KR102483250B1 - Apparatus for safe and management employing image artificial intelligence for controlling to infra of ocean city - Google Patents

Abstract

The present invention relates to an image AI-based integrated safety management apparatus for efficient control of the infrastructure of a marine city. Specifically, according to embodiments of the present invention, an area range for each camera is checked in accordance with a facility size and a service type of a safety management place to select the number of installations to install image monitoring devices, and one between a PTZ-type camera and an ordinary camera is selected and installed in accordance with a collection purpose and a facility environment of the safety management place. In such a state, if monitoring information is collected by multiple image monitoring devices in accordance with manager setting information, an abnormal state in the corresponding monitoring information is checked and provided from the setting format in accordance with one embodiment to perform safety management. Therefore, embodiments provide subsequent maintenance and management for the infrastructure of a tourist spot, sightseeing satisfaction improvement, tourist spot clean management, and related incident/accident response, and disaster/catastrophe response to handle situations which can occur in common in the tourist spot. In other words, the infrastructure of various marine cities including sightseeing is efficiently managed while good views and a variety of content are needed.

Description

Apparatus for safe and management employing image artificial intelligence for controlling to infra of ocean city}

The contents disclosed in this specification relate to the division for safety management in a specific place, and more specifically, when safety management in a specific place is performed, the infrastructure of the place is efficiently managed to effectively deal with various situations. is to respond

Unless otherwise indicated herein, material described in this section is not prior art to the claims in this application, and inclusion in this section is not an admission that it is prior art.

In general, maritime cities include various urban problems due to the characteristics of tourist areas, and tourism-related infrastructure is intensively distributed, causing various situations.

In addition, this problem has been a continuous problem in the past in the tourism industry and urban development projects, but the scope of the problem is diverse and is still not improved.

For example, problems such as traffic congestion and parking problems, coastal facility management problems, marine pollution problems, and disaster/disaster problems are also points that reduce the quality of tourism for tourists.

On the other hand, in general, the safety management system within a city means a technology that can support a system for effectively responding to environmental problems, traffic problems, facility management, disaster/disaster management, etc. occurring in the city in order to reduce the risk of accidents.

However, since it is configured to respond to all situations in the city, there is a limit to applying the system to the details.

In particular, it is generally operated in risk-prone areas and areas requiring intensive management so that the management scope covers the entire city.

As a result, it is not easy to efficiently control the problem due to the poor definition and solution of the problem of the adjacent tourism infrastructure in the ocean, which requires intensive management.

For reference, prior art documents in this background include the following documents.

(Patent Document 0001) KR200494486 B1

For reference, this document 1 relates to the branching of streetlights, and simply provides information on the installation area of streetlights as well as easy maintenance and management of streetlights as well as simple crime prevention.

The disclosed contents enable maintenance of tourist destination infrastructure, improvement of tourism satisfaction, cleanliness management of tourist destinations, response to related incidents/accidents, and disaster/disaster response in order to respond to situations that may commonly occur in tourist destinations.

In other words, while good scenery and various contents are also required, we want to provide an integrated video AI-based safety management device for efficient control of marine city infrastructure that enables efficient management of infrastructure of various marine cities, including tourism.

The video AI-based safety management integrated device for efficient control of marine city infrastructure according to the embodiment is,

First, the area range for each camera is checked according to the facility size and service type of the safety management place, the number of installation units is selected, and video surveillance devices are installed. In addition, either a PTZ type camera or a general camera is selected and installed according to the facility environment and collection purpose of the safety management place.

In this state, when monitoring information is collected from multiple video surveillance devices according to manager setting information, safety management is performed by checking and providing abnormal conditions in the monitoring information from the setting format according to an embodiment. do.

According to the embodiments, to respond to situations that may commonly occur in tourist destinations, maintenance of tourist destination infrastructure, improvement of tourism satisfaction, cleanliness management of tourist destinations, response to related incidents/accidents, and disaster/disaster response are provided.

In other words, it requires good scenery and various contents, but efficiently manages the infrastructure of various marine cities, including tourism.

1 is a diagram for conceptually explaining an image AI-based safety management integrated device for efficient control of marine city infrastructure according to an embodiment.
2 to 4 are views for explaining an information set applied to a video AI-based safety management integrated device (ie, setting format) for efficient control of marine city infrastructure according to an embodiment.
5 is a diagram for explaining a parking lot analysis service applied to an image AI-based safety management integrated device for efficient control of marine city infrastructure according to an embodiment.
6 and 7 are diagrams for explaining a road condition analysis service applied to an image AI-based safety management integrated device for efficient control of marine city infrastructure according to an embodiment.
8 is a diagram showing a system applied to an image AI-based safety management integrated device for efficient control of marine city infrastructure according to an embodiment as a whole.
9 is a block diagram showing the configuration of an integrated video AI-based safety management device for efficient control of marine city infrastructure according to an embodiment.
10 is a flow chart showing the operation of the video AI-based safety management integrated device for efficient control of marine city infrastructure according to an embodiment in order.

1 is a diagram for conceptually explaining an image AI-based safety management integrated device for efficient control of marine city infrastructure according to an embodiment.

As shown in FIG. 1, one embodiment first manages the maintenance of tourist destination infrastructure and improves tourism satisfaction, manages cleanliness of tourist destinations, responds to related incidents/accidents, and disasters/disasters in order to respond to situations that may commonly occur in tourist destinations. provide a response.

In other words, it requires good scenery and various contents, but efficiently manages the infrastructure of various marine cities, including tourism.

Specifically, in this embodiment, first, the area range for each camera is checked according to the size of facilities and service types of the safety management place, the number of installation units is selected, and the video surveillance device is installed. In addition, either a PTZ type camera or a general camera is selected and installed according to the facility environment and collection purpose of the safety management place.

In this state, when monitoring information is collected from various video monitoring devices according to manager setting information, safety management is performed by checking and providing abnormal conditions in the monitoring information from the setting format.

Here, the setting format is as follows.

1. First, a first information set is constructed for each setting object including vehicle, license plate number, and person based on the images of the video surveillance device actually collected for each safety management location.

1) More specifically, when the information set is constructed, external registered public image information is received and image information is collected through web crawling with the registration information processing device. Then, image information of the video surveillance device is collected, and image information is obtained by merging each information into one.

2) Next, labeling is performed by classifying the obtained image information into positive and negative.

3) So, the size, brightness and contrast of the image that has been labeled is corrected, and the gray-scale and threshold are differently adjusted according to the manager setting information to make the image normal. deal with

4) Through this, additional images are secured by providing a function that creates new information through calculations (including right-reversal, brightness control, coordinate movement, rotation, and magnification) for the normally processed image.

2. On the other hand, in the case of such additional securing, the securing is supplemented by creating a second information set, that is, more information such as reversal, rotation, and movement, by recognizing and learning objects in the image for each additionally secured image.

3. Then, based on the first and second information sets, the collected images are divided into frame units, and object labeling in a set sub format (Re-identification (ReID) format of Deep Appearance Descriptor) in each frame unit image Separate the area into a puzzle shape.

In addition, an object-centered value is generated from these objects, features of each detailed region including size, color, and type of the object are calculated, and similar identification objects are searched through a convolution layer to track the motion of the object.

4. Therefore, as a result of the tracking, an abnormal situation is detected by comparing the movement of the object with the set abnormal behavior information for each safety management place.

Hereinafter, detailed operations of this setting format will be described in more detail with reference to drawings.

2 to 4 are views for explaining an information set applied to an image AI-based safety management integrated device (ie, a setting format) for efficient control of marine city infrastructure according to an embodiment.

Specifically, FIG. 2 is a diagram for explaining an information set securing operation according to an embodiment, and FIG. 3 is a diagram for explaining an information labeling method of an image obtained through such an information set. And, FIG. 4 is a diagram for explaining an abnormal behavior detection operation through such an image.

As shown in FIGS. 2 to 4, the information set according to one embodiment is first, as shown in FIG. 2, there are some differences in accuracy depending on the resolution and installation angle of the CCTV image, so the actual CCTV collection is performed to improve accuracy. It learns the image and proceeds to improve the accuracy.

Specifically, the information set securing operation is as follows.

First, based on the collected information, classify into people and vehicles (vehicle type) and save the DB.

Then, set up the process below to construct an object detection-based information set.

First, in terms of personal information protection, an offline-based information set environment is created.

Second, in order to utilize the information of the existing open MS COCO and AIHUB and acquire additional object-specific information sets, each object is scanned with a camera to create an information set to be used as input information for the deep learning model and preprocessed through this information set. Through this process, the CNN network model is trained.

Third, in order to check whether the CNN model learned through the information set operates normally, it is verified through a procedure of comparison and analysis using an already published information set.

In other words, this R-CNN network model is a model that detects and classifies features of text, image, video, and sound information, and is particularly easy to find and classify image patterns.

In addition, since each filter used in R-CNN uses only a part of the information rather than the whole information based on the learned weight, it is not greatly affected by information change and distortion, and has high robustness.

So, based on this process, it is performed in the following way.

1. First, image information crawling is performed.

Specifically, it utilizes the information of MS COCO and AIHUB that have been published.

In addition, image information is obtained through web crawling of Naver and Google.

Finally, CCTV is installed at the demonstration site to obtain image information.

So, image information is created by merging information from existing open MS COCO and AIHUB, image information crawled from the web, and internal information into one.

Next, labeling is performed as shown in FIG. 3 .

2. Labeling work

Labeling is performed by classifying positives and negatives from image information that has not yet been refined.

3. Organize image information

First, the size, brightness, and contrast of the image are corrected, and additionally, the gray-scale and threshold are appropriately adjusted according to the administrator's wishes, that is, according to the administrator's setting information, and the image is processed in general. do.

Next, higher accuracy can be obtained through these operations, and the object and location to be detected in the target image can be known.

4. Increased information

First, the information set can be increased in various ways to increase the actual size of the learning information set.

More information means that overfitting can be reduced.

In addition, when the information set is different from the input value in the actual situation, it is possible to create an information distribution similar to the actual input value through

Specifically, it provides a function that creates new information through various operations (left/right reversal, brightness control, coordinate movement, rotation, magnification, etc.) of the original information, and it is added based on the original information.

Therefore, higher accuracy can be obtained through these operations, and it is possible to know the object and location to be detected in the target image.

4. Infoset creation

An information set of only one embodiment is generated using the above-described information-enhanced image information and labeling information.

On the other hand, in addition to this, an embodiment performs R-CNN-based learning on additional secured information in this way.

Specifically, deep learning based on R-CNN and OBJECT DETECTOR is performed.

In addition, learning about the vehicle (vehicle type), license plate number, and person is performed, respectively.

In this case, it is possible to recognize all objects in one image through GPU-based parallel processing operation.

On the other hand, an embodiment additionally tracks a moving object through these various pieces of information.

Specifically, deep learning-based fast-moving object tracking is performed according to an embodiment.

That is, object tracking is performed by applying the Re-identification (ReID) model with Deep Appearance Descriptor.

In a specific method, the collected image is first divided into frame units, and the object labeling area is separated into a puzzle shape in the form of Re-identification (ReID) of Deep Appearance Descriptor in each frame unit image.

In addition, an object-centered value is generated from these objects, features of each detailed region including size, color, and type of the object are calculated, and similar identification objects are searched through a convolution layer to track the motion of the object.

So, more accurate tracking can be done with the existing ID switching and matching cascade logic.

In addition, it accurately and quickly tracks all objects (recognized objects) in one image through simultaneous parallel processing operations based on CPU and GPU.

Meanwhile, in the next embodiment, abnormal behavior detection is performed as shown in FIG. 4 .

In other words, it detects human abnormal behavior based on 'POSE DETECTOR' linked to deep learning (human) object recognition.

First, an object is recognized based on learning information of context awareness in an input image.

And, through this, for example, an abnormal situation based on an image of a person's abnormal behavior (falling down, assault, etc.) that can occur in a parking lot is detected.

5 is a diagram for explaining a linked service applied to an integrated video AI-based safety management device for efficient control of marine city infrastructure according to an embodiment, and specifically, a diagram for explaining a parking lot analysis service among linked services.

As shown in FIG. 5, the linked service (ie, parking lot analysis service) according to an embodiment first includes parking lot congestion analysis (parking lot management).

In such parking lot management, first, the parking surface coverage range of the camera is checked according to the size of the parking lot, and the number of installation units is selected.

Also, PTZ type cameras (number recognition) and general cameras (vehicle, person recognition) are selected according to the parking lot environment and collection purpose.

Therefore, the location and vehicle number of the vehicle are collected according to the predefined PTZ camera movement value, and the vehicle information of the corresponding surface is inserted into the on-street/off-street public parking lot map, and then the UI is output.

In addition, in the case of a public parking lot, an alarm is displayed for a parked vehicle that has passed a set period of time, and smoke and fire occurring in the parking lot are checked and detected through a set detection sensor.

Additionally, when the vehicle is parked, the camera installed at the entrance of the parking lot recognizes the license plate number and proceeds with the vehicle exit. Then, the vehicle taken out in this way is paid in conjunction with the parking control information processing device, and when unpaid, the parking information processing device is notified, and the payment information unregistered vehicle is paid through the app.

As another example, if the safety management place is a safety / control area, the screen of the video surveillance device is analyzed and the number of people on the screen is stopped for more than the set time for each different safety / control area and if the person is stopped for more than the set time within the set movement variation warnings in each case.

In addition, in the case of an access control area, whether an object appears in the area is checked, and as a result of the check, an image is notified when the object appears. Alternatively, it checks whether an object appears in the boundary area including disaster/disaster situation and late-night time, and notifies when the object appears.

In addition, in another situation, it analyzes the skeleton of a set part including a person's hands, feet, and head on the screen to check for personal accidents (including falls and injuries) and alarms them.

6 and 7 are diagrams for explaining a road condition analysis service as another example of a linked service applied to an image AI-based safety management integrated device for efficient control of marine city infrastructure according to an embodiment.

Specifically, FIG. 6 is a diagram illustrating a vehicle congestion confirmation service, and FIG. 7 is a diagram illustrating a jaywalking confirmation service.

As shown in FIGS. 6 and 7 , the road condition analysis service according to an embodiment first analyzes the vehicle and determines that the road is congested only when the following two conditions are met.

That is, when the safety management place is a road area, the vehicle is analyzed to calculate the density when the vehicle speed is less than the set reference speed and the number of vehicles recognized within the image range is greater than the set reference value, and congestion is calculated when the set reference value is exceeded. sense the situation

As a second example, as shown in FIG. 7, by analyzing the image of a road with a crosswalk, determining the degree of clustering of human objects within the set time, and detecting the jaywalking of people other than the crosswalk when a human object appears within the cluster range. Check.

Thirdly, by analyzing objects on the road, objects other than people and vehicles that have not moved during the set time are identified and determined as unauthorized occupancy.

In addition, potholes among various defects occurring on asphalt pavements are a direct cause of drivers' safety accidents, and automatic pothole recognition in the form of semantic segmentation based on a fully convolutional neural network is performed.

8 is a diagram showing a system applied to an image AI-based safety management integrated device for efficient control of marine city infrastructure according to an embodiment as a whole.

As shown in FIG. 8 , the system of one embodiment includes a plurality of video surveillance devices 100 installed in a safety management place set by an administrator and a management information processing device 200 at a remote location.

Additionally, the system according to an embodiment is an external connection connected to the management information processing device 200, and includes a police station information processing device 300 and the like.

In addition, at this time, each device connects through its own network. In this case, Wi-Fi or LTE may be used as a wireless communication method.

The video surveillance device 100 is installed in a manager-set safety management place, for example, area a and area b, respectively, to collect object information. For example, in the video surveillance device 100 according to an embodiment, the area range for each camera is confirmed according to the size of facilities and service types of the safety management place, and the number of installation units is selected, and the facility environment and collection purpose of the safety management place are determined. According to this, either PTZ type camera or general camera should be selected.

When the object information, that is, monitoring information is collected, the management information processing device 200 performs safety management by confirming and providing abnormal conditions in the corresponding monitoring information from the aforementioned setting format.

9 is a block diagram showing the configuration of an image AI-based safety management integrated device (ie, management information processing device) for efficient control of marine city infrastructure according to an embodiment.

As shown in FIG. 9 , the management information processing device 200 of one embodiment includes an interface unit 201, a control unit 202, a database 203, a key signal input unit 204 and a display unit 205.

The interface unit 201 is connected to the video surveillance device 100 and an external connection, and receives video surveillance information or provides video analysis information to the outside.

When the above-mentioned monitoring information is collected, the control unit 202 performs safety management by checking and providing abnormal conditions in the corresponding monitoring information from the setting format.

The database 203 stores, under the control of the control unit 202, various registration information for monitoring, such as video monitoring location and device information, setting information, monitoring information, and abnormal state information for each different video monitoring device. .

The key signal input unit 204 receives various management setting information for safety management according to manager key manipulation.

Under the control of the control unit 202, the display unit 205 also displays various user interface information for safety management including abnormal state information.

10 is a flow chart sequentially illustrating the operation of the video AI-based safety management integrated device for efficient control of marine city infrastructure according to an embodiment.

As shown in FIG. 10, in one embodiment, manager setting information for safety management is first received according to user key manipulation through a key signal input unit.

Next, through the interface unit, monitoring information provided through the video surveillance device installed for each safety management place set by a plurality of different managers is collected (S1001), and object information is included in the monitoring information.

Meanwhile, as described above, the number of installed video surveillance devices is selected by checking the area range of each camera according to the size of the facility and the type of service of the safety management place. In addition, either a PTZ type camera or a general camera is selected and installed according to the facility environment and collection purpose of the safety management place.

Accordingly, when the control unit collects monitoring information from the video surveillance device through the interface unit according to the manager setting information of the key signal input unit, the control unit checks and provides abnormal conditions in the corresponding monitoring information from the setting format described above, thereby managing safety. is performed (S1002).

Then, under the control of the control unit, registration information and status information including monitoring information and abnormal conditions are stored for each video surveillance device, and user interface information for the abnormal status information is displayed (S1003).

As described above, in one embodiment, the area range for each camera is confirmed according to the size of the facility and the type of service of the safety management place, the number of installation units is selected, and the video surveillance device is installed. In addition, either a PTZ type camera or a general camera is selected and installed according to the facility environment and collection purpose of the safety management place.

In this state, when monitoring information is collected from various video surveillance devices according to manager setting information, safety management is performed by checking and providing abnormal conditions in the monitoring information from the setting format according to an embodiment.

Therefore, through this, one embodiment provides maintenance for tourist destination infrastructure, improvement of tourism satisfaction, cleanliness management of tourist destination, response to related incidents/accidents, and disaster/disaster response to respond to situations that may commonly occur in tourist destinations.

In other words, it requires good scenery and various contents, but efficiently manages the infrastructure of various marine cities, including tourism.

On the other hand, this embodiment, as another example, provides better safety management by searching an object from the following method when used in a place of an intrusion situation.

Specifically, first, a first video surveillance device for recognizing a remote object installed in a traffic area set by an administrator is included. And, in addition to this, a second video monitoring device for recognizing a short-range object installed in the vicinity of the traffic area and spaced apart from the first video monitoring device is included. So, when the long-distance image information is collected by the first video surveillance device, a cut-in vehicle object is extracted from the setting format and mapped to the short-range image information of the second video surveillance device to search for the object, thereby preventing lane intrusion. Provides a management information processing device that tracks and monitors whether

In more detail, this embodiment first makes it possible to respond to an intrusion situation in which regular enforcement is impossible due to the limitations of existing drone operation and enforcement equipment for cameras with less than 210 pixels.

And, in this case, in this embodiment, it is possible to additionally track an object & multi-channel image for video enforcement of a violating vehicle.

In addition, in this embodiment, it is possible to track an object between images by sharing images between cameras such as a long-distance camera (for vehicle detection) and a short-range camera (for detecting a license plate number) and sharing an analysis result of an object to be identified (vehicle).

Specifically, it is as follows.

First, a first video surveillance device for recognizing a remote object installed in a traffic area set by an administrator;

a second video monitoring device for recognizing a short-range object installed in the vicinity of the traffic area and spaced apart from the first video monitoring device; and

When long-distance image information is collected by the first video surveillance device, a cut-in vehicle object is extracted from a set format and mapped to the short-range image information of the second video monitoring device to search for the object, thereby detecting lane intrusion. It includes; a management information processing device that tracks and monitors.

And here, the format operates as follows.

That is, when collecting long-distance image information of the first video surveillance device, the image is divided into frame units, and the object labeling area is separated from each frame unit image to identify the vehicle object (positioning in the video area, center value generation) 1st step;

a second step of estimating the moving path and speed of the vehicle object by calculating the positional movement of the object labeling area separation value before and after each frame;

A third step of confirming whether the road is congested by analyzing each object density and moving speed by classifying them based on lanes and confirming the degree of congestion (including smooth, normal, and congestion);

As a result of the check, a fourth step of setting the relevant lane as the cut-in area value in case of traffic congestion, and not setting the corresponding lane as the cut-in area value in the case of smooth and normal;

If the corresponding lane of the congestion is set as the cut-in area value, cut-in vehicle objects are detected in the congestion situation, and the road is classified through overlay in the collected image of the target road, and illegal in the set cut-in prohibition lane (solid lane). a fifth step of detecting a lane-changing vehicle object;

a sixth step of extracting cut-in vehicle object information and illegal lane-changing vehicle object information by generating an object center value from each of the detected vehicle objects and calculating the area size, color, and type of the vehicle object;

Based on each of the extracted information, a region is divided into a puzzle shape from the collected images of the second video surveillance device, and characteristics of each detailed region including the position, lighting, and resolution of the image are extracted, and the similar identification object is obtained through a convolution layer. A seventh step of searching for ; and

An eighth step of recognizing the vehicle number from the similar identification object by the search and tracking and monitoring whether there is a lane intrusion; includes

On the other hand, in addition, in this embodiment, when objects are extracted as described above, by extracting objects with priority according to various objects, monitoring places, and monitoring purposes from the following method, so that the administrator can smoothly provide what he wants. do.

Specifically, when detecting the above-mentioned cutting-in vehicle object and the illegal lane-changing vehicle object information, video surveillance is performed from the priority of any one or more information set in advance among object information, purpose-specific information, location-specific information, and time-specific information. prioritize differently.

Therefore, corresponding information is extracted and notified according to the priorities to be controlled.

In addition, in the case of detecting the information of the intruding vehicle object and the illegal lane-changing vehicle object, the collected images of the video surveillance device are provided by making the information collectively from the following method, so that information can be quickly and conveniently collected in various situations. make it possible

Specifically, these collected images are as follows.

1) First, a single image is created in the set single image format for the primary collection image.

2) Then, after converting to a set single Mat format image, it is changed to SI (super image) by the set deep learning-based image magnification format.

3) Therefore, by up-scaling the entire single image or entire video corresponding to a plurality of different video surveillance file formats and converting them into high-definition images, the corresponding information can be quickly and conveniently collected in batches.

100: video monitoring device
200: management information processing device

Claims (2)

A key signal input unit that receives manager setting information for safety management according to user key manipulation;
An interface unit that transmits and receives information with a video surveillance device installed for each safety management place set by a plurality of different administrators;
The video surveillance device,
Depending on the facility size and service type of the safety management place, the range of each camera area is confirmed and the number of installation units is selected. ,
When monitoring information is collected from the video monitoring device through the interface unit according to manager setting information of the key signal input unit, a control unit performs safety management by checking and providing abnormal conditions in the corresponding monitoring information from a setting format;
a database for storing registration information and status information including monitoring information and abnormal conditions for each video surveillance device under the control of the control unit; and
a display unit for displaying abnormal state user interface information under the control of the controller; contains,

The setting format is
1. Establish a first information set for each setting object including vehicle, car number, and person based on the images of the actual collected video surveillance device for each safety management place,
1) When the first information set is constructed, external registered public image information is input, image information is collected through web crawling with the registration information processing device, and image information of the video surveillance device is collected, and each Acquiring image information by merging information into one,
2) Labeling is performed by classifying the obtained image information into positive and negative,
3) Correct the size and brightness of the image for which the labeling work was performed, and additionally process the image in general by adjusting the gray-scale and threshold differently according to the manager setting information,
4) Additional images are secured by providing a function that creates new information through calculations (including right-reversal, brightness control, coordinate movement, rotation, and magnification) for the normally processed image,

2. Recognize and learn objects in each image for each additionally secured image to create a second information set,

3. Based on the first and second information sets, the collected image is divided into frame units, and the object labeling area is puzzled in a set sub format (Re-identification (ReID) format of Deep Appearance Descriptor) in each frame unit image. separated into shapes,
Create object-centered values from objects, calculate features for each detailed area, including area size, color, and type of the object, search for similarly identified objects through convolution layers to track the motion of the object,

4. Detecting an abnormal situation by comparing the movement of the object as a result of the tracking with the abnormal behavior information set for each safety management place; Video AI-based safety management integrated device for efficient control of marine city infrastructure. The method of claim 1,
The control unit,
1. If the safety management place is a parking lot, the vehicle location and vehicle number are collected according to the PTZ camera movement value predefined from the setting format, and the vehicle information of the corresponding side is inserted into the registered on-street/off-street public parking lot map After that, output the UI,
In the case of a public parking lot, an alarm is displayed for parked vehicles that have passed the set period,
Smoke and fire occurring in the parking lot are identified and detected through a set detection sensor,
When leaving a parked vehicle, the vehicle number is recognized through the camera installed at the entrance of the parking lot to proceed with exiting the vehicle, and the vehicle is paid in conjunction with the parking control information processing device, and in case of non-payment, the parking information processing device is notified. Pay through the app,

2. In the case where the safety management place is a safety/control area, the screen of the video surveillance device is analyzed and the number of people on the screen is stopped for more than the set time for each different safety/control area and if the person is stopped for more than the set time within the set movement change amount In each case, a warning is given,
In the case of an access control area, whether an object appears in the area is checked, and if the object appears as a result of the above check, a video is notified or an object appears in the boundary area including disaster/disaster situations and late-night hours. Notified when it appears
Analyzing the skeleton of the set part including the person's hands, feet, and head on the screen, checking for personal accidents (including falls and injuries), alarming,

3. When the safety management place is a road area, the vehicle is analyzed and the vehicle speed is less than the set standard speed, and the number of vehicles recognized within the image range is greater than the set standard value, the density is calculated, and congestion occurs when the set standard value is exceeded sense the situation,
By analyzing the image of the road with a crosswalk, determining the crowding of human objects within the set time, and checking the jaywalking of people other than the crosswalk when a human object appears within the cluster range,
Analyzing objects on the road to identify objects other than people and vehicles that have not moved during the set time and determining them as unauthorized occupancy; Video AI-based safety management integrated device for efficient control of marine city infrastructure.

Images