JP7612240B2 - AI-based risk prediction method and system - Google Patents

Description

The present invention relates to an AI-based method and system for predicting dangerous situations, and more particularly to a method and system for predicting the probability of a dangerous situation occurring by determining a critical value according to the surrounding conditions.

As not only crime but also safety accidents increase, CCTVs are being installed in various places to prevent crime and safety accidents in advance or to respond quickly.

Currently, CCTV is commonly used to prevent and respond to crimes by searching CCTV footage (stored on a server or in a specific storage space) of the scene after a dangerous situation has occurred, and identifying the area where the dangerous situation occurred.

Therefore, the current method has the limitation that it is not possible to grasp the occurrence of dangerous situations in real time using CCTV images.

Not only that, but in order to store CCTV footage for a certain period of time, the CCTV footage itself must be transmitted to a server, which occupies a relatively high bandwidth and requires the server to perform a large amount of calculations.

In addition, because CCTV footage is stored as is on the server, personal information (such as faces) is exposed when viewing the CCTV footage, which creates the problem of privacy being violated.

This allows the occurrence of dangerous situations to be identified in real time using real-time CCTV footage, but there is a demand for new methods that are less burdensome not only in terms of protecting personal information but also in terms of data transmission and storage.

In addition, a technical solution is needed that can adaptively notify of dangerous situations depending on the environment in which the CCTV is installed.

The technology behind this invention is disclosed in Korean Patent Publication No. 10-2490062 (published on January 12, 2023).

The present invention aims to provide an AI-based method and system for predicting dangerous situations by extracting information from video, predicting the probability of a dangerous situation occurring using the information, and determining a critical value that can recognize and judge a dangerous situation using information obtained in real time to predict the dangerous situation.

According to an embodiment of the present invention for achieving this technical objective, a method for predicting a dangerous situation using an AI-based dangerous situation prediction system includes the steps of: receiving an image from a camera installed in a monitored area; recognizing one or more objects from the received image; extracting movement information or motion information of the recognized object; predicting a probability of a dangerous situation occurring by applying the movement information or motion information of the object to a trained first learning model; acquiring object density information, meteorological information, and environmental information for the monitored area; applying the object density information, meteorological information, and environmental information to a trained second learning model to determine a threshold value; comparing the probability of a dangerous situation occurring predicted from the first learning model with the threshold value to determine the occurrence of a dangerous situation; and generating an alarm signal if the probability of a dangerous situation occurring is higher than the threshold value.

The method may further include a step of setting the movement information and motion information of the object as input data, and setting a dangerous situation including at least one of an outsider's intrusion, abduction, wandering, falling, fighting, arson, and drowning as output data, and training the first learning model.

The step of acquiring object density information, meteorological information, and environmental information for the monitored area may include acquiring object density information including traffic volume information and detected object density from the image; acquiring meteorological information including sunrise and sunset times, difference between minimum and maximum temperatures, weather, and wind speed; and acquiring environmental information including illuminance information, time, and camera installation position.

The method may further include a step of setting the acquired object density information, meteorological information, and environmental information as input data, setting the danger situation grade as output data, and training the second learning model.

The second learning model may be trained to lower the threshold for generating an alarm signal as the object density decreases, as it gets closer to sunset, as the temperature difference increases, as the weather conditions become worse, as the wind speed increases, as the illuminance decreases, as the camera is installed in the peripheral area, and as it gets closer to night.

According to another embodiment of the present invention, an AI-based dangerous situation prediction system includes an image acquisition unit that receives an image from a camera installed in a monitored area; an object recognition unit that recognizes one or more objects from the received image; a pre-processing unit that extracts movement information or motion information of the recognized object; a learning unit that predicts a probability of a dangerous situation occurring by applying the movement information or motion information of the object to a trained first learning model; an information acquisition unit that acquires object density information, meteorological information, and environmental information about the monitored area; a control unit that applies the object density information, meteorological information, and environmental information to a trained second learning model to determine a threshold value; a determination unit that compares the probability of a dangerous situation occurring predicted from the first learning model with the threshold value to determine the occurrence of a dangerous situation; and an alarm unit that generates an alarm signal if the probability of a dangerous situation occurring is higher than the threshold value.

In this way, according to the present invention, the critical value is determined according to the surrounding conditions, and by recognizing dangerous situations and providing an alarm, false alarms can be reduced and emergency situations can be dealt with quickly.

In addition, by minimizing random errors in the various information received through CCTV images and various sensors, it is possible to predict and prevent current and future dangerous situations for users from various industries and classes in society, thereby ensuring the protection of users in their daily lives.

FIG. 1 is a configuration diagram of an AI-based dangerous situation prediction system according to an embodiment of the present invention.

1 is a flowchart of an AI-based dangerous situation prediction method using a dangerous situation prediction system according to an embodiment of the present invention.

Hereinafter, the present invention will be described in detail with reference to the attached drawings so that those skilled in the art can easily implement the embodiments. However, the present invention may be embodied in various different forms and is not limited to the embodiments described herein. In order to clearly explain the present invention in the drawings, parts that are not relevant to the description are omitted, and similar parts are designated with similar reference numerals throughout the specification.

Throughout the specification, when a part is said to be "connected" to another part, this includes not only when it is "directly connected" to another part, but also when it is "electrically connected" with another element in between. In addition, when a part is said to "include" a certain component, this does not mean that it excludes the other component, but that it can further include the other component, unless otherwise specified to the contrary.

Below, the embodiments of the present invention will be described in detail with reference to the attached drawings so that those skilled in the art can easily implement them.

Below, we will explain the dangerous situation prediction system according to an embodiment of the present invention using Figure 1.

Figure 1 is a configuration diagram of an AI-based dangerous situation prediction system according to an embodiment of the present invention.

As shown in FIG. 1, the risk situation prediction system 100 according to an embodiment of the present invention includes an image acquisition unit 110, an object recognition unit 120, a pre-processing unit 130, a learning unit 140, an information acquisition unit 150, a control unit 160, a judgment unit 170, and an alarm unit 180.

First, the image acquisition unit 110 receives images in real time from a camera installed in the monitored area.

At this time, the image acquisition unit 110 can acquire a two-dimensional RGB image in real time.

Then, the object recognition unit 120 recognizes one or more objects from the input image.

The object recognition unit 120 can recognize and separate the area of a person from the area of an object held by the person from the image input through the image acquisition unit 110.

In other words, the object recognition unit 120 can improve the accuracy of feature information extracted through the recognition of people and individuals, and accurately distinguish between people and individuals.

Then, the pre-processing unit 130 extracts movement information or motion information of the recognized object.

The pre-processing unit 130 extracts object movement information (wandering, entering and exiting, etc.) including the object's movement path from the video, and extracts motion information including behavior recognition information of people or animals (walking, running, falling, etc.).

Then, the learning unit 140 applies the object movement information or motion information to the previously learned first learning model to predict the probability of a dangerous situation occurring.

The learning unit 140 applies the movement information or motion information of the recognized object to the previously learned first learning model to predict the probability of a dangerous situation occurring.

In this case, the first learning model is trained by setting the object's movement information and motion information as input data and setting dangerous situations including at least one of the following as output data: intrusion by an outsider, abduction, wandering, falling, fighting, arson, and drowning.

Then, the information acquisition unit 150 acquires object density information, weather information, and environmental information for the monitored area.

The information acquisition unit 150 can acquire object density information, weather information, and environmental information in real time from the Meteorological Agency server, cameras, multiple sensors, and user terminals.

The control unit 160 then applies the object density information, meteorological information, and environmental information to the second learning model that has already been learned to determine the threshold value.

That is, the control unit 160 determines the appropriate threshold value taking into account the information of the monitoring area acquired in real time.

Then, the judgment unit 170 compares the probability of occurrence of a dangerous situation predicted from the first learning model with the critical value to judge whether or not a dangerous situation has occurred.

The alarm unit 180 generates an alarm signal when the probability of a dangerous situation occurring is higher than a critical value.

The alarm unit 180 can generate an alarm signal to notify the user of a danger through the user terminal.

User terminals include network-connectable terminals such as PCs (Personal Computers), notebook computers, and smart phones.

Hereinafter, a method for predicting a dangerous situation using an AI-based dangerous situation prediction system according to an embodiment of the present invention will be described with reference to FIG. 2.

Figure 2 is a flowchart of an AI-based method for predicting a dangerous situation according to an embodiment of the present invention.

First, as shown in FIG. 2, the image acquisition unit 110 receives an image from a camera installed in the monitored area (step S210).

Then, the object recognition unit 120 recognizes one or more objects from the input image (step S220).

When learning using deep learning, the object recognition unit 120 can segment the person region and the individual region using semantic segmentation and semantic segmentation for region of interest, respectively.

Then, the pre-processing unit 130 extracts movement information or motion information of the object recognized by the object recognition unit 120 (step S230).

For example, when video is input from a surveillance camera installed on a hiking trail, the object recognition unit 120 can recognize the hiking trail background (trees, stones, etc.) and the person. The pre-processing unit 130 can extract the person's movement direction and speed information as movement information, and can extract movement information (sitting, walking, running, falling, etc.).

Then, the learning unit 140 applies the object movement information or motion information extracted from the pre-processing unit 130 to the previously learned first learning model to predict the probability of a dangerous situation occurring (step S240).

Then, the information acquisition unit 150 acquires object density information, weather information, and environmental information (step S250).

The information acquisition unit 150 can acquire environmental information including traffic volume information, object density information, illuminance information, time, and camera installation location (residential area, downtown, park, etc.) from at least one of the cameras, multiple sensors, and user terminals.

The information acquisition unit 150 can acquire meteorological information including sunrise and sunset times, the difference between the minimum and maximum temperatures, the weather (sunny, cloudy, rainy, snowy, etc.) and wind speed from the Japan Meteorological Agency server.

Then, the control unit 160 applies the object density information, meteorological information, and environmental information for the monitored area to the second learning model that has already been learned to determine the threshold value (step S260).

The control unit 160 sets object density information, weather information, and environmental information as input data, sets the danger situation grade as output data, and trains the second learning model.

The second learning model is trained to determine a lower threshold for generating an alarm signal the lower the object density, the closer it is to sunset, the greater the temperature difference, the worse the weather conditions, the faster the wind speed, the lower the illuminance, the closer the camera is to the periphery, and the closer it is to night.

Then, the judgment unit 170 compares the probability of a dangerous situation occurring with the threshold value to judge whether a dangerous situation has occurred (step S270).

The judgment unit 170 can judge that a dangerous situation has occurred if the probability of the dangerous situation occurring is higher than the critical value, and can judge that a dangerous situation has not occurred if the probability of the dangerous situation occurring is lower than the critical value.

If the judgment unit 170 judges that no dangerous situation has occurred, step S210 is repeated to judge whether or not a dangerous situation has occurred.

Then, if the judgment unit 170 judges that a dangerous situation has occurred in the monitored area, the alarm unit 180 generates an alarm signal (step S280).

The alarm unit 180 can generate an alarm signal to notify the user through the user terminal whether or not a dangerous situation has occurred.

In this way, according to the present invention, the critical value is determined according to the surrounding conditions, and by recognizing dangerous situations and providing an alarm, it is possible to reduce false alarms and quickly respond to emergency situations.

In addition, by minimizing random errors in the various information received through CCTV images and various sensors, it is possible to predict and prevent current and future dangerous situations for users from various industries and classes in society, thereby ensuring the protection of users in their daily lives.

The present invention has been described with reference to the embodiments shown in the drawings, but these are merely illustrative, and a person skilled in the art would understand that various modifications and equivalent embodiments are possible. Therefore, the true technical scope of protection of the present invention must be determined by the technical ideas of the claims.

100: Danger situation prediction system 110: Image acquisition unit 120: Object recognition unit 130: Preprocessing unit 140: Learning unit 150: Information acquisition unit 160: Control unit 170: Judgment unit 180: Alarm unit

Claims (10)

A method for predicting a dangerous situation using an AI-based dangerous situation prediction system,
receiving an image from a camera installed in a surveillance area;
recognizing one or more objects from the input image;
extracting movement information or motion information of the recognized object;
applying the movement information or motion information of the object to a first learning model that has been trained to predict a probability of occurrence of a dangerous situation;
acquiring object density information, meteorological information, and environmental information about the monitoring area;
determining a threshold value by applying the object density information, meteorological information, and environmental information to a second learning model that has already been learned;
determining whether a dangerous situation occurs by comparing the probability of occurrence of the dangerous situation predicted from the first learning model with the threshold value;
generating an alarm signal if the probability of the dangerous situation occurring is higher than the critical value;
A method for predicting a dangerous situation, comprising: The method for predicting a dangerous situation according to claim 1, further comprising a step of setting the movement information and motion information of the object as input data, setting a dangerous situation including at least one of an intrusion by an outsider, abduction, wandering, falling, a fight, arson, and drowning as output data, and training the first learning model. The step of acquiring object density information, meteorological information, and environmental information about the monitoring area includes:
acquiring object density information including traffic information and detected object density from the image;
obtaining meteorological information including sunrise and sunset times, minimum and maximum temperature differences, weather and wind speed;
acquiring environmental information including illuminance information, time, and a camera installation position;
The method for predicting a dangerous situation according to claim 1 , The method for predicting a dangerous situation according to claim 1, further comprising the step of setting the acquired object density information, meteorological information, and environmental information as input data, setting a danger situation grade as output data, and training the second learning model. The step of training the second learning model includes:
5. The method of claim 4, further comprising learning to lower the threshold value for generating an alarm signal as the object density is lower, the closer to sunset, the greater the temperature difference, the worse the weather conditions are, the faster the wind speed is, the lower the illuminance is, the closer the camera is installed to the periphery, and the closer it is to night. In an AI-based danger situation prediction system,
an image acquisition unit that receives images from a camera installed in a surveillance area;
an object recognition unit for recognizing one or more objects from the input image;
a pre-processing unit for extracting movement information or motion information of the recognized object;
a learning unit that predicts a probability of a dangerous situation occurring by applying the movement information or motion information of the object to a first learning model that has already been learned;
an information acquisition unit for acquiring object density information, meteorological information, and environmental information about the monitoring area;
a control unit for determining a threshold value by applying the object density information, the meteorological information, and the environmental information to a second learning model that has already been learned;
a determination unit that determines whether a dangerous situation occurs by comparing the probability of occurrence of the dangerous situation predicted from the first learning model with the threshold value;
an alarm unit for generating an alarm signal when the risk occurrence probability is higher than the critical value;
A dangerous situation prediction system including: The learning unit is
The danger situation prediction system of claim 6, wherein the object's movement information and motion information are set as input data, and danger situations including at least one of an outsider's intrusion, abduction, wandering, falling, fighting, arson, and drowning are set as output data to train the first learning model. The information acquisition unit is
Acquiring object density information including traffic volume information and detected object density from the image;
Obtaining meteorological information including sunrise and sunset times, minimum and maximum temperature differences, weather and wind speed;
The dangerous situation prediction system according to claim 6, further comprising: acquiring environmental information including illuminance information, time, and a camera installation position. The control unit is
The system for predicting a dangerous situation according to claim 6, wherein the acquired object density information, meteorological information and environmental information are set as input data, and a danger situation grade is set as output data to train the second learning model. The second learning model is
10. The system of claim 9, further learning to lower the threshold value for generating an alarm signal as the object density is lower, as it is closer to sunset, as the temperature difference is larger, as the weather condition is worse, as the wind speed is higher, as the illuminance is lower, as the camera is installed in the peripheral area, and as it is closer to night.