signal to control the opening/closing of a gate for an authenticated object by communicating with the access control system in the access authentication unit
A camera for multi-event analysis and control, including. According to claim 1,
an event management unit that processes an event for each object in real time with respect to a plurality of objects recognized by the object recognition unit;
Camera for multi-event analysis and control further comprising. 3. The method of claim 2,
The event management department
For multi-event analysis and control, characterized in that an event for individually authenticating a plurality of objects, an event for independently opening and controlling the gate where the object is located among a plurality of gates, and a notification event for gate control are simultaneously performed in real time camera. According to claim 1,
a heartbeat analyzer extracting a skin region from the image transmitted from the camera and detecting a heartbeat signal through a skin color change in the skin region;
Camera for multi-event analysis and control further comprising. According to claim 1,
The object recognition unit
A camera for multi-event analysis and control, characterized in that recognition of a plurality of objects is performed using at least one of an SSD algorithm, a YOLO algorithm, and an AdaBoost algorithm. a camera for acquiring an image;
a management server that receives an image from the camera, controls an access control system connected to the camera and a communication network, and permits access control;
includes,
The management server
an object recognition unit for recognizing at least one object from the image and obtaining an object image of the recognized object;
an access authentication unit for receiving the object image, comparing it with a registered image stored in advance from a database, performing authentication, authorizing access to the authenticated object, and controlling a gate;
A communication unit that receives a pre-stored registration image from a database to perform authentication of a recognized object, or transmits a control signal to control the opening/closing of a gate for an authenticated object by communicating with the access control system in the access authentication unit
A system for multi-event analysis and control further comprising a. 7. The method of claim 6,
an event management unit that processes an event for each object in real time with respect to a plurality of objects recognized by the object recognition unit;
A system for multi-event analysis and control further comprising a. 8. The method of claim 7,
The event management department
For multi-event analysis and control, characterized in that an event for individually authenticating a plurality of objects, an event for independently opening and controlling the gate where the object is located among a plurality of gates, and a notification event for gate control are simultaneously performed in real time system. 7. The method of claim 6,
a heartbeat analyzer extracting a skin region from the image transmitted from the camera and detecting a heartbeat signal through a skin color change in the skin region;
A system for multi-event analysis and control further comprising a. 7. The method of claim 6,
The object recognition unit
A system for multi-event analysis and control, characterized in that recognition of a plurality of objects is performed using at least one of an SSD algorithm, a YOLO algorithm, and an AdaBoost algorithm.

Images